Merge remote-tracking branch 'origin/feat/eip-stopstart' into e2e/fixes-1102

Merge remote-tracking branch 'origin/feat/rds-agent-cmd' into e2e/fixes-1102
Merge remote-tracking branch 'origin/feat/iso-mem-admission' into e2e/fixes-1102
2026-11-02 17:23:32 +00:00 · 2026-11-02 22:46:03 +10:00 · 2026-11-02 22:46:03 +10:00 · 2026-11-02 22:46:03 +10:00 · 2026-11-02 22:44:50 +10:00 · 2026-11-02 22:37:36 +10:00
1030 changed files with 34660 additions and 13585 deletions
@@ -0,0 +1,34 @@
+#!/usr/bin/env bash
+# Usage: wait-for-memory.sh <MiB>
+#
+# Waits until the runner's MemAvailable reaches <MiB>, then returns. Runners on
+# one host share its RAM, and a neighbour's build holds several GiB for a few
+# minutes, so one sample taken during it is not a verdict on the host.
+# Gives up after MEM_WAIT_TIMEOUT_S and names the biggest processes.
+#
+# Env: MEM_WAIT_TIMEOUT_S (900), MEM_WAIT_POLL_S (30), MEMINFO (/proc/meminfo).
+set -uo pipefail
+
+want_mb="${1:?usage: $0 <MiB>}"
+timeout_s="${MEM_WAIT_TIMEOUT_S:-900}"
+poll_s="${MEM_WAIT_POLL_S:-30}"
+meminfo="${MEMINFO:-/proc/meminfo}"
+
+avail_mb() { awk '/^MemAvailable:/ {print int($2 / 1024)}' "$meminfo"; }
+
+start=$SECONDS
+while :; do
+  have=$(avail_mb)
+  if [ -n "$have" ] && [ "$have" -ge "$want_mb" ]; then
+    echo "memory: ${have} MiB available, need ${want_mb} MiB"
+    exit 0
+  fi
+  waited=$(( SECONDS - start ))
+  if [ "$waited" -ge "$timeout_s" ]; then
+    echo "::error::need ${want_mb} MiB available, have ${have:-unknown} MiB after waiting ${waited}s"
+    ps -eo rss,pid,args --sort=-rss | head -11 | cut -c1-200
+    exit 1
+  fi
+  echo "memory: ${have:-unknown} MiB available, need ${want_mb} MiB; waiting (${waited}s of ${timeout_s}s)"
+  sleep "$poll_s"
+done
@@ -0,0 +1,47 @@
+#!/bin/sh
+# wait-for-memory_test.sh — drives wait-for-memory.sh against a fake meminfo
+# that a background writer changes mid-wait, so each verdict is checked without
+# a loaded host.
+set -eu
+
+SCRIPT_DIR=$(CDPATH='' cd -- "$(dirname -- "$0")" && pwd)
+WAIT="$SCRIPT_DIR/wait-for-memory.sh"
+TMP=$(mktemp -d)
+trap 'rm -rf "$TMP"' EXIT
+
+fails=0
+pass() { echo "  ok   - $1"; }
+fail() { echo "  FAIL - $1"; fails=$((fails + 1)); }
+
+meminfo() { printf 'MemTotal:       65536000 kB\nMemAvailable:   %s kB\n' "$(( $1 * 1024 ))" > "$TMP/meminfo"; }
+MEMINFO="$TMP/meminfo"
+MEM_WAIT_POLL_S=1
+export MEMINFO MEM_WAIT_POLL_S
+
+meminfo 30000
+if MEM_WAIT_TIMEOUT_S=5 bash "$WAIT" 26624 > "$TMP/out" 2>&1; then
+  pass "enough memory returns at once"
+else
+  fail "enough memory returns at once: $(cat "$TMP/out")"
+fi
+
+meminfo 21577
+( sleep 2; meminfo 30000 ) &
+if MEM_WAIT_TIMEOUT_S=20 bash "$WAIT" 26624 > "$TMP/out" 2>&1 && grep -q waiting "$TMP/out"; then
+  pass "memory freed mid-wait is admitted"
+else
+  fail "memory freed mid-wait is admitted: $(cat "$TMP/out")"
+fi
+wait
+
+meminfo 21577
+if MEM_WAIT_TIMEOUT_S=2 bash "$WAIT" 26624 > "$TMP/out" 2>&1; then
+  fail "memory that never frees fails: exited 0"
+elif grep -q '::error::need 26624 MiB available, have 21577 MiB' "$TMP/out" && grep -q 'RSS' "$TMP/out"; then
+  pass "memory that never frees fails and names the biggest processes"
+else
+  fail "memory that never frees fails and names the biggest processes: $(cat "$TMP/out")"
+fi
+
+[ "$fails" -eq 0 ] || { echo "wait-for-memory: $fails failed"; exit 1; }
+echo "wait-for-memory: all passed"
@@ -104,7 +104,7 @@ jobs:

      - name: Build microVM artifacts (lb-agent + kernel + initramfs + qboot + SQS store and worker images)
        if: steps.version.outputs.exists != 'true'
-        run: cd spinifex && make build-microvm-image build-sqs-store-microvm-image build-sqs-worker-microvm-image
+        run: cd spinifex && make build-microvm-image build-sqs-worker-microvm-image

      - name: Build amd64 Tarball
        if: steps.version.outputs.exists != 'true'
@@ -1057,9 +1057,6 @@ jobs:
            { echo "::error::br-wan not permitted in /etc/qemu/bridge.conf"; exit 1; }
          command -v sshpass >/dev/null ||
            { echo "::error::sshpass not installed"; exit 1; }
-          free_mb=$(free -m | awk '/^Mem:/ {print $7}')
-          [ "$free_mb" -ge 8192 ] ||
-            { echo "::error::need 8 GiB available, have ${free_mb} MiB"; exit 1; }
          # A 40G image on tmpfs competes with the 8 GiB the VM itself needs.
          fstype=$(df --output=fstype "$TMPDIR" | tail -1 | tr -d ' ')
          [ "$fstype" != tmpfs ] ||
@@ -1090,6 +1087,10 @@ jobs:
              ssh-keygen -t ed25519 -N "" -f "$HOME/.ssh/$k" -C "$k-ci"
          done

+      # The node's 8 GiB, claimed when the VM boots rather than at preflight.
+      - name: Wait for memory
+        run: mulga/spinifex/.github/scripts/wait-for-memory.sh 8192
+
      # Boots the ISO, installs unattended, reads the lease off the serial
      # console, runs the in-guest smoke test, and installs the key for the
      # next step.
@@ -1203,6 +1204,16 @@ jobs:
          sleep 2
          rm -f "$TMPDIR"/tmp.*.qcow2 "$TMPDIR"/tmp.*.log "$TMPDIR"/tmp.*-node.log

+      # The ISO build runs as root inside the workspace, and the next job on
+      # this runner, from any workflow, cannot remove what it leaves there.
+      - name: Release the workspace
+        if: always()
+        run: |
+          sudo umount -R -l "$GITHUB_WORKSPACE"/mulga/scripts/iso-builder/.cache/rootfs 2>/dev/null || true
+          sudo rm -rf --one-file-system "$GITHUB_WORKSPACE"/mulga/scripts/iso-builder/.cache \
+              "$GITHUB_WORKSPACE"/mulga/scripts/iso-builder/dist
+          sudo chown -R "$(id -u):$(id -g)" "$GITHUB_WORKSPACE"
+
  iso-multi:
    name: cell-22 (iso/dhcp/three-node)
    # After cell-21 rather than beside it: same concurrency group, and running
@@ -1291,15 +1302,10 @@ jobs:
          ref: ${{ steps.siblings.outputs.bluebottle }}
          path: mulga/bluebottle

-      # 3 x 8 GiB plus 9 sparse 40G disks, before the instances the cluster
-      # launches on top.
      - name: Preflight
        run: |
          set -euo pipefail
          mkdir -p "$TMPDIR"
-          free_mb=$(free -m | awk '/^Mem:/ {print $7}')
-          [ "$free_mb" -ge 26624 ] ||
-            { echo "::error::need 26 GiB available, have ${free_mb} MiB"; exit 1; }
          fstype=$(df --output=fstype "$TMPDIR" | tail -1 | tr -d ' ')
          [ "$fstype" != tmpfs ] ||
            { echo "::error::$TMPDIR is tmpfs — nine node disks would be in RAM"; exit 1; }
@@ -1322,6 +1328,11 @@ jobs:
      # formation port. Both land with mulga-yxtda and the bridged WAN mode;
      # until then this cell proves the cluster forms and survives, not that a
      # customer can use it.
+      # 3 x 8 GiB plus headroom for the instances the cluster launches on top,
+      # claimed when the nodes boot rather than at preflight.
+      - name: Wait for memory
+        run: mulga/spinifex/.github/scripts/wait-for-memory.sh 26624
+
      - name: Install, form and verify the cluster
        working-directory: mulga/scripts/iso-builder
        run: |
@@ -1354,6 +1365,16 @@ jobs:
          rm -rf /tmp/cell22-artifacts-$GITHUB_RUN_ID /tmp/cell22-$GITHUB_RUN_ID.log \
                 "$WORK_DIR"

+      # The ISO build runs as root inside the workspace, and the next job on
+      # this runner, from any workflow, cannot remove what it leaves there.
+      - name: Release the workspace
+        if: always()
+        run: |
+          sudo umount -R -l "$GITHUB_WORKSPACE"/mulga/scripts/iso-builder/.cache/rootfs 2>/dev/null || true
+          sudo rm -rf --one-file-system "$GITHUB_WORKSPACE"/mulga/scripts/iso-builder/.cache \
+              "$GITHUB_WORKSPACE"/mulga/scripts/iso-builder/dist
+          sudo chown -R "$(id -u):$(id -g)" "$GITHUB_WORKSPACE"
+
  # Predastore throughput ("cell 23"). Predastore has no runners of its own and
  # spinifex depends on it, so the benchmark lives here and checks predastore
  # out as a sibling.
@@ -58,7 +58,7 @@ jobs:
          cache: false

      - name: Build microVM artifacts (lb-agent + kernel + initramfs + qboot + SQS store and worker images)
-        run: cd spinifex && make build-microvm-image build-sqs-store-microvm-image build-sqs-worker-microvm-image
+        run: cd spinifex && make build-microvm-image build-sqs-worker-microvm-image

      - name: Build amd64 Tarball
        env:
@@ -21,12 +21,12 @@ build/microvm/vmlinuz
 build/microvm/initramfs.cpio.gz
 build/microvm/qboot.rom
 build/microvm/qboot.LICENSE
-build/microvm/sqs-store/vmlinuz
-build/microvm/sqs-store/initramfs.cpio.gz
-build/microvm/sqs-store/.stage.*/
 build/microvm/sqs-worker/vmlinuz
 build/microvm/sqs-worker/initramfs.cpio.gz
 build/microvm/sqs-worker/.stage.*/
+build/microvm/lambda-runtime/vmlinuz
+build/microvm/lambda-runtime/initramfs.cpio.gz
+build/microvm/lambda-runtime/.stage.*/
 build/qboot/
 !spinifex/runtime/roles/spinifexui/frontend/dist/
 tmp/
@@ -153,37 +153,46 @@ $(MICROVM_ARTIFACTS) &: $(MICROVM_INPUTS)
 bin/lb-agent:
 	$(MAKE) build-lb-agent

-# The SQS store microVM image: a separate artifact from the LB image above, not
-# installed or released by any target yet.
-SQS_STORE_MICROVM_OUT_DIR := $(MICROVM_OUT_DIR)/sqs-store
-SQS_STORE_MICROVM_ARTIFACTS := $(SQS_STORE_MICROVM_OUT_DIR)/vmlinuz $(SQS_STORE_MICROVM_OUT_DIR)/initramfs.cpio.gz
-SQS_STORE_MICROVM_INPUTS := scripts/build-sqs-store-microvm-image.sh scripts/lib/sqs-store-microvm-publish.sh $(SQS_STORE_MICROVM_OUT_DIR)/init.sh $(SQS_STORE_MICROVM_OUT_DIR)/inittab $(addprefix $(SQS_STORE_MICROVM_OUT_DIR)/,sqs-store-lib.sh sqs-store-prepare sqs-store-run sqs-store-stop) bin/nats-server
-
-$(SQS_STORE_MICROVM_ARTIFACTS) &: $(SQS_STORE_MICROVM_INPUTS)
-	./scripts/build-sqs-store-microvm-image.sh
-
-# The dedicated SQS store server, built from the go.mod-pinned module so the
-# image ships exactly the reviewed version. Static for the musl guest.
+# The dedicated SQS server, built from the go.mod-pinned module so the image
+# ships exactly the reviewed version. Static for the musl guest.
 bin/nats-server: go.mod go.sum
 	@mkdir -p bin
 	GOWORK=off CGO_ENABLED=0 GOOS=linux GOARCH=amd64 go build -trimpath -ldflags '-s -w' -o $@ github.com/nats-io/nats-server/v2

-build-sqs-store-microvm-image: $(SQS_STORE_MICROVM_ARTIFACTS) ## Build the SQS store microVM kernel and initramfs
+# go_local_srcs lists the main module package dirs and files guest package $(1)
+# links, scanned with the guest build settings; a dir's mtime catches a deleted
+# or renamed file. A failed scan names a missing prerequisite, so make stops.
+go_local_srcs = $(shell GOWORK=off CGO_ENABLED=0 GOOS=linux GOARCH=amd64 go list -deps -f '{{if and .Module .Module.Main}}{{.Dir}} {{range .GoFiles}}{{$$.Dir}}/{{.}} {{end}}{{range .EmbedFiles}}{{$$.Dir}}/{{.}} {{end}}{{end}}' $(1) || echo go-list-failed-for-$(1))
+.SECONDEXPANSION:

-# The shared SQS worker microVM image: the store image's /init and kernel with
-# the worker binary instead of nats-server. Not installed or released yet.
+# The SQS microVM image: one guest running its own SQS nats-server and the SQS
+# worker as separate processes. A separate artifact from the LB image above.
 SQS_WORKER_MICROVM_OUT_DIR := $(MICROVM_OUT_DIR)/sqs-worker
 SQS_WORKER_MICROVM_ARTIFACTS := $(SQS_WORKER_MICROVM_OUT_DIR)/vmlinuz $(SQS_WORKER_MICROVM_OUT_DIR)/initramfs.cpio.gz
-SQS_WORKER_MICROVM_INPUTS := scripts/build-sqs-store-microvm-image.sh scripts/lib/sqs-store-microvm-publish.sh $(SQS_STORE_MICROVM_OUT_DIR)/init.sh $(SQS_WORKER_MICROVM_OUT_DIR)/inittab $(addprefix $(SQS_WORKER_MICROVM_OUT_DIR)/,sqs-worker-lib.sh sqs-worker-prepare sqs-worker-ready sqs-worker-stop) bin/sqs-worker
+SQS_WORKER_MICROVM_INPUTS := scripts/build-sqs-store-microvm-image.sh scripts/lib/microvm-image.sh scripts/lib/sqs-store-microvm-publish.sh $(SQS_WORKER_MICROVM_OUT_DIR)/init.sh $(SQS_WORKER_MICROVM_OUT_DIR)/inittab $(addprefix $(SQS_WORKER_MICROVM_OUT_DIR)/,sqs-worker-lib.sh sqs-worker-prepare sqs-worker-store sqs-worker-ready sqs-worker-stop) bin/nats-server bin/sqs-worker

 $(SQS_WORKER_MICROVM_ARTIFACTS) &: $(SQS_WORKER_MICROVM_INPUTS)
-	SQS_MICROVM_PROFILE=worker ./scripts/build-sqs-store-microvm-image.sh
+	./scripts/build-sqs-store-microvm-image.sh

-bin/sqs-worker: go.mod go.sum $(shell find cmd/sqs-worker spinifex/agents/sqs -name '*.go' ! -name '*_test.go')
+bin/sqs-worker: go.mod go.sum $$(call go_local_srcs,./cmd/sqs-worker)
 	@mkdir -p bin
 	GOWORK=off CGO_ENABLED=0 GOOS=linux GOARCH=amd64 go build -trimpath -ldflags '-s -w' -o $@ ./cmd/sqs-worker

-build-sqs-worker-microvm-image: $(SQS_WORKER_MICROVM_ARTIFACTS) ## Build the shared SQS worker microVM kernel and initramfs
+build-sqs-worker-microvm-image: $(SQS_WORKER_MICROVM_ARTIFACTS) ## Build the SQS microVM kernel and initramfs
+
+# The Lambda runtime microVM image. Built on demand; not shipped or installed.
+LAMBDA_RUNTIME_MICROVM_OUT_DIR := $(MICROVM_OUT_DIR)/lambda-runtime
+LAMBDA_RUNTIME_MICROVM_ARTIFACTS := $(LAMBDA_RUNTIME_MICROVM_OUT_DIR)/vmlinuz $(LAMBDA_RUNTIME_MICROVM_OUT_DIR)/initramfs.cpio.gz
+LAMBDA_RUNTIME_MICROVM_INPUTS := scripts/build-lambda-runtime-microvm-image.sh scripts/lib/microvm-image.sh $(addprefix $(LAMBDA_RUNTIME_MICROVM_OUT_DIR)/,init.sh inittab run-agent) bin/lambda-agent
+
+$(LAMBDA_RUNTIME_MICROVM_ARTIFACTS) &: $(LAMBDA_RUNTIME_MICROVM_INPUTS)
+	./scripts/build-lambda-runtime-microvm-image.sh
+
+bin/lambda-agent: go.mod go.sum $$(call go_local_srcs,./cmd/lambda-agent)
+	@mkdir -p bin
+	GOWORK=off CGO_ENABLED=0 GOOS=linux GOARCH=amd64 go build -trimpath -ldflags '-s -w' -o $@ ./cmd/lambda-agent
+
+build-lambda-runtime-microvm-image: $(LAMBDA_RUNTIME_MICROVM_ARTIFACTS) ## Build the Lambda runtime microVM kernel and initramfs (not shipped)

 # qboot is the microvm firmware. The source tarball ships with each release
 # because qboot is GPL-2.0.
@@ -196,13 +205,7 @@ $(QBOOT_ARTIFACTS) &: scripts/build-qboot.sh

 build-microvm-image: $(MICROVM_ARTIFACTS) $(QBOOT_ARTIFACTS) ## Build microVM kernel, initramfs and qboot firmware (incremental — skips when up to date)

-install-sqs-store-microvm: $(SQS_STORE_MICROVM_ARTIFACTS) ## Install the SQS store microVM image to /usr/share/spinifex/microvm/sqs-store/
-	sudo install -d /usr/share/spinifex/microvm/sqs-store
-	sudo rm -f /usr/share/spinifex/microvm/sqs-store/initramfs.cpio.gz
-	sudo install -m 0644 $(SQS_STORE_MICROVM_OUT_DIR)/vmlinuz /usr/share/spinifex/microvm/sqs-store/vmlinuz
-	sudo install -m 0644 $(SQS_STORE_MICROVM_OUT_DIR)/initramfs.cpio.gz /usr/share/spinifex/microvm/sqs-store/initramfs.cpio.gz
-
-install-sqs-worker-microvm: $(SQS_WORKER_MICROVM_ARTIFACTS) ## Install the SQS worker microVM image to /usr/share/spinifex/microvm/sqs-worker/
+install-sqs-worker-microvm: $(SQS_WORKER_MICROVM_ARTIFACTS) ## Install the SQS microVM image to /usr/share/spinifex/microvm/sqs-worker/
 	sudo install -d /usr/share/spinifex/microvm/sqs-worker
 	sudo rm -f /usr/share/spinifex/microvm/sqs-worker/initramfs.cpio.gz
 	sudo install -m 0644 $(SQS_WORKER_MICROVM_OUT_DIR)/vmlinuz /usr/share/spinifex/microvm/sqs-worker/vmlinuz
@@ -252,7 +255,8 @@ test-build-scripts:
 		echo "-- $$t"; \
 		sh "$$t" || exit 1; \
 	done
-	shellcheck -S warning .github/scripts/hypervisor-disk-preflight.sh .github/scripts/hypervisor-disk-preflight_test.sh
+	shellcheck -S warning .github/scripts/hypervisor-disk-preflight.sh .github/scripts/hypervisor-disk-preflight_test.sh \
+		.github/scripts/wait-for-memory.sh .github/scripts/wait-for-memory_test.sh
 	@echo -e "\n....Running shellcheck over build/scripts/**/*.sh + the node scripts...."
 	shellcheck -S warning $$(find build/scripts -name '*.sh' | sort) $(NODE_SCRIPTS)
 	shellcheck -S warning -x scripts/build-oracle-appliance.sh $$(find scripts/appliance -name '*.sh' | sort)
@@ -322,7 +326,7 @@ test-cover:
 # Run unit tests with race detector
 test-race:
 	@echo -e "\n....Running tests with race detector for $(GO_PROJECT_NAME)...."
-	$(_Q)LOG_IGNORE=1 go test $(GOTESTFLAGS) -race -timeout 300s ./spinifex/... ./cmd/... ./internal/... ./contracts/... ./architecture/... $(_RACEQ)
+	$(_Q)LOG_IGNORE=1 SPINIFEX_SQS_R3=1 go test $(GOTESTFLAGS) -race -timeout 300s ./spinifex/... ./cmd/... ./internal/... ./contracts/... ./architecture/... $(_RACEQ)

 # Unit tests for in-repo GitHub Actions (e.g. .github/actions/e2e-analyze).
 # Kept out of `test-cover` so coverage % isn't diluted by CI-only tooling.
@@ -514,7 +518,6 @@ distro-amd64:
 		echo "[distro-amd64] WARNING: microvm artifacts missing — tarball will not include them"; \
 		echo "[distro-amd64]          run 'make build-microvm-image' before 'make distro-amd64'"; \
 	fi
-	./scripts/lib/stage-sqs-store-microvm.sh $(SQS_STORE_MICROVM_OUT_DIR) dist/amd64
 	./scripts/lib/stage-sqs-store-microvm.sh $(SQS_WORKER_MICROVM_OUT_DIR) dist/amd64 sqs-worker
 	tar -czf dist/spinifex-$(VERSION)-linux-amd64.tar.gz -C dist/amd64 .
 	sha256sum dist/spinifex-$(VERSION)-linux-amd64.tar.gz > dist/spinifex-$(VERSION)-linux-amd64.tar.gz.sha256
@@ -534,7 +537,7 @@ distro-arm64:
 distro-clean:
 	rm -rf dist/

-.PHONY: test-package-check build build-ui test-ui-e2e ui-e2e-container ui-baselines build-installer build-lb-agent build-ecs-agent build-system-image build-eks-node-image import-eks-node-image publish-eks-node-image build-ecs-node-image import-ecs-node-image build-rds-postgres-image import-rds-postgres-image build-rds-mariadb-image import-rds-mariadb-image build-microvm-image build-sqs-store-microvm-image build-sqs-worker-microvm-image install-microvm install-sqs-store-microvm install-sqs-worker-microvm go_build preflight test test-cover test-race diff-coverage bench fuzz test-actions test-images test-build-scripts test-harness test-integration generate-aws-model-coverage aws-model-coverage test-segscan-oracle \
+.PHONY: test-package-check build build-ui test-ui-e2e ui-e2e-container ui-baselines build-installer build-lb-agent build-ecs-agent build-system-image build-eks-node-image import-eks-node-image publish-eks-node-image build-ecs-node-image import-ecs-node-image build-rds-postgres-image import-rds-postgres-image build-rds-mariadb-image import-rds-mariadb-image build-microvm-image build-sqs-worker-microvm-image build-lambda-runtime-microvm-image install-microvm install-sqs-worker-microvm go_build preflight test test-cover test-race diff-coverage bench fuzz test-actions test-images test-build-scripts test-harness test-integration generate-aws-model-coverage aws-model-coverage test-segscan-oracle \
 	deploy reinstall clean \
 	install-system install-go install-aws quickinstall \
 	lint fix govulncheck nilaway \
@@ -38,6 +38,7 @@ var allowlist = []allowance{
 	{From: "cmd/aws-model-coverage", To: "domains/ecs/awsapi", Debt: "cmd/aws-model-coverage", Reason: "reads declared service inventories; named exception, must not construct a registration"},
 	{From: "cmd/aws-model-coverage", To: "domains/eks/awsapi", Debt: "cmd/aws-model-coverage", Reason: "reads declared service inventories; named exception, must not construct a registration"},
 	{From: "cmd/aws-model-coverage", To: "domains/iam/awsapi", Debt: "cmd/aws-model-coverage", Reason: "reads declared service inventories; named exception, must not construct a registration"},
+	{From: "cmd/aws-model-coverage", To: "domains/sqs/awsapi", Debt: "cmd/aws-model-coverage", Reason: "reads declared service inventories; named exception, must not construct a registration"},
 	{From: "cmd/aws-model-coverage", To: "domains/sts/awsapi", Debt: "cmd/aws-model-coverage", Reason: "reads declared service inventories; named exception, must not construct a registration"},

 	{From: "runtime/daemon", To: "domains/ec2/awsapi/instance", Debt: "Instance adapter consumers outside EC2", Reason: "runtime/daemon/eks_cp_control.go and rds_deps.go call the instance adapter; needs an EC2 capability"},
@@ -1,6 +1,6 @@
 ARG VERSION=dev

-FROM golang:1.27.3-bookworm@sha256:972b4110ba0f47539a975c31e52c64765e1ff19f2d74469ae4a595c6676e80a5 AS builder
+FROM golang:1.27.3-bookworm@sha256:d9b6a1beb9474adc2a1ce842573639b44b2a0dd49b6669f51d3821baf9ec37c9 AS builder

 ARG VERSION

@@ -0,0 +1,68 @@
+#!/bin/sh
+mount -t proc  proc  /proc
+mount -t sysfs sys   /sys
+mount -t devtmpfs dev /dev
+
+# modprobe needs a usable modules.dep; insmod by explicit path does not, so
+# fall back to it. Handles .ko and .ko.gz.
+load_mod() {
+    modprobe "$1" 2>/dev/null && return 0
+    for ko in $(find /lib/modules -name "$1.ko" -o -name "$1.ko.gz" 2>/dev/null); do
+        insmod "$ko" 2>/dev/null && return 0
+        case "$ko" in
+            *.gz) gunzip -c "$ko" >/tmp/mod.ko 2>/dev/null \
+                  && insmod /tmp/mod.ko 2>/dev/null && return 0 ;;
+        esac
+    done
+    return 1
+}
+# lambda-agent reads its identity from fw_cfg and unloads qemu_fw_cfg before
+# anything else runs. virtio_net needs net_failover and failover, and the
+# insmod fallback resolves no dependencies, so load the chain bottom-up.
+load_mod qemu_fw_cfg || echo "[init] qemu_fw_cfg did not load" >/dev/console
+load_mod failover || true
+load_mod net_failover || true
+load_mod virtio_net || echo "[init] virtio_net did not load" >/dev/console
+
+ip link set lo up
+
+# The guest's addresses come from the same netcfg entry as other system VMs.
+NETCFG=/sys/firmware/qemu_fw_cfg/by_name/opt/spinifex/netcfg/raw
+i=0
+while [ "$i" -lt 25 ] && [ ! -f "$NETCFG" ]; do
+    sleep 0.04
+    i=$((i + 1))
+done
+if [ -f "$NETCFG" ]; then
+    # shellcheck source=/dev/null
+    . "$NETCFG"
+    for n in 0 1 2 3 4 5; do
+        eval mac=\$NIC${n}_MAC
+        [ -z "$mac" ] && continue
+        iface=""
+        for _ in 1 2 3 4 5; do
+            iface=$(ip -o link | awk -v m="$mac" 'tolower($0) ~ tolower(m) {print $2}' | tr -d :)
+            [ -n "$iface" ] && break
+            sleep 0.02
+        done
+        if [ -z "$iface" ]; then
+            echo "[init] no interface for MAC $mac (NIC$n)" >/dev/console
+            continue
+        fi
+        cidr="" gw="" default="" rdst="" rvia=""
+        eval cidr=\$NIC${n}_CIDR
+        eval gw=\$NIC${n}_GW
+        eval default=\$NIC${n}_DEFAULT
+        eval rdst=\$NIC${n}_ROUTE_DST
+        eval rvia=\$NIC${n}_ROUTE_VIA
+        ip link set "$iface" up
+        [ -n "$cidr" ] && ip addr add "$cidr" dev "$iface"
+        [ "$default" = "1" ] && [ -n "$gw" ] && ip route add default via "$gw" dev "$iface"
+        [ -n "$rdst" ] && [ -n "$rvia" ] && ip route add "$rdst" via "$rvia" dev "$iface"
+    done
+else
+    # lambda-agent then fails to reach the host and the guest powers off.
+    echo "[init] fw_cfg netcfg not available" >/dev/console
+fi
+
+exec /sbin/init
@@ -2,9 +2,5 @@
 # (poweroff) handler installed before sysinit runs, so a press can never reach
 # the pre-exec shell PID 1. A press before this line is lost; the host re-sends.
 ::sysinit:/sbin/modprobe tiny_power_button power_signal=12
-::sysinit:/usr/local/lib/sqs-store/sqs-store-prepare
-::respawn:/usr/local/lib/sqs-store/sqs-store-run
+::once:/usr/local/lib/lambda-runtime/run-agent
 ::ctrlaltdel:/sbin/reboot
-# One action: the store stops and its disk unmounts first, and the generic
-# umount -a -r runs only if the store is gone.
-::shutdown:/usr/local/lib/sqs-store/sqs-store-stop
@@ -0,0 +1,6 @@
+#!/bin/sh
+# Runs the Lambda guest agent, then powers the guest off whatever its status.
+/usr/local/bin/lambda-agent >/dev/console 2>&1
+status=$?
+echo "[run-agent] lambda-agent exited with status $status" >/dev/console
+poweroff -f
@@ -1,224 +0,0 @@
-# sqs-store-lib.sh — shared by the SQS store guest's prepare, run and stop
-# steps. Sourced, never executed. Every path and command is overridable so the
-# host test can drive the same code against a fake fw_cfg, sysfs and tools.
-#
-# Guest contract, read from fw_cfg under opt/spinifex/sqs-store/:
-#   data-serial  virtio-blk serial of the already-formatted ext4 data disk
-#   listen-addr  IPv4 address on a configured guest NIC; the only listener
-#   client-nkey  NATS user NKey public key allowed to connect; never a seed
-# Anything absent, malformed or ambiguous refuses the store. Nothing formats.
-
-SQS_FWCFG=${SQS_FWCFG:-/sys/firmware/qemu_fw_cfg/by_name/opt/spinifex/sqs-store}
-SQS_SYSBLOCK=${SQS_SYSBLOCK:-/sys/block}
-SQS_DEVDIR=${SQS_DEVDIR:-/dev}
-# The host test cannot create block nodes, so it relaxes this to -e.
-SQS_NODE_TEST=${SQS_NODE_TEST:--b}
-SQS_RUN=${SQS_RUN:-/run/sqs-store}
-SQS_DATA=${SQS_DATA:-/data}
-SQS_CONSOLE=${SQS_CONSOLE:-/dev/console}
-SQS_MOUNTS=${SQS_MOUNTS:-/proc/mounts}
-SQS_PORT=${SQS_PORT:-4222}
-SQS_STOP_BUDGET_TENTHS=${SQS_STOP_BUDGET_TENTHS:-120}
-NATS_BIN=${NATS_BIN:-/usr/local/bin/nats-server}
-E2FSCK=${E2FSCK:-e2fsck}
-MOUNT=${MOUNT:-mount}
-UMOUNT=${UMOUNT:-umount}
-IP=${IP:-ip}
-
-sqs_log() { echo "[sqs-store] $*" >> "$SQS_CONSOLE"; }
-sqs_fatal() { sqs_log "FATAL: $*"; return 1; }
-
-# sqs_fwcfg <name>: print one fw_cfg value, refusing a missing, empty, oversized
-# or multi-token file. The value is not echoed on failure.
-sqs_fwcfg() {
-    f="$SQS_FWCFG/$1/raw"
-    [ -f "$f" ] || { sqs_fatal "fw_cfg $1 not supplied"; return 1; }
-    [ "$(wc -c < "$f")" -le 128 ] || { sqs_fatal "fw_cfg $1 is oversized"; return 1; }
-    v=$(cat "$f")
-    case "$v" in
-        "" | *[[:space:]]*) sqs_fatal "fw_cfg $1 must be exactly one value"; return 1 ;;
-    esac
-    printf '%s' "$v"
-}
-
-# A volume serial is the volume ID with dashes stripped: vol + 17 hex.
-sqs_valid_serial() {
-    [ "${#1}" -eq 20 ] || return 1
-    case "$1" in vol*) ;; *) return 1 ;; esac
-    case "${1#vol}" in *[!0-9a-f]*) return 1 ;; esac
-}
-
-# A NATS user public NKey: "U" then 55 base32 characters. The checksum is
-# left to the pinned server's own config test.
-sqs_valid_user_nkey() {
-    [ "${#1}" -eq 56 ] || return 1
-    case "$1" in U*) ;; *) return 1 ;; esac
-    case "$1" in *[!A-Z2-7]*) return 1 ;; esac
-}
-
-sqs_valid_ipv4() {
-    case "$1" in *[!0-9.]* | .* | *. | *..*) return 1 ;; esac
-    old_ifs=$IFS; IFS=.
-    # shellcheck disable=SC2086 # split on dots on purpose
-    set -- $1
-    IFS=$old_ifs
-    [ "$#" -eq 4 ] || return 1
-    for o in "$@"; do
-        [ "${#o}" -le 3 ] && [ "$o" -le 255 ] || return 1
-        case "$o" in 0?*) return 1 ;; esac
-    done
-}
-
-# sqs_find_disk <serial>: print the /dev node of the one block device whose
-# sysfs serial matches exactly. Zero or several matches refuse.
-sqs_find_disk() {
-    found="" n=0
-    for b in "$SQS_SYSBLOCK"/*; do
-        [ -r "$b/serial" ] || continue
-        [ "$(cat "$b/serial")" = "$1" ] || continue
-        found=${b##*/}
-        n=$((n + 1))
-    done
-    [ "$n" -eq 1 ] || { sqs_fatal "expected one disk with serial $1, found $n"; return 1; }
-    test "$SQS_NODE_TEST" "$SQS_DEVDIR/$found" || { sqs_fatal "disk $found has no block device node"; return 1; }
-    printf '%s' "$SQS_DEVDIR/$found"
-}
-
-# sqs_has_addr <ipv4>: the address is assigned to a non-loopback interface.
-sqs_has_addr() {
-    "$IP" -o -4 addr show 2>/dev/null \
-        | awk -v a="$1" '$2 != "lo" { split($4, p, "/"); if (p[1] == a) f = 1 } END { exit !f }'
-}
-
-# sqs_alive <pid>: running or stopped, not exited (a zombie counts as exited).
-sqs_alive() { [ -r "/proc/$1/status" ] && ! grep -q '^State:[[:space:]]*Z' "/proc/$1/status" 2>/dev/null; }
-
-# sqs_starttime <pid>: kernel start time (stat field 22). It survives exec, and
-# pairing it with the pid reduces PID-reuse risk; without a pidfd the check and
-# the signal stay separate steps, so it is not an absolute guarantee.
-sqs_starttime() { sed 's/.*) //' "/proc/$1/stat" 2>/dev/null | cut -d' ' -f20; }
-
-# sqs_same <pid> <starttime>: the recorded process is still alive.
-sqs_same() { [ -n "$2" ] && sqs_alive "$1" && [ "$(sqs_starttime "$1")" = "$2" ]; }
-
-# sqs_write_config <addr> <nkey> <file>: the dedicated server's config. Its
-# max_payload admits a 1 MiB SQS message with its headers (ADR-0011:S4); the
-# 1MB NATS default does not.
-sqs_write_config() {
-    cat > "$3" <<EOF
-listen: "$1:$SQS_PORT"
-server_name: "sqs-store"
-max_payload: 2MiB
-jetstream {
-  store_dir: "$SQS_DATA"
-  sync_interval: always
-}
-authorization {
-  users = [ { nkey: "$2" } ]
-}
-EOF
-}
-
-# sqs_prepare: validate the contract, check and mount the data disk, write the
-# server config and mark the store ready. Returns non-zero, unready, on any refusal.
-sqs_prepare() {
-    rm -f "$SQS_RUN/ready"
-    mkdir -p "$SQS_RUN"
-    serial=$(sqs_fwcfg data-serial) || return 1
-    sqs_valid_serial "$serial" || { sqs_fatal "data-serial is not a volume serial"; return 1; }
-    addr=$(sqs_fwcfg listen-addr) || return 1
-    sqs_valid_ipv4 "$addr" || { sqs_fatal "listen-addr is not an IPv4 address"; return 1; }
-    nkey=$(sqs_fwcfg client-nkey) || return 1
-    sqs_valid_user_nkey "$nkey" || { sqs_fatal "client-nkey is not a user public NKey"; return 1; }
-    sqs_has_addr "$addr" || { sqs_fatal "listen-addr $addr is not on any guest NIC"; return 1; }
-
-    # The server checks the config, including the NKey checksum, before the
-    # disk is touched.
-    umask 077
-    sqs_write_config "$addr" "$nkey" "$SQS_RUN/nats.conf" || return 1
-    if ! "$NATS_BIN" -t -c "$SQS_RUN/nats.conf" > "$SQS_RUN/config-test.out" 2>&1; then
-        sqs_fatal "server config test failed: $(sed 's/U[A-Z2-7]\{55\}/<nkey>/g' "$SQS_RUN/config-test.out" | tr '\n' ' ')"
-        return 1
-    fi
-
-    dev=$(sqs_find_disk "$serial") || return 1
-    rc=0
-    "$E2FSCK" -p "$dev" > "$SQS_RUN/fsck.out" 2>&1 || rc=$?
-    sqs_log "e2fsck -p $dev rc=$rc"
-    # 0 clean, 1 corrected: proceed. Anything higher, including a disk that is
-    # not ext4 at all, refuses; the guest never formats.
-    [ "$rc" -le 1 ] || { sqs_fatal "e2fsck refused $dev: $(tr '\n' ' ' < "$SQS_RUN/fsck.out")"; return 1; }
-    mkdir -p "$SQS_DATA"
-    "$MOUNT" -t ext4 -o nodev,nosuid,noexec "$dev" "$SQS_DATA" || { sqs_fatal "mount $dev failed"; return 1; }
-    : > "$SQS_RUN/ready"
-    sqs_log "prepared: $dev mounted at $SQS_DATA, config written for $addr:$SQS_PORT; server not started yet"
-}
-
-# sqs_run: the respawned supervisor body. Its identity is recorded before the
-# stopping check: a stop that misses it set stopping first, so this parks; a
-# stop that reads it signals this process whether or not it has exec'd yet.
-sqs_run() {
-    mkdir -p "$SQS_RUN"
-    echo "$$ $(sqs_starttime $$)" > "$SQS_RUN/pid"
-    if [ -e "$SQS_RUN/stopping" ]; then
-        sqs_log "run: shutdown in progress, not starting"
-        return 2
-    fi
-    if [ ! -e "$SQS_RUN/ready" ]; then
-        sqs_log "run: store not prepared, not starting"
-        return 2
-    fi
-    sqs_log "run: starting pid=$$"
-    # ADR-0011:S11 bounds the store's Go heap; the guest must be larger.
-    GOMEMLIMIT=512MiB exec "$NATS_BIN" -c "$SQS_RUN/nats.conf"
-}
-
-# sqs_stop: first shutdown action. Block any respawn, SIGTERM the recorded run
-# process (a server, or a start not yet exec'd), SIGKILL only after the budget,
-# then sync and unmount. Each signal rechecks the identity; a survivor keeps the disk mounted.
-sqs_stop() {
-    mkdir -p "$SQS_RUN"
-    : > "$SQS_RUN/stopping"
-    sqs_log "stop: begin"
-    rec=$(cat "$SQS_RUN/pid" 2>/dev/null || true)
-    pid=${rec%% *} st=${rec#* }
-    [ "$st" != "$rec" ] || st=""
-    if [ -n "$pid" ] && sqs_same "$pid" "$st"; then
-        kill -TERM "$pid" 2>/dev/null || true
-        sqs_log "stop: SIGTERM sent to run pid=$pid"
-        i=0
-        while sqs_same "$pid" "$st" && [ "$i" -lt "$SQS_STOP_BUDGET_TENTHS" ]; do sleep 0.1; i=$((i + 1)); done
-        if sqs_same "$pid" "$st"; then
-            sqs_log "stop: run pid=$pid still running after $((SQS_STOP_BUDGET_TENTHS / 10))s, SIGKILL"
-            kill -KILL "$pid" 2>/dev/null || true
-            j=0
-            while sqs_same "$pid" "$st" && [ "$j" -lt 30 ]; do sleep 0.1; j=$((j + 1)); done
-        fi
-        if sqs_same "$pid" "$st"; then
-            sqs_log "stop: run pid=$pid STILL ALIVE, leaving $SQS_DATA mounted"
-            return 1
-        fi
-        sqs_log "stop: store exited after $((i / 10)).$((i % 10))s"
-    else
-        sqs_log "stop: no running store"
-    fi
-    sync
-    if grep -q " $SQS_DATA " "$SQS_MOUNTS" 2>/dev/null; then
-        rc=0
-        "$UMOUNT" "$SQS_DATA" || rc=$?
-        sqs_log "stop: umount $SQS_DATA rc=$rc"
-    fi
-}
-
-# sqs_shutdown: the only shutdown action. The generic read-only remount runs
-# only after the store stopped; a surviving writer keeps every mount as it is.
-sqs_shutdown() {
-    if ! sqs_stop; then
-        sync
-        sqs_log "shutdown: store survived, skipping umount -a -r"
-        return 1
-    fi
-    rc=0
-    "$UMOUNT" -a -r || rc=$?
-    echo "[init] shutdown: umount -a -r rc=$rc" >> "$SQS_CONSOLE"
-}
@@ -1,9 +0,0 @@
-#!/bin/sh
-# sysinit: load the disk drivers, then validate the guest contract, check and
-# mount the data disk and write the server config. A refusal leaves the store
-# unprepared, so the run step never starts it.
-. /usr/local/lib/sqs-store/sqs-store-lib.sh
-modprobe virtio_blk 2>/dev/null || true
-modprobe ext4 2>/dev/null || true
-sqs_prepare || sqs_log "store refused; it will not start"
-exit 0
@@ -1,6 +0,0 @@
-#!/bin/sh
-# Respawned by init. A store that must not run parks here instead of exiting,
-# so init does not respawn it in a tight loop.
-. /usr/local/lib/sqs-store/sqs-store-lib.sh
-sqs_run
-exec sleep 2147483647
@@ -1,5 +0,0 @@
-#!/bin/sh
-# The only shutdown action: stop the store and unmount its disk, then run the
-# generic read-only remount only if the store is gone.
-. /usr/local/lib/sqs-store/sqs-store-lib.sh
-sqs_shutdown
@@ -33,6 +33,10 @@ load_mod failover || true
 load_mod net_failover || true
 load_mod virtio_net || true

+# The guest's SQS store listens on 127.0.0.1 and the worker dials it there;
+# the kernel leaves lo down, so bring it up before init starts either.
+ip link set lo up
+
 NETCFG=/sys/firmware/qemu_fw_cfg/by_name/opt/spinifex/netcfg/raw
 i=0
 while [ "$i" -lt 25 ]; do
@@ -87,8 +91,8 @@ for n in 0 1 2 3 4 5; do
    [ -n "$rdst" ] && [ -n "$rvia" ] && ip route add "$rdst" via "$rvia" dev "$iface"
 done

-# The store image takes no lb-agent environment. A CA bundle is optional here:
-# nothing in this image dials out yet.
+# The image takes no lb-agent environment. The worker reads the cluster CA
+# from fw_cfg itself; this copy only serves other tools in the guest.
 CA_CERT=/sys/firmware/qemu_fw_cfg/by_name/opt/spinifex/ca-cert/raw
 mkdir -p /etc/ssl/certs
 [ -f "$CA_CERT" ] && cp "$CA_CERT" /etc/ssl/certs/ca-certificates.crt
@@ -3,6 +3,7 @@
 # the pre-exec shell PID 1. A press before this line is lost; the host re-sends.
 ::sysinit:/sbin/modprobe tiny_power_button power_signal=12
 ::sysinit:/usr/local/lib/sqs-worker/sqs-worker-prepare
+::respawn:/usr/local/lib/sqs-worker/sqs-worker-store
 ::respawn:/usr/local/lib/sqs-worker/sqs-worker-ready
 ::ctrlaltdel:/sbin/reboot
 ::shutdown:/usr/local/lib/sqs-worker/sqs-worker-stop
@@ -1,19 +1,23 @@
-# sqs-worker-lib.sh — shared by the SQS worker guest's readiness and stop
+# sqs-worker-lib.sh — shared by the SQS guest's store, readiness and stop
 # steps. Sourced, never executed. Every path and command is overridable so the
 # host test can drive the same code against test-owned processes and stub tools.
 #
-# store-ready exists only while the running watcher's last check passed. A
-# check proves the store accepted the worker key, not the store server's
-# identity, and a store failure is noticed up to one interval plus call time late.
+# The guest runs its own SQS nats-server on loopback and the worker beside it.
+# The server starts only after boot settled the receipt key; store-ready exists
+# only while the running watcher's last check of that server passed.

 SQSW_RUN=${SQSW_RUN:-/run/sqs-worker}
 SQSW_DATA=${SQSW_DATA:-/data}
 SQSW_CONSOLE=${SQSW_CONSOLE:-/dev/console}
 SQSW_MOUNTS=${SQSW_MOUNTS:-/proc/mounts}
 SQSW_BIN=${SQSW_BIN:-/usr/local/bin/sqs-worker}
+SQSW_LIB=${SQSW_LIB:-/usr/local/lib/sqs-worker/sqs-worker-lib.sh}
+SQSW_NATS_BIN=${SQSW_NATS_BIN:-/usr/local/bin/nats-server}
+SQSW_STORE_PORT=${SQSW_STORE_PORT:-4222}
 SQSW_STOP_BUDGET_TENTHS=${SQSW_STOP_BUDGET_TENTHS:-100}
 SQSW_RESPAWN_GAP=${SQSW_RESPAWN_GAP:-5}
-SQSW_PARK_INTERVAL=${SQSW_PARK_INTERVAL:-30}
+SQSW_PARK=${SQSW_PARK:-sleep 2147483647}
+SQSW_BOOT_RETRY=${SQSW_BOOT_RETRY:-30}
 UMOUNT=${UMOUNT:-umount}

 sqsw_log() { echo "[sqs-worker] $*" >> "$SQSW_CONSOLE"; }
@@ -28,28 +32,102 @@ sqsw_starttime() { sed 's/.*) //' "/proc/$1/stat" 2>/dev/null | cut -d' ' -f20;
 # sqsw_same <pid> <starttime>: the recorded process is still alive.
 sqsw_same() { [ -n "$2" ] && sqsw_alive "$1" && [ "$(sqsw_starttime "$1")" = "$2" ]; }

+# sqsw_write_config <file>: the guest's own server. It listens only on
+# loopback, with no monitor, route or cluster listener. Its max_payload admits
+# a 1 MiB SQS message with its headers (ADR-0011:S4); the 1MB default does not.
+sqsw_write_config() {
+    cat > "$1" <<CONF
+listen: "127.0.0.1:$SQSW_STORE_PORT"
+server_name: "sqs-store"
+max_payload: 2MiB
+jetstream {
+  store_dir: "$SQSW_DATA/jetstream"
+  sync_interval: always
+}
+CONF
+}
+
+# sqsw_prepare_store: write and test the server config once boot has settled
+# the receipt key. Returns non-zero, leaving the store unstarted, otherwise.
+sqsw_prepare_store() {
+    rm -f "$SQSW_RUN/store.conf"
+    [ -s "$SQSW_RUN/receipt-ready" ] || { sqsw_log "store: receipt key not settled, store not prepared"; return 1; }
+    umask 077
+    sqsw_write_config "$SQSW_RUN/store.conf.new" || return 1
+    if ! "$SQSW_NATS_BIN" -t -c "$SQSW_RUN/store.conf.new" > "$SQSW_RUN/config-test.out" 2>&1; then
+        sqsw_log "store: FATAL: server config test failed: $(tr '\n' ' ' < "$SQSW_RUN/config-test.out")"
+        return 1
+    fi
+    mv "$SQSW_RUN/store.conf.new" "$SQSW_RUN/store.conf"
+    sqsw_log "store: config written for 127.0.0.1:$SQSW_STORE_PORT; server not started yet"
+}
+
+# sqsw_park <why>: init respawns a step that exits, so a step that must not run
+# logs once and waits for shutdown instead of looping.
+sqsw_park() {
+    sqsw_log "$1"
+    exec $SQSW_PARK
+}
+
+# sqsw_record <name>: record this process's identity before the stopping
+# check, so a stop that misses the record has already set stopping.
+sqsw_record() { echo "$$ $(sqsw_starttime $$)" > "$SQSW_RUN/$1.pid"; }
+
+# sqsw_boot: settle the disk and receipt key once per guest boot. A transient
+# failure (exit 75) waits and exits so init runs it again; boot never remounts
+# a disk already mounted. Any other failure is final for this guest boot.
+sqsw_boot() {
+    [ ! -e "$SQSW_RUN/boot-refused" ] || sqsw_park "store: boot refused, not starting"
+    [ -s "$SQSW_RUN/receipt-ready" ] && return 0
+    # The boot child records itself before its stopping check, as the other
+    # steps do, so shutdown either sees its record or it never starts.
+    SQSW_RUN="$SQSW_RUN" SQSW_BIN="$SQSW_BIN" sh -c '. "$1"; sqsw_record boot; [ ! -e "$SQSW_RUN/stopping" ] || exit 75; exec "$SQSW_BIN" boot' \
+        sh "$SQSW_LIB" >> "$SQSW_CONSOLE" 2>&1 &
+    child=$!
+    rc=0
+    wait "$child" || rc=$?
+    rm -f "$SQSW_RUN/boot.pid"
+    [ ! -e "$SQSW_RUN/stopping" ] || sqsw_park "store: shutdown during boot, not starting"
+    case "$rc" in
+        0) return 0 ;;
+        75) sqsw_log "store: boot not settled, retrying in ${SQSW_BOOT_RETRY}s"
+            sleep "$SQSW_BOOT_RETRY"
+            exit 1 ;;
+        *) : > "$SQSW_RUN/boot-refused"
+           sqsw_park "store: boot refused (rc=$rc), not starting" ;;
+    esac
+}
+
+# sqsw_store: the respawned server body. It starts only on a prepared config.
+sqsw_store() {
+    mkdir -p "$SQSW_RUN"
+    sqsw_record store
+    [ ! -e "$SQSW_RUN/stopping" ] || sqsw_park "store: shutdown in progress, not starting"
+    sqsw_boot
+    if [ ! -s "$SQSW_RUN/store.conf" ] && ! sqsw_prepare_store; then
+        : > "$SQSW_RUN/boot-refused"
+        sqsw_park "store: not prepared, not starting"
+    fi
+    sqsw_log "store: starting pid=$$"
+    # ADR-0011:S11 bounds the store's Go heap; the guest must be larger.
+    GOMEMLIMIT=512MiB exec "$SQSW_NATS_BIN" -c "$SQSW_RUN/store.conf"
+}
+
 # sqsw_run: the respawned readiness body. A predecessor that was killed or
-# crashed could not remove its marker, so every start removes it first. Its
-# identity is recorded before the stopping check, as the store's run step does.
+# crashed could not remove its marker, so every start removes it first.
 sqsw_run() {
    mkdir -p "$SQSW_RUN"
-    echo "$$ $(sqsw_starttime $$)" > "$SQSW_RUN/ready.pid"
+    sqsw_record ready
    if [ -e "$SQSW_RUN/store-ready" ]; then
        rm -f "$SQSW_RUN/store-ready"
        sqsw_log "ready: removed a store-ready marker left by an earlier watcher"
    fi
-    # Returning here would make init respawn and log at once, forever. Park
-    # instead, logging once: shutdown stops this recorded process, and a key
-    # that appears later lets the watcher start.
-    if [ -e "$SQSW_RUN/stopping" ] || [ ! -s "$SQSW_RUN/public-key" ]; then
-        if [ -e "$SQSW_RUN/stopping" ]; then
-            sqsw_log "ready: shutdown in progress, not starting"
-        else
-            sqsw_log "ready: no worker key; readiness not started, waiting"
-        fi
-        while [ -e "$SQSW_RUN/stopping" ] || [ ! -s "$SQSW_RUN/public-key" ]; do
-            sleep "$SQSW_PARK_INTERVAL"
-        done
+    [ ! -e "$SQSW_RUN/stopping" ] || sqsw_park "ready: shutdown in progress, not starting"
+    [ ! -e "$SQSW_RUN/boot-refused" ] || sqsw_park "ready: boot refused, readiness not started"
+    # Boot may still be retrying; check again after a gap rather than parking.
+    if [ ! -s "$SQSW_RUN/store.conf" ]; then
+        sleep "$SQSW_RESPAWN_GAP"
+        exit 0
    fi
    # Init respawns at once; a watcher that keeps exiting restarts at most
    # once per gap instead of in a tight loop.
@@ -62,37 +140,47 @@ sqsw_run() {
    exec "$SQSW_BIN" ready >> "$SQSW_CONSOLE" 2>&1
 }

-# sqsw_stop: block any respawn, drop the marker, SIGTERM the recorded watcher,
-# SIGKILL only after the budget, and unmount the worker disk only once the
-# watcher is confirmed gone. A survivor keeps the disk mounted.
+# sqsw_end <name>: SIGTERM the recorded process, SIGKILL only after the
+# budget. Returns non-zero while it is still alive.
+sqsw_end() {
+    rec=$(cat "$SQSW_RUN/$1.pid" 2>/dev/null || true)
+    pid=${rec%% *} st=${rec#* }
+    [ "$st" != "$rec" ] || st=""
+    if [ -z "$pid" ] || ! sqsw_same "$pid" "$st"; then
+        sqsw_log "stop: no running $1"
+        return 0
+    fi
+    kill -TERM "$pid" 2>/dev/null || true
+    sqsw_log "stop: SIGTERM sent to $1 pid=$pid"
+    i=0
+    while sqsw_same "$pid" "$st" && [ "$i" -lt "$SQSW_STOP_BUDGET_TENTHS" ]; do sleep 0.1; i=$((i + 1)); done
+    if sqsw_same "$pid" "$st"; then
+        sqsw_log "stop: $1 pid=$pid still running after $((SQSW_STOP_BUDGET_TENTHS / 10))s, SIGKILL"
+        kill -KILL "$pid" 2>/dev/null || true
+        j=0
+        while sqsw_same "$pid" "$st" && [ "$j" -lt 30 ]; do sleep 0.1; j=$((j + 1)); done
+    fi
+    if sqsw_same "$pid" "$st"; then
+        sqsw_log "stop: $1 pid=$pid STILL ALIVE"
+        return 1
+    fi
+    sqsw_log "stop: $1 pid=$pid exited"
+    rm -f "$SQSW_RUN/$1.pid"
+}
+
+# sqsw_stop: block any respawn, drop the marker, stop the watcher, the server
+# and any boot in progress, and unmount the disk only once all are confirmed
+# gone. A survivor keeps the disk mounted.
 sqsw_stop() {
    mkdir -p "$SQSW_RUN"
    : > "$SQSW_RUN/stopping"
    rm -f "$SQSW_RUN/store-ready"
-    rec=$(cat "$SQSW_RUN/ready.pid" 2>/dev/null || true)
-    pid=${rec%% *} st=${rec#* }
-    [ "$st" != "$rec" ] || st=""
-    if [ -n "$pid" ] && sqsw_same "$pid" "$st"; then
-        kill -TERM "$pid" 2>/dev/null || true
-        sqsw_log "stop: SIGTERM sent to readiness watcher pid=$pid"
-        i=0
-        while sqsw_same "$pid" "$st" && [ "$i" -lt "$SQSW_STOP_BUDGET_TENTHS" ]; do sleep 0.1; i=$((i + 1)); done
-        if sqsw_same "$pid" "$st"; then
-            sqsw_log "stop: readiness watcher pid=$pid still running after $((SQSW_STOP_BUDGET_TENTHS / 10))s, SIGKILL"
-            kill -KILL "$pid" 2>/dev/null || true
-            j=0
-            while sqsw_same "$pid" "$st" && [ "$j" -lt 30 ]; do sleep 0.1; j=$((j + 1)); done
-        fi
-        if sqsw_same "$pid" "$st"; then
-            rm -f "$SQSW_RUN/store-ready"
-            sqsw_log "stop: readiness watcher pid=$pid STILL ALIVE, leaving $SQSW_DATA mounted"
-            return 1
-        fi
-        sqsw_log "stop: readiness watcher pid=$pid exited"
-    else
-        sqsw_log "stop: no running readiness watcher"
+    if ! sqsw_end ready || ! sqsw_end store || ! sqsw_end boot; then
+        rm -f "$SQSW_RUN/store-ready"
+        sqsw_log "stop: leaving $SQSW_DATA mounted"
+        return 1
    fi
-    rm -f "$SQSW_RUN/store-ready" "$SQSW_RUN/ready.pid"
+    rm -f "$SQSW_RUN/store-ready"
    sync
    if grep -q " $SQSW_DATA " "$SQSW_MOUNTS" 2>/dev/null; then
        rc=0
@@ -102,11 +190,11 @@ sqsw_stop() {
 }

 # sqsw_shutdown: the only shutdown action. The generic read-only remount runs
-# only after the watcher stopped; a survivor keeps every mount as it is.
+# only after both processes stopped; a survivor keeps every mount as it is.
 sqsw_shutdown() {
    if ! sqsw_stop; then
        sync
-        sqsw_log "shutdown: readiness watcher survived, skipping umount -a -r"
+        sqsw_log "shutdown: a guest process survived, skipping umount -a -r"
        return 1
    fi
    rc=0
@@ -1,8 +1,9 @@
 #!/bin/sh
-# sysinit: load the disk drivers, then select, check and mount the worker disk
-# and load or create the worker's store key on it. Any refusal is logged; the
-# key is never read or created on a disk that is not proven mounted.
+# sysinit: load the disk drivers and clear this boot's markers. The respawned
+# store step mounts the disk, settles the receipt key and starts the store.
+. /usr/local/lib/sqs-worker/sqs-worker-lib.sh
 modprobe virtio_blk 2>/dev/null || true
 modprobe ext4 2>/dev/null || true
-/usr/local/bin/sqs-worker boot >> /dev/console 2>&1 || echo "[sqs-worker] boot refused" >> /dev/console
+mkdir -p "$SQSW_RUN"
+rm -f "$SQSW_RUN/receipt-ready" "$SQSW_RUN/store.conf" "$SQSW_RUN/boot-refused"
 exit 0
@@ -1,6 +1,5 @@
 #!/bin/sh
-# Respawned by init. A watcher that must not run parks here instead of exiting,
-# so init does not respawn it in a tight loop.
+# Respawned by init: the worker, reporting ready only while the local store
+# answers. A watcher that must not run parks instead of exiting.
 . /usr/local/lib/sqs-worker/sqs-worker-lib.sh
 sqsw_run
-exec sleep 2147483647
@@ -1,5 +1,5 @@
 #!/bin/sh
-# The only shutdown action: stop the readiness watcher and unmount the worker
-# disk, then run the generic read-only remount only if the watcher is gone.
+# The only shutdown action: stop the worker, then the store, and unmount the
+# data disk, then run the generic read-only remount only if both are gone.
 . /usr/local/lib/sqs-worker/sqs-worker-lib.sh
 sqsw_shutdown
@@ -0,0 +1,5 @@
+#!/bin/sh
+# Respawned by init: the guest's own SQS nats-server, started only on a
+# prepared config. A store that must not run parks instead of exiting.
+. /usr/local/lib/sqs-worker/sqs-worker-lib.sh
+sqsw_store
@@ -17,6 +17,7 @@ import (
 	iamawsapi "github.com/mulgadc/spinifex/spinifex/domains/iam/awsapi"
 	rdsawsapi "github.com/mulgadc/spinifex/spinifex/domains/rds/awsapi"
 	route53awsapi "github.com/mulgadc/spinifex/spinifex/domains/route53/awsapi"
+	sqsawsapi "github.com/mulgadc/spinifex/spinifex/domains/sqs/awsapi"
 	stsawsapi "github.com/mulgadc/spinifex/spinifex/domains/sts/awsapi"
 	"github.com/mulgadc/spinifex/spinifex/ingress/aws/dispatch"

@@ -66,6 +67,7 @@ func compareAll() ([]awsmodel.OperationCoverage, error) {
 		ecsawsapi.ServiceName:     ecsawsapi.OperationInventory(),
 		eksawsapi.ServiceName:     eksawsapi.OperationInventory(),
 		iamawsapi.ServiceName:     iamawsapi.OperationInventory(),
+		sqsawsapi.ServiceName:     sqsawsapi.OperationInventory(),
 		stsawsapi.ServiceName:     stsawsapi.OperationInventory(),
 	})
 	// The loop below walks the embedded models, so a served service with no
@@ -6,6 +6,7 @@
 //
 //	eks-gateway-fetch -resource addons   # GET /clusters/{cluster}/internal-addons/{accountId}
 //	eks-gateway-fetch -resource recovery -instance-id i-… # GET /clusters/{cluster}/internal-recovery/{accountId}/{instanceId}
+//	eks-gateway-fetch -resource workers  # GET /clusters/{cluster}/internal-workers/{accountId}
 //
 // Flags default to environment variables seeded by cloud-init:
 // EKS_GATEWAY_URL, EKS_GATEWAY_CA, EKS_ACCESS_KEY, EKS_SECRET_KEY, EKS_REGION,
@@ -33,7 +34,7 @@ func main() {
 	flag.StringVar(&cfg.AccountID, "account-id", os.Getenv("EKS_ACCOUNT_ID"), "Cluster account ID")
 	flag.StringVar(&cfg.ClusterName, "cluster", os.Getenv("EKS_CLUSTER_NAME"), "Cluster name")
 	flag.StringVar(&cfg.InstanceID, "instance-id", os.Getenv("EKS_INSTANCE_ID"), "This member's EC2 instance ID (recovery resource)")
-	flag.StringVar(&cfg.Resource, "resource", "addons", "Resource to fetch: addons | recovery")
+	flag.StringVar(&cfg.Resource, "resource", "addons", "Resource to fetch: addons | recovery | workers")
 	flag.Parse()

 	if err := gatewayfetch.Run(context.Background(), cfg); err != nil {
@@ -0,0 +1,32 @@
+// Command eks-pod-identity-webhook is the IRSA mutating admission webhook run
+// on every EKS control-plane server; the implementation is in
+// agents/eks/podidentity. It listens on loopback only and is called by the
+// local kube-apiserver.
+package main
+
+import (
+	"context"
+	"log/slog"
+	"os"
+	"os/signal"
+	"syscall"
+
+	_ "github.com/mulgadc/bluebottle/pkg/fipsboot"
+
+	"github.com/mulgadc/spinifex/spinifex/agents/eks/podidentity"
+)
+
+func main() {
+	log := slog.New(slog.NewTextHandler(os.Stderr, nil))
+	cfg, err := podidentity.ParseServiceConfig(os.Args[1:], os.Getenv, os.Stderr)
+	if err != nil {
+		log.Error("invalid configuration", "error", err)
+		os.Exit(2)
+	}
+	ctx, stop := signal.NotifyContext(context.Background(), syscall.SIGINT, syscall.SIGTERM)
+	defer stop()
+	if err := podidentity.Run(ctx, cfg, log); err != nil {
+		log.Error("pod identity webhook stopped", "error", err)
+		os.Exit(1)
+	}
+}
@@ -0,0 +1,24 @@
+// lambda-agent runs inside a Lambda guest.
+//
+// A normal build reads its per-generation identity from fw_cfg, unloads
+// fw_cfg so nothing started later can read the guest key, and opens the
+// authenticated control connection to the host. It accepts no invocations and
+// never starts the function's bootstrap: the message protocol and grant
+// fencing do not exist yet. The agent never listens for the host.
+//
+// A lambdatestlink build instead relays invocations over an unauthenticated
+// test carrier for local proofs; it is not the guest protocol.
+package main
+
+import (
+	"context"
+	"flag"
+	"os"
+)
+
+func main() { os.Exit(run()) }
+
+func run() int {
+	flag.Parse()
+	return start(context.Background())
+}
@@ -0,0 +1,84 @@
+//go:build !lambdatestlink
+
+package main
+
+import (
+	"context"
+	"crypto/tls"
+	"errors"
+	"fmt"
+	"io"
+	"log/slog"
+
+	"github.com/mulgadc/spinifex/spinifex/agents/lambda/hostlink"
+)
+
+// Exit statuses: the guest refused to start, the control connection failed,
+// or the host closed it.
+const (
+	exitRefused = 2
+	exitLink    = 1
+	exitClosed  = 0
+)
+
+// errNoProtocol is returned when the host sends anything: this build defines no
+// message protocol, so it accepts no work.
+var errNoProtocol = errors.New("host sent data, but this build accepts no invocations")
+
+type startup struct {
+	fwCfgDir  string
+	fwCfgRoot string
+	unload    func() error
+}
+
+func start(ctx context.Context) int {
+	s := startup{fwCfgDir: hostlink.FwCfgDir, fwCfgRoot: hostlink.FwCfgRoot, unload: hostlink.UnloadFwCfg}
+	code, err := s.run(ctx)
+	if err != nil {
+		slog.Error("lambda-agent stopping", "exit", code, "err", err)
+	}
+	return code
+}
+
+// run loads the identity, seals fw_cfg whether or not the load succeeded,
+// then authenticates to the host and holds the connection until it closes.
+func (s startup) run(ctx context.Context) (int, error) {
+	id, loadErr := hostlink.LoadIdentity(s.fwCfgDir)
+	if err := hostlink.SealFwCfg(s.fwCfgRoot, s.unload); err != nil {
+		return exitRefused, errors.Join(loadErr, err)
+	}
+	if loadErr != nil {
+		return exitRefused, loadErr
+	}
+	conn, err := hostlink.Dial(ctx, id)
+	if err != nil {
+		if ctx.Err() != nil {
+			err = fmt.Errorf("%w: %w", err, context.Cause(ctx))
+		}
+		return exitLink, err
+	}
+	defer conn.Close()
+	slog.Info("control connection open; invocations are not enabled in this build", "host", id.HostAddr)
+	if err := awaitHostClose(ctx, conn); err != nil {
+		return exitLink, err
+	}
+	return exitClosed, nil
+}
+
+// awaitHostClose reads until the host closes the connection. Under TLS 1.3 a
+// host that rejects the guest certificate is only seen here, on first read.
+func awaitHostClose(ctx context.Context, conn *tls.Conn) error {
+	stop := context.AfterFunc(ctx, func() { _ = conn.Close() })
+	defer stop()
+	n, err := conn.Read(make([]byte, 1))
+	switch {
+	case n > 0:
+		return errNoProtocol
+	case errors.Is(err, io.EOF):
+		return nil
+	case ctx.Err() != nil:
+		return fmt.Errorf("control connection: %w", context.Cause(ctx))
+	default:
+		return fmt.Errorf("control connection: %w", err)
+	}
+}
@@ -0,0 +1,256 @@
+//go:build !lambdatestlink
+
+package main
+
+import (
+	"context"
+	"crypto/tls"
+	"errors"
+	"net"
+	"os"
+	"path/filepath"
+	"testing"
+	"time"
+
+	lambdav1 "github.com/mulgadc/spinifex/contracts/lambda/v1"
+	"github.com/mulgadc/spinifex/spinifex/runtime/compute/lambda/guestconn"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+const scope lambdav1.Scope = "ap-southeast-2/000000000001/fn-1/env-1/gen-1"
+
+// guest is a fake fw_cfg tree; unload removes it as unloading the module would.
+type guest struct {
+	root     string
+	unloads  int
+	sealedAt chan struct{}
+}
+
+func newGuest(t *testing.T, entries map[string][]byte) *guest {
+	t.Helper()
+	g := &guest{root: filepath.Join(t.TempDir(), "qemu_fw_cfg"), sealedAt: make(chan struct{}, 1)}
+	for name, b := range entries {
+		dir := filepath.Join(g.root, "by_name", lambdav1.FwCfgPrefix, name)
+		require.NoError(t, os.MkdirAll(dir, 0o700))
+		require.NoError(t, os.WriteFile(filepath.Join(dir, "raw"), b, 0o600))
+	}
+	return g
+}
+
+func (g *guest) startup() startup {
+	return startup{
+		fwCfgDir:  filepath.Join(g.root, "by_name", lambdav1.FwCfgPrefix),
+		fwCfgRoot: g.root,
+		unload: func() error {
+			g.unloads++
+			g.sealedAt <- struct{}{}
+			return os.RemoveAll(g.root)
+		},
+	}
+}
+
+type host struct {
+	ln   *net.TCPListener
+	gen  *guestconn.Generation
+	conn chan accepted
+}
+
+func newHost(t *testing.T) *host {
+	t.Helper()
+	ln, err := net.ListenTCP("tcp", &net.TCPAddr{IP: net.IPv4(127, 0, 0, 1)})
+	require.NoError(t, err)
+	t.Cleanup(func() { ln.Close() })
+	gen, err := guestconn.NewGeneration(scope, time.Hour)
+	require.NoError(t, err)
+	return &host{ln: ln, gen: gen, conn: make(chan accepted, 1)}
+}
+
+func (h *host) entries() map[string][]byte { return h.gen.GuestEntries(h.ln.Addr().String()) }
+
+type accepted struct {
+	conn *tls.Conn
+	err  error
+}
+
+// accept runs the host end until ctx ends and reports what it accepted.
+func (h *host) accept(ctx context.Context) {
+	go func() {
+		c, err := h.gen.AcceptGuest(ctx, h.ln, 2*time.Second)
+		h.conn <- accepted{c, err}
+	}()
+}
+
+// guest waits for the host to authenticate the guest.
+func (h *host) guest(t *testing.T) *tls.Conn {
+	t.Helper()
+	a := wait(t, h.conn, "host to authenticate the guest")
+	require.NoError(t, a.err)
+	return a.conn
+}
+
+// wait bounds every test wait so a stalled host or startup fails promptly.
+func wait[T any](t *testing.T, ch <-chan T, what string) T {
+	t.Helper()
+	select {
+	case v := <-ch:
+		return v
+	case <-time.After(10 * time.Second):
+		t.Fatalf("timed out waiting for %s", what)
+		panic("unreachable")
+	}
+}
+
+type result struct {
+	code int
+	err  error
+}
+
+func runAsync(ctx context.Context, s startup) <-chan result {
+	out := make(chan result, 1)
+	go func() {
+		code, err := s.run(ctx)
+		out <- result{code, err}
+	}()
+	return out
+}
+
+func TestStartupAuthenticatesAfterSealingFwCfg(t *testing.T) {
+	ctx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
+	defer cancel()
+	h := newHost(t)
+	h.accept(ctx)
+	g := newGuest(t, h.entries())
+	done := runAsync(ctx, g.startup())
+
+	conn := h.guest(t)
+	select {
+	case <-g.sealedAt:
+	default:
+		t.Fatal("guest dialled before fw_cfg was sealed")
+	}
+	_, err := os.Stat(g.root)
+	assert.ErrorIs(t, err, os.ErrNotExist, "fw_cfg still readable while connected")
+
+	require.NoError(t, conn.Close())
+	res := wait(t, done, "startup to return")
+	assert.Equal(t, exitClosed, res.code)
+	assert.NoError(t, res.err)
+	assert.Equal(t, 1, g.unloads)
+}
+
+func TestStartupRefusesHostData(t *testing.T) {
+	ctx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
+	defer cancel()
+	h := newHost(t)
+	h.accept(ctx)
+	done := runAsync(ctx, newGuest(t, h.entries()).startup())
+
+	conn := h.guest(t)
+	defer conn.Close()
+	_, err := conn.Write([]byte{0})
+	require.NoError(t, err)
+	res := wait(t, done, "startup to return")
+	assert.Equal(t, exitLink, res.code)
+	assert.ErrorIs(t, res.err, errNoProtocol)
+}
+
+// The host pins another generation's guest; the guest learns of the refusal
+// on its first read and fails with a link error, never a clean close.
+func TestStartupFailsWhenHostRefusesGuest(t *testing.T) {
+	ctx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
+	defer cancel()
+	h := newHost(t)
+	h.accept(ctx)
+	other, err := guestconn.NewGeneration(scope, time.Hour)
+	require.NoError(t, err)
+	entries := h.entries()
+	wrong := other.GuestEntries(h.ln.Addr().String())
+	entries[lambdav1.EntryGuestKey], entries[lambdav1.EntryGuestCert] = wrong[lambdav1.EntryGuestKey], wrong[lambdav1.EntryGuestCert]
+	g := newGuest(t, entries)
+
+	res := wait(t, runAsync(ctx, g.startup()), "startup to return")
+	assert.Equal(t, exitLink, res.code)
+	require.Error(t, res.err)
+	assert.Contains(t, res.err.Error(), "bad certificate")
+	assert.Equal(t, 1, g.unloads)
+	select {
+	case a := <-h.conn:
+		require.Error(t, a.err, "host accepted a guest it does not pin")
+	default:
+	}
+}
+
+func TestStartupRefusesWithoutDialling(t *testing.T) {
+	h := newHost(t)
+	cases := map[string]struct {
+		entries map[string][]byte
+		unload  func(g *guest) func() error
+		want    string
+	}{
+		"missing entry": {
+			entries: func() map[string][]byte { e := h.entries(); delete(e, lambdav1.EntryHostCert); return e }(),
+			want:    "read host-cert",
+		},
+		"fw_cfg stays readable": {
+			entries: h.entries(),
+			unload:  func(*guest) func() error { return func() error { return errors.New("module is built in") } },
+			want:    "fw_cfg still readable",
+		},
+		"unload claims success but tree remains": {
+			entries: h.entries(),
+			unload:  func(*guest) func() error { return func() error { return nil } },
+			want:    "unload reported success",
+		},
+	}
+	for name, tc := range cases {
+		t.Run(name, func(t *testing.T) {
+			g := newGuest(t, tc.entries)
+			s := g.startup()
+			if tc.unload != nil {
+				s.unload = tc.unload(g)
+			}
+			code, err := s.run(context.Background())
+			assert.Equal(t, exitRefused, code)
+			assert.ErrorContains(t, err, tc.want)
+		})
+	}
+
+	// Nothing dialled the host in any case above.
+	require.NoError(t, h.ln.SetDeadline(time.Now().Add(50*time.Millisecond)))
+	_, err := h.ln.Accept()
+	var ne net.Error
+	require.ErrorAs(t, err, &ne)
+	assert.True(t, ne.Timeout())
+}
+
+// A missing entry still seals fw_cfg: the key may be present even when
+// another entry is not.
+func TestStartupSealsEvenWhenIdentityIsIncomplete(t *testing.T) {
+	h := newHost(t)
+	e := h.entries()
+	delete(e, lambdav1.EntryHostAddr)
+	g := newGuest(t, e)
+	code, err := g.startup().run(context.Background())
+	assert.Equal(t, exitRefused, code)
+	assert.ErrorContains(t, err, "read host-addr")
+	assert.Equal(t, 1, g.unloads)
+	_, statErr := os.Stat(g.root)
+	assert.ErrorIs(t, statErr, os.ErrNotExist)
+}
+
+func TestStartupStopsWithContext(t *testing.T) {
+	ctx, cancel := context.WithCancelCause(context.Background())
+	defer cancel(nil)
+	h := newHost(t)
+	h.accept(ctx)
+	done := runAsync(ctx, newGuest(t, h.entries()).startup())
+	conn := h.guest(t)
+	defer conn.Close()
+
+	reason := errors.New("guest shutting down")
+	cancel(reason)
+	res := wait(t, done, "startup to return")
+	assert.Equal(t, exitLink, res.code)
+	assert.ErrorIs(t, res.err, reason)
+}
@@ -0,0 +1,93 @@
+//go:build lambdatestlink
+
+package main
+
+import (
+	"context"
+	"errors"
+	"flag"
+	"fmt"
+	"log/slog"
+	"net"
+	"net/http"
+	"os"
+	"os/exec"
+	"syscall"
+	"time"
+
+	"github.com/mulgadc/spinifex/spinifex/agents/lambda/agent"
+	"github.com/mulgadc/spinifex/spinifex/agents/lambda/hostlink"
+	"github.com/mulgadc/spinifex/spinifex/agents/lambda/runtimeapi"
+)
+
+const (
+	runtimeAddr = "127.0.0.1:9001"
+	taskRoot    = "/var/task"
+)
+
+var (
+	linkPath = flag.String("link", os.Getenv("LAMBDA_AGENT_LINK"), "test carrier device path")
+	handler  = flag.String("handler", os.Getenv("_HANDLER"), "function handler passed to the bootstrap")
+)
+
+func start(parent context.Context) int {
+	if *linkPath == "" {
+		slog.Error("no test carrier configured; refusing to start")
+		return 2
+	}
+	// The bootstrap is tenant code; fw_cfg must be unreadable before it runs.
+	if err := hostlink.SealFwCfg(hostlink.FwCfgRoot, hostlink.UnloadFwCfg); err != nil {
+		slog.Error("seal fw_cfg; refusing to start", "err", err)
+		return 2
+	}
+	dev, err := os.OpenFile(*linkPath, os.O_RDWR, 0)
+	if err != nil {
+		slog.Error("open test carrier; refusing to start", "err", err)
+		return 2
+	}
+	defer dev.Close()
+	link := hostlink.NewTestStream(dev)
+
+	srv := runtimeapi.New()
+	ln, err := net.Listen("tcp", runtimeAddr)
+	if err != nil {
+		slog.Error("listen runtime api", "err", err)
+		return 1
+	}
+	hs := &http.Server{Handler: srv.Handler(), ReadHeaderTimeout: 10 * time.Second}
+	go func() { _ = hs.Serve(ln) }()
+	defer hs.Close()
+
+	ctx, cancel := context.WithCancelCause(parent)
+	defer cancel(nil)
+
+	cmd := exec.CommandContext(ctx, taskRoot+"/bootstrap")
+	cmd.Env = []string{
+		"AWS_LAMBDA_RUNTIME_API=" + runtimeAddr,
+		"_HANDLER=" + *handler,
+		"LAMBDA_TASK_ROOT=" + taskRoot,
+		"PATH=/bin:/sbin:/usr/bin",
+	}
+	cmd.Stdout, cmd.Stderr = os.Stdout, os.Stderr
+	cmd.Cancel = func() error { return cmd.Process.Signal(syscall.SIGTERM) }
+	cmd.WaitDelay = 5 * time.Second
+	if err := cmd.Start(); err != nil {
+		slog.Error("start bootstrap", "err", err)
+		return 1
+	}
+	exited := make(chan struct{})
+	go func() {
+		defer close(exited)
+		err := cmd.Wait()
+		cancel(fmt.Errorf("bootstrap exited: %w", err))
+	}()
+
+	err = agent.Run(ctx, link, srv)
+	cancel(errors.New("agent stopping"))
+	<-exited
+	if err != nil {
+		slog.Error("agent stopped", "err", err)
+		return 1
+	}
+	return 0
+}
@@ -1,17 +1,22 @@
-// sqs-worker runs inside the shared SQS worker microVM.
+// sqs-worker runs inside the SQS microVM, next to its own SQS NATS server.
 //
 // Usage:
 //
 //	sqs-worker boot
 //	sqs-worker ready
 //
-// boot selects and mounts the worker's data disk named by fw_cfg, asks the
-// gateway for its current first-store decision with its instance-role
-// credentials, then loads or creates the store key on that disk and
-// publishes only the public key. ready then keeps checking the bound store's
-// JetStream with that key until it is stopped, keeping a readiness marker only
-// while the check passes; it proves the store accepted the worker key, not
-// the store server's identity.
+// boot selects and mounts the guest's data disk named by fw_cfg, asks the
+// gateway for its receipt key decision with its instance-role credentials,
+// then loads the recorded key, or creates one on a new volume and records its
+// fingerprint, before the guest's SQS store first starts on that disk. ready
+// then keeps checking the guest's own loopback JetStream, keeping a readiness
+// marker only while it answers, and serves SQS owner calls against it, pulling
+// them from the gateway with the same credentials.
+//
+// Serving is refused unless the gateway reports a single-node host cluster.
+// A disk whose receipt key is missing or does not match the recorded one
+// refuses to serve; recovering it needs a procedure not yet designed, and
+// nothing deletes or replaces the volume.
 package main

 import (
@@ -23,10 +28,14 @@ import (
 	"syscall"
 	"time"

+	"github.com/nats-io/nats.go/jetstream"
+
 	"github.com/mulgadc/spinifex/internal/gwsign"
 	"github.com/mulgadc/spinifex/internal/sqsworkergw"
 	"github.com/mulgadc/spinifex/spinifex/agents/sqs/workerboot"
 	"github.com/mulgadc/spinifex/spinifex/agents/sqs/workerkey"
+	"github.com/mulgadc/spinifex/spinifex/domains/sqs/workerserve"
+	"github.com/mulgadc/spinifex/spinifex/foundation/state/clustersize"
 )

 // The cluster CA the daemon supplies; the only root the client trusts.
@@ -36,7 +45,11 @@ const caPath = "/sys/firmware/qemu_fw_cfg/by_name/opt/spinifex/ca-cert/raw"
 // no static-key fallback.
 func newClaimer() workerboot.NewClaimer {
 	return func(region, gatewayURL string) (workerkey.Claimer, error) {
-		return newClient(region, gatewayURL)
+		c, err := newClient(region, gatewayURL)
+		if err != nil {
+			return nil, err
+		}
+		return workerboot.GatewayClaimer{Client: c}, nil
 	}
 }

@@ -48,13 +61,6 @@ func newClient(region, gatewayURL string) (*sqsworkergw.Client, error) {
 	return sqsworkergw.New(gatewayURL, region, ca, gwsign.NewInstanceRole())
 }

-type describer struct{ c *sqsworkergw.Client }
-
-func (d describer) DescribeStore(ctx context.Context) (workerboot.Store, error) {
-	st, err := d.c.DescribeStore(ctx)
-	return workerboot.Store(st), err
-}
-
 func main() {
 	if len(os.Args) != 2 || (os.Args[1] != "boot" && os.Args[1] != "ready") {
 		fmt.Fprintln(os.Stderr, "usage: sqs-worker boot|ready")
@@ -66,11 +72,18 @@ func main() {
 		run = ready
 	}
 	if err := run(logf); err != nil {
+		if os.Args[1] == "boot" && workerboot.Transient(err) {
+			logf("boot not settled, will retry: " + err.Error())
+			os.Exit(exitTransient)
+		}
 		logf("FATAL: " + err.Error())
 		os.Exit(1)
 	}
 }

+// exitTransient tells the guest's supervisor that boot may be run again.
+const exitTransient = 75
+
 func config(logf func(string)) workerboot.Config {
 	return workerboot.Config{
 		FwCfgDir:      "/sys/firmware/qemu_fw_cfg/by_name/opt/spinifex/sqs-worker",
@@ -99,19 +112,26 @@ func boot(logf func(string)) error {
 func ready(logf func(string)) error {
 	ctx, cancel := signal.NotifyContext(context.Background(), syscall.SIGTERM, syscall.SIGINT)
 	defer cancel()
+	// The store is one guest, so its streams and buckets are R1.
+	if err := clustersize.Declare(1); err != nil {
+		return err
+	}
 	err := workerboot.Watch(ctx, workerboot.ReadyConfig{
-		Config: config(logf),
-		NewDescriber: func(region, gatewayURL string) (workerboot.StoreDescriber, error) {
-			c, err := newClient(region, gatewayURL)
-			if err != nil {
-				return nil, err
-			}
-			return describer{c}, nil
-		},
+		Config:      config(logf),
 		Checker:     workerboot.NATSChecker{},
 		CallTimeout: 30 * time.Second,
 		Refresh:     30 * time.Second,
 		Refused:     5 * time.Minute,
+		Serve: &workerboot.ServeConfig{
+			Connector: workerboot.NATSConnector{},
+			Run: func(ctx context.Context, region, gatewayURL string, js jetstream.JetStream, receiptKey []byte) error {
+				c, err := newClient(region, gatewayURL)
+				if err != nil {
+					return err
+				}
+				return workerserve.Run(ctx, c, js, receiptKey, logf)
+			},
+		},
 	})
 	if errors.Is(err, context.Canceled) {
 		return nil
@@ -10,7 +10,7 @@ quota state.
 a request/reply route on which a gateway, in the `awsgw-quota-seal` queue
 group, irreversibly seals one account's quota override record.
 `domains/admission/quota` owns it: `awsgw` serves it.
-`accountteardown` is the consumer: it calls this subject to fence every
+`operator/accountlifecycle` is the consumer: it calls this subject to fence every
 override writer at the quota owner before the account record is removed.

 The request names the account in its body (`account_id`) and may carry an
@@ -31,10 +31,20 @@ Any current gateway may answer; the leader-accepted revision is the evidence
 whichever one does, and the override bucket is opened at seal time even on a
 gateway whose quotas were disabled at startup.

+## Seal-quota-override capability
+
+`AccountQuotaOverrideSealCapabilitySubject(node)` (`admission.v1.AccountQuotaOverrideSealCapability.<node>`) is a per-node, read-only request/reply route on which `operator/accountlifecycle` asks one configured gateway whether it supports the quota override seal, immediately before every seal attempt.
+The quota owner (`domains/admission/quota`) owns it: `awsgw` serves it with no queue group, so a request addressed to one node reaches only subscribers configured as that node, and two gateways sharing a node ID both answer.
+The consumer's quota seal reaper probes every configured node that runs `awsgw` and sends no seal request unless every one answers compatible.
+
+`AccountQuotaOverrideSealCapabilityVersion` is a claim about the whole binary's writers, not about the responder: version 1 means every quota override writer in the binary is revision-guarded and refuses a sealed record, and the binary serves the seal subject.
+A gateway binary without this responder has no subscriber, so the probe for that node reads as incomplete (`no-reply`), the same as a stopped or partitioned node.
+The reply proves nothing beyond the moment it was sent: a gateway downgraded or replaced after a successful probe is outside the guarantee.
+
 ### Compatibility boundary

-The subject name carries the contract major; `account_quota_seal_test.go`
-preserves the version-1 JSON shape.
+The subject name carries the contract major; `account_quota_seal_test.go` and `account_quota_seal_capability_test.go`
+preserve the version-1 JSON shapes and the subject builder.
 Producer and consumer are in the coordinated Spinifex release, so the window
 is N and N-1.
 A gateway that predates the subject has no subscription, so the request gets
@@ -0,0 +1,39 @@
+package admissionv1
+
+import "strings"
+
+// AccountQuotaOverrideSealCapabilityVersion is the version this binary reports.
+// Version 1: every quota override writer in the binary is revision-guarded and
+// refuses a sealed record, and the seal subject is served.
+const AccountQuotaOverrideSealCapabilityVersion = 1
+
+// AccountQuotaOverrideSealCapabilitySubject is one gateway's read-only
+// seal-capability route, addressed per node so a probe reaches exactly the
+// node it asks about.
+func AccountQuotaOverrideSealCapabilitySubject(node string) string {
+	return "admission.v1.AccountQuotaOverrideSealCapability." + node
+}
+
+// ValidSealCapabilityNode reports whether node is exactly one NATS subject
+// token. Responder and probe both check it before building a subject, so
+// neither can address a wildcard or another node's route.
+func ValidSealCapabilityNode(node string) bool {
+	return node != "" && !strings.ContainsAny(node, ".*>\t\n\r ")
+}
+
+// AccountQuotaOverrideSealCapabilityRequest asks node whether it supports the
+// quota override seal. CorrelationRef is an opaque caller reference echoed for
+// logs; it grants nothing and is never a fence or authority to write.
+type AccountQuotaOverrideSealCapabilityRequest struct {
+	Node           string `json:"node"`
+	CorrelationRef string `json:"correlation_ref,omitempty"`
+}
+
+// AccountQuotaOverrideSealCapabilityReply reports the responding node's
+// capability version. It carries no account data and proves nothing beyond
+// the moment it was sent.
+type AccountQuotaOverrideSealCapabilityReply struct {
+	Node              string `json:"node"`
+	CapabilityVersion int    `json:"capability_version"`
+	CorrelationRef    string `json:"correlation_ref,omitempty"`
+}
@@ -0,0 +1,43 @@
+package admissionv1
+
+import (
+	"encoding/json"
+	"testing"
+
+	"github.com/stretchr/testify/require"
+)
+
+func TestAccountQuotaOverrideSealCapabilitySubject(t *testing.T) {
+	require.Equal(t, "admission.v1.AccountQuotaOverrideSealCapability.node-1", AccountQuotaOverrideSealCapabilitySubject("node-1"))
+	require.Equal(t, "admission.v1.AccountQuotaOverrideSealCapability.node-2", AccountQuotaOverrideSealCapabilitySubject("node-2"))
+}
+
+func TestAccountQuotaOverrideSealCapabilityRequestJSONContract(t *testing.T) {
+	data, err := json.Marshal(AccountQuotaOverrideSealCapabilityRequest{Node: "node-1", CorrelationRef: "req-1"})
+	require.NoError(t, err)
+	require.JSONEq(t, `{"node":"node-1","correlation_ref":"req-1"}`, string(data))
+
+	data, err = json.Marshal(AccountQuotaOverrideSealCapabilityRequest{Node: "node-1"})
+	require.NoError(t, err)
+	require.JSONEq(t, `{"node":"node-1"}`, string(data))
+}
+
+func TestAccountQuotaOverrideSealCapabilityReplyJSONContract(t *testing.T) {
+	reply := AccountQuotaOverrideSealCapabilityReply{Node: "node-1", CapabilityVersion: 1, CorrelationRef: "req-1"}
+	data, err := json.Marshal(reply)
+	require.NoError(t, err)
+	require.JSONEq(t, `{"node":"node-1","capability_version":1,"correlation_ref":"req-1"}`, string(data))
+
+	data, err = json.Marshal(AccountQuotaOverrideSealCapabilityReply{Node: "node-1", CapabilityVersion: 1})
+	require.NoError(t, err)
+	require.JSONEq(t, `{"node":"node-1","capability_version":1}`, string(data))
+}
+
+func TestValidSealCapabilityNode(t *testing.T) {
+	for _, node := range []string{"node-1", "node_2", "n3"} {
+		require.True(t, ValidSealCapabilityNode(node), node)
+	}
+	for _, node := range []string{"", "a.b", "*", ">", "node-*", "node >", "node\t1", "node\n1"} {
+		require.False(t, ValidSealCapabilityNode(node), "%q", node)
+	}
+}
@@ -126,8 +126,8 @@ type Assign struct {
 	ENIPrivateIP  string `json:"eniPrivateIp,omitempty"`
 	ENISubnetID   string `json:"eniSubnetId,omitempty"`
 	// TaskRoleARN is the task's IAM role (from the taskdef); when set, the agent
-	// serves its credentials at 169.254.170.2 under CredID, defaulting CredID to
-	// the taskID when the scheduler leaves it empty.
+	// serves its credentials at 169.254.170.2 under CredID, a scheduler-minted
+	// secret. An assign without a CredID gets no credentials.
 	CredID      string `json:"credId,omitempty"`
 	TaskRoleARN string `json:"taskRoleArn,omitempty"`
 	// ExecutionRoleARN is the task's execution role (from the taskdef); when set,
@@ -0,0 +1,66 @@
+# Lambda v1 contracts
+
+This package will own the Lambda process contracts declared by ADR-0009:S13.
+It holds the L-GUEST declaration and, in `guest.go`, the connection vocabulary both ends share: the scope, the URI SAN roles, the ALPN identifier and the fw_cfg entry names. The transport and authentication are decided (ADR-0009:S17); the invocation message types and framing are not chosen, so no message types are defined here.
+It is not an AWS tenant-facing API.
+
+## L-GUEST: node executor to guest agent
+
+**Owner:** the Lambda domain.
+
+**Producers and consumers:** the node-local executor (`runtime/compute/lambda`) and the in-guest Lambda agent (`agents/lambda`), in both directions.
+
+**Operations:** bootstrap with scoped material, readiness, invoke, completion, bounded logs, shutdown.
+
+**Transport:** TCP over the microVM's network (ADR-0009:S17).
+The agent dials one connection to a listener the executor owns for the environment, and the guest exposes no listener for host control.
+At most one authenticated control connection is active per environment, and the agent may reconnect; how a prior connection and its in-flight work are handled is open (ADR-0009 M1).
+The network is shared and untrusted: no source address, port or bridge membership is identity or authority.
+Tenant code may be able to reach the listener, but it cannot open the agent's connection or key material, so any connection it makes fails authentication.
+
+**Authentication:** mutual TLS 1.3, bound to the environment ID and realization generation, using material the user runtime cannot read (ADR-0009:S17).
+For each generation the executor creates ephemeral self-signed host and guest certificates whose single URI SAN is `spinifex-lambda://<Region>/<account>/<function ID>/<environment ID>/<generation>/<role>`, with role `host` or `guest`.
+Each end validates the peer certificate against the one pinned peer certificate and separately requires that URI SAN with the opposite role; hostname verification is not relied on.
+ALPN carries the L-GUEST major version, `spinifex-lambda-guest/1`, and an unsupported version fails the handshake.
+Under TLS 1.3 the guest's handshake can complete before the host refuses its certificate, so the guest treats the connection as accepted only after the host's first message.
+
+**Bootstrap material:** the executor passes these fw_cfg entries under `opt/spinifex/lambda/`, and the agent reads them as root and unloads `qemu_fw_cfg` before user code starts.
+
+| Entry | Content |
+|---|---|
+| `guest-key` | Guest private key, PEM |
+| `guest-cert` | Guest certificate, PEM |
+| `host-cert` | The one pinned host certificate, PEM |
+| `expect` | `<Region>/<account>/<function ID>/<environment ID>/<generation>` |
+| `host-addr` | Host listener as `host:port` |
+
+The guest's own interface settings come from `opt/spinifex/netcfg`, the entry other system VMs use.
+The agent unloads `qemu_fw_cfg` even when an entry is missing, and refuses to start if the fw_cfg tree is still readable afterwards.
+Until the message format exists, the agent authenticates and holds the connection but accepts no host message and starts no user code.
+
+**Authority:** one environment's attempt only.
+The user runtime never holds these keys, and sharing the guest with the agent gives user code no platform authority.
+
+**Idempotency and fencing:** an invoke carries the attempt-grant identity and the attempt ID.
+The agent accepts a run command only under a current grant, and a completion under a superseded grant is discarded and reported.
+How a grant is enforced and proven ended is open (ADR-0009 M1).
+
+**Errors and versions:** the agent and executor fail closed on an unknown protocol version or an identity mismatch.
+
+**Readiness:** environment readiness needs authenticated, current-generation agent evidence; a running hypervisor process does not count.
+
+**Payloads:** the contract must carry a synchronous payload and response of up to 6 MB and an asynchronous payload of up to 1 MB, plus framing, over a path whose size bound is declared.
+How payloads move is open (ADR-0009 M3).
+
+### Compatibility boundary
+
+The agent ships inside the runtime image pinned by each published version, so it is outside the coordinated release train.
+The executor supports every L-GUEST major carried by a still-supported runtime image.
+Retiring an image follows a published runtime-deprecation lifecycle that blocks create and update before it blocks invoke.
+
+Evidence required before the first agent image ships: a function on an old image invoked by a new executor, and the reverse during rollback.
+
+## L-RUNTIME
+
+The guest agent serves the AWS Lambda Runtime API, version `2018-06-01`, to the user runtime at `AWS_LAMBDA_RUNTIME_API`.
+Spinifex does not version this AWS contract, and conformance is measured against AWS.
@@ -0,0 +1,54 @@
+// Package lambdav1 holds the Lambda process-contract vocabulary shared by the
+// node executor and the guest agent. See README.md for the declarations.
+package lambdav1
+
+import (
+	"fmt"
+	"strings"
+)
+
+// ALPNv1 names L-GUEST major version 1; a peer offering any other fails the handshake.
+const ALPNv1 = "spinifex-lambda-guest/1"
+
+// FwCfgPrefix prefixes every fw_cfg entry the executor passes the guest.
+const FwCfgPrefix = "opt/spinifex/lambda/"
+
+// fw_cfg entry names under FwCfgPrefix.
+const (
+	EntryGuestKey  = "guest-key"
+	EntryGuestCert = "guest-cert"
+	EntryHostCert  = "host-cert"
+	EntryScope     = "expect"
+	EntryHostAddr  = "host-addr"
+)
+
+// Roles named as the last URI SAN path segment.
+const (
+	RoleHost  = "host"
+	RoleGuest = "guest"
+)
+
+const uriScheme = "spinifex-lambda"
+
+// Scope names one environment generation:
+// Region/account/function ID/environment ID/generation.
+type Scope string
+
+// ParseScope accepts exactly five non-empty, slash-separated segments.
+func ParseScope(s string) (Scope, error) {
+	parts := strings.Split(s, "/")
+	if len(parts) != 5 {
+		return "", fmt.Errorf("lambdav1: scope %q: want 5 segments, got %d", s, len(parts))
+	}
+	for _, p := range parts {
+		if p == "" || strings.ContainsAny(p, " \t\r\n?#%") {
+			return "", fmt.Errorf("lambdav1: scope %q: empty or invalid segment", s)
+		}
+	}
+	return Scope(s), nil
+}
+
+// URI is the single URI SAN a certificate for role in this scope carries.
+func (s Scope) URI(role string) string {
+	return uriScheme + "://" + string(s) + "/" + role
+}
@@ -0,0 +1,20 @@
+package lambdav1
+
+import (
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestParseScope(t *testing.T) {
+	s, err := ParseScope("ap-southeast-2/000000000001/fn-1/env-1/gen-1")
+	require.NoError(t, err)
+	assert.Equal(t, "spinifex-lambda://ap-southeast-2/000000000001/fn-1/env-1/gen-1/host", s.URI(RoleHost))
+	for _, bad := range []string{
+		"", "a/b/c/d", "a/b/c/d/e/f", "a//c/d/e", "a/b/c/d/e/", "a/b/c/d/e?x", "a/b/c/d/e#x", "a b/b/c/d/e",
+	} {
+		_, err := ParseScope(bad)
+		assert.Error(t, err, bad)
+	}
+}
@@ -74,3 +74,65 @@ type ListAccountsResponse struct {
 	Accounts []AccountSummary `json:"accounts"`
 	Count    int              `json:"count"`
 }
+
+// IAMQuiescence is the coordinator's recorded view of IAM quiescence for one
+// fence. Recording only: it admits nothing, and QuiescedScope stays empty until
+// a final inventory re-read exists to catch a late write.
+type IAMQuiescence struct {
+	OperationID   string            `json:"operationId"`
+	Fence         uint64            `json:"fence"`
+	Generation    uint64            `json:"generation"`
+	QuiescedScope []string          `json:"quiescedScope"`
+	Unresolved    []UnresolvedScope `json:"unresolved"`
+	Consumers     []ConsumerBasis   `json:"consumers"`
+	Writers       []AccountedWriter `json:"writers"`
+	Inventory     []InventoryStatus `json:"inventory"`
+	EvaluatedAt   string            `json:"evaluatedAt"`
+}
+
+// UnresolvedScope names a scope or writer that blocks quiescence, with a fixed
+// reason. It is never a timeout.
+type UnresolvedScope struct {
+	Scope  string `json:"scope"`
+	Class  string `json:"class"`
+	Reason string `json:"reason"`
+}
+
+// ConsumerBasis cites an enforcement consumer and the accepted contract that
+// lets it count as having applied the generation.
+type ConsumerBasis struct {
+	Consumer string `json:"consumer"`
+	Basis    string `json:"basis"`
+}
+
+// AccountedWriter is a writer that cannot be fenced and is covered by the
+// owner's inventory instead. Ref names the tracked work that closes the gap.
+type AccountedWriter struct {
+	Writer string `json:"writer"`
+	Basis  string `json:"basis"`
+	Ref    string `json:"ref"`
+}
+
+// InventoryStatus is the owner inventory's status for one kind at evaluation.
+type InventoryStatus struct {
+	Kind   string `json:"kind"`
+	Status string `json:"status"`
+}
+
+// Classes of an UnresolvedScope.
+const (
+	UnresolvedClassFence    = "fence"
+	UnresolvedClassConsumer = "consumer"
+	UnresolvedClassWriter   = "writer"
+)
+
+// Fixed reasons of an UnresolvedScope.
+const (
+	ReasonFenceMismatch                = "fence-mismatch"
+	ReasonIAMReadbackFailed            = "iam-readback-failed"
+	ReasonNoAcceptedEnforcement        = "no-accepted-enforcement-contract"
+	ReasonNoCompleteOwnerInventory     = "no-complete-owner-inventory"
+	ReasonParticipantNotDeclared       = "participant-not-declared"
+	ReasonOwnerInventoryIncomplete     = "owner-inventory-incomplete"
+	ReasonFinalInventoryRereadNotWired = "final-inventory-reread-not-wired"
+)
@@ -152,12 +152,12 @@ Operational commands for inspecting cluster state. These fan out NATS requests t
 | `spx admin account create` | `--name` (required) | Connects to NATS → refuses if an account already holds the name (case- and whitespace-insensitive) → reserves the name in `spinifex-account-names` KV → CAS loop on `spinifex-account-counter:next_id` for sequential 12-digit ID → creates Account record in `spinifex-accounts` KV → creates default VPC → creates `admin` user, AdministratorAccess policy (Action:*, Resource:*) and one access key → prints credentials and writes a `spinifex-<name>` AWS CLI profile |
 | `spx admin account create --remote` | `--name` (required), `--endpoint` (default: this node's AWS gateway), `--region` (default: this node's region), `--ca-bundle` (default: this node's CA), `--client-token` (default: generated), `--source` (default: `spx-cli`) | Same account, created through `POST /admin/CreateAccount` over SigV4 instead of NATS — the path the public signup form uses, so a shell can exercise it. Credentials come from the standard AWS chain (env vars or `AWS_PROFILE`), never from the cluster master key; the caller must be an IAM user in the super-admin account holding `spinifex:CreateAccount`. Reuse the printed `--client-token` to retry: a fresh token would create a second account. |
 | `spx admin account list` | — | Connects to NATS → IAMService.ListAccounts() → prints table with Account ID, Name, Status, Created |
-| `spx admin account list --remote` | `--endpoint` (default: this node's AWS gateway), `--region` (default: this node's region), `--ca-bundle` (default: this node's CA) | Same listing read through `POST /admin/ListAccounts` over SigV4 instead of NATS, so an off-cluster operator or the load-test harness sees its tenants without SSH access to a node. Credentials come from the standard AWS chain; the caller must be an IAM user in the super-admin account holding `spinifex:ListAccounts`. |
+| `spx admin account list --remote` | `--endpoint` (default: this node's AWS gateway), `--region` (default: this node's region), `--ca-bundle` (default: this node's CA) | Same listing read through `POST /admin/ListAccounts` over SigV4 instead of NATS, so an off-cluster operator or the load-test harness sees its tenants without SSH access to a node. Credentials come from the standard AWS credential chain: on a cluster node that means `AWS_PROFILE=spinifex` (there is no default profile, so without it the command fails to find credentials), or equivalent credentials; the caller must be an IAM user in the super-admin account holding `spinifex:ListAccounts`. |
 | `spx admin account quota get <account-id>` | `--endpoint` (default: this node's AWS gateway), `--region` (default: this node's region), `--ca-bundle` (default: this node's CA) | Reads the limits in force for one account through `POST /admin/GetAccountQuota` over SigV4, and reports for each dimension whether it came from the `[quota]` block in `awsgw.toml` (`config`) or from an override stored against the account (`override`). Credentials come from the standard AWS chain; the caller must be an IAM user in the super-admin account holding `spinifex:GetAccountQuota`. |
 | `spx admin account quota set <account-id>` | `--vcpus`, `--vpcs`, `--subnets`, `--eips`, `--volumes`, `--volumes-gib`, `--rds-instances`, `--load-balancers`, `--clear`, plus `--endpoint`, `--region`, `--ca-bundle` | Overrides individual limits for one account through `POST /admin/PutAccountQuota`, leaving every dimension not named inheriting the configured baseline. `-1` uncaps a dimension; `0` is a real limit that denies every request on it. `--clear` removes the whole override record and cannot be combined with a limit flag. The account must already exist. Prints the resulting limits, so a caller never has to guess what it applied. The caller must hold `spinifex:PutAccountQuota`. |
-| `spx admin account delete <account-id>` | `--dry-run`, `--force`, `--yes` (requires `--name`), `--name` (account name, must match the stored record) | Prints the inventory it is about to remove → confirms the account name (prompted unless `--name` is given) → marks the account `TERMINATING`, which blocks its credentials cluster-wide → deletes its resources in dependency order (compute, attachments, storage, network, platform, identity, account), waiting for each stage to actually drain before starting the next → releases the name reservation and quota counter → deletes the account record. Anything that will not delete is reported as **stuck** with the reason, and the account stays `TERMINATING` so the residue keeps an owner. `000000000000` and `000000000001` are refused. `--dry-run` lists and changes nothing. The compute stage reaps ECS clusters, EKS nodegroups and clusters, and RDS instances before it reaps plain instances, because their backing VMs belong to the system account and a tenant-scoped instance listing cannot see them. The storage stage reaps the account's S3 buckets: uploads in flight are aborted, then the objects are removed a page at a time, then the bucket. Only buckets the account owns are ever listed, so the shared platform buckets are out of reach rather than filtered out. Platform-managed serving endpoints are reaped ahead of instances, because only their own delete releases the ENI and attached data volume alongside the VM, and spot instance requests are cancelled first so a live request cannot launch a replacement for something already terminated. The network stage reaps load balancers, then target groups, then NAT gateways and egress-only internet gateways before the subnets, security groups and addresses they hold. The platform stage reaps ACM certificates. |
+| `spx admin account delete <account-id>` | `--dry-run`, `--force`, `--yes` (requires `--name`), `--name` (account name, must match the stored record), `--endpoint`, `--region`, `--ca-bundle`, `--client-token` | Always runs through `POST /admin/DeleteAccount` over SigV4 and follows `DescribeAccountDeletion` to completion; the CLI never tears an account down in its own process, so a CLI delete and an API delete cannot race. Credentials come from the standard AWS chain; the caller must be an IAM user in the super-admin account holding `spinifex:DeleteAccount` and `spinifex:DescribeAccountDeletion`. Prints the inventory it is about to remove → confirms the account name (prompted unless `--name` is given) → marks the account `TERMINATING`, which blocks its credentials cluster-wide → deletes its resources in dependency order (compute, attachments, storage, network, platform, identity, account), waiting for each stage to actually drain before starting the next → seals the account's quota override record so no later write can recreate it, which needs every gateway node the cluster configures to answer a read-only probe first (otherwise nothing is sealed and the account stays `TERMINATING`) → releases the name reservation and quota counter → deletes the account record. Anything that will not delete is reported as **stuck** with the reason, and the account stays `TERMINATING` so the residue keeps an owner. `000000000000` and `000000000001` are refused. `--dry-run` lists and changes nothing. The compute stage reaps ECS clusters, EKS nodegroups and clusters, and RDS instances before it reaps plain instances, because their backing VMs belong to the system account and a tenant-scoped instance listing cannot see them. The storage stage reaps the account's S3 buckets: uploads in flight are aborted, then the objects are removed a page at a time, then the bucket. Only buckets the account owns are ever listed, so the shared platform buckets are out of reach rather than filtered out. Platform-managed serving endpoints are reaped ahead of instances, because only their own delete releases the ENI and attached data volume alongside the VM, and spot instance requests are cancelled first so a live request cannot launch a replacement for something already terminated. The network stage reaps load balancers, then target groups, then NAT gateways and egress-only internet gateways before the subnets, security groups and addresses they hold. The platform stage reaps ACM certificates. |
 | `spx admin account delete --force` | as above | Escalates only for resources the ordinary API refuses: it clears a volume attachment in the control plane without the guest's cooperation (`ec2.ForceDetachVolume`), hard-destroys after a graceful attempt times out, and treats an already-missing resource as deleted. It never reorders the stages and never widens scope beyond the account. Use it once the guest is being destroyed — this is the escape from a volume stuck attached to an instance that will not terminate, where neither can otherwise be deleted. |
-| `spx admin account delete --remote` | as above, plus `--endpoint`, `--region`, `--ca-bundle`, `--client-token` | Same teardown driven through `POST /admin/DeleteAccount` over SigV4 instead of NATS, then follows `DescribeAccountDeletion` to completion, printing each stage as it lands. Credentials come from the standard AWS chain; the caller must be an IAM user in the super-admin account holding `spinifex:DeleteAccount` and `spinifex:DescribeAccountDeletion`. |
+| `spx admin account delete --remote` | as above | Accepted for existing scripts and ignored: every delete already goes through the admin API. |
 | `spx admin principal create <name>` | `--grant` (repeatable or comma-separated; default: every admin method — `CreateAccount`, `DeleteAccount`, `DescribeAccountDeletion`, `GetAccountQuota`, `ListAccounts`, `PutAccountQuota`) | Creates an IAM user in the super-admin account with an inline `spx-admin-methods` policy allowing the requested methods, revokes any access key left from a previous run, and mints one new key. Each method is named rather than granted as `spinifex:*`, so a later addition to the admin surface is not authorised by an existing key. The principal's grants are exactly that one policy: re-running with fewer `--grant` values narrows it, and any other inline policy on the user is removed, so a rename cannot leave a wider grant behind. The secret is printed once; store it in a secret manager or an AWS profile, never in a config file. `admin` is refused as a name — it is the account's own administrator and rotating its key would revoke the bootstrap credential. |
 | `spx admin principal list` | — | Lists every principal in the super-admin account with its access-key count and what it may call. An attached policy is reported by name (`AdministratorAccess (attached)`) rather than expanded, which is how an unscoped principal is spotted. |
 | `spx admin principal revoke <name>` | — | Deletes every access key the named principal holds, leaving the user and its policy in place. Immediate and cluster-wide; re-issue with `principal create`. |
@@ -194,18 +194,32 @@ Operational commands for inspecting cluster state. These fan out NATS requests t

 ### SQS System Instances

-These commands start the first SQS system instances on one node. They do not serve the SQS API, and a bound VM is not a ready service.
+These commands enable SQS on a single-node cluster and start its one system VM.
+A bound VM is not a ready service: `spx admin sqs status` reports readiness.

-Both send a signed `spinifex` admin action (`EnsureSQSWorker`, `EnsureSQSStore`) to the AWS gateway using the standard AWS credential chain, for example `AWS_PROFILE=spinifex`. The caller must be in the super-admin account (`000000000001`) and hold `spinifex:<Action>`. The VMs, volumes, private VPC and ENIs belong to the system account (`000000000000`). `--endpoint`, `--region` and `--ca-bundle` default to this node's gateway, region and `config/ca.pem`; off a node, `--endpoint` and `--region` are required. Each call waits up to 11 minutes.
+The SQS VM runs two processes on one encrypted data volume: the Region's dedicated SQS NATS server, which listens only on the guest's loopback address and is separate from platform NATS, and the worker, which serves SQS owner calls from that local store. The worker starts serving only after the local store answers and the receipt-signing key on the volume matches the fingerprint the platform recorded.
+
+SQS supports only a fresh single-node install. A worker record without a supported launch contract, or any record of the separate SQS store VM from an earlier dev build, cannot be upgraded: `enable` refuses it, and any existing volume is kept, not inspected, emptied or replaced. An existing separate store VM is refused on relaunch with its volume kept, and an existing worker VM is not restarted into a supported mode; upgrading such an install is unsupported.
+
+**Do not put tenant production data in SQS.** SQS-managed per-message encryption is not implemented: `GetQueueAttributes` reports `SqsManagedSseEnabled=false`, and a request to enable SSE-SQS or SSE-KMS is rejected.
+The SQS data volume must be encrypted, which is a separate layer under every queue and not SSE.
+There is no `DeleteQueue` or `PurgeQueue`, so a queue and its unexpired messages cannot be removed once created.
+Retention is checked when a message is received, against the queue's retention at that moment, and nothing deletes an expired message: the store drops a body only at its age backstop of 14 days and 1 hour.
+So a message past a short retention can be delivered again if `MessageRetentionPeriod` is later raised while its body is still stored.
+Account termination does not yet remove SQS queues or messages, and there is no disable: the VM's volume, receipt key and data are kept.
+
+Each sends a signed `spinifex` admin action (`EnableSQS`, `DescribeSQS`, `EnsureSQSWorker`) to the AWS gateway using the standard AWS credential chain, for example `AWS_PROFILE=spinifex`. The caller must be in the super-admin account (`000000000001`) and hold `spinifex:<Action>`. The VM, volume, private VPC and ENI belong to the system account (`000000000000`). `--endpoint`, `--region` and `--ca-bundle` default to this node's gateway, region and `config/ca.pem`; off a node, `--endpoint` and `--region` are required. `ensure-worker` and `enable` wait up to 11 minutes, and `status` up to 1 minute.

 | Command | Flags | Description |
 |---------|-------|-------------|
-| `spx admin sqs ensure-worker` | `--node` (required), `--instance-type` (required), `--endpoint`, `--region`, `--ca-bundle` | Launches the Region's one SQS worker VM on `--node`, or reports the VM already bound there: node, region, instance, instance type and generation. The worker is pinned to its first node; another node is refused with `InvalidParameterValue`, and there is no takeover. It gets a private system VPC with one /24 subnet and no internet gateway, NAT or Elastic IP, and an encrypted 1 GiB system-account data volume on which the guest generates its own store key. The platform records only the public key. A different `--instance-type` is accepted only while no volume, network interface or launch step has been attempted; once any has, it is `IncorrectState`, whether or not the step completed. |
-| `spx admin sqs ensure-store` | `--node` (required, the worker's node), `--instance-type` (required, more than 512 MiB memory), `--volume-size-gib` (required, 1–16384), `--endpoint`, `--region`, `--ca-bundle` | Launches the Region's SQS store VM on the worker's node with an encrypted system-account data volume, or reports the VM already bound there: node, region, instance, instance type and address. Needs the worker bound with its public key recorded, else `IncorrectState`. The store accepts only that key. A different size, type or key than the first call is `IncorrectState`. |
+| `spx admin sqs enable` | `--node` (required), `--instance-type` (required, more than 512 MiB memory), `--volume-size-gib` (required, 1–16384), `--endpoint`, `--region`, `--ca-bundle` | Records SQS as enabled for the Region, pinned to `--node`, then launches or reports the SQS VM there. The desired state is recorded first, so rerunning the command resumes an interrupted enable. More than one node in the cluster is refused before anything is recorded. A worker record with a missing or unsupported launch contract, or a record of the separate store VM, is also refused before anything is recorded or launched; any existing volume is kept, and upgrading it is not supported. |
+| `spx admin sqs status` | `--endpoint`, `--region`, `--ca-bundle` | Shows the requested state, node, generation, VM state, receipt key state, whether a worker is polling this gateway, whether this gateway serves the SQS API, and `Ready`, which is true only when all of them hold. `Unsupported` names a record whose missing or unsupported launch contract blocks enable. |
+| `spx admin sqs ensure-worker` | `--node` (required), `--instance-type` (required, more than 512 MiB memory), `--volume-size-gib` (required, 1–16384), `--endpoint`, `--region`, `--ca-bundle` | Launches the Region's one SQS VM on `--node`, or reports the VM already bound there: node, region, instance, instance type and generation. The VM is pinned to its first node; another node is refused with `InvalidParameterValue`, and there is no takeover. It gets a private system VPC with one /24 subnet and no internet gateway, NAT or Elastic IP, and an encrypted system-account data volume of `--volume-size-gib` that holds the queue data and the receipt-signing key the guest creates on first boot. The platform records only the key's fingerprint. A different `--instance-type` is accepted only while no volume, network interface or launch step has been attempted; once any has, it is `IncorrectState`, whether or not the step completed. A different `--volume-size-gib` than the recorded one is `IncorrectState`. |

 `ServiceUnavailable` never proves that no SQS resources exist, because an earlier attempt may already have created some. Its message says what it does establish. When no node serves the subject, this request was not delivered. When a volume create was not sent, that particular create did not happen. When the node cannot boot the guest yet or its launcher lacks a capability, this request stopped before allocating or recording anything. When the gateway times out waiting for the node's reply, nothing is known: the launch may still be running, so never assume it created nothing. In every case, call the command again with the same arguments. A retry first checks the durable record of the attempt and continues only when that is safe; it can proceed to the first launch after a create that was proven unsent, and otherwise reports the recorded outcome as bound or `IncorrectState`, or returns `ServiceUnavailable` again. `IncorrectState` from a volume, network interface or launch step means that step was recorded as sent but its outcome was never observed. It is never resent, the message names the volume when its ID is known, and no tool reconciles it yet: recovery is manual.

-`bound` says only that the VM was launched for the current record. The worker decides store readiness itself, by checking that the store is bound to its own public key and that a NATS connection with that key succeeds; nothing reports that readiness to the operator. The worker-to-store connection does not authenticate the store server. Real VPC routing, IMDS and gateway TLS from the guest, and worker-to-store traffic on a real interface have not been tested.
+`bound` says only that the VM was launched for the current record. The worker is ready only when the guest's local store answers, and it serves only while its receipt key matches the recorded fingerprint; a missing or different key once one is recorded refuses serving and is never replaced.
+Real VPC routing, IMDS and gateway TLS from the guest have not been tested on a live host.

 ## Private Admin API

@@ -203,12 +203,14 @@ Viperblock responds with an NBD URI that QEMU uses to access the block device.

 A direct-boot microVM keeps the launcher's saved `vm.Config` as a clean base, and `startQEMU` adds data volumes only to a launch-only copy of it (`runtime/compute/vm/lifecycle.go`).
 At each start the daemon takes one snapshot of the instance's already-mounted data requests, probes each NBD or host block export from that snapshot, and builds the command from the same snapshot, so the launched endpoint is the one that was checked and is never written back into the saved base.
-Each data volume becomes a `virtio-blk-device` on the MMIO bus with the same id, node name, I/O thread, serial and `werror=stop,rerror=stop` policy as the PCI device, and takes no PCIe hot-plug port.
+The microVM boots with `pcie=on`, so its NICs are `virtio-net-pci` devices with MSI-X interrupts instead of `virtio-net-device` MMIO devices.
+Each data volume becomes a `virtio-blk-pci` device on the default PCIe bus with the same id, node name, I/O thread, serial and `werror=stop,rerror=stop` policy as the q35 device, and takes no PCIe hot-plug port.
+A microVM started without `pcie=on` keeps the `virtio-*-device` MMIO transports.

 Boot and EFI requests are rejected on this path, and a direct-boot instance with no data requests launches its base unchanged.
 Disk-boot (q35) instances keep the existing PCI hot-plug path.

-Hot-attaching a volume to a running direct-boot microVM is not supported, because the MMIO bus has no PCIe root port to plug into.
+Hot-attaching a volume to a running direct-boot microVM is not supported, because the microVM has no PCIe hot-plug root port to plug into.
 `AttachVolume` refuses it before mounting the volume or changing any instance or volume state, and the EC2 API returns `UnsupportedOperation`.

 This covers only requests that are already mounted when the instance starts; nothing yet supplies data volumes to a system instance, and the path is verified with fake-QEMU unit tests, not with a real EBS volume in a guest.
@@ -276,8 +278,8 @@ storage-provider contract.
 ### SQS message store (internal)

 `domains/sqs/message` defines the store-neutral interface the SQS domain reaches its message store through (ADR-0011:S6), with a JetStream adapter in `domains/sqs/message/jetstream` and a shared conformance suite in `domains/sqs/message/conformance`.
-It is an internal primitive: nothing outside the SQS domain uses it, no daemon or service is wired to it, and it does not itself deliver an SQS API.
-The dedicated JetStream server ADR-0011 describes is not deployed by this code, and Spinifex ships no host service, unit, user or firewall rule for it; the guest that could run it is described under *SQS store system instance (not wired)*.
+It is used only by the SQS worker inside the SQS system instance and does not itself deliver an SQS API.
+Spinifex ships no host service, unit, user or firewall rule for the dedicated JetStream server; the system instance that runs it is described under *SQS system instance (operator-launched)*.

 The interface covers durable and idempotent append, body and metadata reads, conditional state writes and deletes, per-namespace counts, namespace enumeration, an ordered change feed, a bounded paginated scan, namespace deletion and a sweep of tombstoned bodies.
 A namespace carries an immutable message-shard and state-shard allocation chosen when the queue is created, so a later change in shard counts never moves an existing queue.
@@ -296,124 +298,36 @@ If that consumer delivers nothing for about two seconds while the stream holds a

 Coverage is unit and conformance tests on in-process R1 and R3 servers; there is no end-to-end or multi-host validation of this code, and earlier multi-host evidence predates the final feed.

-### SQS store system instance (not wired)
+### SQS system instance (operator-launched)

-Spinifex can build, install, launch, restart and give storage to a dedicated SQS store microVM, but nothing calls any of it.
-No SQS API, domain service, admin command or daemon start-up path creates a store, so none of this is in production use.
+One system-owned microVM per Region runs both the SQS store and the SQS worker as separate processes, on one encrypted data volume pinned to one node (ADR-0011 Amendment 5, ADR-0012 Amendment 6).
+Only an operator activates it with the super-admin `EnsureSQSWorker` or `EnableSQS` actions; nothing starts it at daemon startup, and platform NATS is neither used nor changed for SQS data.
+This first profile is fresh single-node only: a host cluster of more than one node is refused before anything is recorded or allocated, and there is no disable.

-The guest image is built by `make build-sqs-store-microvm-image`, separate from the load balancer image, and bundles the `nats-server` version pinned in `go.mod` plus the ext4 tools.
-Release tarballs carry it as `microvm-sqs-store/`, and `setup.sh` installs it to `/usr/share/spinifex/microvm/sqs-store/`.
-A release without a complete image removes any previously installed one, so a launch is then refused rather than booting a stale image.
+`make build-sqs-worker-microvm-image` builds the one `sqs-worker` image: the static `sqs-worker` binary, the pinned `nats-server`, the ext4 tools and a network `/init`; releases ship it as `microvm-sqs-worker`, which `setup.sh` installs to `/usr/share/spinifex/microvm/sqs-worker`.
+At boot the guest reads three firmware config values under `opt/spinifex/sqs-worker/` (the data disk's serial, its Region and the gateway URL on this node's mgmt bridge) and refuses before touching any disk if one is missing or malformed.
+It selects exactly one disk with that serial, runs `e2fsck -p`, mounts it at `/data` and confirms from `/proc/self/mountinfo` that the whole disk is what is mounted; the guest never formats.

-At boot the guest reads three firmware config values: the data disk's serial, an IPv4 address on one of its interfaces, and one NATS user NKey public key.
-It mounts exactly one matching disk at `/data` after `e2fsck -p`, refuses a missing, duplicate, unformatted or unrepairable disk, and never formats.
-JetStream keeps its data on `/data` with `sync_interval: always` (ADR-0011:S3), `max_payload: 2MiB` admits a full 1 MiB SQS message with its headers (ADR-0011:S4), and the server listens only on that address and accepts only that key.
-On shutdown the store is stopped before `/data` is unmounted; a store that survives SIGKILL leaves every mount in place.
+The store is a dedicated `nats-server` listening only on `127.0.0.1`, with its JetStream directory at `/data/jetstream` and `GOMEMLIMIT=512MiB` (ADR-0011:S11), so nothing outside the VM can reach it and it needs no network identity of its own.
+The store starts only after the receipt key is settled, and the worker connects only after the store answers JetStream account info, so an unavailable store never appears ready.

-A launch uses the versioned subject `system.LaunchSQSStore.v1`, which a daemon without this code does not serve, so it fails with no responder instead of booting another guest.
-The legacy `system.LaunchInstance` path refuses the store kind on both send and receive.
-Every input check runs before capacity is reserved, and the data volume must be an available system-account volume.
-It is mounted through the normal EBS path and detached, never deleted, on terminate, which is the protection standard EBS gives today and nothing stronger.
-CPU mitigations stay enabled for the store, unlike the load balancer.
-A successful launch means the VM is running, not that NATS is ready.
+The receipt-signing key lives on the data disk under `/data/sqs-worker-key`, written to a temporary file, synced, renamed and its directory synced before its fingerprint is recorded.
+`domains/sqs/workerident` holds the platform's receipt-key state on the worker's identity record, beside its generation, claim, pinned node and volume: `pending` is set only when the record first creates its volume, and `recorded` holds the fingerprint.
+Under `pending` the guest may create a key, or reuse a valid key it already wrote before a crash, only while no `/data/jetstream` directory exists; any store data under `pending` refuses, because it proves the volume is not new.
+Under `recorded` a missing, unreadable or different key refuses the boot and is never regenerated; a record with no receipt-key state is refused as not proven new.
+`RecordReceiptKey` writes the fingerprint by compare-and-swap that re-checks the current instance, generation, claim and volume, and a lost reply is answered by reading the record again, never by a new key.
+A replacement worker reuses the pinned volume and keeps its predecessor's receipt-key state.

-The instance record keeps the store's non-secret guest settings so a restart on the node that launched it rewrites the firmware config files.
-A relaunch on any other node is refused, and cross-node recovery never claims a store.
-Both checks exist only in daemons with this code: an older daemon can still claim and relaunch a store elsewhere, and an explicit start served by another node claims the instance before its relaunch is refused.
+The worker reaches the gateway over TLS 1.3 trusting only the cluster CA, signing with instance-role credentials from the metadata service only.
+The internal `spinifex-sqs` actions are `ClaimReceiptKey`, `RecordReceiptKey`, `PollWork` and `CompleteWork`; the gateway admits only an assumed-role session of `sqsWorkerRole` in the system account and resolves the instance through `workerident.Authorize`, a point-in-time check against the current bound record.
+The worker pulls tenant work from its own node's gateway relay and posts results back, so no SQS payload crosses platform NATS; the relay holds a bounded, in-memory set of waiting requests and answers `ServiceUnavailable` when it is full or no worker is polling.

-`domains/sqs/storevolume` prepares a fresh data volume, because the guest never formats.
-It checks for `mkfs.ext4` and `qemu-img` before allocating, creates the volume in `creating`, writes an empty ext4 image to it over NBD on the local node, unpublishes it, and only then makes it `available`.
-While a volume is `creating`, attach, delete, modify, force-detach and snapshot are refused.
-A volume is deleted only after a confirmed unpublish; any publish error, an unconfirmed unpublish or a failed release keeps it and reports its ID, with its state reported as unknown after a failed release.
+`EnsureSQSWorker` takes a `NodeId`, an `InstanceType` and a `VolumeSizeGiB`; the instance type's memory must exceed the store's 512 MiB limit, and a later request for a different type or size is refused as `IncorrectState`.
+`EnableSQS` records the Region's desired state, pinned to one node, and then ensures the worker; `status` reports it ready only when the worker is bound, polling and the tenant API is registered.
+A store launch record or VM from the earlier separate-store design, or a worker record without a supported launch contract, is refused with its volume kept: enable stops before recording anything, and a relaunch of an old store VM is marked failed rather than started.

-Evidence is local only: unit tests, real QEMU boots against disposable scratch disks, and one run of the preparer against the `qemunbdd` provider whose volume kept acknowledged writes across a reboot.
-Viperblock, volume encryption, the EC2 volume record path, multiple nodes and host reachability of the guest address are not tested.
-Credential issuance and rotation, store placement and NATS readiness checks do not exist, and there is no failover: a lost node leaves its store unavailable.
-
-### SQS worker guest (operator-launched)
-
-The shared SQS worker guest holds the SQS store's client credential, so that no platform component does (ADR-0011:S2).
-Only an operator activates it, with the super-admin `EnsureSQSWorker` action described below; nothing starts it at daemon startup, and no worker service runs after boot.
-
-`make build-sqs-worker-microvm-image` builds it with the store image's script and network `/init` under `SQS_MICROVM_PROFILE=worker`; it carries the static `sqs-worker` binary and the ext4 tools, no `nats-server`, and an empty `/data` mountpoint.
-Releases build it beside the store image and ship it as `microvm-sqs-worker`, which `setup.sh` installs to `/usr/share/spinifex/microvm/sqs-worker` with the store image's rules: a release without it removes an earlier one, and an incomplete one is refused.
-At boot `sqs-worker boot` reads three firmware config values under `opt/spinifex/sqs-worker/`: the worker disk's serial, its Region and the gateway URL; none is a first-store decision, because a static value survives a guest reset.
-A missing, empty or malformed value refuses before any disk is touched.
-
-It selects exactly one disk with that serial, runs `e2fsck -p`, mounts it at `/data` and then confirms from `/proc/self/mountinfo` that this disk, as a whole ext4 filesystem, is what is mounted there; an empty mountpoint, a foreign mount or a failed check or mount refuses, and the guest never formats.
-Only then does it ask the gateway for the current decision with `ClaimFirstStore`, signed by `internal/sqsworkergw` with credentials from the instance metadata service only (`gwsign.NewInstanceRole`, no environment, profile or static-key fallback), over TLS 1.3 that trusts only the cluster CA from `opt/spinifex/ca-cert` and never follows a redirect.
-Under that decision `workerkey.Acquire` creates the key on `/data/sqs-worker-key`, owner-only and synced, records only its public key and then uses it, or loads the recorded key; a missing or different key once one is recorded refuses and is never replaced.
-The VM can start before its launch binds the instance, so its first calls are legitimately `AccessDenied`; that denial creates nothing and is asked again, as are a lost reply and a `ServiceUnavailable` or `Throttling` refusal, all within one bound (60 attempts 5 seconds apart), after which a worker that was never bound, or was replaced, refuses the boot.
-A signature or other refusal, or an answer that is neither "create" nor one valid public key, refuses the boot at once.
-
-`domains/sqs/firststore` is the SQS domain's first-store decision: one record per Region in the `spinifex-sqs-first-store` KV bucket moves from `claimed` to `key-recorded` to `launch-intent` under compare-and-swap, bound to the worker's instance ID, claim ID and generation.
-A claim authorizes key creation only while no public key is recorded; once one is, every claim returns only that key, and a store launch intent needs it recorded first.
-An older generation, a different identity at the same generation or an unreadable record authorizes nothing, and the record holds no private key material.
-Reads are stream message gets answered by the stream leader, never direct gets a lagging replica can answer, and every authorizing answer is also confirmed by a compare-and-swap at the revision it read, so a stale or unavailable read authorizes nothing.
-The authenticated caller that will carry this decision must derive the worker's instance ID and generation from authoritative instance state; a generation supplied in a request cannot itself authorize a takeover.
-
-`domains/sqs/workerident` is that authoritative instance state: one record per Region in the `spinifex-sqs-worker-identity` KV bucket holds a monotonic generation, claim ID, pinned node and encrypted data volume, and moves from `launching` to `bound` to `stopped`, or from `launching` to `failed`, under compare-and-swap.
-A launch intent is refused while the previous attempt is `launching` or `bound`, including one whose outcome is unknown, so only one caller ever holds an intent to start a VM on the pinned volume.
-A replacement takes the next generation only after `FailLaunch` or `RecordStopped` records that nothing holds the volume, and each request names the generation it supersedes, so a replayed request cannot take a generation it never observed; there is no automatic takeover.
-The launch path then raises the first-store fence to the successor's generation with `firststore.Fence` before `BeginLaunch`, which refuses a replacement until a leader read shows that fence; the fence only rises, and `firststore` refuses every call from a generation below it.
-Only the current claim ID is recorded, so claim IDs are minted fresh per attempt and a VM launch must be idempotent on generation and claim together.
-Binding writes the instance's index entry first and then binds the record to that instance.
-The index is only a lookup: `Authorize` accepts an instance only while the Region's record is `bound` to exactly that instance, generation and claim, confirmed by a compare-and-swap at the leader-read revision, so an old, orphaned, rewritten or unbound entry authorizes nothing.
-A same-instance reboot keeps its instance ID and identity; a replacement takes a new generation, and the first cut refuses another node or data volume and an unencrypted volume, although the encryption flag is the caller's claim and the launch wiring must verify it from the volume's metadata.
-Both buckets are opened through `OpenDurableKV`, which refuses an existing bucket whose effective stream configuration could expire, evict, lose, lag or refuse records: non-limits retention, an age, per-message or delete-marker TTL, memory storage, asynchronous flush, a size, count or message-size cap, a discard policy other than discard-new, unacknowledged writes, a sealed stream, a mirror or sources, or fewer replicas than the cluster needs.
-`workerkey.Acquire` drives that decision: it syncs a new key to disk, records its public key and only then returns it, so a crash before recording reuses the key already on disk.
-
-`domains/sqs/awsapi/worker` carries the decision to the worker as `ClaimFirstStore` and `RecordWorkerPublicKey`, served under the internal `spinifex-sqs` signing service, which no AWS SDK signs for; the public AmazonSQS action inventory is untouched.
-The gateway admits only an assumed-role session of `sqsWorkerRole` in the system account, then the role's policy, and only then resolves the session name, the instance ID STS set, through `workerident.Authorize`; the request carries no identity, and any parameter beyond the action's own is refused.
-The first-store identity is the current record's instance, claim and generation, and a worker bound to another Region is refused.
-A request that passed `Authorize` just before a confirmed stop and replacement still carries the old generation; the fence was raised before the successor began, so `firststore` refuses it whether or not the successor has claimed yet.
-The adapter never logs a caller-chosen parameter name, because one could carry a secret.
-`RecordWorkerPublicKey` accepts only a valid public user key, so a seed is refused without being recorded, logged or echoed.
-The gateway binds both buckets per request and never creates them, so it provisions nothing and answers `ServiceUnavailable` until a worker launch has created them; the worker launch below ensures the role.
-The private key stays on the worker disk; the guest logs and publishes only the public key to `/run/sqs-worker/public-key`.
-
-`domains/sqs/workerlaunch` realizes the worker on the node it runs on, its only node: a bound record is returned as is; otherwise it takes the intent, then on the first launch marks the intent's one volume create, prepares the volume and records its ID, reads the volume's own record and refuses it unless encrypted, ensures the role, finds or creates the network for the claim and reads the recorded store key, and marks the intent dispatched before launching.
-A dispatched launch is never sent again: a lost reply or error leaves the intent `launching` and dispatched, and every later call returns `ErrOutcomeUnknown` until the outcome is reconciled, since the launch path is not idempotent.
-A failure before dispatch leaves an undispatched intent that the next call resumes with the same claim; a confirmed stop or failure is replaced only after `firststore.Fence` excludes the old generation, reusing the pinned volume.
-The launch passes no store decision or key: the guest asks the gateway at boot.
-The intent exists before any volume, so concurrent first launches allocate nothing for the losers. A Prepare proven to have sent no create withdraws the mark and returns `ServiceUnavailable`, or `UnsupportedOperation` when the cluster storage key is not configured, and a retry may prepare the volume; a create that may have applied is never sent again, and every later call returns `IncorrectState` until it is reconciled. A volume the call knows of but cannot use is reported with its ID in a `VolumeError` and left in place, never deleted. An intent that has sent no create may be re-typed; one that has keeps its type until replaced. The worker role is created with EC2-only trust, and an existing role is checked for that trust before any policy is granted; the instance profile must carry exactly the worker role, and one holding any other role is refused unchanged.
-The daemon's `LaunchSQSWorkerInstance` boots the `sqs-worker` image through the same direct-boot path as the store, with the data volume attached at boot and kept on terminate, a required instance profile, and a required mgmt NIC; it re-reads the volume's own record and refuses one that is not available and encrypted, before anything is allocated.
-Its firmware config adds the gateway URL on this node's mgmt bridge and the cluster CA from the daemon's own trust store; the caller cannot supply either, and no value written is secret.
-A relaunch rebuilds that config from the VM record on the node that launched it and is refused anywhere else, and membership takeover already skips SQS-managed instances.
-`sqsWorkerRole` is ensured in the system account granting exactly `spinifex-sqs:ClaimFirstStore` and `spinifex-sqs:RecordWorkerPublicKey`, from the `domains/sqs/workerrole` contract the gateway gate also reads; an existing role whose trust admits more than the EC2 service is refused.
-
-Evidence is local only: unit tests with a fake firmware config, sysfs and mount table, tests running the boot through the real client against the real `spinifex-sqs` registration and records, and QEMU boots of the built image against disposable scratch disks and a host-side test gateway and metadata service, showing the first boot creates the key, reboots reuse it, a key lost after recording is never replaced, and the seed reaches no log or request.
-The worker's network is a private SQS system VPC per Region, built by `systemvpc.EnsurePrivate` from the `[sqs]` supernet (default `10.240.0.0/14`): one VPC and one /24 subnet, with no internet gateway, NAT gateway, EIP, route table or daemon-to-store node port. Each launch claim gets one ENI in that subnet, in the VPC default security group, found again by its claim on a retry; two ENIs for one claim are refused. The launch intent is marked before the one create is sent, so a create whose ENI is not yet visible is never sent again: the launch fails closed with `IncorrectState` until the ENI appears or the outcome is reconciled. Worker-to-store TCP and the guest's route to the metadata service are not yet proven by any traffic test.
-The super-admin `EnsureSQSWorker` action on the internal `spinifex` operator service takes a `NodeId`, which must be a node of this cluster and a single subject token, and an `InstanceType`, and sends a request to `sqs.worker.Ensure.v1.<node>` with no queue group, so only that node answers and the worker stays pinned.
-That node's daemon re-checks the node the request names and the instance type, and refuses as `ServiceUnavailable` before allocating a volume or network or recording an intent unless the worker image's kernel and initramfs are non-empty files, its mgmt bridge exists and the cluster CA is readable; a call after the image is installed then launches.
-It then runs its one composed launcher for its own Region, one launch at a time, in the daemon's own context with a 30-minute bound, so a caller's timeout does not cancel it.
-A reply lost to that timeout is reported as `ServiceUnavailable`; calling again reports the bound worker, or `IncorrectState` for a dispatched launch whose outcome is unknown, which is never resent; a Region pinned to another node is refused as `InvalidParameterValue`.
-The reply names the bound VM's node, Region, instance ID and generation; bound is not ready, and it says nothing about the guest's boot, its key or the store.
-Stop/detach evidence for `RecordStopped` is not wired, the guest's metadata route and gateway TLS on a real ENI and worker-to-store TCP are unproven, and Viperblock's at-rest key is held by the platform.
-
-The worker identity record holds the instance type of its generation; a request for another type is refused as `IncorrectState`, before dispatch and after binding, the reply names the recorded type, and only a replacement after a confirmed stop may change it.
-Operators call both actions with `spx admin sqs ensure-worker --node --instance-type` and `spx admin sqs ensure-store --node --instance-type --volume-size-gib`, signed for the `spinifex` service with the standard AWS credential chain like the other `spx admin` gateway commands.
-Each command refuses missing inputs before sending anything and reports only the bound VM; a bound VM is not a ready worker or store.
-The super-admin `EnsureSQSStore` action takes a `NodeId`, an `InstanceType` and a `VolumeSizeGiB` and sends a request to `sqs.store.Ensure.v1.<node>` with no queue group; only the node the Region's worker is bound to can launch the store.
-That node refuses an instance type whose memory does not exceed 512 MiB, the store's `GOMEMLIMIT` under ADR-0011:S11; this is a floor, not a validated size, and no type is defaulted.
-It refuses before recording or allocating anything unless the store image and the cluster storage key are present, the worker is bound on this node, and that worker's identity has recorded its public key in the first-store decision, which it then advances to a launch intent.
-The store's own launch record in `spinifex-sqs-store-launch` holds only that public key, the claim, the node and the operator's instance type and size, and marks each step before its uncertain create: the volume, the ENI in the worker's private subnet, and the launch.
-A later request for a different instance type or size is refused as `IncorrectState`, before dispatch and after binding, and the reply names the type the store was launched with.
-A volume preparation that failed before sending any create, such as a missing `mkfs.ext4` or a scratch or format failure, withdraws its mark and returns `ServiceUnavailable`, so a retry may try again once the host is fixed.
-Any other marked step whose result is not visible is reported as `IncorrectState` and never resent; a volume that is not encrypted by its own record is refused as `UnsupportedOperation` with its ID kept, and the daemon re-checks the volume's encryption and the guest's size at launch.
-The daemon never receives the worker's seed; the store's NATS server accepts only that public key, and the reply's bound VM and address are not readiness, which only the worker can check.
-
-After boot, the worker's long-running `sqs-worker ready` step mounts and verifies its disk again and calls the internal `spinifex-sqs` `DescribeStore` action, which admits only the current worker and returns the bound store's address and recorded public key, or pending.
-It loads its key only if it matches that public key, then connects to the store's NATS on port 4222 with it and asks JetStream for account info, each pass bounded to 30 seconds.
-The step runs until shutdown with no overall deadline, so a store the operator launches at any time is picked up: it asks again every 5 seconds while the store is pending, not answering or the reply was lost, and re-checks a ready store every 30 seconds.
-A pass writes only the address and public key to `/run/sqs-worker/store-ready`; any failed pass removes that file, and so does the step's own exit.
-A start that finds shutdown in progress or no worker key logs once and waits in place rather than exiting, so init does not respawn it in a loop; shutdown stops it like a running watcher.
-Init respawns the step, at most once every 5 seconds, and every start removes a marker left by a watcher that was killed or crashed before the new watcher runs.
-A ready store that fails is noticed up to one re-check interval plus one bounded call later, so the marker can be up to about a minute stale.
-Shutdown blocks further respawns, removes the marker, and stops the watcher by its recorded PID and start time, escalating to SIGKILL after a budget; it unmounts the worker disk only once the watcher is confirmed gone, and a survivor leaves every mount in place.
-A mismatched key, a store that refuses the key, a key missing from the disk or an unclassified gateway refusal is asked again only every 5 minutes, so it never spins and still recovers if the operator fixes it; a missing key is never replaced.
-The NKey authenticates the worker to the store, but nothing authenticates the store to the worker: there is no TLS, so a pass means only that something at that address accepted the key.
-This is store readiness for the worker alone; there is no SQS API, and no production sizing or support is claimed.
+Evidence is local: unit tests, the guest shell test and an AWS SDK test that creates a queue and sends, receives and deletes a message through the real worker loop against a real local JetStream store and a restart.
+A real guest boot, its metadata route on a real ENI and the signed path on a live node are unproven, Viperblock's at-rest key is held by the platform, and SQS-managed message encryption is not implemented, so the profile must not hold tenant production data.

 ## Configuration

@@ -60,7 +60,7 @@ intact; ADR-0001 explicitly defers that design.

 | Current path | Standing | Target ownership | Rule |
 |---|---|---|---|
-| `cmd/aws-model-coverage` | Qualification generator | Thin main; reusable code in `internal/tooling` | Produces the dispatch-versus-model inventory; never imported by product packages. |
+| `cmd/aws-model-coverage` | Qualification generator | Thin main; reusable code in `internal/tooling` | Produces the dispatch-versus-model inventory; never imported by product packages. Debt: it reads each domain's declared `OperationInventory` from `domains/<service>/awsapi` (ACM, EC2, ECR, ECS, EKS, ELBv2, IAM, RDS, Route 53, SQS, STS) and never constructs a registration. |
 | `cmd/dhcptest` | Development diagnostic | Thin main; reusable code in `internal/tooling` if extraction is useful | It is a DHCP probe, not a supported `spx admin` command. |
 | `cmd/spx-loadgen` | Qualification/load tool | Thin main; reusable code in `internal/testkit` | It drives test load and is not shipped to nodes. |
 | `tests/e2e/*/cmd/*`, `tests/bench/*/cmd/*` | Test and benchmark-private tools | Remain test-private | They are invoked by a test or benchmark harness, not a product process. |
@@ -369,8 +369,9 @@ while the branch is in flight; they do not describe the umbrella repository's
 | `spinifex#1355` | `spinifex/gateway/ec2.go` (`EC2_Request` in production, the legacy service switch, its `supportedServices` and JSON tables) and the `Quota` field of `GatewayConfig` | `spinifex/domains/ec2/awsapi` (`NewRegistration`, `OperationInventory`, `ServiceName`), composed by `spinifex/runtime/roles/awsgw` (`ec2.go`: `newEC2Registration`, `advertisedEndpoint`; `ec2quota`) | Slice 9b of the EC2 move: EC2 registers through the ADR-0002 seam as service `ec2` and the legacy gateway switch is empty. The EC2 query-protocol error envelope is kept through `ErrorEnvelopeXML`, and the policy denial match is now `*dispatch.PolicyDenialError`. The characterisation tests committed first also pin the Authorize service name and action, the trace span and audit action, the throttle key (account plus action) and the cluster-unavailable gate, which runs before policy and action lookup for gateway-local and NATS-backed actions alike. They and the 9a tests pass unchanged apart from setup: STS and EC2 test setups now register the services they sign as, and the two tests that claimed a legacy service name are removed because no legacy name remains. `EC2_Request` survives only as a test shim. Allowlist changes are two deletions (the `gateway` to `domains/ec2/awsapi` and `domains/ec2/awsapi/instance` rows) and one addition, the per-service model-coverage row `cmd/aws-model-coverage` to `domains/ec2/awsapi`. `startInstanceStatus` stays in `awsgw.go` because ADR-0007:S6 records the raw instance-state uses by file. No change to `dispatch.Actor` or `dispatch.Invocation`. Live on env14: DescribeRegions, DescribeAvailabilityZones and DescribeAccountAttributes, an idempotent CreateVolume replay and its parameter mismatch, the DryRun and InvalidVolume.NotFound errors, a no-policy user refused with UnauthorizedOperation (including on the gateway-local DescribeRegions), an assumed-role session allowed and refused per its policy, and one RunInstances to running then terminated. The gateway request logs carry `service=ec2` and the per-action label. Not exercised live: a throttled request (the default limits are not reachable from one client; the throttle key is pinned by the characterisation tests) and an agent-signed EC2 call. |
 | `spinifex#1363` | `spinifex/gateway` context keys (`ctxService`, `ctxRegion`, `ctxAction`, the principal and access key id keys, and the unread `ctxQueryArgs`), the duplicate `callerPrincipal` and `principalContext` types, `gateway/caller_arn.go` (`buildCallerARN`), and the direct `policyauth`, `accountadmission` and NATS-connection reads in the auth pipeline | `spinifex/ingress/aws/reqctx` (the request record), `credverify.CallerARN`, and the gateway-consumed capabilities `PolicyEvaluator`, `AccountAdmission` and `Readiness`, composed by `spinifex/runtime/roles/awsgw` (`clusterready`) | Slice 11a of the gateway removal: the request-context record and the ADR-0002:S8 capabilities, with no behaviour change. The record is a copy-on-write value under one context key; the principal is sealed once and a second `WithPrincipal` keeps the first caller. The layering rule `S7_RequestRecordPrivateToIngress` lets only `gateway` and `ingress/aws/*` import `reqctx`, with a fault-injection test, and slice 11c moves the package to `ingress/aws/internal/reqctx` and drops the rule. The caller ARN is built once in `credverify` (root cases kept) and filled on `Principal.CallerARN`. The policy evaluator, account admission (`credential.ErrAccountNotActive`, with `accountadmission.ErrNotActive` an alias) and cluster readiness (`Available() bool`) are interfaces the gateway consumes and fails closed on when nil. The step 1 characterisation tests (throttle against both readiness gates and the policy gate for ECS and IAM, `/admin` skipping `GetService` and the second gate, the nil-connection difference between the two gates, five error writers each minting a request id that differs from the audit record, and the credential-scope region reaching only the span) pass unchanged apart from setup. No allowlist rows change, and there is no change to `dispatch.Actor` or `dispatch.Invocation`. Debt for slice 11c: the test-only `defaultCapabilities` seam (a package variable set by a test `init` so about 60 gateway test files can build a config from `IAMService` alone), the nil-connection case of the early readiness gate, the remaining `NATSConn` readers (`DiscoverActiveNodes` and the field that `awsgw` and the integration harness set), and the `IAMService`, `STSService`, `Config`, `RootCAs` and `InternalSuffix` fields of `GatewayConfig`, which no production code reads. Live on env14: DescribeRegions, DescribeAvailabilityZones and DescribeAccountAttributes, an idempotent CreateVolume replay and its parameter mismatch, the DryRun and InvalidVolume.NotFound errors, a bad access key refused with InvalidClientTokenId on a Query service and InvalidClientTokenIdException on a JSON service, a no-policy user refused with UnauthorizedOperation, and an assumed-role session allowed and refused per its policy. Not exercised live: a suspended-account refusal, a throttled request and the NATS-down 503, which the step 1 tests pin. |
 | `spinifex#1378` | `spinifex/gateway` `oidc_discovery.go` and `registry.go` (the `/oidc/eks` routes and the `/v2` mount), the `DiscoverActiveNodes` method and its cache, and the `GatewayConfig` fields `Config`, `RootCAs`, `InternalSuffix`, `Version`, `Commit`, `IAMService`, `STSService`, `NATSConn` and `ExpectedNodes` | `spinifex/runtime/roles/awsgw` (`PublicRoutes`, `NodeDiscovery.ActiveNodes`, `composeGateway`); the gateway keeps one `PublicRoutes` mount hook and the `Admin` registration name supplied by the role |
-| `spinifex#TBD` | `spinifex/gateway` (`GatewayConfig`, the SigV4 and policy middleware, `Request`, the admin dispatch, `AWSOperationInventory`) | `spinifex/ingress/aws/pipeline` (`Config`) and `spinifex/ingress/aws/dispatch` (`AWSOperationInventory`); the gateway layering rows are retired and R2 binds the pipeline |
+| `spinifex#1384` | `spinifex/gateway` (`GatewayConfig`, the SigV4 and policy middleware, `Request`, the admin dispatch, `AWSOperationInventory`) | `spinifex/ingress/aws/pipeline` (`Config`) and `spinifex/ingress/aws/dispatch` (`AWSOperationInventory`); the gateway layering rows are retired and R2 binds the pipeline |
 | `spinifex#1334` | `spinifex/gateway/oidc_discovery.go` (`eksAccountKV`, the discovery-document and JWKS read and construction) | `spinifex/domains/eks/cluster` (`OIDCPublisher`, `OIDCDiscoveryDocument`), composed by `spinifex/runtime/roles/awsgw` | The EKS domain now reads its own account bucket and builds the per-cluster discovery document and JWKS responses (ADR-0005:S4), and the gateway only maps the unauthenticated `/oidc/eks/...` routes onto an `OIDCPublisher` interface it is handed. Responses are byte-identical, pinned by `oidc_discovery_current_behaviour_test.go`, whose assertions are unchanged. The gateway no longer imports `domains/eks/cluster` in production code. `STS` keeps its own `ClusterJWKSResolver` (`OIDCJWKSReader`), untouched. Recorded debt, not repaired here: a stored JWKS is served verbatim with 200 even when it is not valid JSON. |
+| `spinifex#1245`, `spinifex#1390` | — | `contracts/admission/v1` | The quota override seal route (`admission.v1.SealAccountQuotaOverride`, served by `awsgw`) and the per-gateway read-only capability probe (`admission.v1.AccountQuotaOverrideSealCapability.<node>`) are an explicit versioned cross-process contract between the quota owner and `operator/accountlifecycle`. The consumer wiring is new: `NewClusterEngine` takes a `ClusterNodes` value with separate daemon and gateway node sets, `ClusterConfig.NodesRunning` is the single definition of which nodes run a service, and the seal reaper runs in `StageAccount` after the EC2 settings seal. A termination whose gateway probe is incomplete seals nothing and leaves the account `TERMINATING`. Contract, responder, probe and reaper tests pin the subjects, JSON shapes and fail-closed cases; `TestAccountTermination_QuotaOverrideEndsSealed` replaces the former characterization that overrides survive teardown. |
 The EC2 contract row is intentionally different from the directory moves: it creates
 a compatibility boundary. `ec2.cmd.*` retains its deployed one-token NATS
 subscription and permission shape, while callers use
@@ -488,8 +489,8 @@ Enforcement: the owner decided the layering test is blocking now, with a ratchet
 - `domains/rds/instance` → `domains/rds/parametergroup` (legacy edges): `parameterGroupReferences` binds `Dependants` by scanning instance records; this clears when the RDS instance gains its own owner and answers `Dependants` from it. `tags.go` writes the meta record through its generic tag registry; this clears when tagging invokes a parameter-group-owned tag mutation capability, and moving the tagging adapter to `domains/rds/awsapi` alone does not clear it. The Modify handler still calls `propagateParameterGroup` after the owner's write; propagation is instance-side realization and stays legacy until it is designed as asynchronous realization (group generation, instance staleness, retry). `parameterGroupValues` returns the persisted `Record` and overrides to instance code; when the instance owner is extracted, replace them with a consumer-owned resolved-parameters projection.
 - `domains/ecs` → `domains/ecs/taskdefinition` (owner bypass): `tags.go` reads and writes the revision record through its generic tag registry; this clears when ECS tagging invokes an owner-provided tag mutation capability, and moving `tags.go` to `domains/ecs/awsapi` alone does not clear it. `taskDefinitionResolver` returns the persisted `Record` to task and service launch code; when the task and service owners are extracted, replace the record crossing that boundary with a consumer-owned resolved task-definition (launch-spec) projection. Moving packages has not made the record private.
 - `domains/ecs/taskdefinition` (deliberate duplication, not debt): private `getJSON`, `putJSON` and `keysWithPrefix` copy the raw-KV helpers in `domains/ecs` so the extraction keeps ECS's get/put/list error and recovery semantics. Revisit only if a later, tested state abstraction proves it preserves those semantics; do not migrate ECS to `kvstore` merely to remove them.
- `domains/eks/cluster` → `domains/eks/access` (legacy edges): `ResolveTokenReview` opens the EKS bucket in the gateway process and reads records through the exported `access.Get`; this clears when token review asks the owner across a contract rather than opening the bucket itself. `DeleteClusterPrefix` erases access-entry keys as part of its blanket cluster sweep; this clears when cluster delete invokes an owner-provided removal of the cluster's entries. The owner returns the persisted `Record` to the AWS adapter and token review; replace it with consumer-owned projections when those consumers are extracted.
- `domains/eks/cluster` → `domains/eks/addon` (legacy edges): `DeleteClusterPrefix` erases add-on records and manifests in its blanket cluster sweep; this clears when cluster delete invokes an add-on-owned teardown capability. `stageGPUDeviceAddon` is a temporary node-group-to-add-on orchestration edge: best-effort, errors logged and swallowed, never undone on node-group delete; it clears when the node-group owner is extracted and the dependency is designed. `ClusterReconciler` hosts the add-on status subscription; it clears when add-on delivery observation has its own host under the lifecycle correction. Cluster existence is checked in the handler, not through an owner-declared capability; it moves into the owner when cluster admission is defined (today status is ignored). Guest wire types (`AddonStatusReport`, `AddonDeliveryPhase`, `AddonStatusSubject`, `StagedAddonManifest`, `InternalAddonsResponse`) are owned by `contracts/eks/v1`, which `domains/eks/cluster`, `gateway/eks` and `agents/eks/gatewayfetch` import directly.
+- `domains/eks/cluster` → `domains/eks/access` (legacy edges): `ResolveTokenReview` opens the EKS bucket in the gateway process and reads records through the exported `access.Get`; this clears when token review asks the owner across a contract rather than opening the bucket itself. `DeleteClusterChildren` erases access-entry keys as part of its blanket cluster sweep; this clears when cluster delete invokes an owner-provided removal of the cluster's entries. The owner returns the persisted `Record` to the AWS adapter and token review; replace it with consumer-owned projections when those consumers are extracted.
+- `domains/eks/cluster` → `domains/eks/addon` (legacy edges): `DeleteClusterChildren` erases add-on records and manifests in its blanket cluster sweep; this clears when cluster delete invokes an add-on-owned teardown capability. `stageGPUDeviceAddon` is a temporary node-group-to-add-on orchestration edge: best-effort, errors logged and swallowed, never undone on node-group delete; it clears when the node-group owner is extracted and the dependency is designed. `ClusterReconciler` hosts the add-on status subscription; it clears when add-on delivery observation has its own host under the lifecycle correction. Cluster existence is checked in the handler, not through an owner-declared capability; it moves into the owner when cluster admission is defined (today status is ignored). Guest wire types (`AddonStatusReport`, `AddonDeliveryPhase`, `AddonStatusSubject`, `StagedAddonManifest`, `InternalAddonsResponse`) are owned by `contracts/eks/v1`, which `domains/eks/cluster`, `gateway/eks` and `agents/eks/gatewayfetch` import directly.
 - `domains/eks/cluster` and `domains/eks/awsapi` → `domains/eks/addon` (further edges): the AWS adapter reads the persisted `Record` through `addon.Get`/`addon.List`; this clears when the owner returns a view and the add-on adapter moves to `domains/eks/awsapi`. The guest route reaches the owner through the `handlers_eks` NATS RPC (`ListStagedAddonManifests`); this clears with per-resource commands behind EKS registration. The add-on ARN is derived in both the owner and `domains/eks/awsapi/authz.go`; this clears when authorization obtains the ARN from the owner, which an incarnation suffix would require. The owner's `Manifest`/`Report` beside the `contracts/eks/v1` types is deliberate duplication, not debt.
 - `agents/rds/agent` (dependency, narrowed): engine metadata now arrives through the agent-declared `EngineCatalog`, but the agent still imports `domains/rds/instance` for the command protocol, wire and record types (`Command*`, `Parameter`, `EngineHealth*`, `BootstrapMode*`) and API output types. This clears when those move behind a guest/controller wire contract.
 - `domains/rds/awsapi` → `domains/rds/instance` (`InstanceSizing`): the gateway obtains its `Sizing` through the RDS handler package's adapter rather than its own composition root. This clears when the gateway's RDS composition moves to `runtime/roles/awsgw` and builds `Sizing` there.
@@ -688,7 +689,7 @@ series above; "needs decision" is open.
 | `handlers/rds` | — | Done. The remaining RDS instance resource moved to `domains/rds/instance` (package `instance`) in `spinifex#1251` with no API, NATS subject, storage or behaviour change, fulfilling ADR-0001's organise-by-resource condition alongside the already-extracted subnet group, parameter group and engine catalogue. The ADR-0004 blocked-dependency imports (quota admission, the duplicated expected-node fan-out, the IAM ensurer and the Bluebottle master key) remain dependency-rule debt, with their allowlist entries now sourced from `domains/rds/instance`. |
 | `handlers/elbv2` | — | Done. The implementation moved to `domains/elbv2` (package `elbv2`) with no API, NATS subject, storage or behaviour change. Q-30's data-plane boundary remains deferred debt; the amended ADR-0001 row makes it not a relocation prerequisite. Its cross-domain, IAM and `agents/elbv2` imports remain dependency-rule debt, with their allowlist entries now sourced from `domains/elbv2`. |
 | `handlers/iam/mock` | — | Done. Moved with the implementation to `domains/iam/mock`. |
-| `gateway` (root) | Generic HTTP route assembly, SigV4 authentication, the policy gate, request-audit middleware and request-context keys. | Done and deleted. The request pipeline moved into `ingress/aws/pipeline` (as `pipeline.Config`) and the operation inventory into `ingress/aws/dispatch` in `spinifex#TBD`; the `/oidc`, `/v2` and node-discovery code had already moved to `runtime/roles/awsgw` in `spinifex#1378`. The layering test binds the pipeline with R2 and `gateway` is a retired legacy root. |
+| `gateway` (root) | Generic HTTP route assembly, SigV4 authentication, the policy gate, request-audit middleware and request-context keys. | Done and deleted. The request pipeline moved into `ingress/aws/pipeline` (as `pipeline.Config`) and the operation inventory into `ingress/aws/dispatch` in `spinifex#1384`; the `/oidc`, `/v2` and node-discovery code had already moved to `runtime/roles/awsgw` in `spinifex#1378`. The layering test binds the pipeline with R2 and `gateway` is a retired legacy root. |
 | `gateway/bedrock` | — | Done. The whole package moved into `domains/ochre` in `spinifex#1266` with its package-private core and tests intact and no behaviour change. Separating the action adapters into `domains/ochre/awsapi` remains deferred, recorded as `domains/ochre` (action adapters) debt above. The action adapters then moved to `domains/ochre/awsapi` behind one ADR-0002 registration in `spinifex#1351`. |
 | `gateway/{iam,sts,eks,rds}` | — | Done. The action adapters moved to `domains/{iam,sts,eks,rds}/awsapi` in `spinifex#1265` with no API, NATS subject, storage or behaviour change. ADR-0001 does not make ADR-0002 registration a prerequisite for relocating a domain's adapters, as the EC2 adapter moves already showed; the root `gateway` dispatch that still imports them is recorded `gateway` (service registration) debt. |
 | `gateway/{ecs,elbv2}` | — | Done. The action adapters moved to `domains/{ecs,elbv2}/awsapi` in `spinifex#1273` with no API, NATS subject, storage or behaviour change; the root `gateway` dispatch that still imports them is recorded `gateway` (service registration) debt. |
@@ -236,12 +236,12 @@ Each stage runs to empty before the next begins, so nothing is deleted while som
 A large account takes minutes to tear down. The API does not hold the connection open for it: `DeleteAccount` starts the job, returns a `deletionId` immediately, and progress is read from `DescribeAccountDeletion`.

 ```bash
-AWS_PROFILE=spinifex-operator spx admin account delete 000000000042 --remote \
+AWS_PROFILE=spinifex-operator spx admin account delete 000000000042 \
  --name customer@example.com --yes \
  --endpoint https://node1.example.com:9999 --region us-west-1
 ```

-`spx --remote` starts the job and then polls, printing each stage as it lands. Your own client can do the same, or start the job and check back later — the deletion record outlives the account, so `DescribeAccountDeletion` stays answerable after the account is gone.
+`spx admin account delete` always goes through the admin API, so it signs with the standard AWS credential chain. On a cluster node that means `AWS_PROFILE=spinifex` (there is no default profile, so without it the command cannot find credentials), or equivalent credentials for an IAM user holding `spinifex:DeleteAccount` and `spinifex:DescribeAccountDeletion`, and a reachable gateway. The older `--remote` flag is accepted and ignored. The command starts the job and then polls, printing each stage as it lands. Your own client can do the same, or start the job and check back later — the deletion record outlives the account, so `DescribeAccountDeletion` stays answerable after the account is gone.

 One job runs per account. A second `DeleteAccount` with a different token while one is running returns `OperationInProgress`; with the *same* token it returns the running job, which is what makes a retry safe.

@@ -58,7 +58,7 @@ Authentication uses **EKS access entries** (the API authentication mode) — IAM
 **Spinifex specifics**

 - **Access entries are the only auth path.** `authentication_mode` accepts `API` (the default when unset) and `API_AND_CONFIG_MAP`; bare `CONFIG_MAP` is rejected. Spinifex has no `aws-auth` ConfigMap path, so `API_AND_CONFIG_MAP` behaves as `API`: grant access with access entries.
- **Security groups are auto-managed.** Spinifex creates the cluster and node-group security groups deterministically (e.g. `eks-cluster-<name>-nodegroup-sg`); `vpc_config.security_group_ids` is ignored. To expose a workload you add an ingress rule to the auto-managed node-group SG.
+- **The cluster security group is auto-managed.** As on EKS, Spinifex creates one `eks-cluster-sg-<name>-<id>` in your VPC, tagged `kubernetes.io/cluster/<name>=owned` and `aws:eks:cluster-name`, allowing all traffic from itself. It is attached to the workers and the private-endpoint network interface, and `DescribeCluster` reports it as `clusterSecurityGroupId`. A launch template that names security groups replaces it on the workers, as on AWS. Clusters created before this change keep their `-control-plane-sg` and `-nodegroup-sg` until you recreate them. Deleting a cluster deletes the cluster security group and removes only the rules on your other security groups that reference it. `vpc_config.security_group_ids` is ignored.
 - **The worker image is fixed.** Node groups always boot Spinifex's `eks-node` image. `ami_type` is recorded but does not select the image, and the image must be registered on the cluster before you create a node group.
 - **Workers launch with IMDSv2 required and an IMDS hop limit of 2.**
  This matches AWS managed node groups and lets a pod without IRSA reach the node role through the SDK default credential chain.
@@ -96,6 +96,72 @@ Limits:
  The load balancer VM reports target health through its own host's gateway, so while that gateway is down the unhealthy control plane stays in rotation.
 - Not measured: more than one control plane per host, a physical-node failure instead of a stopped gateway, and the target health state as `DescribeTargetHealth` reports it, because the load balancer and control-plane VMs belong to the system account.

+### IAM roles for service accounts in your workloads
+
+A pod gets an IAM role by running under a ServiceAccount that carries the role ARN in an annotation.
+No pod-spec wiring is needed.
+
+```yaml
+apiVersion: v1
+kind: ServiceAccount
+metadata:
+  name: app
+  annotations:
+    eks.amazonaws.com/role-arn: arn:aws:iam::123456789012:role/app-role
+```
+
+Every control plane runs a mutating webhook that acts on pod creation.
+When the pod's ServiceAccount has the annotation, the webhook adds these to each container and init container:
+
+- `AWS_ROLE_ARN` and `AWS_WEB_IDENTITY_TOKEN_FILE`, with a projected token volume named `aws-iam-token` mounted at `/var/run/secrets/eks.amazonaws.com/serviceaccount`.
+- `AWS_REGION`, `AWS_DEFAULT_REGION` and `AWS_STS_REGIONAL_ENDPOINTS`.
+- `AWS_ENDPOINT_URL_STS`, so the SDK calls the Spinifex STS endpoint.
+- `AWS_CA_BUNDLE`, pointing at `/etc/spinifex/ca-bundle/ca.pem`, a ConfigMap named `spinifex-ca-bundle` that is created in every namespace and holds the system roots plus the gateway CA.
+- `SSL_CERT_FILE` with the same path, only when the container does not set it. Clients such as the AWS Load Balancer Controller read this variable instead of `AWS_CA_BUNDLE`.
+
+Your workload still sets `AWS_ENDPOINT_URL` (or the per-service variable) for every service other than STS.
+
+The token audience defaults to `sts.amazonaws.com`.
+The expiry defaults to a value between 79200 and 86400 seconds and is clamped to 600 to 86400 seconds.
+Set them with the `eks.amazonaws.com/audience` and `eks.amazonaws.com/token-expiration` annotations on the ServiceAccount.
+The `token-expiration` annotation on the pod wins over the ServiceAccount.
+
+A container is left alone when it already has the role, region and regional STS settings, which is how the managed add-ons and your own explicit wiring are skipped.
+When some are present, the missing ones and the mount are added.
+Containers named in the pod's `eks.amazonaws.com/skip-containers` annotation (comma separated) are not touched.
+A pod whose ServiceAccount has no role annotation is not changed.
+
+The webhook fails open: if it cannot decide, the pod is created without injection.
+Each injection, skip and failure is logged with the namespace, pod and reason in the control-plane console log.
+Because it fails open, a pod that needs a role but starts without credentials is the symptom to look for, and the console log says why.
+
+The webhook listens on `127.0.0.1:8445` of each control plane, and each API server calls its own.
+Port 8444 is a different listener, the readiness check described above.
+
+Existing clusters get the webhook only when their control planes are replaced by the current `eks-node` image.
+No user-facing operation replaces a control plane, and automatic replacement does not currently work when the replacement reuses the failed member's IP.
+Recreate the cluster to adopt it.
+User-workload injection through a cluster created with an older image is not measured.
+
+### Add-on versions
+
+Add-on versions use the AWS form `v<x.y.z>-eksbuild.<n>`, for example `v1.40.1-eksbuild.1`.
+`DescribeAddonVersions` returns versions in this form, newest first, and every other add-on response reports the version the same way.
+Spinifex ships one EKS build of each upstream version, so `<n>` is `1` today.
+
+`DescribeAddonVersions` honours `addonName`, `kubernetesVersion`, `types`, `publishers`, `owners`, `maxResults` and `nextToken`.
+Each version carries `compatibilities` with the `clusterVersion`, the `platformVersions`, and a `defaultVersion` flag that is true for exactly one version per add-on and cluster version.
+A `kubernetesVersion` that Spinifex does not run returns no add-ons.
+
+`CreateAddon` and `UpdateAddon` accept the AWS form of a version in the catalogue.
+They also accept the bare upstream form, such as `1.40.1`, and Spinifex logs each use at Info.
+This is a Spinifex leniency that AWS does not offer, and it exists so that older scripts keep working.
+Any other value is refused with a message that names `addonVersion` and lists the supported versions, for example `v1.40.1-eksbuild.2`, `1.40` or `v1.40.1`.
+
+In Terraform, pass the AWS form to `aws_eks_addon.addon_version`, either as a literal or from `data.aws_eks_addon_version`.
+The AWS provider validates the format before it calls the API, so it rejects the bare form.
+Leaving `addon_version` unset still installs the catalogue default.
+
 ## Prerequisites

 > [!IMPORTANT]
@@ -178,12 +244,11 @@ The list shows the node group workers only. The control-plane VMs never appear a

 ### 4. Expose a workload

-Workers join an auto-managed security group that only allows intra-cluster traffic. To reach a NodePort service, add one ingress rule to that SG (look it up by its deterministic name):
+Workers join the cluster security group, which only allows intra-cluster traffic. To reach a NodePort service, add one ingress rule to it (`DescribeCluster` reports its id):

 ```bash
-SG=$(aws ec2 describe-security-groups \
-  --filters Name=group-name,Values=eks-cluster-demo-nodegroup-sg \
-  --query 'SecurityGroups[0].GroupId' --output text)
+SG=$(aws eks describe-cluster --name demo \
+  --query 'cluster.resourcesVpcConfig.clusterSecurityGroupId' --output text)

 aws ec2 authorize-security-group-ingress \
  --group-id "$SG" --protocol tcp --port 30080 --cidr 0.0.0.0/0
@@ -225,7 +290,7 @@ The form notes that the authentication mode is API and that the legacy aws-auth

 - The **Access** tab lists IAM principals and their cluster-access policies — add an access entry to let another user or role reach the cluster.
 - The **Addons** tab installs and shows the health of managed addons.
- The **Networking** tab shows the cluster's VPC, subnets, endpoint access, and the auto-managed security groups.
+- The **Networking** tab shows the cluster's VPC, subnets, endpoint access, and the auto-managed cluster security group.

 To run `kubectl` against the cluster, follow the AWS CLI tab's `update-kubeconfig` step.

@@ -308,11 +373,11 @@ Control plane readiness and kubectl authentication above lists which clusters re
 - Confirm the workers can pull images — a default route to an Internet Gateway (public) or NAT Gateway (private).
 - Confirm the node role has `AmazonEKSWorkerNodePolicy`, `AmazonEKS_CNI_Policy`, and `AmazonEC2ContainerRegistryReadOnly`.

-**Workload unreachable from outside the cluster.** The node-group security group only allows intra-cluster traffic by default. Add an ingress rule for your NodePort to `eks-cluster-<name>-nodegroup-sg`, as shown in the CLI tab.
+**Workload unreachable from outside the cluster.** The cluster security group only allows intra-cluster traffic by default. Add an ingress rule for your NodePort to it, as shown in the CLI tab. An Ingress served by the AWS Load Balancer Controller needs no rule: the controller adds it to the cluster security group.

 **`create-cluster` rejected with an authentication-mode error.** `CONFIG_MAP` is the one mode Spinifex rejects, because it has no `aws-auth` ConfigMap path. Set `authentication_mode` to `API` or `API_AND_CONFIG_MAP`, or leave it unset, and grant access with access entries.

-**Addon install fails.** Only addons bundled into the registered `eks-node` image install successfully. If you pin an addon version in Terraform, leave `addon_version` unset so Spinifex selects the catalog default.
+**Addon install fails.** Only addons bundled into the registered `eks-node` image install successfully. If you pin an addon version in Terraform, use the AWS form from `DescribeAddonVersions` (see Add-on versions above), or leave `addon_version` unset so Spinifex selects the catalog default.

 ## Reference Architectures

@@ -17,7 +17,7 @@ tags:

 ## Overview

-The platform serves **405 operations** across the AWS APIs below. Every page names the operations its service implements from the pinned model, generated from the dispatch tables on each build rather than written by hand.
+The platform serves **406 operations** across the AWS APIs below. Every page names the operations its service implements from the pinned model, generated from the dispatch tables on each build rather than written by hand.

 | Service | Operations |
 |---|---:|
@@ -27,9 +27,9 @@ The platform serves **405 operations** across the AWS APIs below. Every page nam
 | [ECS](/coverage/ecs) | 31 |
 | [EKS](/coverage/eks) | 27 |
 | [ELBv2](/coverage/elbv2) | 36 |
-| [IAM](/coverage/iam) | 81 |
+| [IAM](/coverage/iam) | 82 |
 | [RDS](/coverage/rds) | 27 |
 | [Route 53](/coverage/route53) | 13 |
 | [S3](/coverage/s3) | 26 |
 | [STS](/coverage/sts) | 5 |
-| **Total** | **405** |
+| **Total** | **406** |
@@ -2938,7 +2938,8 @@
          "UpdateAccessKey",
          "UpdateAssumeRolePolicy",
          "UpdateRole",
-          "UpdateRoleDescription"
+          "UpdateRoleDescription",
+          "UpdateUser"
        ],
        "implemented": [
          "AddRoleToInstanceProfile",
@@ -3021,7 +3022,8 @@
          "UpdateAccessKey",
          "UpdateAssumeRolePolicy",
          "UpdateRole",
-          "UpdateRoleDescription"
+          "UpdateRoleDescription",
+          "UpdateUser"
        ],
        "stubbed": [],
        "unsupported": [],
@@ -3121,7 +3123,6 @@
          "UpdateServerCertificate",
          "UpdateServiceSpecificCredential",
          "UpdateSigningCertificate",
-          "UpdateUser",
          "UploadSSHPublicKey",
          "UploadServerCertificate",
          "UploadSigningCertificate"
@@ -3978,6 +3979,86 @@
        "extra": []
      }
    },
+    {
+      "service": "sqs",
+      "api_version": "2012-11-05",
+      "opaque": false,
+      "note": "",
+      "operations": {
+        "modelled": [
+          "AddPermission",
+          "CancelMessageMoveTask",
+          "ChangeMessageVisibility",
+          "ChangeMessageVisibilityBatch",
+          "CreateQueue",
+          "DeleteMessage",
+          "DeleteMessageBatch",
+          "DeleteQueue",
+          "GetQueueAttributes",
+          "GetQueueUrl",
+          "ListDeadLetterSourceQueues",
+          "ListMessageMoveTasks",
+          "ListQueueTags",
+          "ListQueues",
+          "PurgeQueue",
+          "ReceiveMessage",
+          "RemovePermission",
+          "SendMessage",
+          "SendMessageBatch",
+          "SetQueueAttributes",
+          "StartMessageMoveTask",
+          "TagQueue",
+          "UntagQueue"
+        ],
+        "registered": [
+          "ChangeMessageVisibility",
+          "ChangeMessageVisibilityBatch",
+          "CreateQueue",
+          "DeleteMessage",
+          "DeleteMessageBatch",
+          "GetQueueAttributes",
+          "GetQueueUrl",
+          "ListQueueTags",
+          "ListQueues",
+          "ReceiveMessage",
+          "SendMessage",
+          "SendMessageBatch",
+          "SetQueueAttributes",
+          "TagQueue",
+          "UntagQueue"
+        ],
+        "implemented": [
+          "ChangeMessageVisibility",
+          "ChangeMessageVisibilityBatch",
+          "CreateQueue",
+          "DeleteMessage",
+          "DeleteMessageBatch",
+          "GetQueueAttributes",
+          "GetQueueUrl",
+          "ListQueueTags",
+          "ListQueues",
+          "ReceiveMessage",
+          "SendMessage",
+          "SendMessageBatch",
+          "SetQueueAttributes",
+          "TagQueue",
+          "UntagQueue"
+        ],
+        "stubbed": [],
+        "unsupported": [],
+        "missing": [
+          "AddPermission",
+          "CancelMessageMoveTask",
+          "DeleteQueue",
+          "ListDeadLetterSourceQueues",
+          "ListMessageMoveTasks",
+          "PurgeQueue",
+          "RemovePermission",
+          "StartMessageMoveTask"
+        ],
+        "extra": []
+      }
+    },
    {
      "service": "sts",
      "api_version": "2011-06-15",
@@ -18,7 +18,7 @@ tags:

 ## Overview

-Spinifex implements **81 operations** in the IAM `2010-05-08` API model.
+Spinifex implements **82 operations** in the IAM `2010-05-08` API model.

 ### Scope

@@ -109,3 +109,4 @@ All IAM operations are account-scoped. The root user of an account bypasses poli
 | `UpdateAssumeRolePolicy` |
 | `UpdateRole` |
 | `UpdateRoleDescription` |
+| `UpdateUser` |
@@ -987,7 +987,7 @@ spot-price: On owned hardware there is no spot-to-on-demand price differential,

 ## iam

-`2010-05-08` — 81 of 180 modelled operations implemented (45.0%), 0 stubbed, 19 not applicable, 0 registered outside the pinned model.
+`2010-05-08` — 82 of 180 modelled operations implemented (45.6%), 0 stubbed, 19 not applicable, 0 registered outside the pinned model.

 ### Not implemented

@@ -1069,7 +1069,6 @@ spot-price: On owned hardware there is no spot-to-on-demand price differential,
 - `UpdateOpenIDConnectProviderThumbprint`
 - `UpdateSAMLProvider`
 - `UpdateServerCertificate`
- `UpdateUser`
 - `UploadServerCertificate`

 ### Not applicable — do not build
@@ -1441,6 +1440,21 @@ request-payment: Requester Pays shifts transfer charges to the caller, which nee
 - `GetBucketRequestPayment`
 - `PutBucketRequestPayment`

+## sqs
+
+`2012-11-05` — 15 of 23 modelled operations implemented (65.2%), 0 stubbed, 0 not applicable, 0 registered outside the pinned model.
+
+### Not implemented
+
+- `AddPermission`
+- `CancelMessageMoveTask`
+- `DeleteQueue`
+- `ListDeadLetterSourceQueues`
+- `ListMessageMoveTasks`
+- `PurgeQueue`
+- `RemovePermission`
+- `StartMessageMoveTask`
+
 ## sts

 `2011-06-15` — 5 of 11 modelled operations implemented (45.5%), 0 stubbed, 0 not applicable, 0 registered outside the pinned model.
@@ -1,7 +1,7 @@
 ---
 title: "Oracle Linux Hosts"
 seoTitle: "Install Spinifex on Oracle Linux 10 — Spinifex Docs"
-description: "Prepare Oracle Linux 10 hosts for Spinifex: UEK R8, SELinux, a NetworkManager br-wan bridge, the NFV SIG repository, IPsec on libreswan and host updates."
+description: "Install Spinifex on Oracle Linux 10: prepare the host with UEK R8, SELinux permissive and a NetworkManager br-wan bridge, then run the single-node install."
 category: "Install"
 tags:
  - install
@@ -17,55 +17,42 @@ resources:

 # Oracle Linux Hosts

-> What differs when the Spinifex host runs Oracle Linux 10 instead of Debian or Ubuntu.
+> Install Spinifex on Oracle Linux 10.

 ## Table of Contents

 - [Overview](#overview)
- [Prerequisites](#prerequisites)
 - [Instructions](#instructions)
+- [What the Installer Changes](#what-the-installer-changes)
 - [Troubleshooting](#troubleshooting)

 ---

 ## Overview

-Spinifex installs on Oracle Linux 10.2 or a later 10.x release, running the UEK R8 kernel, on x86_64.
-The installer, the release tarball and the install steps are the same as on Debian; this page covers what the host needs first and what the installer changes on it.
-Follow it alongside [Single-Node Install](/docs/install).
+Spinifex runs on Oracle Linux 10.2, on the UEK R8 kernel, on x86_64 with an x86-64-v3 CPU (Intel Haswell, AMD Excavator or later).

-**Supported on Oracle Linux:**
+The host must be a dedicated Spinifex host: Spinifex claims every `tap*` device and owns libreswan's `ipsec.service`.

- A single node, or a cluster in which every node runs Oracle Linux; servers beyond the third join as compute nodes, as on Debian.
- Guests, VPCs, routed NAT, Elastic IPs, and Application and Network Load Balancers.
- Whole-GPU passthrough, on x86_64 GPU hosts (see [GPU Passthrough](/docs/gpu-passthrough)).
+**Not supported yet:** clusters mixing Oracle Linux with Debian or Ubuntu, SELinux enforcing, aarch64, Oracle Linux 9 or earlier, the Red Hat Compatible Kernel, a ZFS root, NVIDIA MIG and air-gapped installs.

-**Not supported on Oracle Linux yet:**
+## Instructions

- Clusters that mix Oracle Linux with Debian or Ubuntu nodes, on any release.
- SELinux enforcing, aarch64, Oracle Linux 9 or earlier, the Red Hat Compatible Kernel, a ZFS root, and NVIDIA MIG.
+Work through both tabs in order.

-Open vSwitch and OVN come from the CentOS NFV SIG, because Oracle ships neither for Oracle Linux 10.
-The NFV SIG is a community repository, and Oracle does not support the packages it publishes.
-Spinifex installs `openvswitch3.5` and `ovn25.03`, the versions Debian 13 ships, and its repository file lets the SIG supply nothing else.
-QEMU comes from Oracle's KVM Utilities repository, because only that build has the `microvm` machine type Spinifex load balancers boot on.
+:::tabs
+@tab 1. Prepare Oracle Linux

-An Oracle Linux node must be a dedicated Spinifex host.
-Spinifex claims every `tap*` device and owns libreswan's `ipsec.service`, which it keeps stopped on a single node and runs for its own tunnels on a cluster node, so VPNs, hand-run QEMU guests and libreswan tunnels of your own cannot share the host.
-Crypto policies other than `DEFAULT` and `FIPS` are untested, and so is Secure Boot.
+### Update and check the host

-## Prerequisites
-
-### Operating system and kernel
-
-Install Oracle Linux 10 with Oracle's ISO or a KVM template, then update it to the current minor release and reboot:
+Update to the current minor release and reboot:

 ```bash
 sudo dnf update -y
 sudo reboot
 ```

-After the reboot, check the release, the kernel and the CPU:
+Then check the release, the running kernel and the CPU:

 ```bash
 cat /etc/oracle-release        # Oracle Linux Server release 10.2 or later
@@ -73,17 +60,7 @@ uname -r                       # 6.12.0-...el10uek.x86_64
 grep -c -w avx2 /proc/cpuinfo  # non-zero
 ```

-The kernel must be UEK R8 (`6.12.0-*uek`), and it must be the running kernel: an installed but not booted `kernel-uek` is refused.
-The CPU must be x86-64-v3 or newer (Intel Haswell, AMD Excavator or later), since the NFV SIG and KVM Utilities builds target it.
-The installer names any missing CPU flag.
-
-The installer adds `kernel-uek-modules` for the running kernel, which carries `openvswitch.ko` and pulls in `linux-firmware`, about 800 MB.
-Oracle publishes modules for each kernel separately, so a host running an old kernel may find its modules gone from the repository; the installer then stops and asks for a kernel update and a reboot.
-
-### SELinux
-
-Spinifex ships no SELinux policy yet, so SELinux must be permissive or disabled, both now and at the next boot.
-Oracle's images boot enforcing.
+### Set SELinux permissive

 ```bash
 sudo setenforce 0
@@ -91,14 +68,9 @@ sudo sed -i 's/^SELINUX=.*/SELINUX=permissive/' /etc/selinux/config
 getenforce                     # Permissive
 ```

-The installer and `spx admin preflight` refuse a host that is enforcing in either place.
+### Create the br-wan bridge

-### WAN bridge with NetworkManager
-
-As on every OS, the host's uplink must already sit on a Linux bridge named `br-wan`, with the host address and default route on the bridge (see [Single-Node Install](/docs/install#prerequisites)).
-On Oracle Linux, NetworkManager owns `br-wan`; Spinifex leaves the bridge and its NIC to it and marks the links it creates itself as unmanaged.
-
-Create the bridge from the console, not over SSH on the NIC being moved, since the host loses its address while the profile changes.
+Run this from the console, not over SSH on the NIC being moved.
 Replace `eno1` with the uplink NIC and `"System eno1"` with its current profile (`nmcli connection show`):

 ```bash
@@ -109,118 +81,88 @@ sudo nmcli connection down "System eno1"
 sudo nmcli connection up br-wan
 ```

-For a static address, give the bridge `ipv4.method manual` with `ipv4.addresses`, `ipv4.gateway` and `ipv4.dns` instead of `ipv4.method auto`.
-
-`bridge.stp no` matters: a NetworkManager bridge runs STP by default, which holds the port in its forwarding delay for about 15 seconds after every link up, long enough for DHCP on the bridge to time out.
-Then verify, as on Debian:
+For a static address, use `ipv4.method manual` with `ipv4.addresses`, `ipv4.gateway` and `ipv4.dns`.
+Keep `bridge.stp no`, or DHCP on the bridge can time out.

 ```bash
 ip -br link show br-wan        # UP
 ip route                       # default via ... dev br-wan
 ```

-### Network access during install
+@tab 2. Install Single Node

-Besides `install.mulgadc.com` and the AWS CLI download, the install reaches `yum.oracle.com` for the Oracle Linux repositories, `mirror.stream.centos.org` for the NFV SIG packages, and `www.centos.org` for the NFV SIG signing key.
-An air-gapped install on Oracle Linux is not documented yet.
-
-## Instructions
-
-## Step 1. Install Spinifex
-
-Run the installer as a user with `sudo`, exactly as on Debian:
+### Install Spinifex

 ```bash
 curl -fsSL https://install.mulgadc.com | bash
 ```

-For a single node, continue with [Single-Node Install](/docs/install) from Step 2: `setup-ovn.sh --management`, `spx admin init` and `systemctl start spinifex.target`.
-IPsec stays off on a single node, because there is no peer to encrypt traffic to.
+The installer stops and masks firewalld; see [firewalld](#firewalld) before relying on any rule it held.

-For a cluster, prepare and install every node this way, then continue with [Multi-Node Install](/docs/install-multi-node) from Step 2.
-
-### Clusters
-
-Every node in an Oracle Linux cluster runs Oracle Linux.
-`install-node.sh` refuses a host list that mixes Oracle Linux with Debian or Ubuntu, but `spx admin join` does not check, so keep to one OS yourself when joining by hand.
-
-IPsec encrypts the Geneve tunnels between nodes and is on by default, as on Debian.
-On Oracle Linux it runs on libreswan, the IKE daemon Oracle ships, rather than strongSwan.
-Once the node's IPsec certificate exists, the daemon unmasks and starts `ipsec.service` and `openvswitch-ipsec.service`.
-`ovs-monitor-ipsec` then loads the node certificate, its key and the cluster CA into libreswan's NSS database at `/var/lib/ipsec/nss`, under nicknames that start with `ovs_cert`.
-
-The tunnels are plain ESP (IP protocol 50), as on Debian, negotiated over UDP 500.
-Open UDP 500, UDP 4500 and ESP between nodes: libreswan moves to UDP 4500 if it detects NAT between two nodes.
-The Spinifex host firewall opens all three to cluster peers; a firewall between the nodes must allow them too.
-While IKE to a peer fails, the node drops its Geneve traffic to that peer rather than send it in clear.
-The nft table `inet spinifex_ipsec_guard` drops outbound UDP 6081 from the encap address whose route carries no SA, which also covers the moments pluto flushes the xfrm policy table as it starts and stops.
-It is not persisted: `ipsec.service` puts it back at boot before ovn-controller builds the tunnels, and the daemon re-syncs it every 60 seconds.
-`spx admin init --ipsec=false` leaves the tunnels unencrypted, as on any OS.
-
-## Step 2. Know What the Installer Changed
-
-On Oracle Linux the installer makes these changes, which Debian hosts do not need:
-
- **Repositories.** It enables `ol10_UEKR8` and `ol10_kvm_utils`, and adds `qemu-kvm*` and `qemu-img*` to the `excludepkgs` of `ol10_appstream`, keeping any excludes already set, so a later AppStream build can never replace the KVM Utilities QEMU.
- **NFV SIG.** It writes `/etc/yum.repos.d/spinifex-nfv-sig.repo` with an `includepkgs=` list of the OVS and OVN packages, and imports the SIG key to `/etc/pki/rpm-gpg/RPM-GPG-KEY-CentOS-SIG-NFV` only when its fingerprint is `35154228174901BEFA8E69A621465E289D2A76A7`.
- **OVS and OVN user.** Both run as root, as on Debian, through `OVS_USER_ID` and `OVN_USER_ID` in `/etc/sysconfig/openvswitch` and `/etc/sysconfig/ovn`.
- **DNS.** Where systemd-resolved is installed but not running, it saves `/etc/resolv.conf` to `/var/lib/spinifex-reset/resolv.conf.orig`, starts resolved behind NetworkManager and links `/etc/resolv.conf` to the resolved stub. It then checks that names still resolve and stops, printing the commands that put DNS back, if they do not.
- **chrony.** It appends `confdir /etc/chrony.d` to `/etc/chrony.conf` once, and writes its step policy to `/etc/chrony.d/spinifex.conf`.
- **`spx` under `sudo`.** Oracle Linux's `sudo` path leaves out `/usr/local/bin`, so the installer links `/usr/bin/spx` to `/usr/local/bin/spx`.
-
-The kernel modules list (`/etc/modules-load.d/spinifex.conf`) and the host firewall below apply on every OS.
-
-## Step 3. Replace firewalld Rules You Depend On
-
-The installer stops and masks firewalld, because with it running a packet must pass both firewalld's table and the Spinifex one, which silently breaks routed NAT, Elastic IPs and instance metadata.
-When firewalld was active, the installer turns on the Spinifex nftables policy in its place; pass `--firewall=on` or `--firewall=off` to choose yourself.
-
-firewalld's rules are not saved or migrated.
-The Spinifex policy filters inbound traffic only and opens 53, 443, 3000, 8443 and 9999 to any source.
-It accepts SSH from every IPv4 source until `$trusted_ssh_peers` in `/etc/spinifex/firewall/custom.nft` narrows it.
-Restate any other restriction you had, such as SSH limited to a management subnet, in that file; [Host Firewall](/docs/host-firewall) shows how.
-
-From the moment firewalld stops until the daemon first loads the Spinifex policy, the host accepts every connection.
-Uninstalling Spinifex leaves firewalld unmasked but disabled; it does not turn it back on.
-
-## Step 4. Keep the Host Updated
-
-Do not enable `dnf-automatic`.
-An Open vSwitch package upgrade restarts OVS, which interrupts guest networking, so update `openvswitch3.5`, `ovn25.03*` and `qemu-kvm*` in a maintenance window.
-Stay on the current Oracle Linux 10 minor release, which is what Spinifex is tested on.
-
-Before rebooting into a new kernel, check that its modules are installed, or OVS cannot start after the reboot:
+### Set up OVN networking

 ```bash
-rpm -q kernel-uek-core kernel-uek-modules
+sudo /usr/local/share/spinifex/setup-ovn.sh --management
 ```

-Each `kernel-uek-core` version needs a `kernel-uek-modules` of the same version; install a missing one with `sudo dnf install kernel-uek-modules-<version>`.
+### Initialize

-Track Oracle Linux security advisories (ELSA) for the host, the kernel and QEMU, for example with `dnf updateinfo list --security`.
-Those advisories do not cover the NFV SIG packages, so follow Open vSwitch and OVN security announcements for those.
+```bash
+sudo spx admin init --node node1 --nodes 1
+```

-## Step 5. Uninstall
+### Start services

-`/usr/local/share/spinifex/uninstall-spx.sh` removes Spinifex as on Debian.
-On a cluster node it also stops `openvswitch-ipsec.service` and deletes the `ovs_cert*` entries from libreswan's NSS database, so the node can join a different cluster; other NSS identities and your connection files in `/etc/ipsec.d` stay.
-It lists what it leaves in place on Oracle Linux: the repository changes and the SIG key, `OVS_USER_ID` and `OVN_USER_ID`, the OVS and OVN logrotate edits, the chrony include, `/etc/modules-load.d/spinifex.conf`, the SELinux mode, a masked `ipsec.service`, firewalld disabled, and any GPU passthrough configuration.
-`--purge-deps` is refused on Oracle Linux, because dnf cannot tell the packages Spinifex installed from the host's own; remove them with `dnf remove` yourself.
+```bash
+sudo systemctl start spinifex.target
+```
+
+### Verify
+
+```bash
+export AWS_PROFILE=spinifex
+aws ec2 describe-instance-types
+```
+
+If this returns a list of instance types, Spinifex is installed.
+Continue to [Setting Up Your Cluster](/docs/setting-up-your-cluster) to import an AMI and launch your first instance.
+:::
+
+**For a cluster**, prepare and install every node with the steps above, then continue with [Multi-Node Install](/docs/install-multi-node) from Step 2.
+Every node must run Oracle Linux.
+IPsec between nodes runs on libreswan, so open UDP 500, UDP 4500 and ESP between nodes on any firewall between them.
+
+## What the Installer Changes
+
+On Oracle Linux the installer also:
+
+- Enables `ol10_UEKR8` and `ol10_kvm_utils`, and excludes `qemu-kvm*` and `qemu-img*` from `ol10_appstream`.
+- Adds the NFV SIG repository as `/etc/yum.repos.d/spinifex-nfv-sig.repo`, limited to the OVS and OVN packages.
+- Installs `kernel-uek-modules` for the running kernel, about 800 MB with `linux-firmware`.
+- Switches DNS to systemd-resolved behind NetworkManager where resolved is installed but not running.
+- Links `/usr/bin/spx` to `/usr/local/bin/spx`, so `sudo spx` works.
+
+### firewalld
+
+The installer stops and masks firewalld and turns on the Spinifex nftables policy in its place; firewalld's rules are not migrated.
+The Spinifex policy opens 53, 443, 3000, 8443, 9999 and SSH to any source.
+Restate any restriction you relied on, such as SSH limited to a management subnet, in `/etc/spinifex/firewall/custom.nft`; see [Host Firewall](/docs/host-firewall).
+
+### Updates
+
+Do not enable `dnf-automatic`: an Open vSwitch upgrade restarts OVS and interrupts guest networking, so update `openvswitch3.5`, `ovn25.03*` and `qemu-kvm*` in a maintenance window.
+Before rebooting into a new kernel, check `rpm -q kernel-uek-core kernel-uek-modules` shows matching versions, or OVS cannot start.
+
+### Uninstall
+
+`/usr/local/share/spinifex/uninstall-spx.sh` removes Spinifex and lists what it leaves in place.
+`--purge-deps` is refused on Oracle Linux; remove packages with `dnf remove` yourself.

 ## Troubleshooting

-### Installer stops: SELinux is enforcing
+### Installer stops on SELinux, the kernel or kernel modules

-Set SELinux permissive now and in `/etc/selinux/config`, as in [SELinux](#selinux), then run the installer again.
-Fixing only one of the two is refused too.
-
-### Installer stops: the running kernel is not UEK R8
-
-Install `kernel-uek`, reboot into it, check `uname -r`, then run the installer again.
-
-### Installer stops: kernel-uek-modules is no longer in the repositories
-
-The running kernel is older than anything the repository still carries modules for.
+Set SELinux permissive both now and in `/etc/selinux/config`, or update and boot into the current UEK R8 kernel, then run the installer again:

 ```bash
 sudo dnf update -y kernel-uek
@@ -229,41 +171,30 @@ sudo reboot

 ### Installer stops: a repository is unreachable

-The message names the repository.
-Check the host can reach `yum.oracle.com` and `mirror.stream.centos.org` over HTTPS, including through any proxy dnf uses, then run the installer again.
+Check the host can reach `yum.oracle.com` and `mirror.stream.centos.org` over HTTPS, including through any proxy dnf uses.

 ### DHCP on br-wan times out

-Check STP is off on the bridge:
+STP is on. Turn it off:

 ```bash
-nmcli -g bridge.stp connection show br-wan
+sudo nmcli connection modify br-wan bridge.stp no
+sudo nmcli connection up br-wan
 ```

-If it prints `yes`, run `sudo nmcli connection modify br-wan bridge.stp no` and `sudo nmcli connection up br-wan`.
-
-### `sudo spx` finds no command
-
-Check that `/usr/bin/spx` links to `/usr/local/bin/spx`; run the installer again if it does not, or call `sudo /usr/local/bin/spx` meanwhile.
-
 ### Cluster nodes form no IPsec tunnels

-Check that both IPsec services run, and that libreswan holds the node's certificate and the cluster CA:
-
 ```bash
 systemctl is-active ipsec.service openvswitch-ipsec.service
-sudo certutil -L -d sql:/var/lib/ipsec/nss     # ovs_certkey_<node>, ovs_cert_cacert
 sudo ipsec trafficstatus                       # one line per peer tunnel
 ```

-If the services are inactive, check the daemon log with `journalctl -u spinifex-daemon` for the IPsec reconcile.
-If the certificates are present but no tunnel is listed, check that UDP 500 and ESP pass between the nodes in both directions, and look for `SEC_ERROR_UNKNOWN_ISSUER` in `journalctl -u ipsec`, which means a peer's certificate is from another CA.
-Until the tunnel forms, guest traffic between those two nodes is dropped, which is the guard in [Clusters](#clusters) working.
+If the services are inactive, check `journalctl -u spinifex-daemon`.
+If no tunnel is listed, check UDP 500 and ESP pass between the nodes in both directions.
+Guest traffic between two nodes is dropped until their tunnel forms.

 ### CA certificate not trusted

-`sudo spx admin init` installs the CA automatically. To install it again by hand on Oracle Linux:
-
 ```bash
 sudo cp /etc/spinifex/ca.pem /etc/pki/ca-trust/source/anchors/spinifex-ca.crt
 sudo update-ca-trust
@@ -16,16 +16,16 @@ Create, update, delete and policy association complete synchronously within the
 |---|---|---|
 | Identity | Cluster name plus principal ARN. Key and ARN suffix use `sha256hex(principalARN)` (`domains/eks/access/store.go` `PrincipalARNHash`). ARN `arn:aws:eks:{region}:{account}:access-entry/{cluster}/{hash}` (`foundation/aws/arn/eks.go`); the AWS ARN shape differs and is not verified. | `handlers/eks/service_impl_test.go` `TestAccessEntry_CreateDescribeListDelete`, `domains/eks/access/store_test.go` `TestKeyPaths`, `gateway/eks/authz_test.go` `TestResourceARNsAccessEntryMatchesHandler` |
 | Scope | Account and Region, within one cluster: keys live under `clusters/{cluster}/access-entries/` in the `eks-account-{account}` bucket. | `handlers/eks/access_lifecycle_test.go` `TestAccessEntryPersistedLayout_TopLevelAndNestedFieldNames`, `TestAccessEntryRecord_DecodesHandWrittenJSON`, `handlers/eks/create_cluster_test.go` `TestCreateCluster_SeedsCreatorAdminAccessEntry` |
-| Durable owner and record | `domains/eks/access` (`Owner`) is the only writer of `Record` with nested `AssociatedPolicy` and `Scope`: Create, SeedCreatorAdmin (cluster launch), Update, Associate, Disassociate and Delete. Keys and field names are unchanged by the move. `DeleteClusterPrefix` (`handlers/eks/cluster_state.go`) still erases entries with the cluster, and the gateway process reads records through the exported `access.Get` in `ResolveTokenReview` (`handlers/eks/token_review.go`). | `handlers/eks/access_lifecycle_test.go` `TestAccessEntryPersistedLayout_TopLevelAndNestedFieldNames`, `TestAccessEntryRecord_DecodesHandWrittenJSON`, `domains/eks/access/owner_test.go`, `domains/eks/access/store_test.go`, `handlers/eks/create_cluster_test.go` `TestCreateCluster_SeedsCreatorAdminAccessEntry` |
+| Durable owner and record | `domains/eks/access` (`Owner`) is the only writer of `Record` with nested `AssociatedPolicy` and `Scope`: Create, SeedCreatorAdmin (cluster launch), Update, Associate, Disassociate and Delete. Keys and field names are unchanged by the move. `DeleteClusterChildren` (`domains/eks/cluster/cluster_state.go`) still erases entries with the cluster, and the gateway process reads records through the exported `access.Get` in `ResolveTokenReview` (`handlers/eks/token_review.go`). | `handlers/eks/access_lifecycle_test.go` `TestAccessEntryPersistedLayout_TopLevelAndNestedFieldNames`, `TestAccessEntryRecord_DecodesHandWrittenJSON`, `domains/eks/access/owner_test.go`, `domains/eks/access/store_test.go`, `handlers/eks/create_cluster_test.go` `TestCreateCluster_SeedsCreatorAdminAccessEntry` |
 | AWS adapter | `handlers/eks` keeps request validation (entry type, supported policy, scope), the cluster existence check (`acctKVForCluster`), awserrors mapping and the AWS projection. `gateway/eks/authz.go` uses `access.PrincipalARNHash`. | `handlers/eks/service_impl_test.go` access cases, `gateway/eks/authz_test.go` `TestResourceARNsAccessEntryMatchesHandler` |
 | Desired state | Principal, Kubernetes username and groups, type (`STANDARD` only), tags (set at create only) and cluster-scoped access policy associations. | `handlers/eks/service_impl_test.go` `TestAccessEntry_RejectsNodeType`, `TestAccessPolicy_AssociateRejectsUnsupportedPolicyAndScope`, `TestAccessPolicy_AssociateRejectsNamespaceScope`, `handlers/eks/access_lifecycle_test.go` `TestAccessEntry_UpdateWithNoFieldsKeepsUsernameAndGroups` |
 | Generation | None. Update, associate and disassociate retry against the KV revision; create and delete do not. | |
 | Observed state | Not tracked; the record is the only state. | |
 | Readiness | Usable by token review as soon as the write returns. | `handlers/eks/token_review_test.go` `TestResolveTokenReview_AuthenticatesViaVerifyAndKV`, `TestResolveTokenReview_NoAccessEntryDenies`, `handlers/eks/service_impl_test.go` `TestAccessPolicy_DisassociateRemovesProjectedGroup` |
 | Idempotency | `clientRequestToken` is ignored. Update always advances `modifiedAt`, even when nothing changes. A duplicate create returns `ResourceInUseException`; a repeated delete returns `ResourceNotFoundException`; disassociating an unbound policy succeeds without change. Internal NATS redelivery of create returns `ResourceInUseException` to the redelivered request. | `handlers/eks/service_impl_test.go` `TestAccessEntry_CreateDescribeListDelete`, `TestAccessEntry_DescribeDeleteMissingIsNotFound`, `handlers/eks/access_lifecycle_test.go` `TestAccessPolicy_DisassociateUnassociatedLeavesRecordUnchanged`, `TestAccessPolicy_AssociateWithNoEntryIsNotFound` |
-| Dependencies | Every action requires only that the cluster's meta record exists (`acctKVForCluster`); cluster status is not read, so actions succeed on a `CREATING`, `FAILED` or `DELETING` cluster. Token review does not check the cluster at all. The principal is not checked against IAM. | `handlers/eks/service_impl_test.go` `TestAccessEntry_UnknownClusterIsNotFound`, `handlers/eks/access_lifecycle_test.go` `TestAccessEntryActionsIgnoreClusterLifecycleStatus`, `TestResolveTokenReview_AuthenticatesWithoutClusterMeta` |
+| Dependencies | Every action requires that the cluster's meta record exists (`acctKVForCluster`). `CreateAccessEntry` also refuses a `DELETING` cluster with `ResourceInUseException`, since a recorded delete admits no new dependant; no other status is read, so actions succeed on a `CREATING` or `FAILED` cluster. Token review does not check the cluster at all. The principal is not checked against IAM. | `handlers/eks/service_impl_test.go` `TestAccessEntry_UnknownClusterIsNotFound`, `handlers/eks/access_lifecycle_test.go` `TestAccessEntryActionsIgnoreFailedClusterStatus`, `delete_cluster_async_test.go` `TestS2_EKSDeletingClusterAdmitsNoNewDependants`, `TestResolveTokenReview_AuthenticatesWithoutClusterMeta` |
 | Authorization | Gateway authz derives the access-entry ARN with `PrincipalARNHash` (`gateway/eks/authz.go`). No `iam:PassRole`. The STS step of token review is the existing `eks.VerifyToken` NATS contract served by awsgw. | `gateway/eks/authz_test.go` `TestResourceARNsAccessEntryMatchesHandler` |
-| Deletion ordering | Delete is get-then-delete, not conditional. Cluster delete erases entries in the prefix purge only after all infrastructure teardown succeeds; on failure they survive while the cluster stays `DELETING`. | `handlers/eks/access_lifecycle_test.go` `TestDeleteClusterPrefix_SweepsAccessEntriesScopedToCluster` |
+| Deletion ordering | Delete is get-then-delete, not conditional. Cluster delete erases entries with the cluster's other child records, before the meta, only after all infrastructure teardown succeeds; on failure they survive while the cluster stays `DELETING`. | `handlers/eks/access_lifecycle_test.go` `TestDeleteClusterChildren_SweepsAccessEntriesScopedToCluster` |
 | Restart and interruption | The prefix purge deletes keys in listing order; an interruption after the meta key is gone leaves entries that a new cluster of the same name inherits. Reclaiming a `FAILED` cluster and a failed launch purge with the meta kept, so the recreated cluster inherits the failed attempt's entries. | none |
 | End-to-end | `tests/e2e/eks/eks_test.go` `AccessEntry` and `GetToken`; `tests/integration/cross_tenant_test.go`. No Terraform apply/destroy evidence for `aws_eks_access_entry` or `aws_eks_access_policy_association`. | |

@@ -33,7 +33,7 @@ Create, update, delete and policy association complete synchronously within the

 1. Create is get-then-put with no exclusive create, so concurrent creates for one principal both succeed and the later write wins.
 2. The creator-admin seed at cluster launch is an unconditional put after the cluster meta is final, so it replaces an entry for the same principal created while the cluster was `CREATING`, including its policy associations.
-3. Access actions do not check cluster status (see dependencies row), and opening the bucket creates it for an account that has none.
+3. Access actions other than `CreateAccessEntry` do not check cluster status (see dependencies row), and opening the bucket creates it for an account that has none.
 4. Entries can outlive their cluster (interrupted prefix purge) or carry over to a recreated cluster (reclaim and failed launch keep the meta and do not purge entries separately).
 5. Token review authenticates against an entry without checking the cluster exists or its status.
 6. `ResolveTokenReview` opens the EKS bucket and reads the record in the gateway process through the exported `access.Get`, outside the owner.
@@ -16,11 +16,11 @@ There is no generation, so the record mixes desired fields with the latest obser
 | Item | Present behaviour | Evidence |
 |---|---|---|
 | Identity | Cluster name plus add-on name; one add-on per name per cluster. ARN `arn:aws:eks:{region}:{account}:addon/{cluster}/{addon}` (`foundation/aws/arn/eks.go` `FormatEKSAddon`), deterministic by name, with no per-incarnation suffix. The gateway authorizer derives the same ARN (`domains/eks/awsapi/authz.go` `addonARN`). | `handlers/eks/addon_lifecycle_test.go` `TestAddonPersistedLayout_RecordAndManifestKeysAndFieldNames`, `domains/eks/awsapi/authz_test.go` `TestResourceARNsFidelity` |
-| Scope | Account and Region, within one cluster: keys live under `clusters/{cluster}/addons/` in the `eks-account-{account}` bucket. | `TestAddonPersistedLayout_RecordAndManifestKeysAndFieldNames`, `TestDeleteClusterPrefix_SweepsAddonRecordsAndManifestsScopedToCluster` |
+| Scope | Account and Region, within one cluster: keys live under `clusters/{cluster}/addons/` in the `eks-account-{account}` bucket. | `TestAddonPersistedLayout_RecordAndManifestKeysAndFieldNames`, `TestDeleteClusterChildren_SweepsAddonRecordsAndManifestsScopedToCluster` |
 | Durable record | `clusters/{cluster}/addons/{addon}` holds `Record` (`domains/eks/addon/record.go`): `addonName`, `addonVersion`, `status`, `serviceAccountRoleArn`, `configurationValues`, `health`, `arn`, `tags`, `createdAt`, `modifiedAt` (the optional ones `omitempty`). `clusters/{cluster}/addons/{addon}/manifest` holds `Manifest` (`domains/eks/addon/record.go`): `addonName`, `addonVersion`, `serviceAccountRoleArn`, `configurationValues`. Listing skips the manifest sub-key by segment count. | `TestAddonPersistedLayout_RecordAndManifestKeysAndFieldNames`, `TestAddonRecord_DecodesHandWrittenJSON`, `domains/eks/addon/store_test.go` `TestRecord_RoundTrip`, `TestList_SkipsManifestSubKeys` |
-| Writers | Record: the owner in `domains/eks/addon/owner.go`, reached from the `CreateAddon`, `UpdateAddon` and `DeleteAddon` handlers, the cluster reconciler's report relay and the GPU node group launch: `Owner.Create` (unconditional put), `Owner.Update` and its failure marking (revision CAS), `Owner.Delete` (get then delete), `Owner.ApplyReport` (revision CAS), `Owner.EnsureGPUDevicePlugin` through `Owner.Create`; and the cluster prefix purge (`DeleteClusterPrefix`). Manifest: `StagingInstaller.Install` (unconditional put on create and update), `StagingInstaller.Uninstall` (`domains/eks/addon/delivery.go`), `Owner.Delete` and the prefix purge. | `TestAddonPersistedLayout_RecordAndManifestKeysAndFieldNames`, `TestUpdateAddon_ReportsSuccessfulWhileRecordIsUpdating`, `TestDeleteAddon_ReturnsDeletingAfterErasingRecordAndManifest`, `TestDeleteClusterPrefix_SweepsAddonRecordsAndManifestsScopedToCluster`, `domains/eks/addon/store_test.go` `TestCasUpdate`, `TestDeleteRecord_RemovesRecordAndManifest` |
+| Writers | Record: the owner in `domains/eks/addon/owner.go`, reached from the `CreateAddon`, `UpdateAddon` and `DeleteAddon` handlers, the cluster reconciler's report relay and the GPU node group launch: `Owner.Create` (unconditional put), `Owner.Update` and its failure marking (revision CAS), `Owner.Delete` (get then delete), `Owner.ApplyReport` (revision CAS), `Owner.EnsureGPUDevicePlugin` through `Owner.Create`; and the cluster prefix purge (`DeleteClusterChildren`). Manifest: `StagingInstaller.Install` (unconditional put on create and update), `StagingInstaller.Uninstall` (`domains/eks/addon/delivery.go`), `Owner.Delete` and the prefix purge. | `TestAddonPersistedLayout_RecordAndManifestKeysAndFieldNames`, `TestUpdateAddon_ReportsSuccessfulWhileRecordIsUpdating`, `TestDeleteAddon_ReturnsDeletingAfterErasingRecordAndManifest`, `TestDeleteClusterChildren_SweepsAddonRecordsAndManifestsScopedToCluster`, `domains/eks/addon/store_test.go` `TestCasUpdate`, `TestDeleteRecord_RemovesRecordAndManifest` |
 | Readers | `ListAddons`, `DescribeAddon`, `CreateAddon` (existence check) and the reconciler read the record. `ListStagedAddonManifests` reads every `/manifest` key for the cluster (`addon.ListManifests`); it is served on `eks.ListStagedAddonManifests` (`daemon/daemon.go`) and reached only through the gateway's internal-addons route. | `handlers/eks/addons_test.go` `TestAddon_DescribeListDelete`, `TestListStagedAddonManifests` |
-| Catalogue | Static in-binary catalogue (`domains/eks/addon/catalog.go`): `aws-load-balancer-controller`, `argocd`, `aws-ebs-csi-driver`, and hidden-but-creatable `nvidia-device-plugin` and `spinifex-noop`; one version each, the first being the default. Create and update reject unknown names and versions with `InvalidParameterValue`. | `domains/eks/addon/catalog_test.go`, `handlers/eks/addons_test.go` `TestDescribeAddonVersions_ReturnsCatalog`, `TestCreateAddon_UnknownAddonRejected`, `TestCreateAddon_UnknownVersionRejected`, `TestUpdateAddon_UnknownVersionRejected` |
+| Catalogue | Static in-binary catalogue (`domains/eks/addon/catalog.go`): `aws-load-balancer-controller`, `argocd`, `aws-ebs-csi-driver`, and hidden-but-creatable `nvidia-device-plugin` and `spinifex-noop`; one version each, the first being the default. Versions are stored and used internally in the bare upstream form, and API responses show the AWS form `v<x.y.z>-eksbuild.<n>` (`Spec.AWSVersion`). Create and update accept the AWS form or the exact bare form and reject unknown names and other version strings with `InvalidParameterValue`, which the gateway returns as `InvalidParameterException`. | `domains/eks/addon/catalog_test.go`, `handlers/eks/addons_test.go` `TestDescribeAddonVersions_ReturnsCatalog`, `TestCreateAddon_UnknownAddonRejected`, `TestCreateAddon_UnknownVersionRejected`, `TestUpdateAddon_UnknownVersionRejected` |
 | Desired state | Version, `serviceAccountRoleArn`, `configurationValues` and create-time tags. `resolveConflicts`, `podIdentityAssociations` and `clientRequestToken` are accepted and dropped; `preserve` on delete is ignored. `configurationValues` reaches the guest but the agent discards it (`mulga-eks-addon-sync.sh` `sync_once`), so it has no effect. | `TestAddonPersistedLayout_RecordAndManifestKeysAndFieldNames`, `TestDeleteAddon_ReturnsDeletingAfterErasingRecordAndManifest` |
 | Observed state | `status` and `health` are overwritten by guest reports on the same record as the desired fields. `health` is the latest failure message, projected as one `AddonIssue` with only `message`. Statuses held: `CREATING`, `ACTIVE`, `UPDATING`, `DEGRADED`, `CREATE_FAILED`; `DELETING` is defined but never stored. | `TestAddonRecord_DecodesHandWrittenJSON`, `handlers/eks/addon_status_reconciler_test.go` `TestApplyAddonStatusReport` |
 | Generation | None. Reports carry `version` and `ts`, and neither is read: any report for the add-on name applies to whatever is currently desired. | `TestAddonStatusReport_MismatchedVersionIsAccepted` |
@@ -37,9 +37,9 @@ There is no generation, so the record mixes desired fields with the latest obser
 | Cross-resource dependency | A GPU node group's launch calls `stageGPUDeviceAddon`, which checks the cluster exists and calls `Owner.EnsureGPUDevicePlugin` for `nvidia-device-plugin`; an existing record is a no-op and every other error is logged only. Deleting the node group never unstages it, and an existing record in any status, including `CREATE_FAILED`, blocks re-staging. | `handlers/eks/nodegroup_test.go` `TestCreateNodegroup_GPUNodegroupStagesDevicePluginAddon`, `domains/eks/addon/owner_test.go` `TestOwnerEnsureGPUDevicePlugin_StagesDefaultVersionOnceAndKeepsAnyExistingRecord`, `TestCreateNodegroup_NonGPUNodegroupDoesNotStageDevicePluginAddon`, `tests/e2e/gpu/eks_gpu_test.go` `TestEKSGPUPodExposure` |
 | Idempotency | `clientRequestToken` is ignored. A duplicate create, including an SDK retry with the same token, returns `ResourceInUseException`. A repeated update re-stages and returns another `Successful` with the same id. A repeated delete returns `ResourceNotFoundException`. | `TestCreateAddon_RepeatedClientRequestTokenIsResourceInUse`, `TestUpdateAddon_ReportsSuccessfulWhileRecordIsUpdating`, `TestDeleteAddon_ReturnsDeletingAfterErasingRecordAndManifest`, `handlers/eks/addons_test.go` `TestCreateAddon_CreatesStagingAndDefaultsVersion` |
 | Absent resource | An unknown cluster returns `ResourceNotFoundException` for every action, including the internal manifest listing. A missing add-on returns `ResourceNotFoundException` from describe, update and delete. | `handlers/eks/addons_test.go` `TestCreateAddon_UnknownClusterIsNotFound`, `TestDescribeDeleteAddon_MissingIsNotFound`, `TestUpdateAddon_MissingIsNotFound`, `TestListStagedAddonManifests` |
-| Dependencies | Only the cluster meta's existence is checked; cluster status is not read, so add-ons can be created on a `DELETING` cluster. | `TestAddonActionsIgnoreClusterLifecycleStatus` |
+| Dependencies | `CreateAddon` checks the cluster meta exists and refuses a `DELETING` cluster with `ResourceInUseException`, since a recorded delete admits no new dependant; other statuses are not read. | `TestCreateAddonRefusesOnlyADeletingCluster` |
 | Authorization | Gateway policy uses the add-on ARN above. `serviceAccountRoleArn` requires `iam:PassRole` on `CreateAddon` and `UpdateAddon` (`domains/eks/awsapi/passrole.go`), the same check `CreateCluster`/`CreateNodegroup` apply, including that the role exists. Tag actions on an add-on ARN return `NotImplemented`. | `domains/eks/awsapi/gateway_wire_test.go` `TestPassedRoleARNs_AddonServiceAccountRoleIsChecked`, `domains/eks/awsapi/gateway_passrole_test.go` `TestEKSCreateAddon_PassRole`, `TestEKSUpdateAddon_PassRole`, `domains/eks/awsapi/authz_test.go` `TestResourceARNsFidelity` |
-| Cluster deletion | Add-on records and manifests are erased only by the cluster prefix purge, after the cluster's infrastructure teardown succeeds; there is no per-add-on delete. Reclaiming a failed cluster keeps the meta and does not purge, so the recreated cluster inherits the add-ons. | `TestDeleteClusterPrefix_SweepsAddonRecordsAndManifestsScopedToCluster`, `handlers/eks/cluster_state_test.go` `TestDeleteClusterPrefix_SweepsEveryKey` |
+| Cluster deletion | Add-on records and manifests are erased only by the cluster prefix purge, after the cluster's infrastructure teardown succeeds; there is no per-add-on delete. Reclaiming a failed cluster keeps the meta and does not purge, so the recreated cluster inherits the add-ons. | `TestDeleteClusterChildren_SweepsAddonRecordsAndManifestsScopedToCluster`, `handlers/eks/cluster_state_test.go` `TestDeleteClusterChildren_SweepsEveryKey` |
 | Restart and interruption | Create interrupted between the record put and staging leaves a `CREATING` record with no manifest that nothing re-stages and that blocks a retry. Delete interrupted after unstaging leaves a record in its previous status while the guest removes the software. An update interrupted after the CAS leaves `UPDATING` with the previous manifest. A daemon restart loses nothing durable, but reports sent while no reconciler runs are lost until the next tick. | none |
 | End-to-end | `tests/e2e/eks/eks_test.go` `TestEKS/AddonDelivery` (create to `ACTIVE`, objects present, delete removes the namespace) and `tests/e2e/gpu/eks_gpu_test.go` `TestEKSGPUPodExposure`; `tests/integration/cross_tenant_test.go` `TestCrossTenant_EveryScopedActionIsAccountedFor`. No E2E for update, failure, configuration or the role. No Terraform apply/destroy evidence for `aws_eks_addon`. | |

@@ -0,0 +1,42 @@
+# EKS cluster: delete lifecycle contract
+
+Contract under ADR-0003 S1 for the delete half of an EKS cluster's lifecycle, the half ADR-0003 S8 constrains.
+Create, update and the control-plane reconciler are out of its scope and keep the observed gaps recorded elsewhere.
+Paths are relative to `spinifex/` unless they start with `tests/`, which is relative to the spinifex repo root.
+Test names without a path are in `domains/eks/cluster/`.
+
+## Scope
+
+A cluster is a composite resource: its record at `clusters/{name}/meta` in the `eks-account-{account}` bucket names the dependants a delete must remove.
+In the system account those are the control-plane VMs and their root volumes, the endpoint load balancer with its target groups, the managed control-plane VPC and its network, and the spread placement group.
+In the customer account they are the egress address, the private-endpoint interface and security group, the cluster security groups, a cluster-owned internet gateway, and load balancers the in-cluster controller created.
+Node group workers and CSI-provisioned EBS volumes are never removed by a cluster delete.
+
+## Contract
+
+| Item | Behaviour | Evidence |
+|---|---|---|
+| Request | `DeleteCluster` validates, answers `ResourceNotFoundException` for an absent cluster, answers a `DELETING` cluster from its record, refuses a cluster with node groups with `ResourceInUseException`, records `DELETING` and returns. It waits for no teardown step. | `TestS8_EKSDeleteClusterReturnsBeforeTheTeardownCompletes`, `TestRLC1_EKSDeleteClusterNotFoundOnAbsent`, `TestS8_EKSDeleteClusterRefusesWhileNodegroupsExist`, `tests/e2e/eks` `TestEKS/DeleteCluster` |
+| Obligation | `DELETING` with `DeletingSince` on the record is the tenant's decision. No status writer moves a record off it, and a create launch still running adds its refs to it instead of overwriting it. | `TestS2_EKSDeletingIsStickyForStatusWriters`, `TestS2_EKSLaunchWriteAfterDeleteNeitherRevertsNorRecreates` |
+| Retry | A repeat delete answers `DELETING` from the record and starts nothing. | `TestS8_EKSSecondDeleteAnswersFromTheRecordWithoutASecondTeardown` |
+| Convergence | `convergeClusterDeletion` (`domains/eks/cluster/cluster_teardown.go`), started by the request and backed by `EKSDeletingReaper`, runs under a per-cluster in-process gate and the cross-node teardown lease, renewed on the service context. It acts only on a record that is `DELETING` when re-read. | `TestS8_EKSKickAndSameNodeSweepRunOneTeardown`, `TestS8_EKSTeardownLeaseOutlivesTheRequest`, `TestS7_EKSTeardownActsOnlyOnARecordedDelete` |
+| Retained cleanup | A failed step keeps the record `DELETING` with its refs and the error. The record is removed last, after its child records, and only if no ref appeared that the pass did not tear down. A record with a node group is never removed; a record that is not `DELETING` is refused before anything else is checked. | `TestRLC2_EKSBillableTeardownBeforeKVSweep`, `TestS2_EKSTeardownRemovesTheMetaLast`, `TestS2_EKSTeardownDoesNotSweepARecordChangedDuringTeardown`, `TestS7_EKSTeardownNeverRemovesAClusterWithNodegroupWorkers` |
+| Recovery | A teardown interrupted by a crash or a lost lease is finished by the reaper on any node. Failed passes back off from 90 s, doubling to a 30 min cap, and are never abandoned; `DeleteReapExhausted` only flags a long failure for an operator. | `TestS2_EKSTeardownResumesAfterAnInterruptedConverge`, `TestRLC4_DeletingReaperReDrivesWedgedTeardown`, `TestDeletingReaper_KeepsRetryingPastTheOldAttemptLimit`, `TestDeletingReaper_ResumesAnAlreadyExhaustedRecord` |
+| Visibility | `DescribeCluster` and `ListClusters` show the cluster `DELETING` until the record is removed, then `ResourceNotFoundException`. A failed pass shows one `INTERNAL_FAILURE` health issue naming the step, never the internal error. | `TestDescribeCluster_StuckDeleteShowsTheStepNotTheError`, `TestRLC2_EKSBillableTeardownBeforeKVSweep` |
+| New dependants | `CreateNodegroup` requires `ACTIVE`; `CreateAddon` and `CreateAccessEntry` refuse a `DELETING` cluster with `ResourceInUseException`. | `TestCreateAddonRefusesOnlyADeletingCluster`, `TestS2_EKSDeletingClusterAdmitsNoNewDependants` |
+| Subject budgets | Every `eks.*` request subject is classified by its handler's work in `eksSubjects` (`domains/eks/cluster/service_nats.go`) and keeps the 30 s budget; `eks.DeleteCluster` is request-scaled. A subject the gateway sends or the daemon serves that is not classified fails the table. | `TestS8_EKSSubjectBudgets` |
+| Late child records | No child record outlives the cluster or leaks into a later one of the same name. The teardown sweeps child keys again after removing the meta and stops the cluster's bootstrap subscriber, bootstrap refuses to write into a record that is gone or `DELETING`, the launch re-checks its record around seeding the creator's access entry, and a fresh claim clears whatever an earlier cluster left under the name. | `TestS2_EKSLaunchOvertakenByDeleteLeavesNoChildRecords`, `TestS2_EKSBootstrapWritesNothingIntoADeletingCluster`, `TestS2_EKSFreshClaimInheritsNoStaleChildRecords` |
+| Launch generation | The create launch's writes are fenced on the record's generation (`CreatedAt` and ARN). An overtaken launch never writes a newer cluster of the same name and releases only the instances it started, by ID. One whose record was swept recreates it as `DELETING` with its refs, so the teardown and reaper own the release. One whose record was marked `FAILED` merges its refs and releases what it built. | `TestS2_EKSOvertakenLaunchNeverWritesANewerClusterOfTheSameName`, `TestS2_EKSOvertakenLaunchHandsItsInfraToTheTeardown`, `TestS2_EKSLaunchMarkedFailedReleasesWhatItBuilt` |
+| Running launch | A create launch heartbeats its owner and time on the record (`LaunchOwner`, `LaunchHeartbeat`). The teardown keeps the record while that heartbeat is fresh, so the cluster reads `DELETING` throughout and never `ResourceNotFoundException` then `DELETING` again; the launch hands its refs over at its next write and starts the teardown when it returns. A heartbeat older than 90 s is taken as a crashed launch and no longer waited for, and recreating a swept record stays as the backstop. The launch's OIDC keypair writes are fenced on the record's generation. | `TestS2_EKSDeletedClusterNeverReadsMissingWhileItsLaunchRuns`, `TestS2_EKSCrashedLaunchDoesNotPinTheDelete`, `TestS2_EKSOvertakenLaunchCannotReplaceANewerClustersSigningKey` |
+| Bounded passes | One converger run is bounded at 15 minutes, so a hung step cannot hold the gate and lease; the timed-out pass is recorded and backed off. Waiting on node groups is not counted as a failed pass, and deleting the last node group resumes the delete. | `TestS8_EKSHungTeardownStepReleasesTheGate`, `TestS2_EKSLastNodegroupDeletedResumesTheClusterDelete` |
+| FAILED reclaim | Re-creating a `FAILED` cluster's name tears the old attempt down under the same gate and lease as a delete, and answers `ResourceInUseException` while another holder has it; the per-node teardown bound does not apply to it. | `TestS2_EKSFailedClusterReclaimRunsUnderTheTeardownLease`, `TestS2_EKSFailedClusterReclaimIsNotRefusedForCapacity` |
+| Tenant data | The teardown never deletes a CSI-provisioned volume or terminates a node group worker. | `TestS7_EKSTeardownNeverDeletesCSIVolumesOrWorkers`, `TestS7_EKSTeardownNeverRemovesAClusterWithNodegroupWorkers` |
+
+## Gaps
+
+1. Four `eks.*` handlers still hold their request for work that grows with the resource, and the subject table names them as not conformant: `CreateCluster` (reclaiming a `FAILED` cluster of the same name tears its infra down inline), `DeleteNodegroup` (one awaited terminate per worker), `UpdateNodegroupConfig` (one awaited launch or terminate per worker added or removed) and `RestoreSnapshot` (launches a control-plane VM and fences the old one inline).
+2. AWS's answer to a repeat delete on a `DELETING` cluster has not been observed; `DELETING` from the record is the chosen behaviour.
+3. Control-plane, load balancer and VPC resources in the system account are keyed by the bare cluster name, so a teardown by name can reach another tenant's same-named cluster.
+4. The teardown deletes load balancers the in-cluster controller created in the customer account, which AWS leaves behind.
+5. An overtaken launch's late write can still land under a newer cluster of the same name in the instant between that cluster's claim sweep and the old launch's generation check; it is bounded by one write, since the old launch stops at its next fenced write.
+6. The `FAILED`-name reclaim purge still runs inside the `CreateCluster` request (gap 1).
@@ -4,13 +4,13 @@
 # distroless base (no shell, no package manager) for a small, production-shaped
 # image suitable for pushing to ECR and pulling onto EKS workers.

-FROM golang:1.27.3-alpine@sha256:f2c872bcee949b7be9c490efd02b48a8a2fbf947cdf533f1c58525ec9d8170fd AS build
+FROM golang:1.27.3-alpine@sha256:d200d2dd76b03a99b52651fd1325aff7dcd35422c35c01bac56571b0507f3deb AS build
 WORKDIR /src
 COPY go.mod ./
 COPY main.go ./
 RUN CGO_ENABLED=0 GOOS=linux go build -trimpath -ldflags="-s -w" -o /out/spinifex-demo .

-FROM gcr.io/distroless/static:nonroot@sha256:18f4cc26c9ffa7f20a359298be7deadac5c607c8e03a02b8119f1a5fc2f82c99
+FROM gcr.io/distroless/static:nonroot@sha256:25ac8f0c3b2334a9eb05452266bd4697864a8c8ef0ef71e5e6480d7f92868329
 COPY --from=build /out/spinifex-demo /spinifex-demo
 EXPOSE 8080
 USER nonroot:nonroot
@@ -79,7 +79,8 @@ Terraform manages the cluster, the addons, the ACM cert, and the HTTPS Ingress;
 - The `argocd` and `aws-ebs-csi-driver` bundles must be baked into the `eks-node` AMI — `describe-addon` returns `no baked bundle` otherwise.
 - The EBS-CSI default StorageClass uses `provisioner: ebs.csi.aws.com`; a PVC against it provisions a **Viperblock-backed EBS volume**.
 - If the PVC stays `Pending`, the app still serves (the counter falls back to in-memory) — see [Troubleshooting](#troubleshooting).
- Leave addon versions unset (the AWS provider and the catalog disagree on the version-string format).
+- Addon versions are unset, so Spinifex installs each catalog default.
+  To pin one, use the AWS form that `DescribeAddonVersions` returns, such as `v3.0.23-eksbuild.1`.
 - The Argo CD `Application` is a `kubernetes_manifest`, so the `argoproj.io` CRDs must exist before `apply` — apply the parent and let the addon reach `ACTIVE` first.

 **Prerequisites:**
@@ -471,8 +471,7 @@ resource "aws_eks_cluster" "this" {
  # expects: access_config reads as absent (and its ForceNew
  # bootstrap_cluster_creator_admin_permissions would replace the live cluster on
  # re-add), and vpc_config.security_group_ids reads back as drift. Ignore both so
-  # later applies (e.g. adding the Argo CD NodePort rule) neither replace the
-  # cluster nor detach its security groups.
+  # later applies neither replace the cluster nor detach its security groups.
  lifecycle {
    ignore_changes = [access_config, vpc_config[0].security_group_ids]
  }
@@ -520,8 +519,8 @@ resource "aws_eks_node_group" "workers" {
 #
 # Spinifex wires the controller's AWS credentials and ELB-eligible subnets at the
 # node level, so no IRSA role or subnet tagging is needed here. addon_version is
-# omitted: the AWS provider demands a v-prefixed version that the catalog rejects,
-# so let Spinifex default to its catalog version.
+# omitted so Spinifex installs its catalog default. To pin one, use the AWS form
+# from DescribeAddonVersions, for example v2.11.0-eksbuild.1.
 # ---------------------------------------------------------------------------

 resource "aws_eks_addon" "lbc" {
@@ -542,7 +541,7 @@ resource "aws_eks_addon" "lbc" {
 # Installed through the EKS API; Spinifex stages the bundle host-side and the
 # worker renders it into the K3s auto-deploy dir. The workloads module registers
 # the git repo and creates the Argo CD Application that syncs the demo app.
-# addon_version omitted (see the LBC note).
+# addon_version omitted: the catalog default is installed (see the LBC note).
 # ---------------------------------------------------------------------------

 resource "aws_eks_addon" "argocd" {
@@ -562,7 +561,7 @@ resource "aws_eks_addon" "argocd" {
 #
 # Ships a default gp3 StorageClass (provisioner ebs.csi.aws.com). A PVC in the
 # demo app's git manifests dynamically provisions a Viperblock-backed EBS volume
-# so the app's state survives pod restarts. addon_version omitted (see LBC note).
+# so the app's state survives pod restarts. addon_version omitted: catalog default.
 # ---------------------------------------------------------------------------

 resource "aws_eks_addon" "ebs_csi" {
@@ -642,53 +641,6 @@ resource "aws_acm_certificate" "ingress" {
  }
 }

-# ---------------------------------------------------------------------------
-# Let the ALB reach the workers' NodePort
-#
-# Spinifex auto-manages the nodegroup SG and admits only intra-cluster traffic.
-# The LBC-provisioned ALB lives in this VPC, so admit the VPC CIDR on the
-# NodePort. Look the SG up by its deterministic name and add one rule.
-# ---------------------------------------------------------------------------
-
-data "aws_security_group" "nodegroup" {
-  filter {
-    name   = "group-name"
-    values = ["eks-cluster-${var.cluster_name}-nodegroup-sg"]
-  }
-
-  filter {
-    name   = "vpc-id"
-    values = [aws_vpc.main.id]
-  }
-
-  depends_on = [aws_eks_node_group.workers]
-}
-
-resource "aws_vpc_security_group_ingress_rule" "nodeport_from_vpc" {
-  security_group_id = data.aws_security_group.nodegroup.id
-  cidr_ipv4         = aws_vpc.main.cidr_block
-  from_port         = var.node_port
-  to_port           = var.node_port
-  ip_protocol       = "tcp"
-
-  tags = {
-    Name = "${var.cluster_name}-alb-nodeport"
-  }
-}
-
-# Same, for the Argo CD UI NodePort its ALB forwards to.
-resource "aws_vpc_security_group_ingress_rule" "argocd_nodeport_from_vpc" {
-  security_group_id = data.aws_security_group.nodegroup.id
-  cidr_ipv4         = aws_vpc.main.cidr_block
-  from_port         = var.argocd_node_port
-  to_port           = var.argocd_node_port
-  ip_protocol       = "tcp"
-
-  tags = {
-    Name = "${var.cluster_name}-argocd-nodeport"
-  }
-}
-
 # ---------------------------------------------------------------------------
 # Outputs
 # ---------------------------------------------------------------------------
@@ -64,7 +64,6 @@ The LBC is installed as a managed **addon** (`aws_eks_addon`). Spinifex wires th
 | Node Group | `workers` | `node_desired_size` `t3.large` worker(s) — 1 or 3 |
 | Addon | `aws-load-balancer-controller` | Provisions the ALB from the Ingress |
 | ACM Certificate | `eks-https-ingress` | Self-signed, imported; attached to the HTTPS listener |
-| SG Ingress Rule | `eks-https-alb-nodeport` | Admits the VPC CIDR to the workers' NodePort |
 | K8s Deployment + Service | `spinifex-demo` | Themed demo image, 2 replicas, NodePort |
 | K8s Ingress | `spinifex-demo` | `ingressClassName: alb`, HTTPS via the ACM cert |

@@ -72,9 +71,10 @@ The LBC is installed as a managed **addon** (`aws_eks_addon`). Spinifex wires th

 - **The LBC needs no IRSA or subnet tagging here.** Credentials and ELB-eligible subnets are injected by Spinifex at the node level and into the `alb` `IngressClassParams`. The Ingress just sets `ingressClassName: alb` plus the `alb.ingress.kubernetes.io/*` annotations.
 - **`spinifex.io/managed-ingress = "false"`** disables the K3s built-in traefik/servicelb so the LBC owns ingress, matching AWS parity. Omit the tag (or set it `true`) and the built-in ingress stays on.
- The ALB targets the workers' **NodePort** (`target-type: instance`). The one SG rule admits the VPC CIDR on that port so the in-VPC ALB can reach the workers; the worker SG is otherwise auto-managed.
+- The ALB targets the workers' **NodePort** (`target-type: instance`). The AWS Load Balancer Controller adds the ALB-to-NodePort rule to the cluster security group itself, so the workbook creates no security group rule.
 - Spinifex ACM supports **ImportCertificate only** — the cert is generated by the `tls` provider and imported, not requested.
- **Leave `addon_version` unset** (see the [quickstart/addons notes](/docs/eks-quickstart)) — the AWS provider and the Spinifex catalog disagree on the version string format, so let the server choose.
+- **`addon_version` is unset**, so Spinifex installs its catalog default.
+  To pin a version, use the AWS form that `DescribeAddonVersions` returns, such as `v2.11.0-eksbuild.1`, or read it from `data.aws_eks_addon_version`.

 **Prerequisites:**

@@ -179,16 +179,17 @@ Confirm the addon is `ACTIVE` and the cluster carries `spinifex.io/managed-ingre

 ### HTTPS Loads but Returns 502 / 504

-The ALB reached the workers but the backend didn't answer. Check the demo pods and the NodePort SG rule:
+The ALB reached the workers but the backend didn't answer. Check the demo pods and the rule the controller added to the cluster security group:

 ```bash
 kubectl get pods -o wide
 kubectl get svc spinifex-demo
-aws ec2 describe-security-groups --filters "Name=group-name,Values=eks-cluster-eks-https-nodegroup-sg" \
+SG=$(aws eks describe-cluster --name eks-https --query 'cluster.resourcesVpcConfig.clusterSecurityGroupId' --output text)
+aws ec2 describe-security-groups --group-ids "$SG" \
  --query 'SecurityGroups[0].IpPermissions'
 ```

-The `eks-https-alb-nodeport` rule must admit the VPC CIDR on the NodePort, and the target group health check (`/healthz`) must pass.
+The cluster security group must admit the load balancer's security group on the NodePort, and the target group health check (`/healthz`) must pass.

 ### Addon Fails With "no baked bundle"

@@ -16,7 +16,8 @@
 # The workloads/ module then creates a Deployment, a NodePort Service, and an
 # Ingress (ingressClassName: alb). The LBC reconciles that Ingress into an
 # internet-facing ALB that terminates TLS with the ACM cert and forwards to the
-# workers' NodePort. The cluster's ELB-eligible subnets are injected by Spinifex
+# workers' NodePort, adding the ALB-to-NodePort rule to the cluster security group
+# itself. The cluster's ELB-eligible subnets are injected by Spinifex
 # into the alb IngressClassParams, so the Ingress needs no subnet annotations.
 #
 # Usage:
@@ -446,8 +447,8 @@ resource "aws_eks_node_group" "workers" {
 #
 # Spinifex wires the controller's AWS credentials and ELB-eligible subnets at the
 # node level, so no IRSA role or subnet tagging is needed here. addon_version is
-# omitted: the AWS provider demands a v-prefixed version that the catalog rejects,
-# so let Spinifex default to its catalog version.
+# omitted so Spinifex installs its catalog default. To pin one, use the AWS form
+# from DescribeAddonVersions, for example v2.11.0-eksbuild.1.
 # ---------------------------------------------------------------------------

 resource "aws_eks_addon" "lbc" {
@@ -503,40 +504,6 @@ resource "aws_acm_certificate" "ingress" {
  }
 }

-# ---------------------------------------------------------------------------
-# Let the ALB reach the workers' NodePort
-#
-# Spinifex auto-manages the nodegroup SG and admits only intra-cluster traffic.
-# The LBC-provisioned ALB lives in this VPC, so admit the VPC CIDR on the
-# NodePort. Look the SG up by its deterministic name and add one rule.
-# ---------------------------------------------------------------------------
-
-data "aws_security_group" "nodegroup" {
-  filter {
-    name   = "group-name"
-    values = ["eks-cluster-${var.cluster_name}-nodegroup-sg"]
-  }
-
-  filter {
-    name   = "vpc-id"
-    values = [aws_vpc.main.id]
-  }
-
-  depends_on = [aws_eks_node_group.workers]
-}
-
-resource "aws_vpc_security_group_ingress_rule" "nodeport_from_vpc" {
-  security_group_id = data.aws_security_group.nodegroup.id
-  cidr_ipv4         = aws_vpc.main.cidr_block
-  from_port         = var.node_port
-  to_port           = var.node_port
-  ip_protocol       = "tcp"
-
-  tags = {
-    Name = "${var.cluster_name}-alb-nodeport"
-  }
-}
-
 # ---------------------------------------------------------------------------
 # Outputs
 # ---------------------------------------------------------------------------
@@ -45,7 +45,7 @@ Under the hood it keeps things minimal: a VPC with two public subnets, an IAM ro
 - Creating the EKS cluster and worker IAM roles with faithful managed-policy attachments
 - Provisioning an `aws_eks_cluster` and a managed `aws_eks_node_group`
 - Pointing the Kubernetes provider at the new cluster and deploying a workload from the same `apply`
- Opening one rule on the auto-managed worker security group to expose a NodePort
+- Opening one rule on the cluster security group to expose a NodePort

 **What gets created**

@@ -58,13 +58,13 @@ Under the hood it keeps things minimal: a VPC with two public subnets, an IAM ro
 | ECR Repository | `spinifex-demo` | Holds the demo image the workers pull |
 | EKS Cluster | `eks-quickstart` | Public API endpoint, API auth mode, Kubernetes 1.32 |
 | Node Group | `default` | `node_desired_size` `t3.medium` worker(s) — 1 or 3 |
-| SG Ingress Rule | `eks-quickstart-demo-nodeport` | Opens the demo NodePort on the auto-managed worker SG |
+| SG Ingress Rule | `eks-quickstart-demo-nodeport` | Opens the demo NodePort on the cluster security group |
 | K8s Deployment | `spinifex-demo` | The themed demo image from ECR, 2 replicas |
 | K8s Service | `spinifex-demo` | NodePort publishing the demo on the worker |

 **Spinifex specifics**

- The cluster's security groups are **auto-managed** — `vpc_config.security_group_ids` is ignored. To reach a NodePort, this workbook looks the worker SG up by its deterministic name (`eks-cluster-<name>-nodegroup-sg`) and adds a single ingress rule.
+- The cluster security group is **auto-managed**, as on EKS: Spinifex creates `eks-cluster-sg-<name>-<id>`, attaches it to the workers, and reports it as `vpc_config[0].cluster_security_group_id`. `vpc_config.security_group_ids` is ignored. To reach a NodePort, this workbook adds a single ingress rule to that security group.
 - The worker AMI is always Spinifex's `eks-node` image; `ami_type` is recorded but does **not** select the image.
 - `authentication_mode` can be `"API"` or `"API_AND_CONFIG_MAP"`, and bare `"CONFIG_MAP"` is rejected. Access is granted through access entries, as there is no `aws-auth` ConfigMap path.
 - The workers need outbound internet (here via the IGW) to pull the demo container image.
@@ -182,8 +182,9 @@ aws ec2 describe-instances --filters "Name=tag:spinifex:eks-cluster,Values=eks-q
 # Did the demo pods roll out?
 kubectl get pods -o wide

-# Is the NodePort rule on the worker SG?
-aws ec2 describe-security-groups --filters "Name=group-name,Values=eks-cluster-eks-quickstart-nodegroup-sg" \
+# Is the NodePort rule on the cluster security group?
+SG=$(tofu output -raw cluster_security_group_id)
+aws ec2 describe-security-groups --group-ids "$SG" \
  --query 'SecurityGroups[0].IpPermissions'
 ```

@@ -310,29 +310,15 @@ resource "aws_eks_node_group" "default" {
 }

 # ---------------------------------------------------------------------------
-# Open the NodePort on the auto-managed nodegroup SG
+# Open the NodePort on the cluster security group
 #
-# Spinifex creates the worker SG itself (vpc_config.security_group_ids is
-# ignored) and admits only intra-cluster traffic. To reach the demo NodePort
-# from a browser, look the SG up by its deterministic name and add one rule.
+# Spinifex creates the cluster security group itself, as EKS does, and attaches
+# it to the workers. It admits only intra-cluster traffic. To reach the demo
+# NodePort from a browser, add one rule to the SG the cluster reports.
 # ---------------------------------------------------------------------------

-data "aws_security_group" "nodegroup" {
-  filter {
-    name   = "group-name"
-    values = ["eks-cluster-${var.cluster_name}-nodegroup-sg"]
-  }
-
-  filter {
-    name   = "vpc-id"
-    values = [aws_vpc.main.id]
-  }
-
-  depends_on = [aws_eks_node_group.default]
-}
-
 resource "aws_vpc_security_group_ingress_rule" "nodeport" {
-  security_group_id = data.aws_security_group.nodegroup.id
+  security_group_id = aws_eks_cluster.this.vpc_config[0].cluster_security_group_id
  cidr_ipv4         = var.browse_cidr
  from_port         = var.node_port
  to_port           = var.node_port
@@ -367,6 +353,10 @@ output "region" {
  value = var.region
 }

+output "cluster_security_group_id" {
+  value = aws_eks_cluster.this.vpc_config[0].cluster_security_group_id
+}
+
 output "node_port" {
  value = var.node_port
 }
@@ -123,9 +123,13 @@ Internal and internet-facing ALB and NLB, HTTPS listeners, listener rules, `Modi
 Unhealthy-target routing: an NLB with an HTTP health check on a TCP target group stops sending traffic to a target that fails it while still accepting TCP, sends traffic again once it recovers, and fails open when every target is unhealthy.
 An ALB whose targets all fail their health check keeps serving from them instead of answering 503.

+`TestInternalALBPrivateSubnet` launches an internal ALB in a private subnet of a non-default VPC with no internet gateway and no public route.
+The load balancer must reach active with a healthy target, which needs the owner stamp on its ENI to resolve, and a client inside the VPC must be served by it on every request.
+The client reports on its serial console because the runner has no route into the VPC.
+
 ### `eks`, `ecs`, `ecr` — container services

-`eks`: cluster create/delete, access entries, addon delivery, `GetToken`, kubeconfig artifacts, the EBS CSI volume path, an IRSA web-identity exchange under the stock AWS trust policy (`StringEquals` on the issuer host-path `:sub`/`:aud`, a mismatched `:sub` denied with `AccessDenied`, a `StringLike` `:sub` granted), and an IRSA pod wired with a projected SA token. Worker IMDS is covered too: a managed nodegroup worker reports IMDSv2 required with hop limit 2, a pod with no IRSA resolves to the node role through IMDS, and a nodegroup built from a launch template takes the template's metadata options. Control-plane isolation is covered as well: no control-plane host registers as a Node, a DaemonSet tolerating every taint runs on workers only, metrics-server is Ready, and logs work for every kube-system pod. `ecs`: clusters, container instances, the three network modes (`awsvpc`, `bridge`, `host`), task-role credentials fetched from the container credentials endpoint, a service behind an ELB, and ENI release on terminate. `ecr`: control-plane CRUD plus a real OCI data plane — docker login/build/push/pull round trips, multi-layer images, and raw blob HTTP.
+`eks`: cluster create/delete, access entries, addon delivery, `GetToken`, kubeconfig artifacts, the EBS CSI volume path, an IRSA web-identity exchange under the stock AWS trust policy (`StringEquals` on the issuer host-path `:sub`/`:aud`, a mismatched `:sub` denied with `AccessDenied`, a `StringLike` `:sub` granted), and an IRSA pod whose spec names only a ServiceAccount with the role-arn annotation, where the pod-identity webhook injects the token volume, `AWS_*` environment, STS endpoint and CA bundle and the pod assumes the role, while a pod of an unannotated ServiceAccount is left untouched. The unmodified aws-load-balancer-controller chart 1.11.0 installs with the values an AWS user passes plus the gateway endpoint override, the Shield and WAF switches and the single-subnet feature gate, and reconciles an Ingress into an active ALB. Worker IMDS is covered too: a managed nodegroup worker reports IMDSv2 required with hop limit 2, a pod with no IRSA resolves to the node role through IMDS, and a nodegroup built from a launch template takes the template's metadata options. Control-plane isolation is covered as well: no control-plane host registers as a Node, a DaemonSet tolerating every taint runs on workers only, metrics-server is Ready, and logs work for every kube-system pod. `ecs`: clusters, container instances, the three network modes (`awsvpc`, `bridge`, `host`), task-role credentials fetched from the container credentials endpoint, a service behind an ELB, and ENI release on terminate. `ecr`: control-plane CRUD plus a real OCI data plane — docker login/build/push/pull round trips, multi-layer images, and raw blob HTTP.

 ### `rds` — managed databases

@@ -116,7 +116,7 @@ The Go console serves `script-src 'self'; style-src 'self'`.
 ## Primitives

 `src/components/ui/` holds the shadcn primitives on Base UI.
-Added in Phase 1: breadcrumb, checkbox, command (`cmdk`), dialog, hover-card, popover, resizable (`react-resizable-panels`), scroll-area, slider, switch, table, toggle and toggle-group.
+Added in Phase 1: breadcrumb, checkbox, command (Base UI Autocomplete), dialog, hover-card, popover, resizable (`react-resizable-panels`), scroll-area, slider, switch, table, toggle and toggle-group.
 `button-group` joins buttons edge to edge, as the instance power buttons are.
 `radio-group` is one of a few exclusive choices inside a form, such as a load balancer's type; a raw `<input type="radio">` is not used.

@@ -97,16 +97,9 @@ The deploy script, `deploy-vultr.sh`, is in the Spinifex repository:

 ```bash
 git clone https://github.com/mulgadc/spinifex.git
-cd spinifex
-git switch dev
-cd scripts/terraform/vultr-spx
+cd spinifex/scripts/terraform/vultr-spx
 ```

-> [!NOTE]
-> Vultr support is on the development branch and not yet in a published release.
-> `git switch dev` gives you the deploy script for it, and `--channel dev` in [step 3](#3-deploy) installs the matching Spinifex build, the pre-release published from every change to `dev`.
-> Once Vultr support is in a release, skip `git switch dev` and pass `--channel latest`.
-
 ## 2. Choose a deployment

 | Use | Nodes | Plan | Cost while running |
@@ -143,14 +136,14 @@ Set your API key, then run the command for the deployment you chose:
 export VULTR_API_KEY=<your key>

 # Development and testing: one VM
-./deploy-vultr.sh up --channel dev --plan vx1-g-16c-64g
+./deploy-vultr.sh up --plan vx1-g-16c-64g

 # Production: three VMs
-./deploy-vultr.sh up --channel dev --plan vx1-g-16c-64g --nodes 3
+./deploy-vultr.sh up --plan vx1-g-16c-64g --nodes 3

 # Production on bare metal: one server, or three
-./deploy-vultr.sh up --channel dev --plan vbm-8c-64gb-amd
-./deploy-vultr.sh up --channel dev --plan vbm-8c-64gb-amd --nodes 3
+./deploy-vultr.sh up --plan vbm-8c-64gb-amd
+./deploy-vultr.sh up --plan vbm-8c-64gb-amd --nodes 3
 ```

 One VM takes about 4 minutes and three VMs about 16 minutes.
@@ -4,15 +4,15 @@ go 1.27.3

 require (
 	github.com/aws/aws-sdk-go v1.55.8
-	github.com/aws/aws-sdk-go-v2 v1.47.4
-	github.com/aws/aws-sdk-go-v2/aws/protocol/eventstream v1.7.24
-	github.com/aws/aws-sdk-go-v2/config v1.33.9
-	github.com/aws/aws-sdk-go-v2/credentials v1.20.9
-	github.com/aws/aws-sdk-go-v2/feature/ec2/imds v1.20.3
-	github.com/aws/aws-sdk-go-v2/service/lambda v1.113.3
-	github.com/aws/aws-sdk-go-v2/service/s3 v1.114.4
-	github.com/aws/aws-sdk-go-v2/service/sqs v1.52.6
-	github.com/aws/smithy-go v1.28.6
+	github.com/aws/aws-sdk-go-v2 v1.47.5
+	github.com/aws/aws-sdk-go-v2/aws/protocol/eventstream v1.7.25
+	github.com/aws/aws-sdk-go-v2/config v1.33.12
+	github.com/aws/aws-sdk-go-v2/credentials v1.20.12
+	github.com/aws/aws-sdk-go-v2/feature/ec2/imds v1.20.5
+	github.com/aws/aws-sdk-go-v2/service/lambda v1.113.4
+	github.com/aws/aws-sdk-go-v2/service/s3 v1.114.6
+	github.com/aws/aws-sdk-go-v2/service/sqs v1.52.7
+	github.com/aws/smithy-go v1.28.7
 	github.com/cenkalti/backoff/v4 v4.3.0
 	github.com/charmbracelet/bubbles v1.0.0
 	github.com/charmbracelet/bubbletea v1.3.10
@@ -31,8 +31,8 @@ require (
 	github.com/mulgadc/northstar v1.23.0
 	github.com/mulgadc/predastore v1.23.0
 	github.com/mulgadc/viperblock v1.23.0
-	github.com/nats-io/nats-server/v2 v2.15.1
-	github.com/nats-io/nats.go v1.54.0
+	github.com/nats-io/nats-server/v2 v2.15.2
+	github.com/nats-io/nats.go v1.55.0
 	github.com/nats-io/nkeys v0.4.16
 	github.com/opencontainers/runtime-spec v1.3.0
 	github.com/oracle/oci-go-sdk/v65 v65.128.0
@@ -43,9 +43,9 @@ require (
 	github.com/spf13/pflag v1.0.10
 	github.com/spf13/viper v1.21.0
 	github.com/stretchr/testify v1.12.1
-	go.etcd.io/etcd/api/v3 v3.5.35
-	go.etcd.io/etcd/client/pkg/v3 v3.5.35
-	go.etcd.io/etcd/client/v3 v3.5.35
+	go.etcd.io/etcd/api/v3 v3.7.3
+	go.etcd.io/etcd/client/pkg/v3 v3.7.3
+	go.etcd.io/etcd/client/v3 v3.7.3
 	go.opentelemetry.io/contrib/instrumentation/net/http/otelhttp v0.72.0
 	go.opentelemetry.io/otel v1.47.0
 	go.opentelemetry.io/otel/metric v1.47.0
@@ -53,14 +53,14 @@ require (
 	go.opentelemetry.io/otel/sdk/metric v1.47.0
 	go.opentelemetry.io/otel/trace v1.47.0
 	go.uber.org/goleak v1.3.1-0.20260602155401-5f791e3ef83b
-	go.yaml.in/yaml/v3 v3.0.5
+	go.yaml.in/yaml/v3 v3.0.6
 	golang.org/x/crypto v0.58.0
 	golang.org/x/sync v0.24.0
 	golang.org/x/sys v0.49.0
 	golang.org/x/term v0.47.0
 	golang.org/x/time v0.16.0
 	golang.org/x/tools v0.52.0
-	google.golang.org/grpc v1.83.2
+	google.golang.org/grpc v1.85.0
 	gopkg.in/ini.v1 v1.67.3
 )

@@ -72,17 +72,17 @@ require (
 	github.com/Microsoft/hcsshim v0.15.0-rc.4 // indirect
 	github.com/antithesishq/antithesis-sdk-go v0.8.0-default-no-op // indirect
 	github.com/atotto/clipboard v0.1.4 // indirect
-	github.com/aws/aws-sdk-go-v2/internal/configsources v1.5.7 // indirect
-	github.com/aws/aws-sdk-go-v2/internal/endpoints/v2 v2.8.7 // indirect
-	github.com/aws/aws-sdk-go-v2/internal/v4a v1.5.6 // indirect
-	github.com/aws/aws-sdk-go-v2/service/internal/accept-encoding v1.13.21 // indirect
-	github.com/aws/aws-sdk-go-v2/service/internal/checksum v1.11.7 // indirect
-	github.com/aws/aws-sdk-go-v2/service/internal/presigned-url v1.14.6 // indirect
-	github.com/aws/aws-sdk-go-v2/service/internal/s3shared v1.20.6 // indirect
-	github.com/aws/aws-sdk-go-v2/service/signin v1.10.4 // indirect
-	github.com/aws/aws-sdk-go-v2/service/sso v1.38.4 // indirect
-	github.com/aws/aws-sdk-go-v2/service/ssooidc v1.43.4 // indirect
-	github.com/aws/aws-sdk-go-v2/service/sts v1.51.4 // indirect
+	github.com/aws/aws-sdk-go-v2/internal/configsources v1.5.8 // indirect
+	github.com/aws/aws-sdk-go-v2/internal/endpoints/v2 v2.8.8 // indirect
+	github.com/aws/aws-sdk-go-v2/internal/v4a v1.5.8 // indirect
+	github.com/aws/aws-sdk-go-v2/service/internal/accept-encoding v1.13.23 // indirect
+	github.com/aws/aws-sdk-go-v2/service/internal/checksum v1.11.9 // indirect
+	github.com/aws/aws-sdk-go-v2/service/internal/presigned-url v1.14.8 // indirect
+	github.com/aws/aws-sdk-go-v2/service/internal/s3shared v1.20.8 // indirect
+	github.com/aws/aws-sdk-go-v2/service/signin v1.12.0 // indirect
+	github.com/aws/aws-sdk-go-v2/service/sso v1.39.0 // indirect
+	github.com/aws/aws-sdk-go-v2/service/ssooidc v1.44.0 // indirect
+	github.com/aws/aws-sdk-go-v2/service/sts v1.51.6 // indirect
 	github.com/aymanbagabas/go-osc52/v2 v2.0.1 // indirect
 	github.com/beorn7/perks v1.0.1 // indirect
 	github.com/boltdb/bolt v1.3.1 // indirect
@@ -111,7 +111,7 @@ require (
 	github.com/containerd/plugin v1.1.0 // indirect
 	github.com/containerd/ttrpc v1.2.9 // indirect
 	github.com/containerd/typeurl/v2 v2.3.0 // indirect
-	github.com/coreos/go-semver v0.3.0 // indirect
+	github.com/coreos/go-semver v0.3.1 // indirect
 	github.com/coreos/go-systemd/v22 v22.7.0 // indirect
 	github.com/dgraph-io/badger/v4 v4.9.6 // indirect
 	github.com/dgraph-io/ristretto/v2 v2.4.0 // indirect
@@ -128,7 +128,6 @@ require (
 	github.com/go-playground/validator/v10 v10.30.5 // indirect
 	github.com/go-viper/mapstructure/v2 v2.5.0 // indirect
 	github.com/gofrs/flock v0.10.0 // indirect
-	github.com/gogo/protobuf v1.3.2 // indirect
 	github.com/golang/protobuf v1.5.4 // indirect
 	github.com/google/flatbuffers v25.12.19+incompatible // indirect
 	github.com/google/go-tpm v0.9.8 // indirect
@@ -175,7 +174,6 @@ require (
 	github.com/opencontainers/image-spec v1.1.1 // indirect
 	github.com/pierrec/lz4/v4 v4.1.18 // indirect
 	github.com/pkg/errors v0.9.1 // indirect
-	github.com/pmezard/go-difflib v1.0.1-0.20181226105442-5d4384ee4fb2 // indirect
 	github.com/prometheus/client_golang v1.24.1 // indirect
 	github.com/prometheus/client_model v0.6.3 // indirect
 	github.com/prometheus/common v0.71.0 // indirect
@@ -203,10 +201,9 @@ require (
 	go.opentelemetry.io/otel/log v1.47.0 // indirect
 	go.opentelemetry.io/otel/sdk/log v1.47.0 // indirect
 	go.opentelemetry.io/proto/otlp v1.11.0 // indirect
-	go.uber.org/atomic v1.7.0 // indirect
-	go.uber.org/multierr v1.6.0 // indirect
+	go.uber.org/multierr v1.11.0 // indirect
 	go.uber.org/nilaway v0.0.0-20260721205819-4a0f7653488a // indirect
-	go.uber.org/zap v1.17.0 // indirect
+	go.uber.org/zap v1.27.1 // indirect
 	golang.org/x/exp/typeparams v0.0.0-20260611194520-c48552f49976 // indirect
 	golang.org/x/mod v0.42.0 // indirect
 	golang.org/x/net v0.61.0 // indirect
@@ -22,46 +22,46 @@ github.com/atotto/clipboard v0.1.4 h1:EH0zSVneZPSuFR11BlR9YppQTVDbh5+16AmcJi4g1z
 github.com/atotto/clipboard v0.1.4/go.mod h1:ZY9tmq7sm5xIbd9bOK4onWV4S6X0u6GY7Vn0Yu86PYI=
 github.com/aws/aws-sdk-go v1.55.8 h1:JRmEUbU52aJQZ2AjX4q4Wu7t4uZjOu71uyNmaWlUkJQ=
 github.com/aws/aws-sdk-go v1.55.8/go.mod h1:ZkViS9AqA6otK+JBBNH2++sx1sgxrPKcSzPPvQkUtXk=
-github.com/aws/aws-sdk-go-v2 v1.47.4 h1:bTvk6UdSSSvaQsfLVj1JRDEEqj/PDRwGf1KanaRzau0=
-github.com/aws/aws-sdk-go-v2 v1.47.4/go.mod h1:XkrjAZvr705Nytv6PkPUhoSB2W9+vVHBp5oAWpkh8Ak=
-github.com/aws/aws-sdk-go-v2/aws/protocol/eventstream v1.7.24 h1:t7FxqJ5KHDeKceyD4HtGaG6z7hMSeVGjs5/4c+L5ArM=
-github.com/aws/aws-sdk-go-v2/aws/protocol/eventstream v1.7.24/go.mod h1:5XK6NRVlK8s8yj4zKcM43HLluGA/rAz+azLLvWDMRr0=
-github.com/aws/aws-sdk-go-v2/config v1.33.9 h1:Hv6bcQUJVnmuWaKjqD6Ot8sE75NrWsWV20oGT5aWle0=
-github.com/aws/aws-sdk-go-v2/config v1.33.9/go.mod h1:ivowmrZNfBQLcg4QkaItgi5wae4Bzot9RzIdbY1hfTg=
-github.com/aws/aws-sdk-go-v2/credentials v1.20.9 h1:SNO55PnWhaDafsOcv16lCA0kdoFMl/YlegWoVsU02s0=
-github.com/aws/aws-sdk-go-v2/credentials v1.20.9/go.mod h1:+2reOTRtdD3VSBqpRDNwrZIYy8yMCYa7oK3WKTJ/I1I=
-github.com/aws/aws-sdk-go-v2/feature/ec2/imds v1.20.3 h1:gr3b87OLB29jymWPzp/CAyXwQRJVE+sC4Wm0qBIKHsk=
-github.com/aws/aws-sdk-go-v2/feature/ec2/imds v1.20.3/go.mod h1:BRjf4L/Yx2hY54o+ZibPbLJyMMeSwqYQXwcJfqW66a8=
-github.com/aws/aws-sdk-go-v2/internal/configsources v1.5.7 h1:fSNAt6hs4qLjb9tMZc6vj3PHlOCYGXK2JjJZ7AW2tf4=
-github.com/aws/aws-sdk-go-v2/internal/configsources v1.5.7/go.mod h1:TMav+EuXQu5kNpZZz/6rIC8xx4mCsW+6AKHrjvOSAsE=
-github.com/aws/aws-sdk-go-v2/internal/endpoints/v2 v2.8.7 h1:7NS6kqBjkWIN8P/AEdmtyq6eQGAhN1oHTZdzXclpH6Y=
-github.com/aws/aws-sdk-go-v2/internal/endpoints/v2 v2.8.7/go.mod h1:1PFqtAUsIgF1nEVojzRJSf/v88zjzqrGW+w+UAtkXyQ=
-github.com/aws/aws-sdk-go-v2/internal/v4a v1.5.6 h1:Qmn60RWuk9+bd2ddvJJqwf3Cpac3VvGMOdbGemtUhX4=
-github.com/aws/aws-sdk-go-v2/internal/v4a v1.5.6/go.mod h1:h/YJhe4AVyQxEUMluph10iqV7mjajH7A9L+BHF5ppis=
-github.com/aws/aws-sdk-go-v2/service/internal/accept-encoding v1.13.21 h1:e6YW5uME30TkrElR+4/xcY7NB6CCj2dcv7vtmGXhfhQ=
-github.com/aws/aws-sdk-go-v2/service/internal/accept-encoding v1.13.21/go.mod h1:obpHAMP693+f3/A8PkLhIqfK1C+NWWn8zXzqD3OH520=
-github.com/aws/aws-sdk-go-v2/service/internal/checksum v1.11.7 h1:CwvFlKw0uUScfgOMflXSRZa+ayXgOf6wk7lhyMemQYU=
-github.com/aws/aws-sdk-go-v2/service/internal/checksum v1.11.7/go.mod h1:4SbD8BB+p+QEnG0aX6DoWwrtk9SufyYXRsUv0nve8LE=
-github.com/aws/aws-sdk-go-v2/service/internal/presigned-url v1.14.6 h1:LHHdobjo5le1sjJdolitV7uVvYg3wkaeM1Ju2fDgeZQ=
-github.com/aws/aws-sdk-go-v2/service/internal/presigned-url v1.14.6/go.mod h1:CBGB0evAiaMD3Lk4wPO7LQCcCZLuhzL5Qf/wfEUHQ5w=
-github.com/aws/aws-sdk-go-v2/service/internal/s3shared v1.20.6 h1:mYViDr2azcgxOpRrObZOl631bvavqILoE/L+Kxgbwys=
-github.com/aws/aws-sdk-go-v2/service/internal/s3shared v1.20.6/go.mod h1:OXVdAo9jNhggB+hJRGOyy7WolGTKB6Rg7IIulDeSzPM=
-github.com/aws/aws-sdk-go-v2/service/lambda v1.113.3 h1:oSHldse7GDHpdq81XeAJnhIbsdDKP9bMIt00G8W6uXI=
-github.com/aws/aws-sdk-go-v2/service/lambda v1.113.3/go.mod h1:Y//K046w4uhiR3o77beo/y52Jv2uEY13iJ+3+jicLO8=
-github.com/aws/aws-sdk-go-v2/service/s3 v1.114.4 h1:MrkWRB58Qw5TEkb51QVooVLjlM3IevO2iVXEz74B8pg=
-github.com/aws/aws-sdk-go-v2/service/s3 v1.114.4/go.mod h1:yfIxaKRHasvfy6tNEm3/LpqKsupMzQjMaWj9Y0jMVbo=
-github.com/aws/aws-sdk-go-v2/service/signin v1.10.4 h1:tvpDjuUvgKURf3Yk4hOaGbZ4QdY+070OWcA5d6LUv2A=
-github.com/aws/aws-sdk-go-v2/service/signin v1.10.4/go.mod h1:3ibVKg5Ng82Zby8++jELTD2nHxiPk2DJRLEeW/LKp8M=
-github.com/aws/aws-sdk-go-v2/service/sqs v1.52.6 h1:Suq1LNhMBpsmShEv3vO+J3H8HEeK+GwYMJDYLE4IJSs=
-github.com/aws/aws-sdk-go-v2/service/sqs v1.52.6/go.mod h1:rHGuzFg4AfVnWZa3/533tn5GQd1LzVkTXI9WMT8LqAc=
-github.com/aws/aws-sdk-go-v2/service/sso v1.38.4 h1:ChCtagixSFOxltv0o6E3NT8F7ksvZtZGPwUBynO5Nmk=
-github.com/aws/aws-sdk-go-v2/service/sso v1.38.4/go.mod h1:+FEO9jrVMpc9oN3CdAxJbtlQEu3z8UddDPSedpeM3PY=
-github.com/aws/aws-sdk-go-v2/service/ssooidc v1.43.4 h1:j6c+HA4AoS0EBEjArPjavFEBrucJEryfrQU0TffmACk=
-github.com/aws/aws-sdk-go-v2/service/ssooidc v1.43.4/go.mod h1:RhghaXUk9q66VtOQUX1FdPddQuHnqQZzbB0fEYXmPRA=
-github.com/aws/aws-sdk-go-v2/service/sts v1.51.4 h1:OKEfbf2tcfsF2pBk/S9QqNpo6alaAtiBs4hysCsBjPw=
-github.com/aws/aws-sdk-go-v2/service/sts v1.51.4/go.mod h1:mCd6qGo9b8hTtfjgsmNkFlzwjyce4owd7HDVFnCLXhM=
-github.com/aws/smithy-go v1.28.6 h1:Pj+/sUviNYjow8ELRtNwKHZ9ECeF+7pATX3+shjLIrw=
-github.com/aws/smithy-go v1.28.6/go.mod h1:YE2RhdIuDbA5E5bTdciG9KrW3+TiEONeUWCqxX9i1Fc=
+github.com/aws/aws-sdk-go-v2 v1.47.5 h1:5yA+A+BiXbDq1ZLbYwPViwhEGDMdR0MEhL8Jh2LnLS4=
+github.com/aws/aws-sdk-go-v2 v1.47.5/go.mod h1:iNIkogrRvFwm+1OQIUOP2lPBtlXZboOIUWiCa+pHEMw=
+github.com/aws/aws-sdk-go-v2/aws/protocol/eventstream v1.7.25 h1:MW0Iql1SOTgQPj8mCxrU/eH/K/AIDWi08ItOsXaWRLU=
+github.com/aws/aws-sdk-go-v2/aws/protocol/eventstream v1.7.25/go.mod h1:+ucVYmHYB3JqVueY0E1RHahdXAnKQbIHqybSbvOUbOM=
+github.com/aws/aws-sdk-go-v2/config v1.33.12 h1:IV1YARaXahwCqUIyP8IlWsP85Lil+5ozTozDQXPumRk=
+github.com/aws/aws-sdk-go-v2/config v1.33.12/go.mod h1:h51R06bzMVwNBRZOX5ylHhNDnkB/azZeq/LvxM6q2pQ=
+github.com/aws/aws-sdk-go-v2/credentials v1.20.12 h1:6P7Q7yH/GIwHg6MLKv3ySsBqaCoE1z+C0OTzYMenhEY=
+github.com/aws/aws-sdk-go-v2/credentials v1.20.12/go.mod h1:Kw/SaoW/wc7IqJErqXJIzKo8b7Aelxfbuc6XlbrRnsc=
+github.com/aws/aws-sdk-go-v2/feature/ec2/imds v1.20.5 h1:iXjy2SFhGYNB1jWZBhQqq8HbhsbPF23w+/zbMvPXsTg=
+github.com/aws/aws-sdk-go-v2/feature/ec2/imds v1.20.5/go.mod h1:k0RUiBnyXU0AXRJ9tJmY2jQke6XUIX2PNVnWtvgaPtc=
+github.com/aws/aws-sdk-go-v2/internal/configsources v1.5.8 h1:gzIWc3DTPfPbWKrQnmL6BV6WD7NPD1htxzgdk/bQ/Z4=
+github.com/aws/aws-sdk-go-v2/internal/configsources v1.5.8/go.mod h1:vvUA5cDAZWGo4R14FZPZfXfJCmgsafOWEq78Kg0D+ug=
+github.com/aws/aws-sdk-go-v2/internal/endpoints/v2 v2.8.8 h1:PWv+vG8aCLS8qd+HPdYhK7Vees90aKHWFFQ1AGYD/8E=
+github.com/aws/aws-sdk-go-v2/internal/endpoints/v2 v2.8.8/go.mod h1:xxPZEpflYHJA12366ZGyq66z9dHPCoeCIu99Jck+MPw=
+github.com/aws/aws-sdk-go-v2/internal/v4a v1.5.8 h1:WOp5De9iWdEHDPh2Kb0tBP3vGBm/dVAuRGh4a4tqduo=
+github.com/aws/aws-sdk-go-v2/internal/v4a v1.5.8/go.mod h1:z8W8YgfhbWl62XqcTtK+CxPrE+iHGRVUOtrrbIOp7ZY=
+github.com/aws/aws-sdk-go-v2/service/internal/accept-encoding v1.13.23 h1:MCIR6iZaCKO4UE0hUvoWWLI6JVjppopV64JfAJaMjng=
+github.com/aws/aws-sdk-go-v2/service/internal/accept-encoding v1.13.23/go.mod h1:pphWFS+z0MmIPivj8c/6bATd1Z+Av7Ust75DPdEFsPA=
+github.com/aws/aws-sdk-go-v2/service/internal/checksum v1.11.9 h1:GACgECI422Xt5n4Pojr3O0aocPv8lfH4PJe2Aon5Wv8=
+github.com/aws/aws-sdk-go-v2/service/internal/checksum v1.11.9/go.mod h1:lPfnRkHpn1qWZJLbVEw10rIqSuCee1pNQTyGkUjAYh8=
+github.com/aws/aws-sdk-go-v2/service/internal/presigned-url v1.14.8 h1:+PMpWlJYOn2tm7sm6OZ/ZLd4jStl17T7wjNapd1Rczs=
+github.com/aws/aws-sdk-go-v2/service/internal/presigned-url v1.14.8/go.mod h1:cnG3xZLGQL4c/bt8pjdzlFtZsW7epBBuNS04a5xMjdo=
+github.com/aws/aws-sdk-go-v2/service/internal/s3shared v1.20.8 h1:RS1Wt+K0r4X0ci9Vf4JYTUfHzsQihYA2LgGJeEneKpk=
+github.com/aws/aws-sdk-go-v2/service/internal/s3shared v1.20.8/go.mod h1:adMnTy5kyoUW86jhXnGSVc/mvMQI4GZd1qYWMeaS8M8=
+github.com/aws/aws-sdk-go-v2/service/lambda v1.113.4 h1:wvAdq1sNfwuedO1BX8GjY+oZWzoyeD4hhdjJI+4EPnI=
+github.com/aws/aws-sdk-go-v2/service/lambda v1.113.4/go.mod h1:HgkYN4d72Fawg3MAJQepmNnROFoN11XYu14hgFfChTY=
+github.com/aws/aws-sdk-go-v2/service/s3 v1.114.6 h1:to7+Vr3rLkpYRwVKJ5QRriI0eE306vrVN/6TnVHgxvY=
+github.com/aws/aws-sdk-go-v2/service/s3 v1.114.6/go.mod h1:Jc89ASa2k7HMAskj3bSnt8bf+k+p2IApzkoXnEjkgUw=
+github.com/aws/aws-sdk-go-v2/service/signin v1.12.0 h1:3vu/eUqh0rjh7Zy4N4YKcpZX4Aud+zinB3VsLhhlQFw=
+github.com/aws/aws-sdk-go-v2/service/signin v1.12.0/go.mod h1:wxAk0D2oVBSjuXEIzvcKxTZfM90zticdAbh/VD42L0Y=
+github.com/aws/aws-sdk-go-v2/service/sqs v1.52.7 h1:Eip+mZ0wFZY/10e5w89lWhF09Pa8tbbGwxePP3T2iMc=
+github.com/aws/aws-sdk-go-v2/service/sqs v1.52.7/go.mod h1:bXp38eBNyvSevFIao60mYcC3pQIAkPYWOddhFVv9z8E=
+github.com/aws/aws-sdk-go-v2/service/sso v1.39.0 h1:kIEpD5dUry7VET3Ybl7zj7PPHRB4lb52G9iTz+a1Tqk=
+github.com/aws/aws-sdk-go-v2/service/sso v1.39.0/go.mod h1:UGsEYgWz4Gi+46vDQ5zkFLqUgJziyOE2SXjZjgmfmHc=
+github.com/aws/aws-sdk-go-v2/service/ssooidc v1.44.0 h1:WfFev+HgTodggcNon5HieiLxxwWwb8Br8R2+nDkixvk=
+github.com/aws/aws-sdk-go-v2/service/ssooidc v1.44.0/go.mod h1:gJSdIY6YUzDvqQLcMtPVZgBRHrFwzHryvpE8Igzbxps=
+github.com/aws/aws-sdk-go-v2/service/sts v1.51.6 h1:BVjKrtRUtSYKztBCCfE1oAB+d3ViKcW2cKZfserOqMQ=
+github.com/aws/aws-sdk-go-v2/service/sts v1.51.6/go.mod h1:eSukFLeLkPMm+1OxDR+F0joauZdyBkF47Ow7MqjZiJw=
+github.com/aws/smithy-go v1.28.7 h1:h/n1SHrwVtaDYLz8mR1/Kq4Xh4RaFbFW7bOYckg9mfE=
+github.com/aws/smithy-go v1.28.7/go.mod h1:YE2RhdIuDbA5E5bTdciG9KrW3+TiEONeUWCqxX9i1Fc=
 github.com/aymanbagabas/go-osc52/v2 v2.0.1 h1:HwpRHbFMcZLEVr42D4p7XBqjyuxQH5SMiErDT4WkJ2k=
 github.com/aymanbagabas/go-osc52/v2 v2.0.1/go.mod h1:uYgXzlJ7ZpABp8OJ+exZzJJhRNQ2ASbcXHWsFqH8hp8=
 github.com/beorn7/perks v1.0.1 h1:VlbKKnNfV8bJzeqoa4cOKqO6bYr3WgKZxO8Z16+hsOM=
@@ -128,8 +128,8 @@ github.com/containerd/ttrpc v1.2.9 h1:ha0ak962T0s3CA/RoZ6S6xiWZQF24GrBaEpiGX1uih
 github.com/containerd/ttrpc v1.2.9/go.mod h1:jjtQRwXm4DL3KsHKW8vDiUOV6wO0hi6IPhmJhxU7aEs=
 github.com/containerd/typeurl/v2 v2.3.0 h1:HZHPhRWo5XMy3QGQoPrUzbW/2ckwjfweHmOwlkIrPAQ=
 github.com/containerd/typeurl/v2 v2.3.0/go.mod h1:Qk+PAdUYArVj41TnGi6rJ+48RF0PkcTc4i/taoBcK0w=
-github.com/coreos/go-semver v0.3.0 h1:wkHLiw0WNATZnSG7epLsujiMCgPAc9xhjJ4tgnAxmfM=
-github.com/coreos/go-semver v0.3.0/go.mod h1:nnelYz7RCh+5ahJtPPxZlU+153eP4D4r3EedlOD2RNk=
+github.com/coreos/go-semver v0.3.1 h1:yi21YpKnrx1gt5R+la8n5WgS0kCrsPp33dmEyHReZr4=
+github.com/coreos/go-semver v0.3.1/go.mod h1:irMmmIw/7yzSRPWryHsK7EYSg09caPQL03VsM8rvUec=
 github.com/coreos/go-systemd/v22 v22.7.0 h1:LAEzFkke61DFROc7zNLX/WA2i5J8gYqe0rSj9KI28KA=
 github.com/coreos/go-systemd/v22 v22.7.0/go.mod h1:xNUYtjHu2EDXbsxz1i41wouACIwT7Ybq9o0BQhMwD0w=
 github.com/cpuguy83/go-md2man/v2 v2.0.6/go.mod h1:oOW0eioCTA6cOiMLiUPZOpcVxMig6NIQQ7OS05n1F4g=
@@ -181,8 +181,6 @@ github.com/go-viper/mapstructure/v2 v2.5.0 h1:vM5IJoUAy3d7zRSVtIwQgBj7BiWtMPfmPE
 github.com/go-viper/mapstructure/v2 v2.5.0/go.mod h1:oJDH3BJKyqBA2TXFhDsKDGDTlndYOZ6rGS0BRZIxGhM=
 github.com/gofrs/flock v0.10.0 h1:SHMXenfaB03KbroETaCMtbBg3Yn29v4w1r+tgy4ff4k=
 github.com/gofrs/flock v0.10.0/go.mod h1:FirDy1Ing0mI2+kB6wk+vyyAH+e6xiE+EYA0jnzV9jc=
-github.com/gogo/protobuf v1.3.2 h1:Ov1cvc58UF3b5XjBnZv7+opcTcQFZebYjWzi34vdm4Q=
-github.com/gogo/protobuf v1.3.2/go.mod h1:P1XiOD3dCwIKUDQYPy72D8LYyHL2YPYrpS2s69NZV8Q=
 github.com/golang-jwt/jwt/v5 v5.3.1 h1:kYf81DTWFe7t+1VvL7eS+jKFVWaUnK9cB1qbwn63YCY=
 github.com/golang-jwt/jwt/v5 v5.3.1/go.mod h1:fxCRLWMO43lRc8nhHWY6LGqRcf+1gQWArsqaEUEa5bE=
 github.com/golang/protobuf v1.5.4 h1:i7eJL8qZTpSEXOPTxNKhASYpMn+8e5Q6AdndVa1dWek=
@@ -249,8 +247,6 @@ github.com/josharian/native v1.1.0 h1:uuaP0hAbW7Y4l0ZRQ6C9zfb7Mg1mbFKry/xzDAfmtL
 github.com/josharian/native v1.1.0/go.mod h1:7X/raswPFr05uY3HiLlYeyQntB6OO7E/d2Cu7qoaN2w=
 github.com/jstemmer/go-junit-report/v2 v2.1.0 h1:X3+hPYlSczH9IMIpSC9CQSZA0L+BipYafciZUWHEmsc=
 github.com/jstemmer/go-junit-report/v2 v2.1.0/go.mod h1:mgHVr7VUo5Tn8OLVr1cKnLuEy0M92wdRntM99h7RkgQ=
-github.com/kisielk/errcheck v1.5.0/go.mod h1:pFxgyoBC7bSaBwPgfKdkLd5X25qrDl4LWUI2bnpBCr8=
-github.com/kisielk/gotool v1.0.0/go.mod h1:XhKaO+MFFWcvkIS/tQcRk01m1F5IRFswLeQ+oQHNcck=
 github.com/klauspost/compress v1.20.3 h1:zWNoFqq/lSuYQRim3dUXlXeA0IqwuZGicdG5ZTfY5j4=
 github.com/klauspost/compress v1.20.3/go.mod h1:LUdAzn7YLVvxLpc7y3V1m40wESHTgc1422pwwBSKYuI=
 github.com/klauspost/cpuid/v2 v2.4.1 h1:I9KxadAkvo+QjtcEE2f2tyfEvmIla/T0m+14PrZNl0U=
@@ -317,10 +313,10 @@ github.com/munnerz/goautoneg v0.0.0-20191010083416-a7dc8b61c822 h1:C3w9PqII01/Oq
 github.com/munnerz/goautoneg v0.0.0-20191010083416-a7dc8b61c822/go.mod h1:+n7T8mK8HuQTcFwEeznm/DIxMOiR9yIdICNftLE1DvQ=
 github.com/nats-io/jwt/v2 v2.8.2 h1:XXRgB60MSTnqsRwejQurVDs/hcv2dkt+86GjI+I/bMc=
 github.com/nats-io/jwt/v2 v2.8.2/go.mod h1:Ag/56sq9OblL4JgdYufDd16Egb17Kr/8WwwuO/forVc=
-github.com/nats-io/nats-server/v2 v2.15.1 h1:fK3AeYFIji1Smng60jB0hVoAPZezA/UzsaI3BqXFNEI=
-github.com/nats-io/nats-server/v2 v2.15.1/go.mod h1:yeuYRtKVKIFtmMz2/YPk+6hvmNBnk+WwTUzErBdH6IE=
-github.com/nats-io/nats.go v1.54.0 h1:vsXoOxjHp/GmPUN+EcI7uOf/uB+iAP+kEsAFNQN0yzA=
-github.com/nats-io/nats.go v1.54.0/go.mod h1:y+DZoD1oBOYfZTU681eTUiUjI0vbqYGixNVFHcjHJ0k=
+github.com/nats-io/nats-server/v2 v2.15.2 h1:yLsgt2kslgymvxvjFq1WGWck5hkat2YYfbXvQ145tzU=
+github.com/nats-io/nats-server/v2 v2.15.2/go.mod h1:VXHCZgDglV5wO2B8ZE58yG2HzqxQxYUbQqibUh5Xcr8=
+github.com/nats-io/nats.go v1.55.0 h1:Ng8QKHdvdHIWoaiq0+d/aH1BWkxvcJxWJ0Y/9BPKx3M=
+github.com/nats-io/nats.go v1.55.0/go.mod h1:y+DZoD1oBOYfZTU681eTUiUjI0vbqYGixNVFHcjHJ0k=
 github.com/nats-io/nkeys v0.4.16 h1:rd5oAuLOb8mnAycB0xleuEBNS1pVVnN0fv/FF34Eypg=
 github.com/nats-io/nkeys v0.4.16/go.mod h1:llLgWoI0o4z/Q57q2R1kHfmocyhGV6VG/U18Glg1Afs=
 github.com/nats-io/nuid v1.0.1 h1:5iA8DT8V7q8WK2EScv2padNa/rTESc1KdnPw4TC2paw=
@@ -340,7 +336,6 @@ github.com/pelletier/go-toml/v2 v2.4.3/go.mod h1:2gIqNv+qfxSVS7cM2xJQKtLSTLUE9V8
 github.com/pierrec/lz4/v4 v4.1.14/go.mod h1:gZWDp/Ze/IJXGXf23ltt2EXimqmTUXEy0GFuRQyBid4=
 github.com/pierrec/lz4/v4 v4.1.18 h1:xaKrnTkyoqfh1YItXl56+6KJNVYWlEEPuAQW9xsplYQ=
 github.com/pierrec/lz4/v4 v4.1.18/go.mod h1:gZWDp/Ze/IJXGXf23ltt2EXimqmTUXEy0GFuRQyBid4=
-github.com/pkg/errors v0.8.1/go.mod h1:bwawxfHBFNV+L2hUp1rHADufV3IMtnDRdf1r5NINEl0=
 github.com/pkg/errors v0.9.1 h1:FEBLx1zS214owpjy7qsBeixbURkuhQAwrK5UwLGTwt4=
 github.com/pkg/errors v0.9.1/go.mod h1:bwawxfHBFNV+L2hUp1rHADufV3IMtnDRdf1r5NINEl0=
 github.com/pmezard/go-difflib v1.0.0/go.mod h1:iKH77koFhYxTK1pcRnkKkqfTogsbg7gZNVY4sRDYZ/4=
@@ -409,17 +404,15 @@ github.com/xo/terminfo v0.0.0-20220910002029-abceb7e1c41e h1:JVG44RsyaB9T2KIHavM
 github.com/xo/terminfo v0.0.0-20220910002029-abceb7e1c41e/go.mod h1:RbqR21r5mrJuqunuUZ/Dhy/avygyECGrLceyNeo4LiM=
 github.com/youmark/pkcs8 v0.0.0-20240726163527-a2c0da244d78 h1:ilQV1hzziu+LLM3zUTJ0trRztfwgjqKnBWNtSRkbmwM=
 github.com/youmark/pkcs8 v0.0.0-20240726163527-a2c0da244d78/go.mod h1:aL8wCCfTfSfmXjznFBSZNN13rSJjlIOI1fUNAtF7rmI=
-github.com/yuin/goldmark v1.1.27/go.mod h1:3hX8gzYuyVAZsxl0MRgGTJEmQBFcNTphYh9decYSb74=
-github.com/yuin/goldmark v1.2.1/go.mod h1:3hX8gzYuyVAZsxl0MRgGTJEmQBFcNTphYh9decYSb74=
 github.com/yuin/goldmark v1.4.13/go.mod h1:6yULJ656Px+3vBD8DxQVa3kxgyrAnzto9xy5taEt/CY=
 go.etcd.io/bbolt v1.5.0 h1:S7GAl7Fxv12yohbwFfIbQCGDWbQbtDGPET4P/bD4lxU=
 go.etcd.io/bbolt v1.5.0/go.mod h1:mkltfYE5aUHQxUct9N9V+Kp7aSjFqjgrhcXIS70Lrdk=
-go.etcd.io/etcd/api/v3 v3.5.35 h1:lB//Av9mFrkljv94XDNAY+qiyQA9W6S/TTA2Uwx2jsI=
-go.etcd.io/etcd/api/v3 v3.5.35/go.mod h1:z1FGQolu/tpmQ2MPFwQNt7G3cXBiSkTevZBhfOtEArQ=
-go.etcd.io/etcd/client/pkg/v3 v3.5.35 h1:nUZutZyJVvsKddK6zddXMBxIhmVi0SzPMBRS0FnNM3o=
-go.etcd.io/etcd/client/pkg/v3 v3.5.35/go.mod h1:0MNjMMqajrOI0/l8OLFkJBhE5Vm6BAi7cerKlrO38hw=
-go.etcd.io/etcd/client/v3 v3.5.35 h1:dk0aSLjY5D1eOR0SNnPXWOWCVA9yIzJoB8a912qxmoI=
-go.etcd.io/etcd/client/v3 v3.5.35/go.mod h1:LjtCl6G5li7PgsGjoQIpYXz/OvSbI1pCxpSp1kqo9P4=
+go.etcd.io/etcd/api/v3 v3.7.3 h1:2tumepXe+wi7Wab64uUO71K7ISQyUWLGC1DJbbc59sY=
+go.etcd.io/etcd/api/v3 v3.7.3/go.mod h1:HOR9Qn4gb/1/hVAmYprmlLpvTe6ZgfgNS9Z/+Ty65po=
+go.etcd.io/etcd/client/pkg/v3 v3.7.3 h1:5zMwNu3ycrzCeA5NwOfQbxFRhqq974chq1DrYMua4QI=
+go.etcd.io/etcd/client/pkg/v3 v3.7.3/go.mod h1:/OhzqYrF0U7d1NOk55VMyUQGbYD1WoIe4bJwxb47wOk=
+go.etcd.io/etcd/client/v3 v3.7.3 h1:UrXcetgkuSEocaBlQlulQRJlXLEnUAHuBRbHh0lue1Q=
+go.etcd.io/etcd/client/v3 v3.7.3/go.mod h1:DcPxpqcpV8WIaRi/2waPCx0/Uq8rQUweDJaDYZk1gEE=
 go.opentelemetry.io/auto/sdk v1.2.1 h1:jXsnJ4Lmnqd11kwkBV2LgLoFMZKizbCi5fNZ/ipaZ64=
 go.opentelemetry.io/auto/sdk v1.2.1/go.mod h1:KRTj+aOaElaLi+wW1kO/DZRXwkF4C5xPbEe3ZiIhN7Y=
 go.opentelemetry.io/contrib/bridges/otelslog v0.21.0 h1:/CdgE52i1OEPHhsGUqrggYNRaZy4v4YgVenGd52dRBI=
@@ -456,26 +449,22 @@ go.opentelemetry.io/otel/trace v1.47.0 h1:JOjX/Oci8K94QHddo+bbfya/Ai/nf6/dt9ZfrF
 go.opentelemetry.io/otel/trace v1.47.0/go.mod h1:jNaSLa2PZEYFG6fRjJABAu+bw4FS08uDmPg28lTghu0=
 go.opentelemetry.io/proto/otlp v1.11.0 h1:5rrYs0Ykyj50sdU/JU0x8etU+LubXWb+gED6TbEdMIk=
 go.opentelemetry.io/proto/otlp v1.11.0/go.mod h1:SmVizdCOAm3XBtG1g1NnOdhW6jtddT72hLMhv8VwA8E=
-go.uber.org/atomic v1.7.0 h1:ADUqmZGgLDDfbSL9ZmPxKTybcoEYHgpYfELNoN+7hsw=
-go.uber.org/atomic v1.7.0/go.mod h1:fEN4uk6kAWBTFdckzkM89CLk9XfWZrxpCo0nPH17wJc=
 go.uber.org/goleak v1.3.1-0.20260602155401-5f791e3ef83b h1:FHYZT12VzXnb6P399kd1i9Ai7aLF+CFSUsOFylDt7Lk=
 go.uber.org/goleak v1.3.1-0.20260602155401-5f791e3ef83b/go.mod h1:CoHD4mav9JJNrW/WLlf7HGZPjdw8EucARQHekz1X6bE=
 go.uber.org/mock v0.6.0 h1:hyF9dfmbgIX5EfOdasqLsWD6xqpNZlXblLB/Dbnwv3Y=
 go.uber.org/mock v0.6.0/go.mod h1:KiVJ4BqZJaMj4svdfmHM0AUx4NJYO8ZNpPnZn1Z+BBU=
-go.uber.org/multierr v1.6.0 h1:y6IPFStTAIT5Ytl7/XYmHvzXQ7S3g/IeZW9hyZ5thw4=
-go.uber.org/multierr v1.6.0/go.mod h1:cdWPpRnG4AhwMwsgIHip0KRBQjJy5kYEpYjJxpXp9iU=
+go.uber.org/multierr v1.11.0 h1:blXXJkSxSSfBVBlC76pxqeO+LN3aDfLQo+309xJstO0=
+go.uber.org/multierr v1.11.0/go.mod h1:20+QtiLqy0Nd6FdQB9TLXag12DsQkrbs3htMFfDN80Y=
 go.uber.org/nilaway v0.0.0-20260721205819-4a0f7653488a h1:CsRd824LAzMmlpu9nYv2PHOqPQYv4Zjd4kGM+g0ahS0=
 go.uber.org/nilaway v0.0.0-20260721205819-4a0f7653488a/go.mod h1:i9Hi8GeZFLBtUCXP73ZNSHKZ4K2JtxT0pLmo3sbJzlo=
-go.uber.org/zap v1.17.0 h1:MTjgFu6ZLKvY6Pvaqk97GlxNBuMpV4Hy/3P6tRGlI2U=
-go.uber.org/zap v1.17.0/go.mod h1:MXVU+bhUf/A7Xi2HNOnopQOrmycQ5Ih87HtOu4q5SSo=
+go.uber.org/zap v1.27.1 h1:08RqriUEv8+ArZRYSTXy1LeBScaMpVSTBhCeaZYfMYc=
+go.uber.org/zap v1.27.1/go.mod h1:GB2qFLM7cTU87MWRP2mPIjqfIDnGu+VIO4V/SdhGo2E=
 go.yaml.in/yaml/v2 v2.4.4 h1:tuyd0P+2Ont/d6e2rl3be67goVK4R6deVxCUX5vyPaQ=
 go.yaml.in/yaml/v2 v2.4.4/go.mod h1:gMZqIpDtDqOfM0uNfy0SkpRhvUryYH0Z6wdMYcacYXQ=
 go.yaml.in/yaml/v3 v3.0.4/go.mod h1:DhzuOOF2ATzADvBadXxruRBLzYTpT36CKvDb3+aBEFg=
-go.yaml.in/yaml/v3 v3.0.5 h1:N6y/pJk8buWs9NY5ERU2HSMfm+IuD/OtfdAnq6kESPw=
-go.yaml.in/yaml/v3 v3.0.5/go.mod h1:HVTZu1O7/Vkt2N+BFy8Zza+lnLsABggaTM2ZpNIGuKg=
+go.yaml.in/yaml/v3 v3.0.6 h1:zcDfvjXEqTQmpjniUBLc91xYJMB2c4QOFxEF/LCEckM=
+go.yaml.in/yaml/v3 v3.0.6/go.mod h1:HVTZu1O7/Vkt2N+BFy8Zza+lnLsABggaTM2ZpNIGuKg=
 golang.org/x/crypto v0.0.0-20190308221718-c2843e01d9a2/go.mod h1:djNgcEr1/C05ACkg1iLfiJU5Ep61QUkGW8qpdssI0+w=
-golang.org/x/crypto v0.0.0-20191011191535-87dc89f01550/go.mod h1:yigFU9vqHzYiE8UmvKecakEJjdnWj3jj499lnFckfCI=
-golang.org/x/crypto v0.0.0-20200622213623-75b288015ac9/go.mod h1:LzIPMQfyMNhhGPhUkYOs5KpL4U8rLKemX1yGLhDgUto=
 golang.org/x/crypto v0.0.0-20210921155107-089bfa567519/go.mod h1:GvvjBRRGRdwPK5ydBHafDWAxML/pGHZbMvKqRZ5+Abc=
 golang.org/x/crypto v0.58.0 h1:COkYLr4k7nDI4r6QtMML2AfzeIswWHAJthLS6K/M/54=
 golang.org/x/crypto v0.58.0/go.mod h1:Xh+kl5A+M0gMiWDytH2Hhr5Z7WbNCpv/eB/EFgfSk+w=
@@ -483,33 +472,24 @@ golang.org/x/exp v0.0.0-20241108190413-2d47ceb2692f h1:XdNn9LlyWAhLVp6P/i8QYBW+h
 golang.org/x/exp v0.0.0-20241108190413-2d47ceb2692f/go.mod h1:D5SMRVC3C2/4+F/DB1wZsLRnSNimn2Sp/NPsCrsv8ak=
 golang.org/x/exp/typeparams v0.0.0-20260611194520-c48552f49976 h1:GTD/WuaexTazIG/SxLOz4rEKZPDVilmVVC2nz4xhwfE=
 golang.org/x/exp/typeparams v0.0.0-20260611194520-c48552f49976/go.mod h1:PqrXSW65cXDZH0k4IeUbhmg/bcAZDbzNz3byBpKCsXo=
-golang.org/x/mod v0.2.0/go.mod h1:s0Qsj1ACt9ePp/hMypM3fl4fZqREWJwdYDEqhRiZZUA=
-golang.org/x/mod v0.3.0/go.mod h1:s0Qsj1ACt9ePp/hMypM3fl4fZqREWJwdYDEqhRiZZUA=
 golang.org/x/mod v0.6.0-dev.0.20220419223038-86c51ed26bb4/go.mod h1:jJ57K6gSWd91VN4djpZkiMVwK6gcyfeH4XE8wZrZaV4=
 golang.org/x/mod v0.8.0/go.mod h1:iBbtSCu2XBx23ZKBPSOrRkjjQPZFPuis4dIYUhu/chs=
 golang.org/x/mod v0.42.0 h1:ICeL3t2L30ljSPHHxDnR4fqVcBt/zR2zJAb9juavmQg=
 golang.org/x/mod v0.42.0/go.mod h1:u5fzIrRInkL8MaO1cbRkUHT64v303YIT5KEqpNr7jRs=
-golang.org/x/net v0.0.0-20190404232315-eb5bcb51f2a3/go.mod h1:t9HGtf8HONx5eT2rtn7q6eTqICYqUVnKs3thJo3Qplg=
 golang.org/x/net v0.0.0-20190620200207-3b0461eec859/go.mod h1:z5CRVTTTmAJ677TzLLGU+0bjPO0LkuOLi4/5GtJWs/s=
-golang.org/x/net v0.0.0-20200226121028-0de0cce0169b/go.mod h1:z5CRVTTTmAJ677TzLLGU+0bjPO0LkuOLi4/5GtJWs/s=
-golang.org/x/net v0.0.0-20201021035429-f5854403a974/go.mod h1:sp8m0HH+o8qH0wwXwYZr8TS3Oi6o0r6Gce1SSxlDquU=
 golang.org/x/net v0.0.0-20210226172049-e18ecbb05110/go.mod h1:m0MpNAwzfU5UDzcl9v0D8zg8gWTRqZa9RBIspLL5mdg=
 golang.org/x/net v0.0.0-20220722155237-a158d28d115b/go.mod h1:XRhObCWvk6IyKnWLug+ECip1KBveYUHfp+8e9klMJ9c=
 golang.org/x/net v0.6.0/go.mod h1:2Tu9+aMcznHK/AK1HMvgo6xiTLG5rD5rZLDS+rp2Bjs=
 golang.org/x/net v0.61.0 h1:RR+6j/BTBCrggOLugxa+66R8FvoN6MHlzDlZIbN9YFw=
 golang.org/x/net v0.61.0/go.mod h1:WgDCOTH8iwtB66T4/GfI94AHo/hhdb7Hah9NDseQfiI=
 golang.org/x/sync v0.0.0-20190423024810-112230192c58/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
-golang.org/x/sync v0.0.0-20190911185100-cd5d95a43a6e/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
-golang.org/x/sync v0.0.0-20201020160332-67f06af15bc9/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
 golang.org/x/sync v0.0.0-20220722155255-886fb9371eb4/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
 golang.org/x/sync v0.1.0/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
 golang.org/x/sync v0.24.0 h1:iqiDrGoDgyuvAKrhLQthUEzU/R378N78gXbBjxed7oA=
 golang.org/x/sync v0.24.0/go.mod h1:sUUOizhqBxiL6pEWpqNLUiaJn1ShEbZ6BBqskPbjZm0=
 golang.org/x/sys v0.0.0-20190215142949-d0b11bdaac8a/go.mod h1:STP8DvDyc/dI5b8T5hshtkjS+E42TnysNCUPdjciGhY=
-golang.org/x/sys v0.0.0-20190412213103-97732733099d/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
 golang.org/x/sys v0.0.0-20200116001909-b77594299b42/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
 golang.org/x/sys v0.0.0-20200223170610-d5e6a3e2c0ae/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
-golang.org/x/sys v0.0.0-20200930185726-fdedc70b468f/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
 golang.org/x/sys v0.0.0-20201119102817-f84b799fce68/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
 golang.org/x/sys v0.0.0-20210615035016-665e8c7367d1/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
 golang.org/x/sys v0.0.0-20210630005230-0f9fa26af87c/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
@@ -542,8 +522,6 @@ golang.org/x/time v0.16.0 h1:vMb6ptszcQMkcwiRTAuNNU50gom6++Q/6gY2hDM6VDE=
 golang.org/x/time v0.16.0/go.mod h1:rVKOqvZeKvrDKTQiAHJ7wmwP0RzleSphoEA9RcdLA0s=
 golang.org/x/tools v0.0.0-20180917221912-90fa682c2a6e/go.mod h1:n7NCudcB/nEzxVGmLbDWY5pfWTLqBcC2KZ6jyYvM4mQ=
 golang.org/x/tools v0.0.0-20191119224855-298f0cb1881e/go.mod h1:b+2E5dAYhXwXZwtnZ6UAqBI28+e2cm9otk0dWdXHAEo=
-golang.org/x/tools v0.0.0-20200619180055-7c47624df98f/go.mod h1:EkVYQZoAsY45+roYkvgYkIh4xh/qjgUK9TdY2XT94GE=
-golang.org/x/tools v0.0.0-20210106214847-113979e3529a/go.mod h1:emZCQorbCU4vsT4fOWvOPXz4eW1wZW4PmDk9uLelYpA=
 golang.org/x/tools v0.1.12/go.mod h1:hNGJHUnrk76NpqgfD5Aqm5Crs+Hm0VOH/i9J2+nxYbc=
 golang.org/x/tools v0.6.0/go.mod h1:Xwgl3UAJ/d3gWutnCtw505GrjyAbvKui8lOU390QaIU=
 golang.org/x/tools v0.52.0 h1:ryxg4U63oaUYk7uOLTF37n43f4qgtGvqV4Ioit1BaEs=
@@ -555,17 +533,14 @@ golang.org/x/tools/go/packages/packagestest v0.1.1-deprecated/go.mod h1:RVAQXBGN
 golang.org/x/vuln v1.6.0 h1:FeMO9Rm/HwyduOztbvKcOw+zvDEPr4I4aQNSfevFcKY=
 golang.org/x/vuln v1.6.0/go.mod h1:bWlG2493/sjR7ksvicBgMrznH3eYQEyK8ifUYBrqUbg=
 golang.org/x/xerrors v0.0.0-20190717185122-a985d3407aa7/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
-golang.org/x/xerrors v0.0.0-20191011141410-1b5146add898/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
-golang.org/x/xerrors v0.0.0-20191204190536-9bdfabe68543/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
-golang.org/x/xerrors v0.0.0-20200804184101-5ec99f83aff1/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
 gonum.org/v1/gonum v0.17.0 h1:VbpOemQlsSMrYmn7T2OUvQ4dqxQXU+ouZFQsZOx50z4=
 gonum.org/v1/gonum v0.17.0/go.mod h1:El3tOrEuMpv2UdMrbNlKEh9vd86bmQ6vqIcDwxEOc1E=
 google.golang.org/genproto/googleapis/api v0.0.0-20260928230214-8a89bd6388cc h1:rZHRz0ogq4Pid1IKtA/ivHi8XONVF5tqUQXA2TtgpkE=
 google.golang.org/genproto/googleapis/api v0.0.0-20260928230214-8a89bd6388cc/go.mod h1:K/n0XLazCJHjD73YuMyNJsN38BPPy1kfFnKxUXaThFU=
 google.golang.org/genproto/googleapis/rpc v0.0.0-20260928230214-8a89bd6388cc h1:4bNTbnb44EqGVy9HaQxSY2jnafSUdgV3qHtedvNpDKg=
 google.golang.org/genproto/googleapis/rpc v0.0.0-20260928230214-8a89bd6388cc/go.mod h1:OaIUM3+LpYcK2GXM4FTmhWoIq371Owdr+Cc7/BsYHHc=
-google.golang.org/grpc v1.83.2 h1:EManeRomTObA0BU7I8vXgg/78uE5MJ9M8B39EX2WscU=
-google.golang.org/grpc v1.83.2/go.mod h1:YPI1hK3kDked6iHvgX3tR0y+nX/qpMFKhPgFsokw1S8=
+google.golang.org/grpc v1.85.0 h1:BPBCfuVRQ1qvBhKDOq9uvQllMvO59rKsYhXpp7Dd8vw=
+google.golang.org/grpc v1.85.0/go.mod h1:0rQrOD58yjal/0EaVbGBANindixsCZNQPDV3VKtaKuQ=
 google.golang.org/protobuf v1.36.12 h1:pJOKDDOyeXErUroCihFAd5LQuwXBSpVnKGrj5o/fwxc=
 google.golang.org/protobuf v1.36.12/go.mod h1:HTf+CrKn2C3g5S8VImy6tdcUvCska2kB7j23XfzDpco=
 gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405/go.mod h1:Co6ibVJAznAaIkqp8huTwlJQCZ016jof/cbN4VW5Yz0=
@@ -575,7 +550,7 @@ gopkg.in/yaml.v2 v2.2.8/go.mod h1:hI93XBmqTisBFMUTm0b8Fm+jr3Dg1NNxqwp+5A1VGuI=
 gopkg.in/yaml.v2 v2.4.0 h1:D8xgwECY7CYvx+Y2n4sBz93Jn9JRvxdiyyo8CTfuKaY=
 gopkg.in/yaml.v2 v2.4.0/go.mod h1:RDklbk79AGWmwhnvt/jBztapEOGDOx6ZbXqjP6csGnQ=
 gopkg.in/yaml.v3 v3.0.0-20200313102051-9f266ea9e77c/go.mod h1:K4uyk7z7BCEPqu6E+C64Yfv1cQ7kz7rIZviUmN+EgEM=
-gopkg.in/yaml.v3 v3.0.0-20210107192922-496545a6307b/go.mod h1:K4uyk7z7BCEPqu6E+C64Yfv1cQ7kz7rIZviUmN+EgEM=
+gopkg.in/yaml.v3 v3.0.1 h1:fxVm/GzAzEWqLHuvctI91KS9hhNmmWOoWu0XTYJS7CA=
 gopkg.in/yaml.v3 v3.0.1/go.mod h1:K4uyk7z7BCEPqu6E+C64Yfv1cQ7kz7rIZviUmN+EgEM=
 pgregory.net/rapid v1.3.0 h1:vBvO0VSqti75J1jjYqpgPNBLKMd1+gxa9fYo7vk/Exc=
 pgregory.net/rapid v1.3.0/go.mod h1:dPlE4OBBxgXPqkP79flB6sJL1dx5azpI7HQ9MY9Z7uk=
@@ -19,5 +19,6 @@ var modelFiles = map[Service]string{
 	RDS:                    "models/rds-2014-10-31.json.gz",
 	Route53:                "models/route-53-2013-04-01.json.gz",
 	S3:                     "models/s3-2006-03-01.json.gz",
+	SQS:                    "models/sqs-2012-11-05.json.gz",
 	STS:                    "models/sts-2011-06-15.json.gz",
 }
@@ -36,6 +36,7 @@ const (
 	RDS                    Service = "rds"
 	Route53                Service = "route53"
 	S3                     Service = "s3"
+	SQS                    Service = "sqs"
 	STS                    Service = "sts"
 )

@@ -8,7 +8,7 @@ import (
 )

 func TestServices(t *testing.T) {
-	want := []Service{ACM, EC2, ECR, ECS, EKS, ElasticLoadBalancingV2, IAM, RDS, Route53, S3, STS}
+	want := []Service{ACM, EC2, ECR, ECS, EKS, ElasticLoadBalancingV2, IAM, RDS, Route53, S3, SQS, STS}
 	if got := Services(); !reflect.DeepEqual(got, want) {
 		t.Fatalf("Services() = %v, want %v", got, want)
 	}
@@ -1,6 +1,6 @@
 // Package sqsworkergw is the shared SQS worker guest's SigV4 client for the
 // gateway's internal spinifex-sqs service. It sends only public values: the
-// worker's identity is the signing session, and the store seed never leaves
+// worker's identity is the signing session, and the receipt key never leaves
 // the guest.
 package sqsworkergw

@@ -10,20 +10,21 @@ import (
 	"crypto/sha256"
 	"crypto/tls"
 	"crypto/x509"
+	"encoding/base64"
 	"encoding/hex"
+	"encoding/json"
 	"encoding/xml"
 	"errors"
 	"fmt"
 	"io"
 	"maps"
-	"net"
 	"net/http"
 	"net/url"
 	"strings"
 	"time"

 	"github.com/mulgadc/bluebottle/pkg/tlsconfig"
-	"github.com/nats-io/nkeys"
+	"github.com/mulgadc/spinifex/spinifex/domains/sqs/relay"
 )

 const (
@@ -33,9 +34,12 @@ const (
 	callTimeout    = 30 * time.Second
 	maxResponse    = 64 << 10

-	actionClaim    = "ClaimFirstStore"
-	actionDescribe = "DescribeStore"
-	actionRecord   = "RecordWorkerPublicKey"
+	actionClaim    = "ClaimReceiptKey"
+	actionRecord   = "RecordReceiptKey"
+	actionPoll     = "PollWork"
+	actionComplete = "CompleteWork"
+	// maxPollResponse fits the gateway's 16 MiB poll batch as base64 XML.
+	maxPollResponse = 32 << 20
 )

 // Signer signs a request in place for service and region.
@@ -43,7 +47,7 @@ type Signer interface {
 	Sign(r *http.Request, payloadHash, service, region string) error
 }

-// Client calls the internal first-store actions for this worker.
+// Client calls the internal SQS worker actions for this worker.
 type Client struct {
 	baseURL    string
 	region     string
@@ -114,80 +118,72 @@ func (e *TransientError) Error() string   { return e.Err.Error() }
 func (e *TransientError) Unwrap() error   { return e.Err }
 func (e *TransientError) Retryable() bool { return true }

-type claimResult struct {
-	CreateKey bool   `xml:"CreateKey"`
-	PublicKey string `xml:"PublicKey"`
+// ReceiptKey is the worker volume's receipt key initialization as the
+// gateway reports it: pending authorizes creating or resuming one, otherwise
+// Fingerprint names the key the disk must already hold.
+type ReceiptKey struct {
+	Pending     bool
+	Fingerprint string
+	// HostNodes is the gateway's declared cluster node count; a gateway
+	// that does not report one reads as zero.
+	HostNodes int
 }

-// Claim asks for this worker's current first-store decision. A decision
-// that is neither "create" nor one valid public key is refused here.
-func (c *Client) Claim(ctx context.Context) (bool, string, error) {
+type claimResult struct {
+	Pending     bool   `xml:"Pending"`
+	Fingerprint string `xml:"Fingerprint"`
+	HostNodes   int    `xml:"HostNodes"`
+}
+
+// ClaimReceiptKey asks for the receipt key decision. A reply that is neither
+// pending nor one valid fingerprint is refused here.
+func (c *Client) ClaimReceiptKey(ctx context.Context) (ReceiptKey, error) {
 	var res claimResult
 	if err := c.call(ctx, actionClaim, nil, &res); err != nil {
-		return false, "", err
+		return ReceiptKey{}, err
 	}
-	switch {
-	case res.CreateKey && res.PublicKey == "":
-		return true, "", nil
-	case !res.CreateKey && nkeys.IsValidPublicUserKey(res.PublicKey):
-		return false, res.PublicKey, nil
+	if res.Pending == (res.Fingerprint != "") || (res.Fingerprint != "" && !ValidFingerprint(res.Fingerprint)) {
+		return ReceiptKey{}, errors.New("spinifex-sqs ClaimReceiptKey: response is not a valid decision")
 	}
-	return false, "", errors.New("spinifex-sqs ClaimFirstStore: response is not a valid decision")
+	return ReceiptKey(res), nil
 }

-// Store is the Region's store launch as the gateway reports it. Bound means
-// a VM was launched for it, not that it is ready.
-type Store struct {
-	Bound     bool
-	Address   string
-	PublicKey string
-}
-
-type describeResult struct {
-	State     string `xml:"State"`
-	Address   string `xml:"Address"`
-	PublicKey string `xml:"PublicKey"`
-}
-
-// DescribeStore reports this worker's store. A bound reply must carry an
-// IPv4 address and a valid public key; a pending one must carry neither.
-func (c *Client) DescribeStore(ctx context.Context) (Store, error) {
-	var res describeResult
-	if err := c.call(ctx, actionDescribe, nil, &res); err != nil {
-		return Store{}, err
+// RecordReceiptKey records the key's fingerprint. Recording the same one
+// again succeeds, so a lost reply is safe to repeat.
+func (c *Client) RecordReceiptKey(ctx context.Context, fingerprint string) error {
+	if !ValidFingerprint(fingerprint) {
+		return errors.New("spinifex-sqs RecordReceiptKey: refusing to send a value that is not a fingerprint")
 	}
-	switch {
-	case res.State == "pending" && res.Address == "" && res.PublicKey == "":
-		return Store{}, nil
-	case res.State == "bound" && isIPv4(res.Address) && nkeys.IsValidPublicUserKey(res.PublicKey):
-		return Store{Bound: true, Address: res.Address, PublicKey: res.PublicKey}, nil
-	}
-	return Store{}, errors.New("spinifex-sqs DescribeStore: response is not a valid store state")
+	return c.call(ctx, actionRecord, url.Values{"Fingerprint": {fingerprint}}, nil)
 }

-func isIPv4(s string) bool {
-	ip := net.ParseIP(s)
-	return ip != nil && ip.To4() != nil && !strings.Contains(s, ":")
-}
-
-// RecordPublicKey records the worker's public key. Anything other than a
-// user public key is refused before it is sent.
-func (c *Client) RecordPublicKey(ctx context.Context, publicKey string) error {
-	if !nkeys.IsValidPublicUserKey(publicKey) {
-		return errors.New("spinifex-sqs RecordWorkerPublicKey: refusing to send a value that is not a user public key")
+// ValidFingerprint reports a lowercase hex SHA-256 fingerprint.
+func ValidFingerprint(s string) bool {
+	if len(s) != 2*sha256.Size || strings.ToLower(s) != s {
+		return false
 	}
-	return c.call(ctx, actionRecord, url.Values{"PublicKey": {publicKey}}, nil)
+	_, err := hex.DecodeString(s)
+	return err == nil
 }

 func (c *Client) call(ctx context.Context, action string, params url.Values, out any) error {
 	form := url.Values{"Action": {action}, "Version": {apiVersion}}
 	maps.Copy(form, params)
-	body := []byte(form.Encode())
-	req, err := http.NewRequestWithContext(ctx, http.MethodPost, c.baseURL+"/", bytes.NewReader(body))
+	return c.do(ctx, action, "", []byte(form.Encode()), "application/x-www-form-urlencoded; charset=utf-8", maxResponse, out)
+}
+
+// do signs and sends one request. rawQuery, when set, carries the action and
+// its arguments; the body hash is signed either way.
+func (c *Client) do(ctx context.Context, action, rawQuery string, body []byte, contentType string, limit int64, out any) error {
+	target := c.baseURL + "/"
+	if rawQuery != "" {
+		target += "?" + rawQuery
+	}
+	req, err := http.NewRequestWithContext(ctx, http.MethodPost, target, bytes.NewReader(body))
 	if err != nil {
 		return fmt.Errorf("spinifex-sqs %s: build request: %w", action, err)
 	}
-	req.Header.Set("Content-Type", "application/x-www-form-urlencoded; charset=utf-8")
+	req.Header.Set("Content-Type", contentType)
 	sum := sha256.Sum256(body)
 	if err := c.signer.Sign(req, hex.EncodeToString(sum[:]), signingService, c.region); err != nil {
 		return &TransientError{Err: fmt.Errorf("spinifex-sqs %s: sign: %w", action, err)}
@@ -197,7 +193,7 @@ func (c *Client) call(ctx context.Context, action string, params url.Values, out
 		return &TransientError{Err: fmt.Errorf("spinifex-sqs %s: send: %w", action, err)}
 	}
 	defer resp.Body.Close()
-	respBody, err := io.ReadAll(io.LimitReader(resp.Body, maxResponse))
+	respBody, err := io.ReadAll(io.LimitReader(resp.Body, limit))
 	if err != nil {
 		return &TransientError{Err: fmt.Errorf("spinifex-sqs %s: read response: %w", action, err)}
 	}
@@ -210,6 +206,44 @@ func (c *Client) call(ctx context.Context, action string, params url.Values, out
 	return decodeResult(action, respBody, out)
 }

+// PollWork long-polls the gateway for owner calls queued for this worker.
+// An empty list is a poll that found nothing.
+func (c *Client) PollWork(ctx context.Context) ([]relay.Work, error) {
+	var res struct {
+		Items []string `xml:"Items>member"`
+	}
+	if err := c.do(ctx, actionPoll, "", []byte(url.Values{"Action": {actionPoll}, "Version": {apiVersion}}.Encode()),
+		"application/x-www-form-urlencoded; charset=utf-8", maxPollResponse, &res); err != nil {
+		return nil, err
+	}
+	out := make([]relay.Work, 0, len(res.Items))
+	for _, item := range res.Items {
+		raw, err := base64.StdEncoding.DecodeString(item)
+		if err != nil {
+			return nil, errors.New("spinifex-sqs PollWork: item is not base64")
+		}
+		var w relay.Work
+		if err := json.Unmarshal(raw, &w); err != nil || w.ID == "" || w.Op == "" {
+			return nil, errors.New("spinifex-sqs PollWork: item is not a work item")
+		}
+		out = append(out, w)
+	}
+	return out, nil
+}
+
+// CompleteWork posts an encoded result for work item id and returns the
+// gateway's outcome name. Any error leaves delivery unknown.
+func (c *Client) CompleteWork(ctx context.Context, id string, result []byte) (string, error) {
+	q := url.Values{"Action": {actionComplete}, "Version": {apiVersion}, "WorkId": {id}}
+	var res struct {
+		Outcome string `xml:"Outcome"`
+	}
+	if err := c.do(ctx, actionComplete, q.Encode(), result, "application/json", maxResponse, &res); err != nil {
+		return "", err
+	}
+	return res.Outcome, nil
+}
+
 func decodeResult(action string, body []byte, out any) error {
 	want := action + "Result"
 	dec := xml.NewDecoder(bytes.NewReader(body))
@@ -10,8 +10,6 @@ import (
 	"strings"
 	"sync/atomic"
 	"testing"
-
-	"github.com/nats-io/nkeys"
 )

 type nopSigner struct{ err error }
@@ -24,15 +22,10 @@ func (s nopSigner) Sign(r *http.Request, _, service, region string) error {
 	return nil
 }

-func userKey(t *testing.T) (pub, seed string) {
-	t.Helper()
-	kp, err := nkeys.CreateUser()
-	if err != nil {
-		t.Fatal(err)
-	}
-	p, _ := kp.PublicKey()
-	s, _ := kp.Seed()
-	return p, string(s)
+var fp = strings.Repeat("ab", 32)
+
+func claimBody(inner string) string {
+	return `<ClaimReceiptKeyResponse><ClaimReceiptKeyResult>` + inner + `</ClaimReceiptKeyResult></ClaimReceiptKeyResponse>`
 }

 // server answers with body and status, counting and keeping requests.
@@ -73,63 +66,56 @@ func TestNew_RefusesWithoutHTTPSRegionCAOrSigner(t *testing.T) {
 }

 func TestClaim_SignsForTheInternalServiceAndRegion(t *testing.T) {
-	c, _, seen := server(t, 200, `<ClaimFirstStoreResponse><ClaimFirstStoreResult><CreateKey>true</CreateKey></ClaimFirstStoreResult></ClaimFirstStoreResponse>`)
-	create, pub, err := c.Claim(context.Background())
-	if err != nil || !create || pub != "" {
-		t.Fatalf("claim = %v %q %v", create, pub, err)
+	c, _, seen := server(t, 200, claimBody(`<Pending>true</Pending><HostNodes>1</HostNodes>`))
+	got, err := c.ClaimReceiptKey(context.Background())
+	if err != nil || !got.Pending || got.Fingerprint != "" || got.HostNodes != 1 {
+		t.Fatalf("claim = %+v %v", got, err)
 	}
-	if got := (*seen)[0]; !strings.HasPrefix(got, "test spinifex-sqs/ap-southeast-2|") || !strings.Contains(got, "Action=ClaimFirstStore") {
-		t.Fatalf("request = %q", got)
+	if req := (*seen)[0]; !strings.HasPrefix(req, "test spinifex-sqs/ap-southeast-2|") || !strings.Contains(req, "Action=ClaimReceiptKey") {
+		t.Fatalf("request = %q", req)
 	}
 }

-// Only "create" or exactly one valid public key is a decision.
+// Only pending alone, or exactly one valid fingerprint, is a decision.
 func TestClaim_RefusesAnythingButOneDecision(t *testing.T) {
-	pub, seed := userKey(t)
-	cases := map[string]string{
-		"create with a key": `<CreateKey>true</CreateKey><PublicKey>` + pub + `</PublicKey>`,
-		"neither":           `<CreateKey>false</CreateKey>`,
-		"seed as key":       `<CreateKey>false</CreateKey><PublicKey>` + seed + `</PublicKey>`,
-		"no result":         ``,
-	}
-	for name, inner := range cases {
+	for name, inner := range map[string]string{
+		"pending with a key":  `<Pending>true</Pending><Fingerprint>` + fp + `</Fingerprint>`,
+		"neither":             `<Pending>false</Pending>`,
+		"uppercase key":       `<Pending>false</Pending><Fingerprint>` + strings.ToUpper(fp) + `</Fingerprint>`,
+		"short key":           `<Pending>false</Pending><Fingerprint>` + fp[:62] + `</Fingerprint>`,
+		"not the claim reply": "",
+	} {
 		t.Run(name, func(t *testing.T) {
-			body := `<ClaimFirstStoreResponse><ClaimFirstStoreResult>` + inner + `</ClaimFirstStoreResult></ClaimFirstStoreResponse>`
+			body := claimBody(inner)
 			if inner == "" {
 				body = `<Other/>`
 			}
 			c, _, _ := server(t, 200, body)
-			_, _, err := c.Claim(context.Background())
-			if err == nil {
+			if _, err := c.ClaimReceiptKey(context.Background()); err == nil {
 				t.Fatal("accepted a non-decision")
 			}
-			if strings.Contains(err.Error(), seed) {
-				t.Fatal("the error echoes a seed")
-			}
 		})
 	}
-	t.Run("existing", func(t *testing.T) {
-		c, _, _ := server(t, 200, `<ClaimFirstStoreResponse><ClaimFirstStoreResult><CreateKey>false</CreateKey><PublicKey>`+pub+`</PublicKey></ClaimFirstStoreResult></ClaimFirstStoreResponse>`)
-		create, got, err := c.Claim(context.Background())
-		if err != nil || create || got != pub {
-			t.Fatalf("claim = %v %q %v", create, got, err)
-		}
-	})
+	c, _, _ := server(t, 200, claimBody(`<Pending>false</Pending><Fingerprint>`+fp+`</Fingerprint>`))
+	got, err := c.ClaimReceiptKey(context.Background())
+	if err != nil || got.Pending || got.Fingerprint != fp {
+		t.Fatalf("recorded claim = %+v %v", got, err)
+	}
 }

-func TestRecordPublicKey_SendsOnlyAPublicKey(t *testing.T) {
-	pub, seed := userKey(t)
-	c, n, seen := server(t, 200, `<RecordWorkerPublicKeyResponse/>`)
-	if err := c.RecordPublicKey(context.Background(), seed); err == nil || strings.Contains(err.Error(), seed) {
-		t.Fatalf("seed err = %v", err)
+func TestRecordReceiptKey_SendsOnlyAFingerprint(t *testing.T) {
+	c, n, seen := server(t, 200, `<RecordReceiptKeyResponse/>`)
+	const raw = "not-a-fingerprint-and-maybe-key-material"
+	if err := c.RecordReceiptKey(context.Background(), raw); err == nil || strings.Contains(err.Error(), raw) {
+		t.Fatalf("invalid err = %v", err)
 	}
 	if n.Load() != 0 {
-		t.Fatal("a seed was sent")
+		t.Fatal("a non-fingerprint was sent")
 	}
-	if err := c.RecordPublicKey(context.Background(), pub); err != nil {
+	if err := c.RecordReceiptKey(context.Background(), fp); err != nil {
 		t.Fatal(err)
 	}
-	if !strings.Contains((*seen)[0], "PublicKey="+pub) {
+	if !strings.Contains((*seen)[0], "Fingerprint="+fp) {
 		t.Fatalf("request = %q", (*seen)[0])
 	}
 }
@@ -151,7 +137,7 @@ func TestErrors_ClassifiedWithoutEchoingTheBody(t *testing.T) {
 	}
 	for _, tc := range cases {
 		c, _, _ := server(t, tc.status, `<ErrorResponse><Error><Code>`+tc.code+`</Code><Message>secret-ish detail</Message></Error></ErrorResponse>`)
-		_, _, err := c.Claim(context.Background())
+		_, err := c.ClaimReceiptKey(context.Background())
 		api, ok := errors.AsType[*APIError](err)
 		if !ok || api.Code != tc.code || api.Retryable() != tc.retryable || api.Pending() != (tc.code == "AccessDenied") {
 			t.Fatalf("%s: err = %v", tc.code, err)
@@ -163,7 +149,7 @@ func TestErrors_ClassifiedWithoutEchoingTheBody(t *testing.T) {

 	c, _, _ := server(t, 200, "")
 	c.baseURL = "https://127.0.0.1:1"
-	_, _, err := c.Claim(context.Background())
+	_, err := c.ClaimReceiptKey(context.Background())
 	tr, ok := errors.AsType[*TransientError](err)
 	if !ok || !tr.Retryable() {
 		t.Fatalf("unreachable err = %v, want TransientError", err)
@@ -181,55 +167,10 @@ func TestCall_DoesNotFollowRedirects(t *testing.T) {
 	}))
 	t.Cleanup(srv.Close)
 	c.baseURL = srv.URL
-	if _, _, err := c.Claim(context.Background()); err == nil {
+	if _, err := c.ClaimReceiptKey(context.Background()); err == nil {
 		t.Fatal("a redirect was accepted as a decision")
 	}
 	if followed.Load() {
 		t.Fatal("the signed request followed a redirect")
 	}
 }
-
-// Only "pending" with nothing else, or "bound" with an IPv4 address and one
-// valid public key, is a store state.
-func TestDescribeStore_AcceptsOnlyAValidState(t *testing.T) {
-	pub, seed := userKey(t)
-	wrap := func(inner string) string {
-		return `<DescribeStoreResponse><DescribeStoreResult>` + inner + `</DescribeStoreResult></DescribeStoreResponse>`
-	}
-	t.Run("pending", func(t *testing.T) {
-		c, _, seen := server(t, 200, wrap(`<State>pending</State>`))
-		st, err := c.DescribeStore(context.Background())
-		if err != nil || st.Bound {
-			t.Fatalf("store = %+v %v", st, err)
-		}
-		if !strings.Contains((*seen)[0], "Action=DescribeStore") {
-			t.Fatalf("request = %q", (*seen)[0])
-		}
-	})
-	t.Run("bound", func(t *testing.T) {
-		c, _, _ := server(t, 200, wrap(`<State>bound</State><Address>10.240.0.9</Address><PublicKey>`+pub+`</PublicKey>`))
-		st, err := c.DescribeStore(context.Background())
-		if err != nil || !st.Bound || st.Address != "10.240.0.9" || st.PublicKey != pub {
-			t.Fatalf("store = %+v %v", st, err)
-		}
-	})
-	for name, inner := range map[string]string{
-		"pending with address": `<State>pending</State><Address>10.240.0.9</Address>`,
-		"bound without key":    `<State>bound</State><Address>10.240.0.9</Address>`,
-		"bound with seed":      `<State>bound</State><Address>10.240.0.9</Address><PublicKey>` + seed + `</PublicKey>`,
-		"bound with hostname":  `<State>bound</State><Address>store.example</Address><PublicKey>` + pub + `</PublicKey>`,
-		"bound with IPv6":      `<State>bound</State><Address>::ffff:10.240.0.9</Address><PublicKey>` + pub + `</PublicKey>`,
-		"unknown state":        `<State>ready</State>`,
-	} {
-		t.Run(name, func(t *testing.T) {
-			c, _, _ := server(t, 200, wrap(inner))
-			_, err := c.DescribeStore(context.Background())
-			if err == nil {
-				t.Fatal("accepted an invalid store state")
-			}
-			if strings.Contains(err.Error(), seed) {
-				t.Fatal("the error echoes a seed")
-			}
-		})
-	}
-}
@@ -0,0 +1,7 @@
+// Package deadpid gives tests a PID that no process can hold.
+package deadpid
+
+// PID is above the kernel's PID_MAX_LIMIT (4194304), so kill(2) on it always
+// fails with ESRCH. A plausible-looking literal such as 99999 is a real PID on
+// a busy host, and signalling it terminates whatever process holds it.
+const PID = 1 << 30
@@ -11,7 +11,6 @@ import (
 	"crypto/x509"
 	"crypto/x509/pkix"
 	"encoding/pem"
-	"errors"
 	"math/big"
 	"net"
 	"os"
@@ -70,7 +69,7 @@ func Start(t *testing.T, selfID uint64, members []Member) *Server {
 	})
 	s := &Server{Endpoint: "https://" + lis.Addr().String(), CertDir: dir, members: members, selfID: selfID}
 	gs := grpc.NewServer(grpc.Creds(creds))
-	pb.RegisterClusterServer(gs, &cluster{s})
+	pb.RegisterClusterServer(gs, &cluster{s: s})
 	pb.RegisterMaintenanceServer(gs, &maintenance{s: s})
 	go func() { _ = gs.Serve(lis) }()
 	t.Cleanup(gs.Stop)
@@ -91,7 +90,11 @@ func (s *Server) FailWith(err error) {
 	s.failAll = err
 }

-type cluster struct{ s *Server }
+type cluster struct {
+	pb.UnimplementedClusterServer
+
+	s *Server
+}

 func (c *cluster) MemberList(context.Context, *pb.MemberListRequest) (*pb.MemberListResponse, error) {
 	c.s.mu.Lock()
@@ -116,18 +119,6 @@ func (c *cluster) MemberRemove(_ context.Context, req *pb.MemberRemoveRequest) (
 	return &pb.MemberRemoveResponse{Header: &pb.ResponseHeader{MemberId: c.s.selfID}}, nil
 }

-func (c *cluster) MemberAdd(context.Context, *pb.MemberAddRequest) (*pb.MemberAddResponse, error) {
-	return nil, errors.New("not implemented")
-}
-
-func (c *cluster) MemberUpdate(context.Context, *pb.MemberUpdateRequest) (*pb.MemberUpdateResponse, error) {
-	return nil, errors.New("not implemented")
-}
-
-func (c *cluster) MemberPromote(context.Context, *pb.MemberPromoteRequest) (*pb.MemberPromoteResponse, error) {
-	return nil, errors.New("not implemented")
-}
-
 type maintenance struct {
 	pb.UnimplementedMaintenanceServer

@@ -0,0 +1,108 @@
+package testutil
+
+import (
+	"context"
+	"sync"
+
+	"github.com/nats-io/nats.go/jetstream"
+)
+
+// StaleReadKV wraps a bucket so that the first Get of a key after a write to it
+// returns the entry as it stood before that write, as a lagging replica does
+// when it answers a direct get. Later Gets see the current entry.
+type StaleReadKV struct {
+	jetstream.KeyValue
+
+	mu    sync.Mutex
+	stale map[string]jetstream.KeyValueEntry
+}
+
+var _ jetstream.KeyValue = (*StaleReadKV)(nil)
+
+// NewStaleReadKV wraps kv.
+func NewStaleReadKV(kv jetstream.KeyValue) *StaleReadKV {
+	return &StaleReadKV{KeyValue: kv, stale: map[string]jetstream.KeyValueEntry{}}
+}
+
+// remember captures the entry about to be overwritten.
+func (k *StaleReadKV) remember(ctx context.Context, key string) jetstream.KeyValueEntry {
+	prev, err := k.KeyValue.Get(ctx, key)
+	if err != nil {
+		return nil
+	}
+	return prev
+}
+
+func (k *StaleReadKV) arm(key string, prev jetstream.KeyValueEntry) {
+	if prev == nil {
+		return
+	}
+	k.mu.Lock()
+	k.stale[key] = prev
+	k.mu.Unlock()
+}
+
+func (k *StaleReadKV) Get(ctx context.Context, key string) (jetstream.KeyValueEntry, error) {
+	k.mu.Lock()
+	prev, ok := k.stale[key]
+	delete(k.stale, key)
+	k.mu.Unlock()
+	if ok {
+		return prev, nil
+	}
+	return k.KeyValue.Get(ctx, key)
+}
+
+func (k *StaleReadKV) Update(ctx context.Context, key string, value []byte, last uint64) (uint64, error) {
+	prev := k.remember(ctx, key)
+	rev, err := k.KeyValue.Update(ctx, key, value, last)
+	if err == nil {
+		k.arm(key, prev)
+	}
+	return rev, err
+}
+
+func (k *StaleReadKV) Put(ctx context.Context, key string, value []byte) (uint64, error) {
+	prev := k.remember(ctx, key)
+	rev, err := k.KeyValue.Put(ctx, key, value)
+	if err == nil {
+		k.arm(key, prev)
+	}
+	return rev, err
+}
+
+// RejectingKV wraps a bucket so that writes for which reject returns an error
+// fail with it, and every other operation passes through.
+type RejectingKV struct {
+	jetstream.KeyValue
+
+	reject func(key string, value []byte) error
+}
+
+var _ jetstream.KeyValue = (*RejectingKV)(nil)
+
+// NewRejectingKV wraps kv. reject runs before each Put, Create and Update.
+func NewRejectingKV(kv jetstream.KeyValue, reject func(key string, value []byte) error) *RejectingKV {
+	return &RejectingKV{KeyValue: kv, reject: reject}
+}
+
+func (k *RejectingKV) Put(ctx context.Context, key string, value []byte) (uint64, error) {
+	if err := k.reject(key, value); err != nil {
+		return 0, err
+	}
+	return k.KeyValue.Put(ctx, key, value)
+}
+
+func (k *RejectingKV) Create(ctx context.Context, key string, value []byte, opts ...jetstream.KVCreateOpt) (uint64, error) {
+	if err := k.reject(key, value); err != nil {
+		return 0, err
+	}
+	return k.KeyValue.Create(ctx, key, value, opts...)
+}
+
+func (k *RejectingKV) Update(ctx context.Context, key string, value []byte, last uint64) (uint64, error) {
+	if err := k.reject(key, value); err != nil {
+		return 0, err
+	}
+	return k.KeyValue.Update(ctx, key, value, last)
+}
@@ -7,17 +7,14 @@ package pagedstore

 import (
 	"context"
-	"sort"
-	"strconv"

 	"github.com/aws/aws-sdk-go/aws"
 	"github.com/aws/aws-sdk-go/service/s3"
 	"github.com/mulgadc/spinifex/spinifex/providers/objectstore"
 )

-// Store wraps a MemoryObjectStore and splits ListObjectsV2's answer into
-// pages of at most PageSize objects, sorted by key so pagination is
-// deterministic across runs (MemoryObjectStore itself iterates a map).
+// Store wraps a MemoryObjectStore whose ListObjectsV2 pages like S3, capping
+// every page at PageSize entries so small fixtures span several pages.
 type Store struct {
 	*objectstore.MemoryObjectStore

@@ -28,51 +25,20 @@ type Store struct {
 	Calls int
 }

-// New returns a paging store with the given per-page object limit.
+// New returns an empty Store that serves at most pageSize entries per page.
 func New(pageSize int) *Store {
 	return &Store{MemoryObjectStore: objectstore.NewMemoryObjectStore(), PageSize: pageSize}
 }

 var _ objectstore.ObjectStore = (*Store)(nil)

-// ListObjectsV2 delegates to MemoryObjectStore for prefix/delimiter matching,
-// then slices the sorted result into one page per call, resuming from the
-// offset encoded in ContinuationToken.
+// ListObjectsV2 lowers the caller's MaxKeys to PageSize and lets the memory
+// store page, so tokens, ordering and common prefixes follow S3's rules.
 func (s *Store) ListObjectsV2(ctx context.Context, input *s3.ListObjectsV2Input) (*s3.ListObjectsV2Output, error) {
 	s.Calls++
-
-	full, err := s.MemoryObjectStore.ListObjectsV2(ctx, input)
-	if err != nil {
-		return nil, err
+	page := *input
+	if page.MaxKeys == nil || *page.MaxKeys > int64(s.PageSize) {
+		page.MaxKeys = aws.Int64(int64(s.PageSize))
 	}
-
-	sort.Slice(full.Contents, func(i, j int) bool {
-		return aws.StringValue(full.Contents[i].Key) < aws.StringValue(full.Contents[j].Key)
-	})
-
-	start := 0
-	if tok := aws.StringValue(input.ContinuationToken); tok != "" {
-		start, err = strconv.Atoi(tok)
-		if err != nil {
-			return nil, err
-		}
-	}
-	end := min(start+s.PageSize, len(full.Contents))
-	if start > end {
-		start = end
-	}
-
-	page := &s3.ListObjectsV2Output{
-		Contents:       full.Contents[start:end],
-		CommonPrefixes: full.CommonPrefixes,
-		Name:           input.Bucket,
-		KeyCount:       aws.Int64(int64(end - start)),
-	}
-	if end < len(full.Contents) {
-		page.IsTruncated = aws.Bool(true)
-		page.NextContinuationToken = aws.String(strconv.Itoa(end))
-	} else {
-		page.IsTruncated = aws.Bool(false)
-	}
-	return page, nil
+	return s.MemoryObjectStore.ListObjectsV2(ctx, &page)
 }
@@ -6,7 +6,7 @@ Node installation, dev-environment lifecycle, guest image builds, and verificati

 | Script | What it does |
 |---|---|
-| `setup.sh` | The production binary installer, served from `install.mulgadc.com`. Installs deps, users, sudoers, the host firewall policy, files, systemd units, logrotate and udev rules, and loads `openvswitch`, `geneve`, `vhost_net` and `tun`, listing them in `/etc/modules-load.d/spinifex.conf` for later boots. It masks firewalld wherever it is installed, since the Spinifex nftables policy is the only supported node firewall. It writes a needrestart drop-in so an apt upgrade never restarts the Spinifex, OVS or OVN units; restarting them after an upgrade is left to the operator. An unset `--firewall` keeps the mode recorded in `/etc/spinifex/firewall/mode`, and otherwise arms the policy when firewalld was active. Where systemd-resolved is installed but not running, resolvconf is absent and NetworkManager runs, it hands host DNS to resolved: it saves `/etc/resolv.conf` once to `/var/lib/spinifex-reset/resolv.conf.orig`, links `/etc/resolv.conf` to the resolved stub, and stops unless resolved has upstream servers and the package repository host still resolves. It links `/usr/bin/spx` to `/usr/local/bin/spx` where sudo's PATH leaves out `/usr/local/bin`, and takes the nbdkit plugin directory from `nbdkit --dump-config`, stopping when nbdkit cannot say. `SETUP_STAGES` runs a subset; `ISO_BUILD=1` runs it inside the ISO builder's chroot. |
+| `setup.sh` | The production binary installer, served from `install.mulgadc.com`. Installs deps, users, sudoers, the host firewall policy, files, systemd units, logrotate and udev rules, and loads `openvswitch`, `geneve`, `vhost_net` and `tun`, listing them in `/etc/modules-load.d/spinifex.conf` for later boots. It masks firewalld wherever it is installed, since the Spinifex nftables policy is the only supported node firewall. It writes a needrestart drop-in so an apt upgrade never restarts the Spinifex, OVS or OVN units; restarting them after an upgrade is left to the operator. An unset `--firewall` keeps the mode recorded in `/etc/spinifex/firewall/mode`, and otherwise arms the policy when firewalld was active. Where systemd-resolved is installed but not running, resolvconf is absent and NetworkManager runs, it hands host DNS to resolved: it saves `/etc/resolv.conf` once to `/var/lib/spinifex-reset/resolv.conf.orig`, writes `/etc/NetworkManager/conf.d/90-spinifex-dns.conf` (`dns=systemd-resolved`, `rc-manager=unmanaged`) so NetworkManager never rewrites `/etc/resolv.conf`, links `/etc/resolv.conf` to the resolved stub, and stops unless the link holds, resolved has upstream servers and the package repository host still resolves. A rerun after that handover adds a missing drop-in, reloads NetworkManager and relinks the stub. It links `/usr/bin/spx` to `/usr/local/bin/spx` where sudo's PATH leaves out `/usr/local/bin`, and takes the nbdkit plugin directory from `nbdkit --dump-config`, stopping when nbdkit cannot say. `SETUP_STAGES` runs a subset; `ISO_BUILD=1` runs it inside the ISO builder's chroot. |
 | `setup-ovn.sh` | Bootstraps a compute node for OVN VPC networking: installs OVN/OVS, creates `br-int`, configures the WAN bridge (bridged, DHCP, or routed-NAT uplink), sets the chassis identity, and applies overlay sysctls. It runs OVS and OVN as root, setting `OVS_USER_ID` and `OVN_USER_ID` where `/etc/sysconfig` holds them, and hooks the socket-permission helper onto `ovs-vswitchd`, `ovn-controller`, `openvswitch-ipsec` and, on EL, `ovn-northd`. `--management` also starts the OVN central services, writes their `ovn-ctl` flags to `/etc/default/ovn-central` (Debian) or `/etc/sysconfig/ovn-northd` (EL), and records the role in `/etc/spinifex/ovn-role`. With clustered databases it returns only once this node's NB and SB members are back in their RAFT clusters under a known leader, and fails after `OVN_RAFT_REJOIN_WAIT` seconds (default 60) otherwise, so reruns one node at a time never take two members of a cluster down. A later run without `--management` refuses while that file, or the `OVN_CTL_OPTS` line older nodes carry in `/etc/default/ovn-central`, records the role. It adds `maxsize 100M` to the OVS and OVN logrotate files, plus `rotate 5` where the file sets no count, and rotates as root, matching the daemons. On a host whose IPsec stack is libreswan it enables, stops and masks `openvswitch-ipsec.service` and `ipsec.service` for the daemon to unmask when IPsec is on, unless `openvswitch-ipsec.service` is already running; elsewhere it enables and starts `openvswitch-ipsec.service`, unless the daemon has masked it to keep IPsec off. The links it creates persist through systemd-networkd where that is enabled, and otherwise through `/etc/spinifex/hostnet.json`, which `spinifex-hostnet.service` (`spx admin hostnet apply`) replays at boot and on every OVS start, starting `ovn-controller` again when an OVS stop took it down; the first run records the choice in `/etc/spinifex/hostnet-backend`. Where NetworkManager runs, it leaves the links Spinifex owns unmanaged. |
 | `install-node.sh` | Forms a multi-node cluster across hosts that are already running Spinifex: rebuilds the OVN database in clustered form, runs `spx admin init` on the first host and `spx admin join` on the rest, then verifies. Before touching any host it refuses hosts of different OS families (`distro_os_family`) or `spx` versions, and `--ipsec on` when any host uses libreswan. The scripted form of [install-multi-node](../docs/install/install-multi-node/README.md). Destroys the joining nodes' logical network state and crypto — see its `--help`. |
 | `dev-install.sh` | Full local dev setup through the production installer: builds from source, packs a tarball, runs `setup.sh`, initialises the cluster, and starts services under systemd. |
@@ -0,0 +1,97 @@
+#!/bin/bash
+# build-lambda-runtime-microvm-image.sh — build the Lambda runtime microVM kernel
+# + initramfs: a guest running the Lambda guest agent, with virtio-net for its
+# host control connection. It is not shipped.
+#
+# Inputs:
+#   $REPO_ROOT/build/microvm/lambda-runtime/{init.sh,inittab,run-agent}
+#   $REPO_ROOT/bin/lambda-agent  (make bin/lambda-agent)
+#
+# Outputs:
+#   $REPO_ROOT/build/microvm/lambda-runtime/vmlinuz
+#   $REPO_ROOT/build/microvm/lambda-runtime/initramfs.cpio.gz
+#
+# Rootfs strategy and module stripping: scripts/lib/microvm-image.sh.
+# shellcheck disable=SC2034 # variables are consumed by the sourced library
+set -euo pipefail
+
+REPO_ROOT="$(cd "$(dirname "$0")/.." && pwd)"
+BUILD_DIR="$REPO_ROOT/build/microvm/lambda-runtime"
+INIT_SH="$BUILD_DIR/init.sh"
+INITTAB="$BUILD_DIR/inittab"
+RUN_AGENT="$BUILD_DIR/run-agent"
+LAMBDA_AGENT_BIN="$REPO_ROOT/bin/lambda-agent"
+TAG="[build-lambda-runtime-microvm-image]"
+
+ALPINE_VERSION="${ALPINE_VERSION:-3.24}"
+ALPINE_MIRROR="${ALPINE_MIRROR:-https://dl-cdn.alpinelinux.org/alpine}"
+# kmod: the image's own depmod/modprobe rebuild module metadata. busybox ships ip.
+ALPINE_PACKAGES="busybox linux-virt kmod ca-certificates"
+# tiny_power_button turns the ACPI power button into a signal busybox init
+# handles as poweroff.
+MICROVM_EXTRA_MODULES="${MICROVM_EXTRA_MODULES:-tiny_power_button}"
+# virtio_net needs failover and net_failover, which the virtio* glob misses.
+MICROVM_EXTRA_KEEP_GLOBS=("*failover*.ko*")
+MICROVM_GATE_MODULES="qemu_fw_cfg virtio_net"
+
+# shellcheck source=lib/microvm-image.sh
+source "$REPO_ROOT/scripts/lib/microvm-image.sh"
+
+echo "$TAG repo root: $REPO_ROOT"
+echo "$TAG build dir: $BUILD_DIR"
+
+microvm_check_tools
+
+for f in "$INIT_SH" "$INITTAB" "$RUN_AGENT"; do
+    if [ ! -f "$f" ]; then
+        echo "ERROR: $f not found — cannot build initramfs" >&2
+        exit 1
+    fi
+done
+if [ ! -x "$LAMBDA_AGENT_BIN" ]; then
+    echo "ERROR: $LAMBDA_AGENT_BIN not found — run 'make bin/lambda-agent'" >&2
+    exit 1
+fi
+if ! file -b "$LAMBDA_AGENT_BIN" | grep -q "statically linked"; then
+    echo "ERROR: $LAMBDA_AGENT_BIN is not statically linked — it cannot run on the musl guest" >&2
+    exit 1
+fi
+
+microvm_init_dirs
+microvm_make_rootfs
+microvm_empty_dev
+
+echo "$TAG installing init, inittab, run-agent and lambda-agent..."
+install -m 0755 "$INIT_SH" "$CHROOT_DIR/init"
+install -m 0644 "$INITTAB" "$CHROOT_DIR/etc/inittab"
+mkdir -p "$CHROOT_DIR/usr/local/lib/lambda-runtime" "$CHROOT_DIR/usr/local/bin"
+install -m 0755 "$RUN_AGENT" "$CHROOT_DIR/usr/local/lib/lambda-runtime/run-agent"
+install -m 0755 "$LAMBDA_AGENT_BIN" "$CHROOT_DIR/usr/local/bin/lambda-agent"
+# Function-code mountpoint and scratch space ship empty.
+mkdir -p "$CHROOT_DIR/var/task" "$CHROOT_DIR/tmp" \
+    "$CHROOT_DIR/proc" "$CHROOT_DIR/sys" "$CHROOT_DIR/dev" "$CHROOT_DIR/etc/ssl/certs"
+# The bootstrap runs unprivileged, so / must be traversable (mktemp -d makes it
+# 0700) and nothing it runs from may be group-writable.
+chmod 0755 "$CHROOT_DIR" "$CHROOT_DIR/var/task" "$CHROOT_DIR/usr/local/lib/lambda-runtime"
+
+microvm_locate_kernel
+microvm_strip_modules
+microvm_strip_rootfs
+microvm_stage_kernel
+microvm_pack_initramfs
+
+# Nothing reaches $BUILD_DIR until the staged pair passes these checks.
+for out in vmlinuz initramfs.cpio.gz; do
+    if [ ! -s "$STAGE_DIR/$out" ]; then
+        echo "ERROR: staged $out is missing or empty" >&2
+        exit 1
+    fi
+done
+if ! gzip -t "$STAGE_DIR/initramfs.cpio.gz"; then
+    echo "ERROR: staged initramfs.cpio.gz is not valid gzip" >&2
+    exit 1
+fi
+mv -f "$STAGE_DIR/vmlinuz" "$BUILD_DIR/vmlinuz"
+mv -f "$STAGE_DIR/initramfs.cpio.gz" "$BUILD_DIR/initramfs.cpio.gz"
+echo ""
+echo "$TAG done."
@@ -1,44 +1,37 @@
 #!/bin/bash
-# build-sqs-store-microvm-image.sh — build the SQS store microVM kernel +
-# initramfs: the dedicated nats-server, ext4 support, the store's guest steps
-# and a clean power-down. The guest never formats a disk; build/microvm/
-# sqs-store/sqs-store-lib.sh states its fw_cfg contract. The LB image
-# (build-microvm-image.sh) is separate and unchanged.
-#
-# SQS_MICROVM_PROFILE=worker builds the shared SQS worker guest instead: the
-# same network /init and kernel, build/microvm/sqs-worker's steps and
-# $REPO_ROOT/bin/sqs-worker, with no nats-server.
+# build-sqs-store-microvm-image.sh — build the SQS microVM kernel + initramfs:
+# one guest running the dedicated SQS nats-server and the SQS worker as separate
+# processes, ext4 support, its guest steps and a clean power-down. The guest
+# never formats a disk. The LB image (build-microvm-image.sh) is separate.
 #
 # Inputs:
-#   $REPO_ROOT/build/microvm/sqs-store/init.sh, inittab
-#   $REPO_ROOT/build/microvm/sqs-store/sqs-store-{lib.sh,prepare,run,stop}
+#   $REPO_ROOT/build/microvm/sqs-worker/init.sh, inittab
+#   $REPO_ROOT/build/microvm/sqs-worker/sqs-worker-{lib.sh,prepare,ready,stop}
 #   $REPO_ROOT/bin/nats-server  (make bin/nats-server, the go.mod-pinned version)
+#   $REPO_ROOT/bin/sqs-worker
 #
 # Outputs:
-#   $REPO_ROOT/build/microvm/sqs-store/vmlinuz
-#   $REPO_ROOT/build/microvm/sqs-store/initramfs.cpio.gz
+#   $REPO_ROOT/build/microvm/sqs-worker/vmlinuz
+#   $REPO_ROOT/build/microvm/sqs-worker/initramfs.cpio.gz
 #
 # Rootfs strategy (first match wins):
 #   1. apk available on host → install directly into a temp chroot (Alpine CI/CD)
 #   2. docker available      → run alpine container, export filesystem
 #   3. podman available      → same via podman
+# shellcheck disable=SC2034 # variables are consumed by the sourced library
 set -euo pipefail

 REPO_ROOT="$(cd "$(dirname "$0")/.." && pwd)"
-BUILD_DIR="$REPO_ROOT/build/microvm/sqs-store"
+BUILD_DIR="$REPO_ROOT/build/microvm/sqs-worker"
 INIT_SH="$BUILD_DIR/init.sh"
 INITTAB="$BUILD_DIR/inittab"
 NATS_SERVER_BIN="$REPO_ROOT/bin/nats-server"
+SQS_WORKER_BIN="$REPO_ROOT/bin/sqs-worker"

-PROFILE="${SQS_MICROVM_PROFILE:-store}"
+PROFILE="${SQS_MICROVM_PROFILE:-worker}"
 case "$PROFILE" in
-    store) ;;
-    worker)
-        BUILD_DIR="$REPO_ROOT/build/microvm/sqs-worker"
-        INITTAB="$BUILD_DIR/inittab"
-        SQS_WORKER_BIN="$REPO_ROOT/bin/sqs-worker"
-        ;;
-    *) echo "ERROR: SQS_MICROVM_PROFILE must be store or worker, not $PROFILE" >&2; exit 1 ;;
+    worker) ;;
+    *) echo "ERROR: SQS_MICROVM_PROFILE must be worker, not $PROFILE" >&2; exit 1 ;;
 esac
 TAG="[build-sqs-$PROFILE-microvm-image]"

@@ -52,32 +45,28 @@ ALPINE_PACKAGES="busybox linux-virt iproute2 ca-certificates e2fsprogs kmod"
 # button (QMP system_powerdown) into a signal busybox init handles as poweroff.
 MICROVM_EXTRA_MODULES="${MICROVM_EXTRA_MODULES:-ext4 tiny_power_button}"

+# Module file globs kept on top of virtio*/*fw_cfg*: virtio_net deps for live migration.
+MICROVM_EXTRA_KEEP_GLOBS=("*failover*.ko*")
+# Modules the image modprobe must resolve, beyond the extras.
+MICROVM_GATE_MODULES="qemu_fw_cfg virtio_net"
+
+# shellcheck source=lib/microvm-image.sh
+source "$REPO_ROOT/scripts/lib/microvm-image.sh"
+
 echo "$TAG repo root: $REPO_ROOT"
 echo "$TAG build dir: $BUILD_DIR"
 mkdir -p "$BUILD_DIR"

-# --- Verify non-negotiable tools ---
-# fakeroot is mandatory: device nodes are written into the cpio archive inside a
-# faked-root environment so no real device node is ever created on the host
-# filesystem (which previously required sudo mknod and risked clobbering host
-# /dev/null on cleanup).
-for tool in cpio gzip find fakeroot; do
-    if ! command -v "$tool" >/dev/null 2>&1; then
-        echo "ERROR: required tool not found: $tool" >&2
-        exit 1
-    fi
-done
+microvm_check_tools

-if [ "$PROFILE" = worker ]; then
-    if [ ! -x "$SQS_WORKER_BIN" ]; then
-        echo "ERROR: $SQS_WORKER_BIN not found — run 'make bin/sqs-worker'" >&2
-        exit 1
-    fi
-    if ! file -b "$SQS_WORKER_BIN" | grep -q "statically linked"; then
-        echo "ERROR: $SQS_WORKER_BIN is not statically linked — it cannot run on the musl guest" >&2
-        exit 1
-    fi
-else
+if [ ! -x "$SQS_WORKER_BIN" ]; then
+    echo "ERROR: $SQS_WORKER_BIN not found — run 'make bin/sqs-worker'" >&2
+    exit 1
+fi
+if ! file -b "$SQS_WORKER_BIN" | grep -q "statically linked"; then
+    echo "ERROR: $SQS_WORKER_BIN is not statically linked — it cannot run on the musl guest" >&2
+    exit 1
+fi
 # --- Verify the dedicated nats-server is the go.mod-pinned build ---
 # A stale or host-installed binary would ship an unreviewed server version.
 NATS_SERVER_PIN=$(awk '$1 == "github.com/nats-io/nats-server/v2" { print $2; exit }' "$REPO_ROOT/go.mod")
@@ -99,102 +88,12 @@ if [ "$nats_server_version" != "$NATS_SERVER_PIN" ]; then
    exit 1
 fi
 echo "$TAG nats-server: $nats_server_version (matches go.mod)"
-fi

-# --- Create chroot ---
-CHROOT_DIR=$(mktemp -d)
-# Guard: every destructive op below targets "$CHROOT_DIR/...". An empty or
-# unexpected value would aim rm -rf at the host (e.g. /dev). Refuse anything that
-# is not a freshly-created temp dir.
-case "${CHROOT_DIR:?mktemp -d returned empty}" in
-    /tmp/*|/var/tmp/*|"${TMPDIR:-/nonexistent}"/*) ;;
-    *) echo "ERROR: refusing to use unexpected CHROOT_DIR: $CHROOT_DIR" >&2; exit 1 ;;
-esac
-# Outputs are written here first, beside $BUILD_DIR so publishing is a rename.
-STAGE_DIR=$(mktemp -d "$BUILD_DIR/.stage.XXXXXX")
-CONTAINER_TOOL=""
-CONTAINER_CID=""
-cleanup() {
-    echo "$TAG cleaning up chroot: $CHROOT_DIR"
-    rm -rf "$CHROOT_DIR"
-    rm -rf "${STAGE_DIR:?}"
-    if [ -n "$CONTAINER_CID" ] && [ -n "$CONTAINER_TOOL" ]; then
-        echo "$TAG cleaning up container: $CONTAINER_CID"
-        $CONTAINER_TOOL rm -f "$CONTAINER_CID" >/dev/null 2>&1 || true
-    fi
-}
-trap cleanup EXIT
+microvm_init_dirs

-# --- Populate rootfs ---
-# Strategy 1: native apk (Alpine hosts, CI)
-if command -v apk >/dev/null 2>&1; then
-    echo "$TAG rootfs: native apk"
-    mkdir -p \
-        "$CHROOT_DIR/etc/apk" \
-        "$CHROOT_DIR/lib/apk/db" \
-        "$CHROOT_DIR/var/cache/apk"
-    cat > "$CHROOT_DIR/etc/apk/repositories" <<EOF
-${ALPINE_MIRROR}/v${ALPINE_VERSION}/main
-${ALPINE_MIRROR}/v${ALPINE_VERSION}/community
-EOF
-    apk add \
-        --root "$CHROOT_DIR" \
-        --arch x86_64 \
-        --no-cache \
-        --allow-untrusted \
-        --initdb \
-        $ALPINE_PACKAGES
+microvm_make_rootfs

-# Strategy 2: docker -> BuildKit local export.
-# Do NOT use `docker export` here. On Docker's overlay2 driver it leaks the
-# container's overlay mount (the mount survives `docker rm`), orphaning layers
-# under /var/lib/docker that only a daemon restart reclaims; on a busy CI host
-# this grows without bound. `buildx --output type=local` writes the image
-# filesystem straight to a directory and releases every mount, so nothing leaks.
-elif command -v docker >/dev/null 2>&1 && docker info >/dev/null 2>&1; then
-    echo "$TAG rootfs: docker buildx --output type=local (alpine:${ALPINE_VERSION})"
-    BUILD_CTX=$(mktemp -d)
-    cat > "$BUILD_CTX/Dockerfile" <<DOCKERFILE
-FROM alpine:${ALPINE_VERSION}
-RUN apk add --no-cache ${ALPINE_PACKAGES}
-DOCKERFILE
-    docker buildx build \
-        --platform linux/amd64 \
-        --output "type=local,dest=${CHROOT_DIR}" \
-        "$BUILD_CTX"
-    rm -rf "$BUILD_CTX"
-
-# Strategy 3: podman -> run + export. Podman's store does not exhibit the docker
-# overlay2 export-mount leak, and podman has no buildx.
-elif command -v podman >/dev/null 2>&1; then
-    echo "$TAG rootfs: podman export (alpine:${ALPINE_VERSION})"
-    CONTAINER_TOOL=podman
-    CONTAINER_CID=$(podman run -d \
-        "alpine:${ALPINE_VERSION}" \
-        sh -c "apk add --no-cache ${ALPINE_PACKAGES}")
-
-    exit_code=$(podman wait "$CONTAINER_CID")
-    if [ "$exit_code" != "0" ]; then
-        echo "ERROR: package installation failed in container (exit $exit_code)" >&2
-        exit 1
-    fi
-
-    podman export "$CONTAINER_CID" | tar -x -C "$CHROOT_DIR"
-
-else
-    echo "ERROR: no rootfs build tool available." >&2
-    echo "       Install one of: apk (Alpine), docker, podman" >&2
-    exit 1
-fi
-
-# --- /dev: empty mountpoint only ---
-# tar extraction without root silently creates 0-byte regular files for device
-# nodes. Drop them and leave an empty /dev mountpoint. The two static nodes the
-# kernel needs (/dev/console to wire init's stdio, /dev/null) are written into
-# the cpio archive under fakeroot at pack time — NOT created on the host. init
-# mounts devtmpfs over /dev as its first action for everything else at runtime.
-rm -rf "${CHROOT_DIR:?}/dev"
-mkdir "$CHROOT_DIR/dev"
+microvm_empty_dev

 # --- Place init script ---
 echo "$TAG installing init.sh as /init..."
@@ -213,39 +112,23 @@ fi
 echo "$TAG installing inittab..."
 install -m 0644 "$INITTAB" "$CHROOT_DIR/etc/inittab"

-if [ "$PROFILE" = worker ]; then
-    echo "$TAG installing worker guest steps..."
-    for step in sqs-worker-lib.sh sqs-worker-prepare sqs-worker-ready sqs-worker-stop; do
-        if [ ! -f "$BUILD_DIR/$step" ]; then
-            echo "ERROR: $BUILD_DIR/$step not found — cannot build initramfs" >&2
-            exit 1
-        fi
-    done
-    mkdir -p "$CHROOT_DIR/usr/local/lib/sqs-worker" "$CHROOT_DIR/usr/local/bin"
-    install -m 0755 "$BUILD_DIR/sqs-worker-lib.sh" "$BUILD_DIR/sqs-worker-prepare" "$BUILD_DIR/sqs-worker-ready" "$BUILD_DIR/sqs-worker-stop" \
-        "$CHROOT_DIR/usr/local/lib/sqs-worker/"
-    install -m 0755 "$SQS_WORKER_BIN" "$CHROOT_DIR/usr/local/bin/sqs-worker"
-    # The mountpoint ships empty; the worker proves its disk is mounted here.
-    mkdir -p "$CHROOT_DIR/data"
-else
-# --- Place the store's guest steps ---
-echo "$TAG installing store guest steps..."
-mkdir -p "$CHROOT_DIR/usr/local/lib/sqs-store"
-for step in sqs-store-lib.sh sqs-store-prepare sqs-store-run sqs-store-stop; do
+echo "$TAG installing worker guest steps..."
+for step in sqs-worker-lib.sh sqs-worker-prepare sqs-worker-store sqs-worker-ready sqs-worker-stop; do
    if [ ! -f "$BUILD_DIR/$step" ]; then
        echo "ERROR: $BUILD_DIR/$step not found — cannot build initramfs" >&2
        exit 1
    fi
 done
-install -m 0644 "$BUILD_DIR/sqs-store-lib.sh" "$CHROOT_DIR/usr/local/lib/sqs-store/"
-install -m 0755 "$BUILD_DIR/sqs-store-prepare" "$BUILD_DIR/sqs-store-run" "$BUILD_DIR/sqs-store-stop" \
-    "$CHROOT_DIR/usr/local/lib/sqs-store/"
-
+mkdir -p "$CHROOT_DIR/usr/local/lib/sqs-worker" "$CHROOT_DIR/usr/local/bin"
+install -m 0755 "$BUILD_DIR/sqs-worker-lib.sh" "$BUILD_DIR/sqs-worker-prepare" "$BUILD_DIR/sqs-worker-store" "$BUILD_DIR/sqs-worker-ready" "$BUILD_DIR/sqs-worker-stop" \
+    "$CHROOT_DIR/usr/local/lib/sqs-worker/"
+install -m 0755 "$SQS_WORKER_BIN" "$CHROOT_DIR/usr/local/bin/sqs-worker"
+# The mountpoint ships empty; the worker proves its disk is mounted here.
+mkdir -p "$CHROOT_DIR/data"
 # --- Copy nats-server binary ---
 echo "$TAG copying nats-server binary..."
 mkdir -p "$CHROOT_DIR/usr/local/bin"
 install -m 0755 "$NATS_SERVER_BIN" "$CHROOT_DIR/usr/local/bin/nats-server"
-fi

 # --- Create required directories for init ---
 mkdir -p \
@@ -254,30 +137,8 @@ mkdir -p \
    "$CHROOT_DIR/dev" \
    "$CHROOT_DIR/etc/ssl/certs"

-# --- Locate kernel ---
-KERNEL_IMG=$(find "$CHROOT_DIR/boot" -name "vmlinuz*" | sort | tail -1)
-if [ -z "$KERNEL_IMG" ]; then
-    echo "ERROR: no vmlinuz found in $CHROOT_DIR/boot" >&2
-    exit 1
-fi
-echo "$TAG kernel image: $KERNEL_IMG"

-KERNEL_VER=$(basename "$KERNEL_IMG" | sed 's/vmlinuz-//')
-KERNEL_CONFIG="$CHROOT_DIR/boot/config-${KERNEL_VER}"
-
-# --- Assert fw_cfg module ---
-if ! find "$CHROOT_DIR/lib/modules" \
-        -name "qemu_fw_cfg.ko*" \
-        -o -name "fw_cfg_sysfs.ko*" \
-    | grep -q .; then
-    if ! grep -q "CONFIG_FW_CFG_SYSFS=y" "$KERNEL_CONFIG" 2>/dev/null; then
-        echo "ERROR: qemu_fw_cfg module missing from initramfs — fw_cfg reads will fail at boot" >&2
-        exit 1
-    fi
-    echo "$TAG fw_cfg: built-in (CONFIG_FW_CFG_SYSFS=y)"
-else
-    echo "$TAG fw_cfg: module found in lib/modules"
-fi
+microvm_locate_kernel

 # --- Assert ext4 userspace tools (data volume) ---
 for fs_tool in mkfs.ext4 e2fsck; do
@@ -288,227 +149,10 @@ for fs_tool in mkfs.ext4 e2fsck; do
 done
 echo "$TAG e2fsprogs: mkfs.ext4 + e2fsck present"

-# --- Strip kernel modules (Phase 2 keep-list approach) ---
-# Only modules needed by a virtio-mmio microvm guest are kept:
-#   virtio*.ko*    — virtio bus, net, blk, rng, console, and helpers
-#   *fw_cfg*.ko*   — QEMU fw_cfg sysfs driver (delivers boot blobs)
-#   *failover*.ko* — net_failover + failover (virtio_net deps for live migration)
-#   $MICROVM_EXTRA_MODULES — ext4 by default, plus dependency closure
-# All other loadable modules are removed. Drivers compiled into the kernel
-# (CONFIG_VIRTIO_MMIO=y etc.) are unaffected — they live in vmlinuz, not here.
-
-# module_path prints a module's kernel/ path from the stock modules.dep, by
-# name or modules.alias; it fails when the module is not loadable.
-module_path() {
-    local kdir="$1" name="${2//-/_}" path alias_target
-    path=$(awk -F: -v n="$name" '{ f = $1; sub(/.*\//, "", f); sub(/\.ko(\.gz|\.xz|\.zst)?$/, "", f);
-        gsub(/-/, "_", f); if (f == n) { print $1; exit } }' "$kdir/modules.dep")
-    if [ -z "$path" ] && [ -f "$kdir/modules.alias" ]; then
-        alias_target=$(awk -v n="$2" '$1 == "alias" && $2 == n { print $3; exit }' "$kdir/modules.alias")
-        [ -n "$alias_target" ] && [ "$alias_target" != "$2" ] && path=$(module_path "$kdir" "$alias_target")
-    fi
-    [ -n "$path" ] && printf '%s\n' "$path"
-}
-
-# module_builtin reports whether a module name is compiled into the kernel.
-module_builtin() {
-    local kdir="$1" name="${2//-/_}"
-    [ -f "$kdir/modules.builtin" ] || return 1
-    awk -v n="$name" '{ f = $0; sub(/.*\//, "", f); sub(/\.ko$/, "", f); gsub(/-/, "_", f);
-        if (f == n) { found = 1; exit } } END { exit !found }' "$kdir/modules.builtin"
-}
-
-# image_kmod runs the image's own kmod (depmod, modprobe) through its musl
-# loader, so module metadata matches the guest's modprobe without a host kmod.
-image_kmod() {
-    "$CHROOT_DIR/lib/ld-musl-x86_64.so.1" --library-path "$CHROOT_DIR/lib:$CHROOT_DIR/usr/lib" \
-        --argv0 "$1" "$CHROOT_DIR/bin/kmod" "${@:2}"
-}
-
-echo "$TAG stripping kernel modules to virtio+fw_cfg+${MICROVM_EXTRA_MODULES} only..."
-for kver_dir in "$CHROOT_DIR/lib/modules"/*/; do
-    kver="$(basename "$kver_dir")"
-    kernel_dir="$kver_dir/kernel"
-    [ -d "$kernel_dir" ] || continue
-
-    if [ ! -s "$kver_dir/modules.dep" ]; then
-        echo "ERROR: stock ${kver_dir}modules.dep missing — cannot resolve module dependencies" >&2
-        exit 1
-    fi
-
-    # Extra modules start the closure; the glob families are kept as found.
-    keep_list=$(mktemp)
-    : > "$keep_list"
-    for name in $MICROVM_EXTRA_MODULES; do
-        if ! path=$(module_path "$kver_dir" "$name"); then
-            if module_builtin "$kver_dir" "$name"; then continue; fi
-            echo "ERROR: module $name is neither loadable nor built into $kver" >&2
-            exit 1
-        fi
-        echo "$path" >> "$keep_list"
-    done
-
-    # Close over modules.dep hard dependencies and modules.softdep pre:/post:
-    # names until the set stops growing. A softdep must be loadable or built in.
-    while :; do
-        before=$(sort -u "$keep_list" | wc -l)
-        for path in $(sort -u "$keep_list"); do
-            awk -F': ' -v p="$path" '$1 == p { n = split($2, d, " "); for (i = 1; i <= n; i++) print d[i] }' \
-                "$kver_dir/modules.dep" >> "$keep_list"
-            mod=$(basename "$path"); mod="${mod%%.ko*}"; mod="${mod//-/_}"
-            [ -f "$kver_dir/modules.softdep" ] || continue
-            for soft in $(awk -v n="$mod" '{ m = $2; gsub(/-/, "_", m) }
-                    $1 == "softdep" && m == n { for (i = 3; i <= NF; i++) if ($i != "pre:" && $i != "post:") print $i }' \
-                    "$kver_dir/modules.softdep"); do
-                if soft_path=$(module_path "$kver_dir" "$soft"); then
-                    echo "$soft_path" >> "$keep_list"
-                elif ! module_builtin "$kver_dir" "$soft"; then
-                    echo "ERROR: softdep $soft of $mod is neither loadable nor built into $kver" >&2
-                    exit 1
-                fi
-            done
-        done
-        after=$(sort -u "$keep_list" | wc -l)
-        [ "$before" -eq "$after" ] && break
-    done
-
-    extra_names=$(sort -u "$keep_list" | while read -r p; do m=$(basename "$p"); m="${m%%.ko*}"; echo "${m//-/_}"; done)
-    (cd "$kver_dir" && find kernel -name "virtio*.ko*" -o -name "*fw_cfg*.ko*" \
-        -o -name "*failover*.ko*") >> "$keep_list"
-
-    # Copy the keeper set out, preserving each kernel/ relative path.
-    tmp_keep=$(mktemp -d)
-    while read -r rel; do
-        rel="${rel#kernel/}"
-        mkdir -p "$tmp_keep/$(dirname "$rel")"
-        cp "$kernel_dir/$rel" "$tmp_keep/$rel"
-    done < <(sort -u "$keep_list")
-    rm -f "$keep_list"
-
-    # Replace kernel module tree with the keeper set.
-    rm -rf "$kernel_dir"
-    mkdir -p "$kernel_dir"
-    # Restore each kept file into the same relative path.
-    (cd "$tmp_keep" && find . -name "*.ko*" | while read -r f; do
-        dest_dir="$kernel_dir/$(dirname "${f#./}")"
-        mkdir -p "$dest_dir"
-        cp "$f" "$dest_dir/"
-    done)
-    rm -rf "$tmp_keep"
-
-    # Decompress kept modules to plain .ko. Alpine ships .ko.gz, but a host
-    # kmod that lacks gzip support (e.g. Debian trixie's kmod 34) silently
-    # produces an EMPTY modules.dep from compressed modules — the guest's
-    # modprobe then no-ops and qemu_fw_cfg never loads. Plain .ko sidesteps the
-    # host-kmod dependency entirely and the guest needs no decompressor either.
-    find "$kernel_dir" -name "*.ko.gz" -exec gunzip -f {} +
-
-    # Independent closure gate: each extra-closure module's own .modinfo depends=
-    # must name a kept module or a built-in, so a resolver bug fails here.
-    kept_names=$(find "$kernel_dir" -name "*.ko" -exec basename {} .ko \; | tr '-' '_' | sort -u)
-    while read -r ko; do
-        ko_name=$(basename "$ko" .ko); ko_name="${ko_name//-/_}"
-        grep -qx "$ko_name" <<< "$extra_names" || continue
-        deps=$(tr '\0' '\n' < "$ko" | grep -a -m1 '^depends=' | cut -d= -f2 | tr ',' ' ' || true)
-        for dep in $deps; do
-            dep="${dep//-/_}"
-            if ! grep -qx "$dep" <<< "$kept_names" && ! module_builtin "$kver_dir" "$dep"; then
-                echo "ERROR: $(basename "$ko") depends on $dep, which is neither kept nor built in" >&2
-                exit 1
-            fi
-        done
-    done < <(find "$kernel_dir" -name "*.ko")
-    for name in $MICROVM_EXTRA_MODULES; do
-        if ! grep -qx "${name//-/_}" <<< "$kept_names" && ! module_builtin "$kver_dir" "$name"; then
-            echo "ERROR: required module $name missing from the stripped module tree" >&2
-            exit 1
-        fi
-    done
-    echo "$TAG extra module closure: $(echo "$extra_names" | tr '\n' ' ')"
-    echo "$TAG modules kept: $(echo "$kept_names" | tr '\n' ' ')"
-
-    # Rebuild module metadata with the image's own kmod, the binary the guest's
-    # modprobe runs, so no host depmod is needed. Stock indexes still name .ko.gz
-    # files, so they are removed first and a failed rebuild cannot ship them.
-    for f in dep dep.bin alias alias.bin softdep weakdep symbols symbols.bin \
-        builtin.bin builtin.alias.bin devname; do
-        rm -f "$kver_dir/modules.$f"
-    done
-    if ! image_kmod depmod -b "$CHROOT_DIR" "$kver"; then
-        echo "ERROR: image depmod failed for $kver" >&2
-        exit 1
-    fi
-    if [ ! -s "$kver_dir/modules.dep" ] || [ ! -s "$kver_dir/modules.dep.bin" ]; then
-        echo "ERROR: ${kver_dir}modules.dep or modules.dep.bin missing or empty after depmod" >&2
-        exit 1
-    fi
-    # Resolution gate: the image's modprobe must resolve every required module
-    # to plain .ko files present in the stripped tree.
-    for name in $MICROVM_EXTRA_MODULES qemu_fw_cfg virtio_net; do
-        module_builtin "$kver_dir" "$name" && continue
-        if ! plan=$(image_kmod modprobe -d "$CHROOT_DIR" -S "$kver" --show-depends "$name"); then
-            echo "ERROR: image modprobe cannot resolve $name for $kver" >&2
-            exit 1
-        fi
-        while read -r verb path _; do
-            [ "$verb" = "insmod" ] || continue
-            case "$path" in
-                "$CHROOT_DIR/lib/modules/$kver/kernel/"*.ko) [ -f "$path" ] && continue ;;
-            esac
-            echo "ERROR: modprobe plan for $name names $path, not a kept plain .ko" >&2
-            exit 1
-        done <<< "$plan"
-        echo "$TAG modprobe $name: $(awk '{ n = split($2, a, "/"); printf "%s ", a[n] }' <<< "$plan")"
-    done
-done
-
-# --- Strip package-manager artifacts (not needed at runtime) ---
-echo "$TAG stripping package manager artifacts..."
-rm -rf \
-    "$CHROOT_DIR/lib/apk" \
-    "$CHROOT_DIR/var/cache/apk" \
-    "$CHROOT_DIR/etc/apk" \
-    "$CHROOT_DIR/usr/share/man" \
-    "$CHROOT_DIR/usr/share/doc" \
-    "$CHROOT_DIR/usr/share/apk" \
-    "$CHROOT_DIR/usr/include" \
-    "$CHROOT_DIR/usr/lib/pkgconfig"
-
-# --- Strip debug symbols from binaries ---
-# nats-server in /usr/local/bin is left as built (-s -w), so the shipped bytes
-# are the ones whose version was checked above.
-echo "$TAG stripping debug symbols from binaries..."
-find "$CHROOT_DIR/usr/sbin" "$CHROOT_DIR/usr/bin" \
-     "$CHROOT_DIR/sbin" "$CHROOT_DIR/bin" \
-    -type f 2>/dev/null | while read -r bin; do
-    strip --strip-all "$bin" 2>/dev/null || true
-done
-
-# --- Copy vmlinuz out before stripping /boot from chroot ---
-VMLINUZ_OUT="$STAGE_DIR/vmlinuz"
-echo "$TAG copying kernel: $VMLINUZ_OUT"
-cp "$KERNEL_IMG" "$VMLINUZ_OUT"
-
-# Drop /boot from the chroot — vmlinuz, Alpine's stock initramfs-virt, System.map
-# and kernel config all live here (~25 MiB). None are needed at runtime: the
-# guest already has the kernel loaded by QEMU's -kernel, and Alpine's initramfs
-# is unused because we replace it with our own /init. Leaving /boot in the cpio
-# nearly doubles uncompressed initramfs size and forces a 512 MiB guest just to
-# survive decompression.
-rm -rf "${CHROOT_DIR:?}/boot"
-
-# --- Build initramfs ---
-INITRAMFS_OUT="$STAGE_DIR/initramfs.cpio.gz"
-echo "$TAG building initramfs: $INITRAMFS_OUT"
-# Create the static device nodes and pack the archive inside a single fakeroot
-# session: mknod/cpio see faked device entries and write real char-device
-# records into the cpio, while the host filesystem never gets an actual node.
-fakeroot sh -c '
-    cd "$1" || exit 1
-    mknod -m 600 dev/console c 5 1
-    mknod -m 666 dev/null    c 1 3
-    find . | cpio --quiet -o -H newc | gzip -9 > "$2"
-' _ "$CHROOT_DIR" "$INITRAMFS_OUT"
+microvm_strip_modules
+microvm_strip_rootfs
+microvm_stage_kernel
+microvm_pack_initramfs

 # --- Gate the staged pair, then publish it ---
 # Nothing reaches $BUILD_DIR until every gate passes, so a failed build cannot
@@ -11,12 +11,14 @@ set -eu
 # cron can branch without re-parsing, then disables itself so it runs exactly
 # once.
 #
-#   server      → eks-token-webhook, k3s (server), k3s-first-boot (bootstrap
-#                 publish), mulga-eks-state-report
+#   server      → eks-token-webhook, k3s (server), eks-pod-identity-webhook,
+#                 k3s-first-boot (bootstrap publish), mulga-eks-state-report,
+#                 mulga-eks-node-gc
 #   server-join → eks-token-webhook, k3s (server, joins the first server's etcd
-#                 quorum), mulga-eks-state-report. NO k3s-first-boot: the first
-#                 server already publishes the cluster-identical bootstrap
-#                 artifacts; a join re-publish only races the bootstrap bus.
+#                 quorum), eks-pod-identity-webhook, mulga-eks-state-report,
+#                 mulga-eks-node-gc. NO k3s-first-boot: the first server
+#                 already publishes the cluster-identical bootstrap artifacts;
+#                 a join re-publish only races the bootstrap bus.
 #   agent       → k3s-agent
 #
 # Paths are overridable via env so the selection logic is unit-testable (bats)
@@ -80,6 +82,14 @@ svc_start() {
    esac
 }

+# IRSA injector: every server runs one, called by its own apiserver. It is
+# optional (failurePolicy Ignore), so it goes last and a fault only costs
+# injection, never the control plane.
+svc_optional() {
+    svc_enable "$1" && svc_start "$1" ||
+        log "WARN: optional service $1 failed to enable or start; continuing without it"
+}
+
 svc_disable() {
    case "${INIT_SYSTEM}" in
        openrc) rc-update del "$1" default 2>/dev/null || true ;;
@@ -140,6 +150,7 @@ case "${ROLE}" in
        # writer owns. HA multi-server addon delivery is tracked separately
        # Server-join nodes do not run this setup.
        svc_enable mulga-eks-addon-sync
+        svc_enable mulga-eks-node-gc
        svc_start eks-token-webhook
        svc_start mulga-eks-k3s-recovery
        svc_start k3s
@@ -147,6 +158,8 @@ case "${ROLE}" in
        svc_start k3s-first-boot
        svc_start mulga-eks-state-report
        svc_start mulga-eks-addon-sync
+        svc_start mulga-eks-node-gc
+        svc_optional eks-pod-identity-webhook
        ;;
    server-join)
        log "configuring server-join role"
@@ -158,11 +171,14 @@ case "${ROLE}" in
        svc_enable k3s
        svc_enable konnectivity-server
        svc_enable mulga-eks-state-report
+        svc_enable mulga-eks-node-gc
        svc_start eks-token-webhook
        svc_start mulga-eks-k3s-recovery
        svc_start k3s
        svc_start konnectivity-server
        svc_start mulga-eks-state-report
+        svc_start mulga-eks-node-gc
+        svc_optional eks-pod-identity-webhook
        ;;
    agent)
        log "configuring agent role"
@@ -22,6 +22,7 @@ for tool in rc-update rc-service; do
    cat > "${STUB_OPENRC}/${tool}" <<EOF
 #!/bin/sh
 echo "${tool} \$*" >> "\${CALLS}"
+case "\$*" in *"\${STUB_FAIL:-@none@}"*) exit 1 ;; esac
 exit 0
 EOF
    chmod +x "${STUB_OPENRC}/${tool}"
@@ -32,6 +33,7 @@ mkdir -p "${STUB_SYSTEMD}"
 cat > "${STUB_SYSTEMD}/systemctl" <<'EOF'
 #!/bin/sh
 echo "systemctl $*" >> "${CALLS}"
+case "$*" in *"${STUB_FAIL:-@none@}"*) exit 1 ;; esac
 exit 0
 EOF
 chmod +x "${STUB_SYSTEMD}/systemctl"
@@ -129,6 +131,7 @@ for INIT in openrc systemd; do
    check "server(${INIT}): enables k3s-first-boot" "yes" "$(present "${CALLS}" "$(enable_cmd "${INIT}" k3s-first-boot)")"
    check "server(${INIT}): starts k3s" "yes" "$(present "${CALLS}" "$(start_cmd "${INIT}" k3s)")"
    check "server(${INIT}): enables konnectivity-server" "yes" "$(present "${CALLS}" "$(enable_cmd "${INIT}" konnectivity-server)")"
+    check "server(${INIT}): enables pod-identity-webhook" "yes" "$(present "${CALLS}" "$(enable_cmd "${INIT}" eks-pod-identity-webhook)")"
    check "server(${INIT}): enables k3s-recovery" "yes" "$(present "${CALLS}" "$(enable_cmd "${INIT}" mulga-eks-k3s-recovery)")"
    # Recovery is enabled (so it also runs pre-k3s on later boots) AND started
    # inline now — a restore-snapshot DR seed's directive must apply on this
@@ -141,8 +144,38 @@ for INIT in openrc systemd; do
            END { print (rn > 0 && kn > 0 && rn < kn) ? "yes" : "no" }
        ' "${CALLS}"
    )"
+    check "server(${INIT}): enables node-gc" "yes" "$(present "${CALLS}" "$(enable_cmd "${INIT}" mulga-eks-node-gc)")"
+    check "server(${INIT}): starts node-gc" "yes" "$(present "${CALLS}" "$(start_cmd "${INIT}" mulga-eks-node-gc)")"
    check "server(${INIT}): no k3s-agent" "no" "$(present "${CALLS}" "$(enable_cmd "${INIT}" k3s-agent)")"
    check "server(${INIT}): self-disables" "yes" "$(present "${CALLS}" "$(disable_cmd "${INIT}" eks-node-role)")"
+    check "server(${INIT}): pod-identity-webhook is the last call" "$(start_cmd "${INIT}" eks-pod-identity-webhook)" "$(grep -v "$(disable_cmd "${INIT}" eks-node-role)" "${CALLS}" | tail -n 1)"
+    check "server(${INIT}): pod-identity-webhook enabled after addon-sync start" "yes" "$(
+        awk -v a="$(start_cmd "${INIT}" mulga-eks-addon-sync)" -v w="$(enable_cmd "${INIT}" eks-pod-identity-webhook)" '
+            index($0, a) == 1 { an = NR }
+            index($0, w) == 1 { wn = NR }
+            END { print (an > 0 && wn > an) ? "yes" : "no" }
+        ' "${CALLS}"
+    )"
+
+    # A broken optional webhook must leave the control plane booting.
+    for FAILING in enable start; do
+        run_case "${INIT}" server_${FAILING}_fail
+        printf 'SPINIFEX_K3S_ROLE=server\nEKS_CLUSTER_NAME=alpha\n' > "${EKS_NODE_ENVFILE}"
+        if [ "${FAILING}" = enable ]; then
+            STUB_FAIL="$(enable_cmd "${INIT}" eks-pod-identity-webhook)"
+            export STUB_FAIL
+            STUB_FAIL="${STUB_FAIL#rc-update add }"; STUB_FAIL="${STUB_FAIL#systemctl enable }"
+        else
+            STUB_FAIL="$(start_cmd "${INIT}" eks-pod-identity-webhook)"
+            export STUB_FAIL
+            STUB_FAIL="${STUB_FAIL#rc-service }"; STUB_FAIL="${STUB_FAIL#systemctl start --no-block }"
+        fi
+        invoke
+        unset STUB_FAIL
+        check "server(${INIT}): webhook ${FAILING} failure still exits 0" 0 "${rc}"
+        check "server(${INIT}): webhook ${FAILING} failure still starts mulga-eks-addon-sync" "yes" "$(present "${CALLS}" "$(start_cmd "${INIT}" mulga-eks-addon-sync)")"
+        check "server(${INIT}): webhook ${FAILING} failure still self-disables" "yes" "$(present "${CALLS}" "$(disable_cmd "${INIT}" eks-node-role)")"
+    done

    # --- server-join ---
    run_case "${INIT}" server_join
@@ -153,12 +186,16 @@ for INIT in openrc systemd; do
    check "server-join(${INIT}): enables webhook" "yes" "$(present "${CALLS}" "$(enable_cmd "${INIT}" eks-token-webhook)")"
    check "server-join(${INIT}): enables k3s" "yes" "$(present "${CALLS}" "$(enable_cmd "${INIT}" k3s)")"
    check "server-join(${INIT}): enables state-report" "yes" "$(present "${CALLS}" "$(enable_cmd "${INIT}" mulga-eks-state-report)")"
+    check "server-join(${INIT}): enables node-gc" "yes" "$(present "${CALLS}" "$(enable_cmd "${INIT}" mulga-eks-node-gc)")"
+    check "server-join(${INIT}): starts node-gc" "yes" "$(present "${CALLS}" "$(start_cmd "${INIT}" mulga-eks-node-gc)")"
    check "server-join(${INIT}): enables konnectivity-server" "yes" "$(present "${CALLS}" "$(enable_cmd "${INIT}" konnectivity-server)")"
+    check "server-join(${INIT}): enables pod-identity-webhook" "yes" "$(present "${CALLS}" "$(enable_cmd "${INIT}" eks-pod-identity-webhook)")"
    check "server-join(${INIT}): enables k3s-recovery" "yes" "$(present "${CALLS}" "$(enable_cmd "${INIT}" mulga-eks-k3s-recovery)")"
    check "server-join(${INIT}): starts k3s-recovery" "yes" "$(present "${CALLS}" "$(start_cmd "${INIT}" mulga-eks-k3s-recovery)")"
    # Join servers must NOT re-publish bootstrap: the first server already
    # published the cluster-identical artifacts, and a join re-publish would
    # only race the bootstrap bus. This is deliberate, not an oversight.
+    check "server-join(${INIT}): pod-identity-webhook is the last call" "$(start_cmd "${INIT}" eks-pod-identity-webhook)" "$(grep -v "$(disable_cmd "${INIT}" eks-node-role)" "${CALLS}" | tail -n 1)"
    check "server-join(${INIT}): no k3s-first-boot" "yes" "$(absent_all "${CALLS}" "k3s-first-boot")"
    check "server-join(${INIT}): no k3s-agent" "yes" "$(absent_all "${CALLS}" "$(enable_cmd "${INIT}" k3s-agent)")"
    check "server-join(${INIT}): self-disables" "yes" "$(present "${CALLS}" "$(disable_cmd "${INIT}" eks-node-role)")"
@@ -170,7 +207,7 @@ for INIT in openrc systemd; do
    check "agent(${INIT}): exit 0" 0 "${rc}"
    check "agent(${INIT}): role file" "agent" "$(cat "${ROLE_FILE}")"
    check "agent(${INIT}): enables k3s-agent" "yes" "$(present "${CALLS}" "$(enable_cmd "${INIT}" k3s-agent)")"
-    check "agent(${INIT}): no server services" "yes" "$(absent_all "${CALLS}" "eks-token-webhook" "k3s-first-boot" "$(enable_cmd "${INIT}" k3s)")"
+    check "agent(${INIT}): no server services" "yes" "$(absent_all "${CALLS}" "eks-token-webhook" "eks-pod-identity-webhook" "k3s-first-boot" "mulga-eks-node-gc" "$(enable_cmd "${INIT}" k3s)")"

    # --- agent (inferred from agent.env, no explicit role) ---
    run_case "${INIT}" agent_inferred
@@ -0,0 +1,26 @@
+#!/sbin/openrc-run
+# eks-pod-identity-webhook — IRSA mutating admission webhook. Binds 127.0.0.1:8445
+# only; the local kube-apiserver is the sole client. Supervised and respawned so
+# it can wait out k3s writing its CAs without ever delaying k3s or the token
+# webhook: it exits until the CAs exist and supervise-daemon retries.
+
+description="EKS pod identity webhook (Mulga)"
+command="/usr/local/bin/eks-pod-identity-webhook"
+command_args="--addr=127.0.0.1:8445"
+pidfile="/run/eks-pod-identity-webhook.pid"
+supervisor="supervise-daemon"
+respawn_delay=5
+respawn_max=0
+output_log="/dev/console"
+error_log="/dev/console"
+
+if [ -f /etc/spinifex-eks/first-boot.env ]; then
+    # shellcheck disable=SC1091
+    . /etc/spinifex-eks/first-boot.env
+    export EKS_REGION EKS_ADDON_GATEWAY_URL EKS_GATEWAY_CA
+fi
+
+depend() {
+    need net
+    after k3s
+}
@@ -22,9 +22,9 @@ APK_PACKAGES="ca-certificates curl jq iptables ip6tables conntrack-tools cgroup-
 # at a pinned tag (its GitHub releases ship only container images, and its go.mod
 # carries replace directives that block `go install pkg@version`). The agent half
 # is the matching proxy-agent container image, pinned in k3s.initd.
-BUILD_COMMANDS="CGO_ENABLED=0 GOFIPS140=v1.0.0 go build -o bin/eks-token-webhook ./cmd/eks-token-webhook && CGO_ENABLED=0 GOFIPS140=v1.0.0 go build -o bin/eks-gateway-publish ./cmd/eks-gateway-publish && CGO_ENABLED=0 GOFIPS140=v1.0.0 go build -o bin/eks-gateway-fetch ./cmd/eks-gateway-fetch && CGO_ENABLED=0 GOFIPS140=v1.0.0 go build -o bin/eks-webhook-cert ./cmd/eks-webhook-cert && CGO_ENABLED=0 GOFIPS140=v1.0.0 go build -o bin/eks-konnectivity-cert ./cmd/eks-konnectivity-cert && CGO_ENABLED=0 GOFIPS140=v1.0.0 go build -o bin/eks-etcd-member ./cmd/eks-etcd-member && CGO_ENABLED=0 GOFIPS140=v1.0.0 go build -o bin/ecr-credential-provider ./cmd/ecr-credential-provider && rm -rf /tmp/anp-build && git clone --depth 1 --branch v0.30.3 https://github.com/kubernetes-sigs/apiserver-network-proxy /tmp/anp-build && ( cd /tmp/anp-build && CGO_ENABLED=0 GOFIPS140=v1.0.0 go build -o /tmp/anp-build/konnectivity-server ./cmd/server ) && cp /tmp/anp-build/konnectivity-server bin/konnectivity-server && rm -rf /tmp/anp-build"
+BUILD_COMMANDS="CGO_ENABLED=0 GOFIPS140=v1.0.0 go build -o bin/eks-token-webhook ./cmd/eks-token-webhook && CGO_ENABLED=0 GOFIPS140=v1.0.0 go build -o bin/eks-gateway-publish ./cmd/eks-gateway-publish && CGO_ENABLED=0 GOFIPS140=v1.0.0 go build -o bin/eks-gateway-fetch ./cmd/eks-gateway-fetch && CGO_ENABLED=0 GOFIPS140=v1.0.0 go build -o bin/eks-webhook-cert ./cmd/eks-webhook-cert && CGO_ENABLED=0 GOFIPS140=v1.0.0 go build -o bin/eks-konnectivity-cert ./cmd/eks-konnectivity-cert && CGO_ENABLED=0 GOFIPS140=v1.0.0 go build -o bin/eks-etcd-member ./cmd/eks-etcd-member && CGO_ENABLED=0 GOFIPS140=v1.0.0 go build -o bin/eks-pod-identity-webhook ./cmd/eks-pod-identity-webhook && CGO_ENABLED=0 GOFIPS140=v1.0.0 go build -o bin/ecr-credential-provider ./cmd/ecr-credential-provider && rm -rf /tmp/anp-build && git clone --depth 1 --branch v0.30.3 https://github.com/kubernetes-sigs/apiserver-network-proxy /tmp/anp-build && ( cd /tmp/anp-build && CGO_ENABLED=0 GOFIPS140=v1.0.0 go build -o /tmp/anp-build/konnectivity-server ./cmd/server ) && cp /tmp/anp-build/konnectivity-server bin/konnectivity-server && rm -rf /tmp/anp-build"

-INSTALL_BINARIES="bin/eks-token-webhook:/usr/local/bin/eks-token-webhook bin/eks-gateway-publish:/usr/local/bin/eks-gateway-publish bin/eks-gateway-fetch:/usr/local/bin/eks-gateway-fetch bin/eks-webhook-cert:/usr/local/bin/eks-webhook-cert bin/eks-konnectivity-cert:/usr/local/bin/eks-konnectivity-cert bin/eks-etcd-member:/usr/local/bin/eks-etcd-member bin/konnectivity-server:/usr/local/bin/konnectivity-server bin/ecr-credential-provider:/usr/local/bin/ecr-credential-provider"
+INSTALL_BINARIES="bin/eks-token-webhook:/usr/local/bin/eks-token-webhook bin/eks-gateway-publish:/usr/local/bin/eks-gateway-publish bin/eks-gateway-fetch:/usr/local/bin/eks-gateway-fetch bin/eks-webhook-cert:/usr/local/bin/eks-webhook-cert bin/eks-konnectivity-cert:/usr/local/bin/eks-konnectivity-cert bin/eks-etcd-member:/usr/local/bin/eks-etcd-member bin/eks-pod-identity-webhook:/usr/local/bin/eks-pod-identity-webhook bin/konnectivity-server:/usr/local/bin/konnectivity-server bin/ecr-credential-provider:/usr/local/bin/ecr-credential-provider"

 # All role services ship; none but the selector are enabled at bake time.
 # eks-node-role picks server vs agent at first boot and enables the rest.
@@ -36,12 +36,15 @@ scripts/images/eks-node/k3s-prestart.sh:/usr/local/sbin/k3s-prestart \
 scripts/images/eks-node/k3s-agent.initd:/etc/init.d/k3s-agent \
 scripts/images/eks-node/k3s-agent-prestart.sh:/usr/local/sbin/k3s-agent-prestart \
 scripts/images/eks-node/eks-token-webhook.initd:/etc/init.d/eks-token-webhook \
+scripts/images/eks-node/eks-pod-identity-webhook.initd:/etc/init.d/eks-pod-identity-webhook \
 scripts/images/eks-node/k3s-first-boot.initd:/etc/init.d/k3s-first-boot \
 scripts/images/eks-node/k3s-first-boot.sh:/usr/local/sbin/k3s-first-boot \
 scripts/images/eks-node/mulga-eks-state-report.initd:/etc/init.d/mulga-eks-state-report \
 scripts/images/eks-node/mulga-eks-state-report.sh:/usr/local/sbin/mulga-eks-state-report \
 scripts/images/eks-node/mulga-eks-addon-sync.initd:/etc/init.d/mulga-eks-addon-sync \
 scripts/images/eks-node/mulga-eks-addon-sync.sh:/usr/local/sbin/mulga-eks-addon-sync \
+scripts/images/eks-node/mulga-eks-node-gc.initd:/etc/init.d/mulga-eks-node-gc \
+scripts/images/eks-node/mulga-eks-node-gc.sh:/usr/local/sbin/mulga-eks-node-gc \
 scripts/images/eks-node/konnectivity-server.initd:/etc/init.d/konnectivity-server \
 scripts/images/eks-node/konnectivity-server-prestart.sh:/usr/local/sbin/konnectivity-server-prestart \
 scripts/images/eks-node/konnectivity/konnectivity-agent.yaml:/usr/share/spinifex-eks/konnectivity/konnectivity-agent.yaml \
@@ -0,0 +1,27 @@
+#!/sbin/openrc-run
+# mulga-eks-node-gc — server-role service that deletes the Node of every
+# nodegroup worker the nodegroup has terminated, every NODE_GC_INTERVAL
+# seconds. Enabled on every server role by eks-node-role.sh.
+
+description="Mulga EKS terminated-worker Node cleanup"
+command="/usr/local/sbin/mulga-eks-node-gc"
+command_background="yes"
+pidfile="/run/mulga-eks-node-gc.pid"
+output_log="/var/log/mulga-eks-node-gc.log"
+error_log="/var/log/mulga-eks-node-gc.log"
+
+# A Node goes NotReady about 40s after its worker stops, so 30s deletes it
+# within a couple of minutes of the terminate.
+export NODE_GC_INTERVAL="${NODE_GC_INTERVAL:-30}"
+
+depend() {
+    need net
+    after k3s
+}
+
+start_pre() {
+    if [ ! -f /etc/spinifex-eks/first-boot.env ]; then
+        eerror "/etc/spinifex-eks/first-boot.env not present — cloud-init did not seed gateway creds"
+        return 1
+    fi
+}
@@ -0,0 +1,89 @@
+#!/bin/sh
+set -eu
+
+# mulga-eks-node-gc — deletes the Node of a nodegroup worker whose instance the
+# nodegroup has terminated (DeleteNodegroup, scale-down, reclaim), as the AWS
+# cloud controller deletes a Node whose instance no longer exists. K3s runs no
+# AWS cloud provider: its own reports every instance as existing, so without
+# this a terminated worker stays a NotReady Node forever, its pods stay bound to
+# it, and once most Nodes are NotReady the node lifecycle controller stops
+# evicting at all.
+#
+# Pull (host→VM): GET /clusters/{cluster}/internal-workers/{accountId}, the
+# instance ID of every worker the cluster's nodegroups still own.
+#
+# A Node is deleted only when all of these hold, so a live worker is never
+# touched: it carries the eks.amazonaws.com/nodegroup label, its providerID is
+# aws:///<az>/<instance-id>, it is not Ready, and its instance is not in the
+# fetched set. A failed fetch deletes nothing.
+#
+# Runs on every server; deletes are idempotent. Loops every NODE_GC_INTERVAL
+# seconds in service mode, one-shot otherwise.
+
+ENVFILE=${ENVFILE:-/etc/spinifex-eks/first-boot.env}
+[ -f "${ENVFILE}" ] || { logger -t mulga-eks-node-gc "${ENVFILE} missing — exiting"; exit 0; }
+# set -a so the sourced creds are exported to the eks-gateway-fetch child.
+set -a
+# shellcheck disable=SC1090
+. "${ENVFILE}"
+set +a
+
+: "${EKS_GATEWAY_URL:?}"
+: "${EKS_ACCOUNT_ID:?}"
+: "${EKS_CLUSTER_NAME:?}"
+
+KUBECONFIG=${KUBECONFIG:-/etc/rancher/k3s/k3s.yaml}
+export KUBECONFIG
+
+# One "name|providerID|nodegroup|Ready" row per Node.
+NODE_ROWS='jsonpath={range .items[*]}{.metadata.name}{"|"}{.spec.providerID}{"|"}{.metadata.labels.eks\.amazonaws\.com/nodegroup}{"|"}{.status.conditions[?(@.type=="Ready")].status}{"\n"}{end}'
+
+log() {
+    logger -t mulga-eks-node-gc "$*"
+    echo "[mulga-eks-node-gc] $*" > "${CONSOLE:-/dev/console}" 2>/dev/null || true
+}
+
+# stale_nodes LIVE_FILE — reads Node rows on stdin and prints the name of each
+# Node to delete: a nodegroup worker, not Ready, whose instance is not listed
+# in LIVE_FILE.
+stale_nodes() {
+    awk -F'|' -v live="$1" '
+        BEGIN { while ((getline id < live) > 0) keep[id] = 1 }
+        $3 == "" || $4 == "True" { next }
+        $2 !~ /^aws:\/\/\/[^\/]+\/i-[0-9a-f]+$/ { next }
+        { n = split($2, p, "/"); if (!(p[n] in keep)) print $1 }
+    '
+}
+
+sync_once() {
+    # Nodes are listed before the workers are fetched. A worker is recorded on
+    # its nodegroup before it boots, so every Node listed here is already in a
+    # set fetched after.
+    if ! rows=$(kubectl get nodes -o "${NODE_ROWS}" 2>/dev/null); then
+        log "kubectl get nodes failed; skipping this pass"
+        return 0
+    fi
+    live=$(mktemp)
+    if ! eks-gateway-fetch -resource workers > "${live}" 2>/dev/null; then
+        log "fetch of the cluster's nodegroup workers failed; deleting nothing"
+        rm -f "${live}"
+        return 0
+    fi
+    printf '%s\n' "${rows}" | stale_nodes "${live}" | while read -r node; do
+        if kubectl delete node "${node}" --ignore-not-found --wait=false >/dev/null 2>&1; then
+            log "deleted Node ${node}: its instance is no longer a nodegroup worker"
+        else
+            log "delete of Node ${node} failed; retrying next pass"
+        fi
+    done
+    rm -f "${live}"
+}
+
+if [ -n "${NODE_GC_INTERVAL:-}" ] && [ "${NODE_GC_INTERVAL}" -gt 0 ] 2>/dev/null; then
+    while true; do
+        sync_once
+        sleep "${NODE_GC_INTERVAL}"
+    done
+else
+    sync_once
+fi
@@ -0,0 +1,107 @@
+#!/bin/sh
+# Self-contained POSIX test for mulga-eks-node-gc.sh. kubectl, eks-gateway-fetch
+# and logger are stubbed on PATH: kubectl serves canned Node rows and records
+# every delete, and eks-gateway-fetch serves a canned worker set or fails.
+#
+# Run: sh scripts/images/eks-node/mulga-eks-node-gc_test.sh
+set -eu
+
+SCRIPT_DIR=$(cd "$(dirname "$0")" && pwd)
+AGENT="${SCRIPT_DIR}/mulga-eks-node-gc.sh"
+WORK=$(mktemp -d)
+trap 'rm -rf "${WORK}"' EXIT
+
+STUBBIN="${WORK}/bin"
+mkdir -p "${STUBBIN}"
+
+cat > "${STUBBIN}/kubectl" <<'EOF'
+#!/bin/sh
+case "$1 $2" in
+    "get nodes")
+        [ "${NODES_RC:-0}" -eq 0 ] || exit "${NODES_RC}"
+        printf '%s\n' "${NODES_BODY:-}" ;;
+    "delete node") echo "$3" >> "${DELETED}" ;;
+esac
+EOF
+
+cat > "${STUBBIN}/eks-gateway-fetch" <<'EOF'
+#!/bin/sh
+echo "$*" >> "${FETCHED}"
+[ "${FETCH_RC:-0}" -eq 0 ] || exit "${FETCH_RC}"
+printf '%s' "${LIVE_BODY:-}"
+EOF
+
+cat > "${STUBBIN}/logger" <<'EOF'
+#!/bin/sh
+:
+EOF
+
+chmod +x "${STUBBIN}"/*
+PATH="${STUBBIN}:${PATH}"
+export PATH
+
+ENVFILE="${WORK}/first-boot.env"
+cat > "${ENVFILE}" <<'EOF'
+EKS_GATEWAY_URL=https://gw.invalid:9999
+EKS_ACCOUNT_ID=000000000001
+EKS_CLUSTER_NAME=demo3
+EOF
+CONSOLE="${WORK}/console"
+DELETED="${WORK}/deleted"
+FETCHED="${WORK}/fetched"
+export ENVFILE CONSOLE DELETED FETCHED
+
+PASS=0
+FAIL=0
+check() {
+    if [ "$2" = "$3" ]; then
+        PASS=$((PASS + 1))
+    else
+        FAIL=$((FAIL + 1))
+        printf 'FAIL: %s\n  want: %s\n  got:  %s\n' "$1" "$2" "$3"
+    fi
+}
+
+run_agent() {
+    : > "${DELETED}"
+    : > "${FETCHED}"
+    sh "${AGENT}"
+}
+
+deleted() { tr '\n' ' ' < "${DELETED}" | sed 's/ $//'; }
+
+# Every way a Node can look: only a NotReady nodegroup worker whose instance the
+# cluster's nodegroups no longer own is deleted.
+NODES_BODY='live|aws:///z1/i-0aaa|ng1|True
+gone|aws:///z1/i-0bbb|ng1|Unknown
+gone-never-reported|aws:///z1/i-0ddd|ng2|
+owned-notready|aws:///z1/i-0ccc|ng1|False
+gone-but-ready|aws:///z1/i-0eee|ng1|True
+self-managed|aws:///z1/i-0fff||Unknown
+k3s-provider|k3s://k3s-node|ng1|Unknown'
+LIVE_BODY='i-0aaa
+i-0ccc
+'
+export NODES_BODY LIVE_BODY
+
+run_agent
+check "deletes only NotReady nodegroup Nodes whose instance is gone" "gone gone-never-reported" "$(deleted)"
+check "fetches the workers resource" "-resource workers" "$(cat "${FETCHED}")"
+
+# A cluster with no nodegroups left: every NotReady nodegroup Node goes.
+LIVE_BODY='' run_agent
+check "empty worker set deletes every NotReady nodegroup Node" "gone gone-never-reported owned-notready" "$(deleted)"
+
+# A failed fetch is not an empty set: nothing is deleted.
+FETCH_RC=1 run_agent
+check "failed fetch deletes nothing" "" "$(deleted)"
+
+# Without a Node listing the workers are not fetched, so a set fetched before a
+# Node registered can never be compared against it.
+NODES_RC=1 run_agent
+check "failed node listing deletes nothing" "" "$(deleted)"
+check "failed node listing does not fetch" "" "$(cat "${FETCHED}")"
+
+echo "---"
+echo "PASS=${PASS} FAIL=${FAIL}"
+[ "${FAIL}" -eq 0 ]
@@ -44,9 +44,11 @@ ln -sf /usr/local/bin/k3s /usr/local/bin/ctr
 # role's services are baked; the selector enables the right ones at first boot.
 chmod 0755 /etc/init.d/eks-node-role /etc/init.d/k3s /etc/init.d/k3s-agent \
    /etc/init.d/eks-token-webhook /etc/init.d/k3s-first-boot /etc/init.d/mulga-eks-state-report \
-    /etc/init.d/mulga-eks-addon-sync /etc/init.d/konnectivity-server
+    /etc/init.d/mulga-eks-addon-sync /etc/init.d/mulga-eks-node-gc /etc/init.d/konnectivity-server \
+    /etc/init.d/eks-pod-identity-webhook
 chmod 0755 /usr/local/sbin/eks-node-role /usr/local/sbin/k3s-first-boot \
-    /usr/local/sbin/mulga-eks-state-report /usr/local/sbin/mulga-eks-addon-sync
+    /usr/local/sbin/mulga-eks-state-report /usr/local/sbin/mulga-eks-addon-sync \
+    /usr/local/sbin/mulga-eks-node-gc
 # start_pre helpers shared with the systemd wrappers that will eventually
 # replace these runscripts.
 chmod 0755 /usr/local/sbin/k3s-prestart /usr/local/sbin/k3s-agent-prestart \
@@ -0,0 +1,25 @@
+#!/bin/sh
+# Every /etc/init.d script that INSTALL_FILES ships at 0644 must be chmod'd
+# 0755 by setup.sh, or OpenRC refuses to enable it ("is not executable") and
+# eks-node-role, running under set -eu, aborts before k3s starts.
+#
+# Run: sh scripts/images/eks-node/setup_initd_test.sh
+set -eu
+
+SCRIPT_DIR=$(cd "$(dirname "$0")" && pwd)
+MANIFEST="${SCRIPT_DIR}/manifest.conf"
+SETUP="${SETUP_SH:-${SCRIPT_DIR}/setup.sh}"
+
+fail=0
+for dst in $(grep -o ':/etc/init\.d/[A-Za-z0-9._-]*' "${MANIFEST}" | cut -d: -f2); do
+    if ! grep -E '^[[:space:]]*(chmod 0755 )?.*'"${dst}"'([[:space:]\\]|$)' "${SETUP}" \
+        | grep -q . ; then
+        echo "FAIL: ${dst} is installed 0644 and never chmod'd 0755 in setup.sh"
+        fail=1
+    fi
+done
+
+if [ "${fail}" -ne 0 ]; then
+    exit 1
+fi
+echo "PASS: every shipped init script is chmod'd 0755"
@@ -0,0 +1,391 @@
+#!/bin/bash
+# microvm-image.sh — shared steps for the microVM image builders, sourced by
+# build-sqs-store-microvm-image.sh and build-lambda-runtime-microvm-image.sh.
+#
+# The caller sets before use:
+#   TAG                      log prefix
+#   BUILD_DIR                output directory; STAGE_DIR is created beside it
+#   ALPINE_VERSION, ALPINE_MIRROR, ALPINE_PACKAGES
+#   MICROVM_EXTRA_MODULES    modules kept by name, with their dependency closure
+#   MICROVM_GATE_MODULES     modules the image modprobe must resolve, beyond the extras
+#   MICROVM_EXTRA_KEEP_GLOBS bash array of module file globs kept on top of virtio*/*fw_cfg*
+# Functions set CHROOT_DIR, STAGE_DIR, KERNEL_IMG and KERNEL_VER for the caller.
+
+
+microvm_check_tools() {
+    # --- Verify non-negotiable tools ---
+    # fakeroot is mandatory: device nodes are written into the cpio archive inside a
+    # faked-root environment so no real device node is ever created on the host
+    # filesystem (which previously required sudo mknod and risked clobbering host
+    # /dev/null on cleanup).
+    for tool in cpio gzip find fakeroot; do
+        if ! command -v "$tool" >/dev/null 2>&1; then
+            echo "ERROR: required tool not found: $tool" >&2
+            exit 1
+        fi
+    done
+}
+
+microvm_init_dirs() {
+    # --- Create chroot ---
+    CHROOT_DIR=$(mktemp -d)
+    # Guard: every destructive op below targets "$CHROOT_DIR/...". An empty or
+    # unexpected value would aim rm -rf at the host (e.g. /dev). Refuse anything that
+    # is not a freshly-created temp dir.
+    case "${CHROOT_DIR:?mktemp -d returned empty}" in
+        /tmp/*|/var/tmp/*|"${TMPDIR:-/nonexistent}"/*) ;;
+        *) echo "ERROR: refusing to use unexpected CHROOT_DIR: $CHROOT_DIR" >&2; exit 1 ;;
+    esac
+    # Outputs are written here first, beside $BUILD_DIR so publishing is a rename.
+    STAGE_DIR=$(mktemp -d "$BUILD_DIR/.stage.XXXXXX")
+    CONTAINER_TOOL=""
+    CONTAINER_CID=""
+    trap microvm_cleanup EXIT
+}
+
+microvm_cleanup() {
+        echo "$TAG cleaning up chroot: $CHROOT_DIR"
+        rm -rf "$CHROOT_DIR"
+        rm -rf "${STAGE_DIR:?}"
+        if [ -n "$CONTAINER_CID" ] && [ -n "$CONTAINER_TOOL" ]; then
+            echo "$TAG cleaning up container: $CONTAINER_CID"
+            $CONTAINER_TOOL rm -f "$CONTAINER_CID" >/dev/null 2>&1 || true
+        fi
+}
+
+# microvm_make_rootfs populates $CHROOT_DIR from ALPINE_PACKAGES.
+microvm_make_rootfs() {
+    # --- Populate rootfs ---
+    # Strategy 1: native apk (Alpine hosts, CI)
+    if command -v apk >/dev/null 2>&1; then
+        echo "$TAG rootfs: native apk"
+        mkdir -p \
+            "$CHROOT_DIR/etc/apk" \
+            "$CHROOT_DIR/lib/apk/db" \
+            "$CHROOT_DIR/var/cache/apk"
+        cat > "$CHROOT_DIR/etc/apk/repositories" <<EOF
+${ALPINE_MIRROR}/v${ALPINE_VERSION}/main
+${ALPINE_MIRROR}/v${ALPINE_VERSION}/community
+EOF
+        apk add \
+            --root "$CHROOT_DIR" \
+            --arch x86_64 \
+            --no-cache \
+            --allow-untrusted \
+            --initdb \
+            $ALPINE_PACKAGES
+
+    # Strategy 2: docker -> BuildKit local export.
+    # Do NOT use `docker export` here. On Docker's overlay2 driver it leaks the
+    # container's overlay mount (the mount survives `docker rm`), orphaning layers
+    # under /var/lib/docker that only a daemon restart reclaims; on a busy CI host
+    # this grows without bound. `buildx --output type=local` writes the image
+    # filesystem straight to a directory and releases every mount, so nothing leaks.
+    elif command -v docker >/dev/null 2>&1 && docker info >/dev/null 2>&1; then
+        echo "$TAG rootfs: docker buildx --output type=local (alpine:${ALPINE_VERSION})"
+        BUILD_CTX=$(mktemp -d)
+        cat > "$BUILD_CTX/Dockerfile" <<DOCKERFILE
+FROM alpine:${ALPINE_VERSION}
+RUN apk add --no-cache ${ALPINE_PACKAGES}
+DOCKERFILE
+        docker buildx build \
+            --platform linux/amd64 \
+            --output "type=local,dest=${CHROOT_DIR}" \
+            "$BUILD_CTX"
+        rm -rf "$BUILD_CTX"
+
+    # Strategy 3: podman -> run + export. Podman's store does not exhibit the docker
+    # overlay2 export-mount leak, and podman has no buildx.
+    elif command -v podman >/dev/null 2>&1; then
+        echo "$TAG rootfs: podman export (alpine:${ALPINE_VERSION})"
+        CONTAINER_TOOL=podman
+        CONTAINER_CID=$(podman run -d \
+            "alpine:${ALPINE_VERSION}" \
+            sh -c "apk add --no-cache ${ALPINE_PACKAGES}")
+
+        exit_code=$(podman wait "$CONTAINER_CID")
+        if [ "$exit_code" != "0" ]; then
+            echo "ERROR: package installation failed in container (exit $exit_code)" >&2
+            exit 1
+        fi
+
+        podman export "$CONTAINER_CID" | tar -x -C "$CHROOT_DIR"
+
+    else
+        echo "ERROR: no rootfs build tool available." >&2
+        echo "       Install one of: apk (Alpine), docker, podman" >&2
+        exit 1
+    fi
+}
+
+microvm_empty_dev() {
+    # --- /dev: empty mountpoint only ---
+    # tar extraction without root silently creates 0-byte regular files for device
+    # nodes. Drop them and leave an empty /dev mountpoint. The two static nodes the
+    # kernel needs (/dev/console to wire init's stdio, /dev/null) are written into
+    # the cpio archive under fakeroot at pack time — NOT created on the host. init
+    # mounts devtmpfs over /dev as its first action for everything else at runtime.
+    rm -rf "${CHROOT_DIR:?}/dev"
+    mkdir "$CHROOT_DIR/dev"
+}
+
+# microvm_locate_kernel finds the kernel and asserts fw_cfg is loadable or built in.
+microvm_locate_kernel() {
+    # --- Locate kernel ---
+    KERNEL_IMG=$(find "$CHROOT_DIR/boot" -name "vmlinuz*" | sort | tail -1)
+    if [ -z "$KERNEL_IMG" ]; then
+        echo "ERROR: no vmlinuz found in $CHROOT_DIR/boot" >&2
+        exit 1
+    fi
+    echo "$TAG kernel image: $KERNEL_IMG"
+
+    KERNEL_VER=$(basename "$KERNEL_IMG" | sed 's/vmlinuz-//')
+    KERNEL_CONFIG="$CHROOT_DIR/boot/config-${KERNEL_VER}"
+
+    # --- Assert fw_cfg module ---
+    if ! find "$CHROOT_DIR/lib/modules" \
+            -name "qemu_fw_cfg.ko*" \
+            -o -name "fw_cfg_sysfs.ko*" \
+        | grep -q .; then
+        if ! grep -q "CONFIG_FW_CFG_SYSFS=y" "$KERNEL_CONFIG" 2>/dev/null; then
+            echo "ERROR: qemu_fw_cfg module missing from initramfs — fw_cfg reads will fail at boot" >&2
+            exit 1
+        fi
+        echo "$TAG fw_cfg: built-in (CONFIG_FW_CFG_SYSFS=y)"
+    else
+        echo "$TAG fw_cfg: module found in lib/modules"
+    fi
+}
+
+# --- Strip kernel modules (Phase 2 keep-list approach) ---
+# Only modules needed by a virtio-mmio microvm guest are kept:
+#   virtio*.ko*    — virtio bus, net, blk, rng, console, and helpers
+#   *fw_cfg*.ko*   — QEMU fw_cfg sysfs driver (delivers boot blobs)
+#   $MICROVM_EXTRA_KEEP_GLOBS — caller-named module file globs
+#   $MICROVM_EXTRA_MODULES — caller-named modules, plus dependency closure
+# All other loadable modules are removed. Drivers compiled into the kernel
+# (CONFIG_VIRTIO_MMIO=y etc.) are unaffected — they live in vmlinuz, not here.
+
+# module_path prints a module's kernel/ path from the stock modules.dep, by
+# name or modules.alias; it fails when the module is not loadable.
+module_path() {
+    local kdir="$1" name="${2//-/_}" path alias_target
+    path=$(awk -F: -v n="$name" '{ f = $1; sub(/.*\//, "", f); sub(/\.ko(\.gz|\.xz|\.zst)?$/, "", f);
+        gsub(/-/, "_", f); if (f == n) { print $1; exit } }' "$kdir/modules.dep")
+    if [ -z "$path" ] && [ -f "$kdir/modules.alias" ]; then
+        alias_target=$(awk -v n="$2" '$1 == "alias" && $2 == n { print $3; exit }' "$kdir/modules.alias")
+        [ -n "$alias_target" ] && [ "$alias_target" != "$2" ] && path=$(module_path "$kdir" "$alias_target")
+    fi
+    [ -n "$path" ] && printf '%s\n' "$path"
+}
+
+# module_builtin reports whether a module name is compiled into the kernel.
+module_builtin() {
+    local kdir="$1" name="${2//-/_}"
+    [ -f "$kdir/modules.builtin" ] || return 1
+    awk -v n="$name" '{ f = $0; sub(/.*\//, "", f); sub(/\.ko$/, "", f); gsub(/-/, "_", f);
+        if (f == n) { found = 1; exit } } END { exit !found }' "$kdir/modules.builtin"
+}
+
+# image_kmod runs the image's own kmod (depmod, modprobe) through its musl
+# loader, so module metadata matches the guest's modprobe without a host kmod.
+image_kmod() {
+    "$CHROOT_DIR/lib/ld-musl-x86_64.so.1" --library-path "$CHROOT_DIR/lib:$CHROOT_DIR/usr/lib" \
+        --argv0 "$1" "$CHROOT_DIR/bin/kmod" "${@:2}"
+}
+
+microvm_strip_modules() {
+    echo "$TAG stripping kernel modules to virtio+fw_cfg+${MICROVM_EXTRA_MODULES} only..."
+    for kver_dir in "$CHROOT_DIR/lib/modules"/*/; do
+        kver="$(basename "$kver_dir")"
+        kernel_dir="$kver_dir/kernel"
+        [ -d "$kernel_dir" ] || continue
+
+        if [ ! -s "$kver_dir/modules.dep" ]; then
+            echo "ERROR: stock ${kver_dir}modules.dep missing — cannot resolve module dependencies" >&2
+            exit 1
+        fi
+
+        # Extra modules start the closure; the glob families are kept as found.
+        keep_list=$(mktemp)
+        : > "$keep_list"
+        for name in $MICROVM_EXTRA_MODULES; do
+            if ! path=$(module_path "$kver_dir" "$name"); then
+                if module_builtin "$kver_dir" "$name"; then continue; fi
+                echo "ERROR: module $name is neither loadable nor built into $kver" >&2
+                exit 1
+            fi
+            echo "$path" >> "$keep_list"
+        done
+
+        # Close over modules.dep hard dependencies and modules.softdep pre:/post:
+        # names until the set stops growing. A softdep must be loadable or built in.
+        while :; do
+            before=$(sort -u "$keep_list" | wc -l)
+            for path in $(sort -u "$keep_list"); do
+                awk -F': ' -v p="$path" '$1 == p { n = split($2, d, " "); for (i = 1; i <= n; i++) print d[i] }' \
+                    "$kver_dir/modules.dep" >> "$keep_list"
+                mod=$(basename "$path"); mod="${mod%%.ko*}"; mod="${mod//-/_}"
+                [ -f "$kver_dir/modules.softdep" ] || continue
+                for soft in $(awk -v n="$mod" '{ m = $2; gsub(/-/, "_", m) }
+                        $1 == "softdep" && m == n { for (i = 3; i <= NF; i++) if ($i != "pre:" && $i != "post:") print $i }' \
+                        "$kver_dir/modules.softdep"); do
+                    if soft_path=$(module_path "$kver_dir" "$soft"); then
+                        echo "$soft_path" >> "$keep_list"
+                    elif ! module_builtin "$kver_dir" "$soft"; then
+                        echo "ERROR: softdep $soft of $mod is neither loadable nor built into $kver" >&2
+                        exit 1
+                    fi
+                done
+            done
+            after=$(sort -u "$keep_list" | wc -l)
+            [ "$before" -eq "$after" ] && break
+        done
+
+        extra_names=$(sort -u "$keep_list" | while read -r p; do m=$(basename "$p"); m="${m%%.ko*}"; echo "${m//-/_}"; done)
+        find_args=(-name "virtio*.ko*" -o -name "*fw_cfg*.ko*")
+        for glob in ${MICROVM_EXTRA_KEEP_GLOBS[@]+"${MICROVM_EXTRA_KEEP_GLOBS[@]}"}; do
+            find_args+=(-o -name "$glob")
+        done
+        (cd "$kver_dir" && find kernel "${find_args[@]}") >> "$keep_list"
+
+        # Copy the keeper set out, preserving each kernel/ relative path.
+        tmp_keep=$(mktemp -d)
+        while read -r rel; do
+            rel="${rel#kernel/}"
+            mkdir -p "$tmp_keep/$(dirname "$rel")"
+            cp "$kernel_dir/$rel" "$tmp_keep/$rel"
+        done < <(sort -u "$keep_list")
+        rm -f "$keep_list"
+
+        # Replace kernel module tree with the keeper set.
+        rm -rf "$kernel_dir"
+        mkdir -p "$kernel_dir"
+        # Restore each kept file into the same relative path.
+        (cd "$tmp_keep" && find . -name "*.ko*" | while read -r f; do
+            dest_dir="$kernel_dir/$(dirname "${f#./}")"
+            mkdir -p "$dest_dir"
+            cp "$f" "$dest_dir/"
+        done)
+        rm -rf "$tmp_keep"
+
+        # Decompress kept modules to plain .ko. Alpine ships .ko.gz, but a host
+        # kmod that lacks gzip support (e.g. Debian trixie's kmod 34) silently
+        # produces an EMPTY modules.dep from compressed modules — the guest's
+        # modprobe then no-ops and qemu_fw_cfg never loads. Plain .ko sidesteps the
+        # host-kmod dependency entirely and the guest needs no decompressor either.
+        find "$kernel_dir" -name "*.ko.gz" -exec gunzip -f {} +
+
+        # Independent closure gate: each extra-closure module's own .modinfo depends=
+        # must name a kept module or a built-in, so a resolver bug fails here.
+        kept_names=$(find "$kernel_dir" -name "*.ko" -exec basename {} .ko \; | tr '-' '_' | sort -u)
+        while read -r ko; do
+            ko_name=$(basename "$ko" .ko); ko_name="${ko_name//-/_}"
+            grep -qx "$ko_name" <<< "$extra_names" || continue
+            deps=$(tr '\0' '\n' < "$ko" | grep -a -m1 '^depends=' | cut -d= -f2 | tr ',' ' ' || true)
+            for dep in $deps; do
+                dep="${dep//-/_}"
+                if ! grep -qx "$dep" <<< "$kept_names" && ! module_builtin "$kver_dir" "$dep"; then
+                    echo "ERROR: $(basename "$ko") depends on $dep, which is neither kept nor built in" >&2
+                    exit 1
+                fi
+            done
+        done < <(find "$kernel_dir" -name "*.ko")
+        for name in $MICROVM_EXTRA_MODULES; do
+            if ! grep -qx "${name//-/_}" <<< "$kept_names" && ! module_builtin "$kver_dir" "$name"; then
+                echo "ERROR: required module $name missing from the stripped module tree" >&2
+                exit 1
+            fi
+        done
+        echo "$TAG extra module closure: $(echo "$extra_names" | tr '\n' ' ')"
+        echo "$TAG modules kept: $(echo "$kept_names" | tr '\n' ' ')"
+
+        # Rebuild module metadata with the image's own kmod, the binary the guest's
+        # modprobe runs, so no host depmod is needed. Stock indexes still name .ko.gz
+        # files, so they are removed first and a failed rebuild cannot ship them.
+        for f in dep dep.bin alias alias.bin softdep weakdep symbols symbols.bin \
+            builtin.bin builtin.alias.bin devname; do
+            rm -f "$kver_dir/modules.$f"
+        done
+        if ! image_kmod depmod -b "$CHROOT_DIR" "$kver"; then
+            echo "ERROR: image depmod failed for $kver" >&2
+            exit 1
+        fi
+        if [ ! -s "$kver_dir/modules.dep" ] || [ ! -s "$kver_dir/modules.dep.bin" ]; then
+            echo "ERROR: ${kver_dir}modules.dep or modules.dep.bin missing or empty after depmod" >&2
+            exit 1
+        fi
+        # Resolution gate: the image's modprobe must resolve every required module
+        # to plain .ko files present in the stripped tree.
+        for name in $MICROVM_EXTRA_MODULES $MICROVM_GATE_MODULES; do
+            module_builtin "$kver_dir" "$name" && continue
+            if ! plan=$(image_kmod modprobe -d "$CHROOT_DIR" -S "$kver" --show-depends "$name"); then
+                echo "ERROR: image modprobe cannot resolve $name for $kver" >&2
+                exit 1
+            fi
+            while read -r verb path _; do
+                [ "$verb" = "insmod" ] || continue
+                case "$path" in
+                    "$CHROOT_DIR/lib/modules/$kver/kernel/"*.ko) [ -f "$path" ] && continue ;;
+                esac
+                echo "ERROR: modprobe plan for $name names $path, not a kept plain .ko" >&2
+                exit 1
+            done <<< "$plan"
+            echo "$TAG modprobe $name: $(awk '{ n = split($2, a, "/"); printf "%s ", a[n] }' <<< "$plan")"
+        done
+    done
+}
+
+# microvm_strip_rootfs drops package-manager artifacts and debug symbols.
+microvm_strip_rootfs() {
+    # --- Strip package-manager artifacts (not needed at runtime) ---
+    echo "$TAG stripping package manager artifacts..."
+    rm -rf \
+        "$CHROOT_DIR/lib/apk" \
+        "$CHROOT_DIR/var/cache/apk" \
+        "$CHROOT_DIR/etc/apk" \
+        "$CHROOT_DIR/usr/share/man" \
+        "$CHROOT_DIR/usr/share/doc" \
+        "$CHROOT_DIR/usr/share/apk" \
+        "$CHROOT_DIR/usr/include" \
+        "$CHROOT_DIR/usr/lib/pkgconfig"
+
+    # --- Strip debug symbols from binaries ---
+    echo "$TAG stripping debug symbols from binaries..."
+    find "$CHROOT_DIR/usr/sbin" "$CHROOT_DIR/usr/bin" \
+         "$CHROOT_DIR/sbin" "$CHROOT_DIR/bin" \
+        -type f 2>/dev/null | while read -r bin; do
+        strip --strip-all "$bin" 2>/dev/null || true
+    done
+}
+
+# microvm_stage_kernel copies vmlinuz to $STAGE_DIR, then drops /boot from the chroot.
+microvm_stage_kernel() {
+    # --- Copy vmlinuz out before stripping /boot from chroot ---
+    VMLINUZ_OUT="$STAGE_DIR/vmlinuz"
+    echo "$TAG copying kernel: $VMLINUZ_OUT"
+    cp "$KERNEL_IMG" "$VMLINUZ_OUT"
+
+    # Drop /boot from the chroot — vmlinuz, Alpine's stock initramfs-virt, System.map
+    # and kernel config all live here (~25 MiB). None are needed at runtime: the
+    # guest already has the kernel loaded by QEMU's -kernel, and Alpine's initramfs
+    # is unused because we replace it with our own /init. Leaving /boot in the cpio
+    # nearly doubles uncompressed initramfs size and forces a 512 MiB guest just to
+    # survive decompression.
+    rm -rf "${CHROOT_DIR:?}/boot"
+}
+
+microvm_pack_initramfs() {
+    # --- Build initramfs ---
+    INITRAMFS_OUT="$STAGE_DIR/initramfs.cpio.gz"
+    echo "$TAG building initramfs: $INITRAMFS_OUT"
+    # Create the static device nodes and pack the archive inside a single fakeroot
+    # session: mknod/cpio see faked device entries and write real char-device
+    # records into the cpio, while the host filesystem never gets an actual node.
+    fakeroot sh -c '
+        cd "$1" || exit 1
+        mknod -m 600 dev/console c 5 1
+        mknod -m 666 dev/null    c 1 3
+        find . | cpio --quiet -o -H newc | gzip -9 > "$2"
+    ' _ "$CHROOT_DIR" "$INITRAMFS_OUT"
+}
@@ -1,293 +0,0 @@
-#!/bin/sh
-# sqs-store-guest_test.sh — the SQS store guest steps start the store only on
-# a valid contract and one checked, mounted disk, never format, and stop the
-# store before unmounting it.
-#
-# Drives build/microvm/sqs-store/sqs-store-lib.sh against a fake fw_cfg, sysfs
-# and stub tools. The stop cases signal only test-owned children.
-# The case settings below are read by the sourced library, not by this file.
-# shellcheck disable=SC2034
-set -eu
-
-ROOT=$(CDPATH='' cd -- "$(dirname -- "$0")/../.." && pwd)
-LIB=${SQS_GUEST_LIB:-"$ROOT/build/microvm/sqs-store/sqs-store-lib.sh"}
-TMP=$(mktemp -d)
-CHILDREN=""
-cleanup() {
-    for p in $CHILDREN; do kill -KILL "$p" 2>/dev/null || true; done
-    rm -rf "$TMP"
-}
-trap cleanup EXIT
-
-fails=0
-pass() { echo "  ok   - $1"; }
-fail() { echo "  FAIL - $1"; fails=$((fails + 1)); }
-check() { if eval "$2"; then pass "$1"; else fail "$1"; fi; }
-
-echo "-- build/microvm/sqs-store/sqs-store-lib.sh"
-
-SERIAL=vol0123456789abcdef0
-NKEY=UAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAB
-ADDR=10.11.12.13
-
-# Stub tools record each call to calls.log; their exit codes come from files.
-mkdir -p "$TMP/bin"
-cat > "$TMP/bin/e2fsck" <<'EOF'
-#!/bin/sh
-echo "e2fsck $*" >> "$T/calls.log"
-exit "$(cat "$T/fsck.rc")"
-EOF
-cat > "$TMP/bin/mount" <<'EOF'
-#!/bin/sh
-echo "mount $*" >> "$T/calls.log"
-exit "$(cat "$T/mount.rc")"
-EOF
-cat > "$TMP/bin/umount" <<'EOF'
-#!/bin/sh
-echo "umount $*" >> "$T/calls.log"
-EOF
-for f in mkfs mkfs.ext4 mke2fs; do
-    printf '#!/bin/sh\necho "mkfs %s $*" >> "$T/calls.log"\n' "$f" > "$TMP/bin/$f"
-done
-cat > "$TMP/bin/ip" <<'EOF'
-#!/bin/sh
-cat "$T/ip.out"
-EOF
-cat > "$TMP/bin/nats-server" <<'EOF'
-#!/bin/sh
-echo "nats-server $*" >> "$T/calls.log"
-[ "$1" = -t ] || echo "GOMEMLIMIT=${GOMEMLIMIT:-unset}" >> "$T/env.log"
-[ "$1" = -t ] && exit "$(cat "$T/configtest.rc")"
-exit 0
-EOF
-chmod +x "$TMP/bin"/*
-
-# fresh <name>: a case directory with a valid contract and one matching disk
-# among others. Callers then break one thing.
-fresh() {
-    T="$TMP/$1"
-    export T
-    mkdir -p "$T/fwcfg" "$T/sys" "$T/dev" "$T/run" "$T/data"
-    for k in data-serial listen-addr client-nkey; do mkdir -p "$T/fwcfg/$k"; done
-    printf '%s\n' "$SERIAL" > "$T/fwcfg/data-serial/raw"
-    printf '%s\n' "$ADDR" > "$T/fwcfg/listen-addr/raw"
-    printf '%s\n' "$NKEY" > "$T/fwcfg/client-nkey/raw"
-    disk vda vol00000000000000000
-    disk vdb "$SERIAL"
-    echo 0 > "$T/fsck.rc"
-    echo 0 > "$T/mount.rc"
-    echo 0 > "$T/configtest.rc"
-    : > "$T/calls.log"
-    : > "$T/console"
-    printf '1: lo    inet 127.0.0.1/8 scope host lo\n2: eth0    inet %s/24 brd 10.11.12.255 scope global eth0\n' "$ADDR" > "$T/ip.out"
-}
-disk() { mkdir -p "$T/sys/$1"; printf '%s\n' "$2" > "$T/sys/$1/serial"; : > "$T/dev/$1"; }
-
-# prepare: run sqs_prepare in a subshell with every path pointed at the case.
-prepare() {
-    (
-        SQS_FWCFG="$T/fwcfg" SQS_SYSBLOCK="$T/sys" SQS_DEVDIR="$T/dev" SQS_NODE_TEST=-e
-        SQS_RUN="$T/run" SQS_DATA="$T/data" SQS_CONSOLE="$T/console"
-        E2FSCK="$TMP/bin/e2fsck" MOUNT="$TMP/bin/mount" IP="$TMP/bin/ip" NATS_BIN="$TMP/bin/nats-server"
-        PATH="$TMP/bin:$PATH"
-        export PATH
-        # shellcheck source=../../build/microvm/sqs-store/sqs-store-lib.sh
-        . "$LIB"
-        sqs_prepare
-    ) > /dev/null 2>&1
-}
-refused() { [ ! -e "$T/run/ready" ] && grep -q FATAL "$T/console"; }
-never_mounted() { ! grep -q '^mount ' "$T/calls.log"; }
-never_formatted() { ! grep -q '^mkfs ' "$T/calls.log"; }
-never_checked() { ! grep -q '^e2fsck ' "$T/calls.log"; }
-
-# --- A valid contract prepares the store ---
-fresh valid
-check "valid contract is prepared" 'prepare && [ -e "$T/run/ready" ]'
-check "the matching disk is checked, then mounted at the data dir" \
-    'grep -qx "e2fsck -p $T/dev/vdb" "$T/calls.log" && grep -qx "mount -t ext4 -o nodev,nosuid,noexec $T/dev/vdb $T/data" "$T/calls.log"'
-check "config listens only on the supplied address" \
-    'grep -c "listen" "$T/run/nats.conf" | grep -qx 1 && grep -qx "listen: \"$ADDR:4222\"" "$T/run/nats.conf"'
-check "config admits only the supplied user NKey" 'grep -q "users = \[ { nkey: \"$NKEY\" } \]" "$T/run/nats.conf"'
-check "config syncs JetStream on the data disk" \
-    'grep -q "store_dir: \"$T/data\"" "$T/run/nats.conf" && grep -q "sync_interval: always" "$T/run/nats.conf"'
-check "config admits a full SQS message (2MiB max_payload)" \
-    'grep -c "max_payload" "$T/run/nats.conf" | grep -qx 1 && grep -qx "max_payload: 2MiB" "$T/run/nats.conf"'
-check "config is tested by the server before ready" 'grep -qx "nats-server -t -c $T/run/nats.conf" "$T/calls.log"'
-check "no auth material on the console" '! grep -q "$NKEY" "$T/console"'
-check "prepare reports the disk and config, not a listener" 'grep -q "prepared: $T/dev/vdb mounted" "$T/console" && ! grep -qi "listening" "$T/console"'
-
-fresh fsck-corrected; echo 1 > "$T/fsck.rc"
-check "e2fsck exit 1 (corrected) proceeds" 'prepare && [ -e "$T/run/ready" ]'
-
-# --- Disk refusals: nothing mounted, nothing formatted ---
-fresh no-disk; rm -rf "$T/sys/vdb"
-check "missing disk is refused before any check or mount" '! prepare && refused && never_checked && never_mounted'
-
-fresh wrong-disk; printf 'vol0fffffffffffffff0\n' > "$T/fwcfg/data-serial/raw"
-check "disk with another serial is refused" '! prepare && refused && never_checked && never_mounted'
-
-fresh two-disks; disk vdc "$SERIAL"
-check "two disks with the serial are refused" '! prepare && refused && never_checked && never_mounted'
-
-fresh bad-serial; printf 'vol-0123456789abcdef\n' > "$T/fwcfg/data-serial/raw"
-check "malformed serial is refused" '! prepare && refused && never_mounted'
-
-fresh no-node; rm -f "$T/dev/vdb"
-check "disk without a device node is refused" '! prepare && refused && never_mounted'
-
-for rc in 2 4 8; do
-    fresh "fsck-$rc"; echo "$rc" > "$T/fsck.rc"
-    check "e2fsck exit $rc is refused without mount or format" '! prepare && refused && never_mounted && never_formatted'
-done
-
-fresh mount-fail; echo 32 > "$T/mount.rc"
-check "failed mount is refused" '! prepare && refused'
-
-# --- Auth and listener refusals happen before the disk is touched ---
-fresh no-nkey; rm -rf "$T/fwcfg/client-nkey"
-check "missing NKey is refused before the disk" '! prepare && refused && never_checked && never_mounted'
-
-for bad in seed lower short two empty; do
-    fresh "nkey-$bad"
-    case $bad in
-        seed) v=SUAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA ;;
-        lower) v=uaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaab ;;
-        short) v=UAAAA ;;
-        two) v="$NKEY
-$NKEY" ;;
-        empty) v="" ;;
-    esac
-    printf '%s\n' "$v" > "$T/fwcfg/client-nkey/raw"
-    check "NKey ($bad) is refused before the disk, without echoing it" \
-        '! prepare && refused && never_checked && { [ -z "$v" ] || ! grep -qF "$v" "$T/console"; }'
-done
-
-fresh nkey-checksum; echo 1 > "$T/configtest.rc"
-check "NKey the server rejects is refused before the disk" '! prepare && refused && never_checked && never_mounted'
-
-fresh no-addr; rm -rf "$T/fwcfg/listen-addr"
-check "missing listen address is refused" '! prepare && refused && never_mounted'
-
-fresh addr-absent; printf '1: lo    inet 127.0.0.1/8 scope host lo\n2: eth0    inet 10.11.12.99/24 scope global eth0\n' > "$T/ip.out"
-check "listen address not on a guest NIC is refused" '! prepare && refused && never_mounted'
-
-fresh addr-loopback; printf '127.0.0.1\n' > "$T/fwcfg/listen-addr/raw"
-check "loopback listen address is refused" '! prepare && refused && never_mounted'
-
-for bad in 0.0.0.0x 1.2.3 1.2.3.256 01.2.3.4; do
-    fresh "addr-$bad"; printf '%s\n' "$bad" > "$T/fwcfg/listen-addr/raw"
-    check "malformed listen address $bad is refused" '! prepare && refused && never_mounted'
-done
-
-# --- Run: the store starts only when prepared and not stopping ---
-run_step() {
-    (
-        SQS_RUN="$T/run" SQS_CONSOLE="$T/console" NATS_BIN="$TMP/bin/nats-server"
-        # shellcheck source=../../build/microvm/sqs-store/sqs-store-lib.sh
-        . "$LIB"
-        sqs_run
-    ) > /dev/null 2>&1
-}
-fresh run-ready; prepare; : > "$T/calls.log"
-check "a prepared store starts the server" 'run_step && grep -qx "nats-server -c $T/run/nats.conf" "$T/calls.log"'
-fresh run-memlimit; prepare; : > "$T/env.log"
-check "the server runs under the ADR-0011:S11 heap limit" 'run_step && grep -qx "GOMEMLIMIT=512MiB" "$T/env.log"'
-fresh run-unready
-check "an unprepared store never starts the server" '! run_step && ! grep -q "^nats-server" "$T/calls.log"'
-fresh run-stopping; prepare; : > "$T/calls.log"; : > "$T/run/stopping"
-check "a stopping store never starts the server" '! run_step && ! grep -q "^nats-server" "$T/calls.log"'
-
-# --- Stop: SIGTERM the verified store, then unmount ---
-STORE="$TMP/bin/fake-store"
-cp "$(command -v sleep)" "$STORE"
-stop_step() {
-    (
-        SQS_RUN="$T/run" SQS_DATA="$T/data" SQS_CONSOLE="$T/console" SQS_MOUNTS="$T/mounts"
-        UMOUNT="$TMP/bin/umount" NATS_BIN="$STORE" SQS_STOP_BUDGET_TENTHS=10
-        # shellcheck source=../../build/microvm/sqs-store/sqs-store-lib.sh
-        . "$LIB"
-        sqs_stop
-    ) > /dev/null 2>&1
-}
-gone() { [ ! -r "/proc/$1/status" ] || grep -q '^State:[[:space:]]*Z' "/proc/$1/status"; }
-# record <pid>: write the identity sqs_run would record for that process.
-# shellcheck source=../../build/microvm/sqs-store/sqs-store-lib.sh
-record() { echo "$1 $( . "$LIB"; sqs_starttime "$1" )" > "$T/run/pid"; }
-
-fresh stop-orderly; echo "dev $T/data ext4 rw 0 0" > "$T/mounts"
-"$STORE" 60 & pid=$!; CHILDREN="$CHILDREN $pid"; record "$pid"
-stop_step
-check "stop sends SIGTERM and the store exits" 'gone "$pid" && grep -q "SIGTERM sent" "$T/console" && ! grep -q SIGKILL "$T/console"'
-check "stop blocks respawn and unmounts after the store exits" \
-    '[ -e "$T/run/stopping" ] && grep -qx "umount $T/data" "$T/calls.log"'
-
-fresh stop-hung; echo "dev $T/data ext4 rw 0 0" > "$T/mounts"
-(trap '' TERM; exec "$STORE" 60) & pid=$!; CHILDREN="$CHILDREN $pid"; record "$pid"
-stop_step
-check "store ignoring SIGTERM is killed after the budget, then unmounted" \
-    'gone "$pid" && grep -q SIGKILL "$T/console" && grep -qx "umount $T/data" "$T/calls.log"'
-
-fresh stop-reused; echo "dev $T/data ext4 rw 0 0" > "$T/mounts"
-"$STORE" 60 & pid=$!; CHILDREN="$CHILDREN $pid"; echo "$pid 1" > "$T/run/pid"
-stop_step
-check "a recorded PID with another start time is never signalled" '! gone "$pid" && grep -q "no running store" "$T/console"'
-kill -KILL "$pid"
-
-# A start held after its stopping check and before its exec: the run step
-# blocks writing "run: starting" to a FIFO console until the test reads it.
-fresh stop-pending-start; echo "dev $T/data ext4 rw 0 0" > "$T/mounts"
-: > "$T/run/ready"
-mkfifo "$T/run-console"
-printf '#!/bin/sh\n: > "%s/started"\nexec "%s" 60\n' "$T" "$STORE" > "$TMP/bin/pending-nats"
-chmod +x "$TMP/bin/pending-nats"
-SQS_RUN="$T/run" SQS_CONSOLE="$T/run-console" NATS_BIN="$TMP/bin/pending-nats" \
-    sh -c '. "$1"; sqs_run' sh "$LIB" & pid=$!; CHILDREN="$CHILDREN $pid"
-i=0
-while [ "$i" -lt 100 ] && ! grep -q "^$pid " "$T/run/pid" 2>/dev/null; do i=$((i + 1)); sleep 0.05; done
-check "pending start is held before its exec" 'grep -q "^$pid " "$T/run/pid" && ! gone "$pid" && [ ! -e "$T/started" ]'
-stop_step
-# Release the FIFO: a start that survived the stop would exec the server now.
-cat "$T/run-console" > /dev/null 2>&1 &
-reader=$!; CHILDREN="$CHILDREN $reader"
-i=0
-while [ "$i" -lt 20 ] && ! gone "$pid"; do i=$((i + 1)); sleep 0.05; done
-check "stop ends a pending start before it can exec the server" 'gone "$pid" && [ ! -e "$T/started" ]'
-check "stop unmounts only after the pending start is gone" \
-    'grep -q "store exited" "$T/console" && grep -qx "umount $T/data" "$T/calls.log"'
-kill -KILL "$reader" 2>/dev/null || true
-
-# --- Shutdown: the generic remount runs only once the store is gone ---
-# shutdown_step [survivor-pid]: with a pid, that process is reported alive
-# whatever happens to it, as a writer stuck past SIGKILL would be.
-shutdown_step() {
-    (
-        SQS_RUN="$T/run" SQS_DATA="$T/data" SQS_CONSOLE="$T/console" SQS_MOUNTS="$T/mounts"
-        UMOUNT="$TMP/bin/umount" SQS_STOP_BUDGET_TENTHS=5
-        # shellcheck source=../../build/microvm/sqs-store/sqs-store-lib.sh
-        . "$LIB"
-        if [ -n "${1:-}" ]; then sqs_same() { [ "$1" = "$survivor" ]; }; survivor=$1; fi
-        sqs_shutdown
-    ) > /dev/null 2>&1
-}
-
-fresh shutdown-orderly; echo "dev $T/data ext4 rw 0 0" > "$T/mounts"
-"$STORE" 60 & pid=$!; CHILDREN="$CHILDREN $pid"; record "$pid"
-check "orderly shutdown succeeds" 'shutdown_step'
-check "orderly shutdown unmounts the data disk, then runs umount -a -r" \
-    '[ "$(grep "^umount" "$T/calls.log" | tr "\n" "|")" = "umount $T/data|umount -a -r|" ]'
-
-fresh shutdown-survivor; echo "dev $T/data ext4 rw 0 0" > "$T/mounts"
-"$STORE" 60 & pid=$!; CHILDREN="$CHILDREN $pid"; record "$pid"
-check "shutdown with a surviving store reports failure" '! shutdown_step "$pid"'
-check "a surviving store leaves the data disk and every other mount alone" \
-    '! grep -q "^umount" "$T/calls.log" && grep -q "STILL ALIVE" "$T/console" && grep -q "skipping umount -a -r" "$T/console"'
-
-fresh stop-unmounted; : > "$T/mounts"
-stop_step
-check "stop with no data mount does not unmount" '! grep -q "^umount" "$T/calls.log"'
-
-if [ "$fails" -ne 0 ]; then
-    echo "  $fails failure(s)"
-    exit 1
-fi
@@ -11,13 +11,11 @@ STAGE_DIR="${1:?usage: sqs-store-microvm-publish.sh STAGE_DIR OUT_DIR}"
 OUT_DIR="${2:?usage: sqs-store-microvm-publish.sh STAGE_DIR OUT_DIR}"
 VMLINUZ="$STAGE_DIR/vmlinuz"
 INITRAMFS="$STAGE_DIR/initramfs.cpio.gz"
-PROFILE="${3:-store}"
+PROFILE="${3:-worker}"
 case "$PROFILE" in
-    store)
-        payload="usr/local/bin/nats-server usr/local/lib/sqs-store/sqs-store-lib.sh usr/local/lib/sqs-store/sqs-store-prepare usr/local/lib/sqs-store/sqs-store-run usr/local/lib/sqs-store/sqs-store-stop" ;;
    worker)
-        payload="usr/local/bin/sqs-worker usr/local/lib/sqs-worker/sqs-worker-lib.sh usr/local/lib/sqs-worker/sqs-worker-prepare usr/local/lib/sqs-worker/sqs-worker-ready usr/local/lib/sqs-worker/sqs-worker-stop data" ;;
-    *) echo "ERROR: profile must be store or worker, not $PROFILE" >&2; exit 1 ;;
+        payload="usr/local/bin/nats-server usr/local/bin/sqs-worker usr/local/lib/sqs-worker/sqs-worker-lib.sh usr/local/lib/sqs-worker/sqs-worker-prepare usr/local/lib/sqs-worker/sqs-worker-store usr/local/lib/sqs-worker/sqs-worker-ready usr/local/lib/sqs-worker/sqs-worker-stop data" ;;
+    *) echo "ERROR: profile must be worker, not $PROFILE" >&2; exit 1 ;;
 esac
 tag="[build-sqs-$PROFILE-microvm-image]"

@@ -56,12 +54,7 @@ for want_ko in ext4 tiny_power_button; do
    fi
 done
 if grep -qE '(^|/)(lb-agent|haproxy|nginx)$' <<< "$archive_list"; then
-    echo "ERROR: LB data-plane binaries found in the SQS store image" >&2
-    exit 1
-fi
-# The worker holds the store client key; it must never carry the store server.
-if [ "$PROFILE" = worker ] && grep -qE '(^|/)nats-server$' <<< "$archive_list"; then
-    echo "ERROR: nats-server found in the SQS worker image" >&2
+    echo "ERROR: LB data-plane binaries found in the SQS image" >&2
    exit 1
 fi
 echo "$tag content gate: PASS"
@@ -1,5 +1,5 @@
 #!/bin/sh
-# sqs-store-microvm-publish_test.sh — a staged SQS store microVM pair reaches
+# sqs-store-microvm-publish_test.sh — a staged SQS microVM pair reaches
 # the output directory only when every gate passes, and a failed build cannot
 # leave a pair that make then takes as up to date.
 #
@@ -28,9 +28,10 @@ done
 stage() {
    dir=$1 omit=${2:-}
    root="$TMP/root.$$.$(date +%s%N)"
-    for p in init etc/inittab usr/local/bin/nats-server sbin/mkfs.ext4 sbin/e2fsck \
-        usr/local/lib/sqs-store/sqs-store-lib.sh usr/local/lib/sqs-store/sqs-store-prepare \
-        usr/local/lib/sqs-store/sqs-store-run usr/local/lib/sqs-store/sqs-store-stop \
+    for p in init etc/inittab usr/local/bin/nats-server usr/local/bin/sqs-worker data sbin/mkfs.ext4 sbin/e2fsck \
+        usr/local/lib/sqs-worker/sqs-worker-lib.sh usr/local/lib/sqs-worker/sqs-worker-prepare \
+        usr/local/lib/sqs-worker/sqs-worker-store usr/local/lib/sqs-worker/sqs-worker-ready \
+        usr/local/lib/sqs-worker/sqs-worker-stop \
        dev/console dev/null lib/modules/6.12-virt/kernel/fs/ext4/ext4.ko \
        lib/modules/6.12-virt/kernel/drivers/acpi/tiny-power-button.ko; do
        [ "$p" = "$omit" ] && continue
@@ -78,7 +79,7 @@ else
 fi

 # --- Each failing gate keeps the previous valid pair exactly ---
-for case in nats-server store-stop ext4 size; do
+for case in nats-server worker-store ext4 size; do
    d="$TMP/keep-$case"
    stage "$d/prev"
    bash "$PUBLISH" "$d/prev" "$d/out" >/dev/null 2>&1
@@ -86,7 +87,7 @@ for case in nats-server store-stop ext4 size; do
    max=50
    case $case in
        nats-server) stage "$d/stage" usr/local/bin/nats-server ;;
-        store-stop) stage "$d/stage" usr/local/lib/sqs-store/sqs-store-stop ;;
+        worker-store) stage "$d/stage" usr/local/lib/sqs-worker/sqs-worker-store ;;
        ext4) stage "$d/stage" lib/modules/6.12-virt/kernel/fs/ext4/ext4.ko ;;
        size) stage "$d/stage"; max=0 ;;
    esac
@@ -1,7 +1,8 @@
 #!/bin/sh
-# sqs-worker-guest_test.sh — the SQS worker guest's readiness watcher is
-# supervised by init, never leaves a store-ready marker behind a dead watcher,
-# and shutdown unmounts the worker disk only once the watcher is gone.
+# sqs-worker-guest_test.sh — the SQS guest's store starts only on a config
+# prepared after the receipt key settled, its readiness watcher never leaves a
+# store-ready marker behind a dead watcher, and shutdown unmounts the data disk
+# only once both the watcher and the store are gone.
 #
 # Drives build/microvm/sqs-worker/sqs-worker-lib.sh with test-owned processes
 # and a stub umount. Every signal goes only to this test's own children.
@@ -12,6 +13,7 @@ set -eu
 ROOT=$(CDPATH='' cd -- "$(dirname -- "$0")/../.." && pwd)
 LIB=${SQSW_GUEST_LIB:-"$ROOT/build/microvm/sqs-worker/sqs-worker-lib.sh"}
 INITTAB="$ROOT/build/microvm/sqs-worker/inittab"
+GUEST_INIT="$ROOT/build/microvm/sqs-worker/init.sh"
 TMP=$(mktemp -d)
 CHILDREN=""
 cleanup() {
@@ -41,26 +43,56 @@ echo "$$" >> "$T/watchers"
 echo "10.0.0.5 UPUB" > "$T/run/store-ready"
 exec sleep 60
 EOF
+# The fake server checks its config, or notes how it was started and stays.
+cat > "$TMP/bin/nats" <<'EOF'
+#!/bin/sh
+if [ "$1" = -t ]; then [ -z "${NATS_BAD:-}" ]; exit; fi
+echo "$GOMEMLIMIT $*" >> "$T/servers"
+exec sleep 60
+EOF
+# The fake boot takes its exit codes in order from $T/boot-rcs, and like the
+# real one writes the receipt marker only on success; "hold" blocks like a slow call.
+cat > "$TMP/bin/boot" <<'EOF'
+#!/bin/sh
+echo "$*" >> "$T/boots"
+rc=$(head -n1 "$T/boot-rcs"); sed -i 1d "$T/boot-rcs"
+if [ "$rc" = hold ]; then echo "$$" > "$T/boot-child"; exec sleep 60; fi
+[ "$rc" != 0 ] || echo FP > "$T/run/receipt-ready"
+exit "$rc"
+EOF
 chmod +x "$TMP/bin"/*

 fresh() {
    T="$TMP/$1"
    export T
    mkdir -p "$T/run" "$T/data"
-    echo UPUBLICKEY > "$T/run/public-key"
+    echo FP > "$T/run/receipt-ready"
+    echo "listen: 127.0.0.1:4222" > "$T/run/store.conf"
    for f in calls.log console mounts; do : > "$T/$f"; done
 }
 gone() { [ ! -r "/proc/$1/status" ] || grep -q '^State:[[:space:]]*Z' "/proc/$1/status" 2>/dev/null; }
 waitfor() { i=0; while [ "$i" -lt 100 ] && ! eval "$1"; do i=$((i + 1)); sleep 0.05; done; eval "$1"; }
 # start_run: one init respawn of the readiness step, in the background.
 start_run() {
-    SQSW_RUN="$T/run" SQSW_CONSOLE="$T/console" SQSW_BIN="$TMP/bin/watcher" SQSW_RESPAWN_GAP="${GAP:-0}" SQSW_PARK_INTERVAL="${PARK:-30}" \
+    SQSW_RUN="$T/run" SQSW_CONSOLE="$T/console" SQSW_BIN="$TMP/bin/watcher" SQSW_RESPAWN_GAP="${GAP:-0}" SQSW_PARK="sleep 60" \
        sh -c '. "$1"; sqsw_run' sh "$LIB" & pid=$!
    CHILDREN="$CHILDREN $pid"
 }
+# start_store: one init respawn of the store step, in the background.
+start_store() {
+    SQSW_RUN="$T/run" SQSW_DATA="$T/data" SQSW_CONSOLE="$T/console" SQSW_NATS_BIN="$TMP/bin/nats" \
+        SQSW_BIN="$TMP/bin/boot" SQSW_BOOT_RETRY=0 SQSW_PARK="sleep 60" SQSW_LIB="$LIB" \
+        sh -c '. "$1"; sqsw_store' sh "$LIB" & spid=$!
+    CHILDREN="$CHILDREN $spid"
+}
+# prepare_store: the sysinit step's store preparation, after boot.
+prepare_store() {
+    SQSW_RUN="$T/run" SQSW_DATA="$T/data" SQSW_CONSOLE="$T/console" SQSW_NATS_BIN="$TMP/bin/nats" \
+        sh -c '. "$1"; sqsw_prepare_store' sh "$LIB"
+}
 # record <pid>: write the identity sqs_run would record for that process.
 # shellcheck source=../../build/microvm/sqs-worker/sqs-worker-lib.sh
-record() { echo "$1 $( . "$LIB"; sqsw_starttime "$1" )" > "$T/run/ready.pid"; }
+record() { echo "$1 $( . "$LIB"; sqsw_starttime "$1" )" > "$T/run/${2:-ready}.pid"; }
 # shutdown_step [survivor-pid]: with a pid, that process is reported alive
 # whatever happens to it, as a watcher stuck past SIGKILL would be.
 shutdown_step() {
@@ -74,8 +106,86 @@ shutdown_step() {
    ) > /dev/null 2>&1
 }

-check "init respawns the readiness step rather than running it once" \
-    'grep -qx "::respawn:/usr/local/lib/sqs-worker/sqs-worker-ready" "$INITTAB" && ! grep -q "::once:" "$INITTAB"'
+check "init respawns the store and readiness steps rather than running them once" \
+    'grep -qx "::respawn:/usr/local/lib/sqs-worker/sqs-worker-store" "$INITTAB" &&
+     grep -qx "::respawn:/usr/local/lib/sqs-worker/sqs-worker-ready" "$INITTAB" && ! grep -q "::once:" "$INITTAB"'
+# The store binds 127.0.0.1, which a fresh guest kernel leaves down.
+lo_at=$(grep -n -x "ip link set lo up" "$GUEST_INIT" | cut -d: -f1 | head -n 1)
+init_at=$(grep -n -x "exec /sbin/init" "$GUEST_INIT" | cut -d: -f1 | head -n 1)
+check "guest init brings loopback up before init starts the store" \
+    '[ -n "$lo_at" ] && [ -n "$init_at" ] && [ "$lo_at" -lt "$init_at" ]'
+
+# --- The store is prepared only after boot settled the receipt key ---
+fresh prepare-ok; rm -f "$T/run/store.conf"
+check "a settled receipt key prepares the store" 'prepare_store && [ -s "$T/run/store.conf" ]'
+check "the store listens only on loopback, with no auth, monitor, route or cluster listener" \
+    'grep -qx "listen: \"127.0.0.1:4222\"" "$T/run/store.conf" && grep -qx "  store_dir: \"$T/data/jetstream\"" "$T/run/store.conf" &&
+     ! grep -Eq "authorization|http|monitor|cluster|routes|leafnodes|websocket|mqtt" "$T/run/store.conf"'
+fresh prepare-no-key; rm -f "$T/run/store.conf" "$T/run/receipt-ready"
+check "a receipt key boot did not settle leaves the store unprepared" '! prepare_store && [ ! -e "$T/run/store.conf" ]'
+fresh prepare-bad; rm -f "$T/run/store.conf"
+check "a config the server rejects leaves the store unprepared" '! NATS_BAD=1 prepare_store && [ ! -e "$T/run/store.conf" ]'
+
+# --- The store step starts the server only on a prepared config ---
+fresh store-run
+start_store
+check "a prepared store starts the dedicated server with its heap bound" \
+    'waitfor "[ -s \"\$T/servers\" ]" && grep -qx "512MiB -c $T/run/store.conf" "$T/servers"'
+kill -KILL "$spid" 2>/dev/null || true
+fresh store-refused; : > "$T/run/boot-refused"
+start_store; waitfor "grep -q 'store: boot refused' \"\$T/console\"" || true
+check "a store whose boot was refused parks instead of starting or respawning" \
+    '! gone "$spid" && [ ! -e "$T/servers" ] && [ ! -e "$T/boots" ]'
+kill -KILL "$spid" 2>/dev/null || true
+
+# --- Boot is supervised: a transient failure is retried, a refusal is final ---
+fresh boot-transient; rm -f "$T/run/store.conf" "$T/run/receipt-ready"; printf '75\n0\n' > "$T/boot-rcs"
+start_store
+check "a transient boot failure exits so init runs the step again, starting nothing" \
+    'waitfor "gone $spid" && [ ! -e "$T/servers" ] && [ ! -e "$T/run/boot-refused" ]'
+start_store
+check "the respawn after recovery boots again and starts the store" \
+    'waitfor "[ -s \"\$T/servers\" ]" && [ "$(wc -l < "$T/boots")" -eq 2 ] && [ -s "$T/run/store.conf" ]'
+kill -KILL "$spid" 2>/dev/null || true
+fresh boot-final; rm -f "$T/run/store.conf" "$T/run/receipt-ready"; printf '1\n0\n' > "$T/boot-rcs"
+start_store; waitfor "grep -q 'boot refused (rc=1)' \"\$T/console\"" || true
+check "a final boot refusal parks the store and records the refusal" \
+    '! gone "$spid" && [ -e "$T/run/boot-refused" ] && [ ! -e "$T/servers" ]'
+kill -KILL "$spid" 2>/dev/null || true
+start_store; waitfor "grep -q 'store: boot refused, not starting' \"\$T/console\"" || true
+check "a later respawn never boots again after a final refusal" '[ "$(wc -l < "$T/boots")" -eq 1 ]'
+kill -KILL "$spid" 2>/dev/null || true
+fresh boot-settled; rm -f "$T/run/store.conf"; : > "$T/boot-rcs"
+start_store
+check "a respawn after the key settled starts the store without booting again" \
+    'waitfor "[ -s \"\$T/servers\" ]" && [ ! -e "$T/boots" ]'
+kill -KILL "$spid" 2>/dev/null || true
+fresh boot-held; rm -f "$T/run/store.conf" "$T/run/receipt-ready"; echo hold > "$T/boot-rcs"
+echo "/dev/vdb $T/data ext4 rw 0 0" > "$T/mounts"
+start_store; waitfor '[ -s "$T/boot-child" ]' || true
+bchild=$(cat "$T/boot-child" 2>/dev/null); CHILDREN="$CHILDREN $bchild"
+shutdown_step
+check "shutdown stops a boot still in progress before it unmounts the disk" \
+    'waitfor "gone $bchild" && grep -q "umount $T/data" "$T/calls.log"'
+kill -KILL "$spid" "$bchild" 2>/dev/null || true
+fresh boot-held-survivor; rm -f "$T/run/store.conf" "$T/run/receipt-ready"; echo hold > "$T/boot-rcs"
+echo "/dev/vdb $T/data ext4 rw 0 0" > "$T/mounts"
+start_store; waitfor '[ -s "$T/boot-child" ]' || true
+bchild=$(cat "$T/boot-child" 2>/dev/null); CHILDREN="$CHILDREN $bchild"
+check "a boot that survives SIGKILL fails shutdown and keeps the disk mounted" \
+    '! shutdown_step "$bchild" && ! grep -q umount "$T/calls.log" && grep -q "boot pid=$bchild STILL ALIVE" "$T/console"'
+kill -KILL "$spid" "$bchild" 2>/dev/null || true
+# Shutdown sets stopping after the store's own check but before the boot child
+# runs: the child must still record itself and never start sqs-worker boot.
+fresh boot-window; rm -f "$T/run/store.conf" "$T/run/receipt-ready"; echo 0 > "$T/boot-rcs"
+: > "$T/run/stopping"
+SQSW_RUN="$T/run" SQSW_CONSOLE="$T/console" SQSW_BIN="$TMP/bin/boot" SQSW_BOOT_RETRY=0 SQSW_PARK="sleep 60" SQSW_LIB="$LIB" \
+    sh -c '. "$1"; sqsw_boot' sh "$LIB" & wpid=$!
+CHILDREN="$CHILDREN $wpid"
+waitfor "grep -q 'shutdown during boot' \"\$T/console\"" || true
+check "a boot child started after shutdown began never runs boot" \
+    'grep -q "shutdown during boot" "$T/console" && [ ! -e "$T/boots" ] && [ ! -e "$T/run/boot-refused" ]'
+kill -KILL "$wpid" 2>/dev/null || true

 # --- A watcher killed after readiness: its marker must not outlive it ---
 fresh killed-after-ready
@@ -103,8 +213,7 @@ kill -KILL "$pid" 2>/dev/null || true
 # A run that cannot start parks rather than exiting, so init does not respawn
 # it in a loop; a second is many park intervals, and it must log only once.
 fresh run-stopping; : > "$T/run/stopping"; echo stale > "$T/run/store-ready"
-PARK=0.05 start_run; waitfor "grep -q 'shutdown in progress' \"\$T/console\"" || true
-sleep 1
+start_run; waitfor "grep -q 'shutdown in progress' \"\$T/console\"" || true
 check "a start during shutdown parks instead of exiting and respawning" '! gone "$pid"'
 check "a parked start during shutdown starts no watcher and removes the marker" \
    '[ ! -e "$T/watchers" ] && [ ! -e "$T/run/store-ready" ]'
@@ -112,26 +221,24 @@ check "a parked start during shutdown logs once" '[ "$(grep -c "shutdown in prog
 shutdown_step
 check "shutdown stops the parked start" 'waitfor "gone $pid"'

-fresh run-no-key; rm -f "$T/run/public-key"
-PARK=0.05 start_run; waitfor "grep -q 'no worker key' \"\$T/console\"" || true
-sleep 1
-check "a guest without a worker key parks instead of exiting and respawning" '! gone "$pid" && [ ! -e "$T/watchers" ]'
-check "a parked guest without a worker key logs once" '[ "$(grep -c "no worker key" "$T/console")" -eq 1 ]'
-echo UPUBLICKEY > "$T/run/public-key"
-check "a key that appears later starts the watcher" 'waitfor "[ -s \"\$T/watchers\" ]"'
-kill -KILL "$pid" 2>/dev/null || true
-
-fresh run-no-key-stop; rm -f "$T/run/public-key"
-PARK=0.05 start_run; waitfor "grep -q 'no worker key' \"\$T/console\"" || true
+fresh run-unprepared; rm -f "$T/run/store.conf"
+start_run
+check "a watcher whose store is not prepared yet exits to be respawned and never reports ready" \
+    'waitfor "gone $pid" && [ ! -e "$T/watchers" ] && [ ! -e "$T/run/store-ready" ]'
+fresh run-refused; rm -f "$T/run/store.conf"; : > "$T/run/boot-refused"
+start_run; waitfor "grep -q 'ready: boot refused' \"\$T/console\"" || true
+check "a guest whose boot was refused parks its watcher and never reports ready" \
+    '! gone "$pid" && [ ! -e "$T/watchers" ] && [ ! -e "$T/run/store-ready" ]'
 shutdown_step
-check "shutdown stops a start parked for a missing key" 'waitfor "gone $pid" && [ ! -e "$T/watchers" ]'
+check "shutdown stops a watcher parked after a refused boot" 'waitfor "gone $pid" && [ ! -e "$T/watchers" ]'

-# --- Shutdown: unmount only once the watcher is confirmed gone ---
+# --- Shutdown: unmount only once the watcher and the store are gone ---
 fresh shutdown-orderly; echo "dev $T/data ext4 rw 0 0" > "$T/mounts"
-start_run; waitfor "[ -s \"\$T/run/store-ready\" ]"
+start_store; start_run; waitfor "[ -s \"\$T/run/store-ready\" ]"; waitfor "[ -s \"\$T/servers\" ]"
 check "orderly shutdown succeeds" 'shutdown_step'
-check "orderly shutdown confirms the watcher exited, removes the marker, then unmounts" \
-    'gone "$pid" && grep -q "readiness watcher pid=$pid exited" "$T/console" && [ ! -e "$T/run/store-ready" ] &&
+check "orderly shutdown stops the watcher, then the store, removes the marker, then unmounts" \
+    'gone "$pid" && gone "$spid" && [ ! -e "$T/run/store-ready" ] &&
+     [ "$(grep "exited" "$T/console" | sed "s/.*stop: \([a-z]*\) pid.*/\1/" | tr "\n" "|")" = "ready|store|" ] &&
     [ "$(grep "^umount" "$T/calls.log" | tr "\n" "|")" = "umount $T/data|umount -a -r|" ]'
 check "shutdown blocks any later respawn" '[ -e "$T/run/stopping" ]'

@@ -151,10 +258,16 @@ check "a surviving watcher leaves every mount alone and is never reported stoppe
 check "a surviving watcher still leaves no readiness marker" '[ ! -e "$T/run/store-ready" ]'
 kill -KILL "$pid" 2>/dev/null || true

+fresh shutdown-store-survivor; echo "dev $T/data ext4 rw 0 0" > "$T/mounts"
+(trap '' TERM; exec sleep 60) & pid=$!; CHILDREN="$CHILDREN $pid"; record "$pid" store
+check "shutdown with a surviving store reports failure and leaves every mount alone" \
+    '! shutdown_step "$pid" && ! grep -q "^umount" "$T/calls.log" && grep -q "store pid=$pid STILL ALIVE" "$T/console"'
+kill -KILL "$pid" 2>/dev/null || true
+
 fresh shutdown-reused; echo "dev $T/data ext4 rw 0 0" > "$T/mounts"
 (exec sleep 60) & pid=$!; CHILDREN="$CHILDREN $pid"; echo "$pid 1" > "$T/run/ready.pid"
 shutdown_step
-check "a recorded PID with another start time is never signalled" '! gone "$pid" && grep -q "no running readiness watcher" "$T/console"'
+check "a recorded PID with another start time is never signalled" '! gone "$pid" && grep -q "no running ready" "$T/console"'
 kill -KILL "$pid" 2>/dev/null || true

 if [ "$fails" -ne 0 ]; then
@@ -1,11 +1,10 @@
 #!/bin/sh
-# stage-sqs-store-microvm.sh <image-dir> <dist-root> [sqs-store|sqs-worker]:
-# copy a built SQS microVM image into a release tree as <dist-root>/microvm-<name>.
+# stage-sqs-store-microvm.sh <image-dir> <dist-root> [sqs-worker]: copy the
+# built SQS microVM image into a release tree as <dist-root>/microvm-sqs-worker.
 # Without a complete image it stages nothing and says so, as the LB image does.
 set -eu
-src=$1 name=${3:-sqs-store}
+src=$1 name=${3:-sqs-worker}
 case "$name" in
-    sqs-store) label="SQS store" ;;
    sqs-worker) label="SQS worker" ;;
    *) echo "[distro] unknown SQS microVM image '$name'" >&2; exit 2 ;;
 esac
@@ -100,7 +100,8 @@ profile_binaries() {
 ./cmd/eks-gateway-fetch:usr/local/bin/eks-gateway-fetch \
 ./cmd/eks-webhook-cert:usr/local/bin/eks-webhook-cert \
 ./cmd/eks-konnectivity-cert:usr/local/bin/eks-konnectivity-cert \
-./cmd/eks-etcd-member:usr/local/bin/eks-etcd-member"
+./cmd/eks-etcd-member:usr/local/bin/eks-etcd-member \
+./cmd/eks-pod-identity-webhook:usr/local/bin/eks-pod-identity-webhook"
            ;;
        ecs) echo "./cmd/ecs-agent:usr/local/bin/ecs-agent" ;;
        *)   echo "" ;;
@@ -7,7 +7,7 @@
 # Debian is the build host regardless of what is being built — mkosi bootstraps
 # the target distribution from its own archive, so an Ubuntu image builds fine
 # from here (see ubuntu-keyring below).
-FROM debian:trixie-slim@sha256:62bd423aeee912639ca8ee681c2b762cb3e25137e0a92b20be455898a9f034ad
+FROM debian:trixie-slim@sha256:f6b08b09a05725f9431502d3b5714f5c4576e82c1a84fa1debe0c24a7e0c3ec4

 # The runner user is uid 1000; matching it keeps bind-mounted build output
 # owned by the invoking user instead of root.
@@ -1014,6 +1014,9 @@ RESOLV_CONF="${SPINIFEX_RESOLV_CONF:-/etc/resolv.conf}"
 RESOLVED_STUB=/run/systemd/resolve/stub-resolv.conf
 # node-reset.sh keeps /var/lib/spinifex-reset, so the copy survives until uninstall.
 RESOLV_CONF_SAVED="${SPINIFEX_RESOLV_CONF_SAVED:-/var/lib/spinifex-reset/resolv.conf.orig}"
+# NetworkManager started while resolv.conf is a plain file rewrites it on its
+# next DNS update, replacing the stub link; this drop-in stops that.
+NM_DNS_DROPIN="${SPINIFEX_NM_DNS_DROPIN:-/etc/NetworkManager/conf.d/90-spinifex-dns.conf}"

 # systemd-resolved ships resolvconf as a link to resolvectl, which only works
 # while resolved runs, so that one is not a resolvconf backend.
@@ -1066,10 +1069,52 @@ dns_handover_failed() {
    fatal "Host DNS through systemd-resolved does not work: $1.
    To put host DNS back as it was before this installer:
      sudo systemctl disable --now systemd-resolved
+      sudo rm -f $NM_DNS_DROPIN
      sudo cp -a --remove-destination $RESOLV_CONF_SAVED $RESOLV_CONF
      sudo systemctl restart NetworkManager"
 }

+# Prints the rc-manager NetworkManager would use if it is not unmanaged, which
+# means a file read after the Spinifex drop-in overrides it.
+nm_rc_manager_override() {
+    local config
+    config=$($SUDO NetworkManager --print-config) || return 1
+    printf '%s\n' "$config" | awk -F= '
+        /^\[/ { main = ($0 == "[main]"); next }
+        main && $1 == "rc-manager" { rc = $2 }
+        END { if (rc != "unmanaged") print "rc-manager=" (rc == "" ? "symlink" : rc) }'
+}
+
+# Writes the NetworkManager drop-in and checks nothing overrides it. Returns 1
+# when it was already in place.
+install_nm_dns_dropin() {
+    local want override changed=1
+    want=$(printf '[main]\ndns=systemd-resolved\nrc-manager=unmanaged')
+    if [ "$(cat "$NM_DNS_DROPIN" 2>/dev/null)" != "$want" ]; then
+        $SUDO mkdir -p "$(dirname "$NM_DNS_DROPIN")"
+        printf '%s\n' "$want" | $SUDO tee "$NM_DNS_DROPIN" > /dev/null || dns_handover_failed "cannot write $NM_DNS_DROPIN"
+        info "  $NM_DNS_DROPIN: NetworkManager leaves $RESOLV_CONF alone"
+        changed=0
+    fi
+    override=$(nm_rc_manager_override) || dns_handover_failed "cannot read NetworkManager's configuration with 'NetworkManager --print-config'"
+    if [ -n "$override" ]; then
+        dns_handover_failed "NetworkManager reads $override from a file after $NM_DNS_DROPIN, so it would replace the $RESOLV_CONF link. Remove that setting ('NetworkManager --print-config' shows the files it reads) and re-run this installer"
+    fi
+    return "$changed"
+}
+
+# Links the stub and checks the link is still there a second later, relinking a
+# bounded number of times in case a write NetworkManager had queued lands first.
+link_resolved_stub() {
+    local i
+    for i in 1 2 3 4 5; do
+        $SUDO ln -sfn "$RESOLVED_STUB" "$RESOLV_CONF"
+        sleep 1
+        [ "$(readlink "$RESOLV_CONF")" = "$RESOLVED_STUB" ] && return 0
+    done
+    dns_handover_failed "$RESOLV_CONF is not a link to $RESOLVED_STUB 5 seconds after linking it, so something keeps rewriting it"
+}
+
 # Checks resolved has upstream servers and still resolves the package
 # repository host, the one name this host is known to have resolved.
 check_resolved_dns() {
@@ -1103,8 +1148,13 @@ hand_dns_to_resolved() {
    [ "${ISO_BUILD:-0}" = "1" ] && return 0

    if systemctl is-active --quiet systemd-resolved.service; then
-        # A handover made by an earlier run is checked again on every run.
+        # A handover made by an earlier run is checked again, and repaired, on
+        # every run: installs before the drop-in could lose the link.
        if [ -e "$RESOLV_CONF_SAVED" ] || [ -L "$RESOLV_CONF_SAVED" ]; then
+            if install_nm_dns_dropin && systemctl is-active --quiet NetworkManager.service; then
+                $SUDO systemctl reload NetworkManager.service || dns_handover_failed "NetworkManager did not reload"
+            fi
+            [ "$(readlink "$RESOLV_CONF")" = "$RESOLVED_STUB" ] || link_resolved_stub
            check_resolved_dns
        fi
        return 0
@@ -1131,10 +1181,15 @@ hand_dns_to_resolved() {
        $SUDO cp -a "$RESOLV_CONF" "$RESOLV_CONF_SAVED" || fatal "Cannot save $RESOLV_CONF to $RESOLV_CONF_SAVED"
        info "  saved $RESOLV_CONF to $RESOLV_CONF_SAVED"
    fi
+    install_nm_dns_dropin || true
    $SUDO systemctl enable --now systemd-resolved.service > /dev/null 2>&1 \
        || dns_handover_failed "systemd-resolved did not start"
    $SUDO systemctl restart NetworkManager.service || dns_handover_failed "NetworkManager did not restart"
-    $SUDO ln -sfn "$RESOLVED_STUB" "$RESOLV_CONF"
+    # Best effort: a link with no DHCP server can hold startup past the timeout.
+    if command -v nm-online > /dev/null; then
+        nm-online -s -q --timeout=15 || true
+    fi
+    link_resolved_stub
    check_resolved_dns
    info "  $RESOLV_CONF -> $RESOLVED_STUB"
 }
@@ -1530,6 +1585,15 @@ download_spinifex() {
    EXTRACT_DIR="$SPINIFEX_TMPDIR"
 }

+# install.mulgadc.com serves the latest release's setup.sh, which cannot know what a
+# newer --channel or --version tarball needs installed, so it runs the tarball's own.
+hand_off_to_packaged_setup() {
+    [ -z "${INSTALL_SPINIFEX_TARBALL:-}" ] && [ -f "$EXTRACT_DIR/setup.sh" ] || return 0
+    info "Running the setup.sh shipped in the downloaded release"
+    export INSTALL_SPINIFEX_TARBALL="$TARBALL" SPINIFEX_HANDOFF_TMPDIR="$SPINIFEX_TMPDIR"
+    exec bash "$EXTRACT_DIR/setup.sh" "$@"
+}
+
 # OL's sudo secure_path leaves out /usr/local/bin, so `sudo spx` would find
 # nothing there. A link in /usr/bin changes sudo for nothing else on the host.
 link_spx_for_sudo() {
@@ -1624,13 +1688,12 @@ install_files() {
        $SUDO cp "$EXTRACT_DIR/microvm/"* /usr/share/spinifex/microvm/
        info "  /usr/share/spinifex/microvm/*"
    fi
-    install_sqs_store_image
    install_sqs_worker_image
 }

-# The SQS images ship beside the load balancer's, not inside it: an older
-# installer copies microvm/* without -r and would stop at a subdirectory.
-install_sqs_store_image() { install_sqs_microvm_image sqs-store "SQS store"; }
+# The SQS image ships beside the load balancer's, not inside it: an older
+# installer copies microvm/* without -r and would stop at a subdirectory. A
+# store image left by an earlier install is not touched: nothing boots it.
 install_sqs_worker_image() { install_sqs_microvm_image sqs-worker "SQS worker"; }

 # install_sqs_microvm_image <name> <label>
@@ -2178,6 +2241,7 @@ EOF

 # --- Main ---
 main() {
+    local args=("$@")
    while [ $# -gt 0 ]; do
        case "$1" in
            --firewall=*) INSTALL_SPINIFEX_FIREWALL="${1#*=}" ;;
@@ -2207,18 +2271,20 @@ main() {
    detect_os
    detect_arch

-    # Orchestration (stop running services, fix stale /run/spinifex/nbd perms)
-    # only applies in full-install mode on a live host.
-    if [ "${ISO_BUILD:-0}" != "1" ] && [ -z "${SETUP_STAGES:-}" ]; then
-        handle_upgrade
-    fi
-
    # Stages that need EXTRACT_DIR: files, directories, systemd, logrotate, udev.
-    # Only download/extract when at least one such stage is enabled.
+    # Only download/extract when at least one such stage is enabled, and before
+    # anything on the host changes, so the hand-off finds it untouched.
    if stage_enabled files || stage_enabled directories \
        || stage_enabled systemd || stage_enabled logrotate \
        || stage_enabled udev; then
        download_spinifex
+        hand_off_to_packaged_setup "${args[@]}"
+    fi
+
+    # Orchestration (stop running services, fix stale /run/spinifex/nbd perms)
+    # only applies in full-install mode on a live host.
+    if [ "${ISO_BUILD:-0}" != "1" ] && [ -z "${SETUP_STAGES:-}" ]; then
+        handle_upgrade
    fi

    stage_enabled deps       && install_deps
@@ -2248,6 +2314,7 @@ main() {
    fi

    [ -n "${SPINIFEX_TMPDIR:-}" ] && rm -rf "$SPINIFEX_TMPDIR"
+    [ -n "${SPINIFEX_HANDOFF_TMPDIR:-}" ] && rm -rf "$SPINIFEX_HANDOFF_TMPDIR"

    # Post-install orchestration (service restart, summary, newgrp) only in
    # full-install mode on a live host.
@@ -0,0 +1,88 @@
+#!/bin/sh
+# setup_handoff_test.sh — tests that setup.sh runs the setup.sh shipped in the
+# tarball it downloaded, before it changes anything on the host.
+#
+# setup.sh is sourced with INSTALL_SPINIFEX_LIB_ONLY=1 and main runs with the
+# host-touching functions stubbed; the packaged setup.sh is a fixture that
+# prints what it was given.
+set -eu
+
+SCRIPT_DIR=$(CDPATH='' cd -- "$(dirname -- "$0")" && pwd)
+SETUP="$SCRIPT_DIR/setup.sh"
+TMP=$(mktemp -d)
+trap 'rm -rf "$TMP"' EXIT
+
+fails=0
+pass() { echo "  ok   - $1"; }
+fail() { echo "  FAIL - $1"; fails=$((fails + 1)); }
+
+echo "-- setup.sh hand-off to the packaged setup.sh"
+
+mkdir -p "$TMP/extract"
+cat > "$TMP/packaged-setup.sh" <<'EOF'
+echo "PACKAGED args=[$*] tarball=${INSTALL_SPINIFEX_TARBALL:-} handoff=${SPINIFEX_HANDOFF_TMPDIR:-}"
+EOF
+
+# run <packaged: yes|no> [main args...] — runs main with every step that would
+# touch the host stubbed to print its name, and sets OUT.
+run() {
+    packaged="$1"
+    shift
+    rm -f "$TMP/extract/setup.sh"
+    [ "$packaged" = yes ] && cp "$TMP/packaged-setup.sh" "$TMP/extract/setup.sh"
+    OUT=$(INSTALL_SPINIFEX_LIB_ONLY=1 TMPROOT="$TMP" bash -c '
+        . "$1"; shift
+        info() { :; }
+        setup_sudo() { :; }
+        resolve_firewall_mode() { :; }
+        detect_os() { :; }
+        detect_arch() { :; }
+        download_spinifex() {
+            SPINIFEX_TMPDIR="$TMPROOT/extract"
+            TARBALL="$SPINIFEX_TMPDIR/spinifex.tar.gz"
+            EXTRACT_DIR="$SPINIFEX_TMPDIR"
+        }
+        handle_upgrade() { echo HANDLE_UPGRADE; exit 0; }
+        main "$@"
+    ' _ "$SETUP" "$@" 2>&1) || true
+    OUT=$(printf '%s\n' "$OUT" | sed '/^$/d')
+}
+
+run yes --channel dev --firewall off
+want="PACKAGED args=[--channel dev --firewall off] tarball=$TMP/extract/spinifex.tar.gz handoff=$TMP/extract"
+if [ "$OUT" = "$want" ]; then
+    pass "a downloaded tarball's setup.sh runs with the same arguments, before services are stopped"
+else
+    fail "downloaded tarball: got [$OUT], want [$want]"
+fi
+
+run no --channel dev
+if [ "$OUT" = HANDLE_UPGRADE ]; then
+    pass "a tarball without setup.sh is installed by the running script"
+else
+    fail "no packaged setup.sh: got [$OUT]"
+fi
+
+run_local() {
+    rm -f "$TMP/extract/setup.sh"
+    cp "$TMP/packaged-setup.sh" "$TMP/extract/setup.sh"
+    OUT=$(INSTALL_SPINIFEX_LIB_ONLY=1 INSTALL_SPINIFEX_TARBALL=/somewhere/spinifex.tar.gz bash -c '
+        . "$1"
+        info() { :; }
+        EXTRACT_DIR="$2"; TARBALL="$2/spinifex.tar.gz"; SPINIFEX_TMPDIR="$2"
+        hand_off_to_packaged_setup --channel dev
+        echo STAYED
+    ' _ "$SETUP" "$TMP/extract" 2>&1) || true
+}
+run_local
+if [ "$OUT" = STAYED ]; then
+    pass "a local tarball, including the one a hand-off passes on, is not handed off again"
+else
+    fail "local tarball: got [$OUT]"
+fi
+
+if [ "$fails" -gt 0 ]; then
+    echo "setup_handoff_test.sh: $fails failure(s)"
+    exit 1
+fi
+echo "setup_handoff_test.sh: ok"
@@ -20,9 +20,10 @@ fail() { echo "  FAIL - $1"; fails=$((fails + 1)); }
 echo "-- setup.sh host DNS, chrony, needrestart and nbdkit"

 mkdir -p "$TMP/sys"
-for tool in bash awk sed grep readlink basename dirname cp ln mkdir cat head tail wc tee rm chmod install; do
+for tool in bash awk sed grep readlink basename dirname cp ln mkdir cat head tail wc tee rm chmod install mv; do
    ln -s "$(command -v "$tool")" "$TMP/sys/$tool"
 done
+ln -s "$(command -v sleep)" "$TMP/sys/realsleep"

 # stub <name> <body> — a stub command; $S is the fixture's state directory.
 stub() {
@@ -30,10 +31,17 @@ stub() {
    chmod +x "$TMP/bin/$1"
 }

+# nm_settle — waits for the fixture's NetworkManager to finish writing.
+nm_settle() {
+    n=0
+    while [ -e "$TMP/state/nm-pending" ] && [ $n -lt 500 ]; do "$TMP/sys/realsleep" 0.01; n=$((n + 1)); done
+}
+
 # host — a fresh fixture: resolved installed and inactive, resolvconf the
 # resolvectl link, NetworkManager running with no DNS settings, a dnf host
 # whose first enabled repository is yum.oracle.com.
 host() {
+    nm_settle
    rm -rf "${TMP:?}/state" "${TMP:?}/root" "${TMP:?}/bin"
    mkdir -p "$TMP/state/active" "$TMP/root/etc" "$TMP/root/usr/lib/systemd/system" "$TMP/bin"
    : > "$TMP/root/usr/lib/systemd/system/systemd-resolved.service"
@@ -42,11 +50,28 @@ host() {
    printf '[main]\n# dns=default\nconfigure-and-quit=no\n' > "$TMP/state/nm-config"
    printf 'Global:\nLink 2 (br-wan): 192.168.1.1\n' > "$TMP/state/resolvectl-dns"
    printf '[ol10_UEKR8]\nbaseurl = https://mirror.example/uek/\nenabled = 0\n[ol10_baseos_latest]\nbaseurl = https://yum.oracle.com/repo/OracleLinux/OL10/baseos/latest/x86_64/\nenabled = 1\n' > "$TMP/state/dnf-dump"
+    # A (re)started NetworkManager whose rc-manager is not unmanaged writes
+    # resolv.conf as a plain file on its first DNS update, which can land just
+    # after the installer links the stub.
    stub systemctl '[ "$1" = is-active ] && { [ -e "$S/active/$3" ]; exit; }
 echo "$*" >> "$S/calls"
 [ "$1 $2" = "enable --now" ] && : > "$S/active/$3"
+case "$1 $2" in
+    "restart NetworkManager.service" | "reload NetworkManager.service")
+        NetworkManager --print-config | grep -qx rc-manager=unmanaged && exit 0
+        : > "$S/nm-pending"
+        (
+            n=0
+            while [ ! -L "$S/../root/etc/resolv.conf" ] && [ $n -lt 300 ]; do realsleep 0.01; n=$((n + 1)); done
+            printf "nameserver 192.168.1.1\n" > "$S/nm-resolv.conf"
+            mv -f "$S/nm-resolv.conf" "$S/../root/etc/resolv.conf"
+            rm -f "$S/nm-pending"
+        ) > /dev/null 2>&1 &
+        ;;
+esac
 exit 0'
-    stub NetworkManager 'cat "$S/nm-config"'
+    stub NetworkManager 'cat "$S/nm-config"; cat "$S/../root/nm/90-spinifex-dns.conf" "$S/nm-config-late" 2>/dev/null; :'
+    stub nm-online ':'
    stub resolvectl '[ "$1" = dns ] && cat "$S/resolvectl-dns"'
    stub resolvectl-real ':'
    ln -s resolvectl "$TMP/bin/resolvconf"
@@ -60,10 +85,17 @@ run() {
    RC=0
    OUT=$(PATH="$TMP/bin:$TMP/sys" DISTRO_ROOT="$TMP/root" \
        SPINIFEX_RESOLV_CONF="$TMP/root/etc/resolv.conf" SPINIFEX_RESOLV_CONF_SAVED="$TMP/root/saved/resolv.conf.orig" \
+        SPINIFEX_NM_DNS_DROPIN="$TMP/root/nm/90-spinifex-dns.conf" \
        INSTALL_SPINIFEX_LIB_ONLY=1 bash -c '. "$1"; shift; for f in "$@"; do $f; done' _ "$SETUP" "$@" 2>&1) || RC=$?
 }

 calls() { cat "$TMP/state/calls" 2>/dev/null || true; }
+# linked — whether resolv.conf is still the stub link once NetworkManager settles.
+linked() {
+    nm_settle
+    [ "$(readlink "$TMP/root/etc/resolv.conf")" = /run/systemd/resolve/stub-resolv.conf ]
+}
+dropin() { cat "$TMP/root/nm/90-spinifex-dns.conf" 2>/dev/null || echo "<none>"; }
 saved() { cat "$TMP/root/saved/resolv.conf.orig" 2>/dev/null || echo "<none>"; }
 expect_untouched() {
    if [ "$RC" = 0 ] && [ -z "$(calls)" ] && [ ! -L "$TMP/root/etc/resolv.conf" ] && [ "$(saved)" = "<none>" ]; then
@@ -86,7 +118,7 @@ expect_fatal() { # expect_fatal <desc> <text>...
 # --------------------------------------------------------------------------
 host
 run hand_dns_to_resolved
-if [ "$RC" = 0 ] && [ "$(readlink "$TMP/root/etc/resolv.conf")" = /run/systemd/resolve/stub-resolv.conf ] \
+if [ "$RC" = 0 ] && linked \
    && [ "$(saved)" = "$(printf 'nameserver 192.168.1.1\nsearch example.lan')" ] \
    && [ "$(calls)" = "$(printf 'enable --now systemd-resolved.service\nrestart NetworkManager.service')" ] \
    && [ "$(cat "$TMP/state/looked-up")" = yum.oracle.com ]; then
@@ -98,12 +130,44 @@ fi
 : > "$TMP/state/calls"
 rm -f "$TMP/state/looked-up"
 run hand_dns_to_resolved
-if [ "$RC" = 0 ] && [ -z "$(calls)" ] && [ "$(cat "$TMP/state/looked-up")" = yum.oracle.com ]; then
+if [ "$RC" = 0 ] && [ -z "$(calls)" ] && linked && [ "$(cat "$TMP/state/looked-up")" = yum.oracle.com ]; then
    pass "a rerun after a handover checks DNS again and changes nothing"
 else
    fail "rerun after handover: rc=$RC calls='$(calls)': $OUT"
 fi

+if [ "$(dropin)" = "$(printf '[main]\ndns=systemd-resolved\nrc-manager=unmanaged')" ]; then
+    pass "the handover tells NetworkManager to send DNS to resolved and leave resolv.conf alone"
+else
+    fail "NetworkManager drop-in: '$(dropin)'"
+fi
+
+# An install made before the drop-in, whose NetworkManager replaced the link.
+host
+mkdir -p "$TMP/root/saved"
+echo "the original" > "$TMP/root/saved/resolv.conf.orig"
+: > "$TMP/state/active/systemd-resolved.service"
+run hand_dns_to_resolved
+if [ "$RC" = 0 ] && linked && [ "$(calls)" = "reload NetworkManager.service" ] \
+    && [ "$(dropin)" = "$(printf '[main]\ndns=systemd-resolved\nrc-manager=unmanaged')" ]; then
+    pass "a rerun after a handover whose link NetworkManager replaced adds the drop-in, reloads NM and relinks the stub"
+else
+    fail "rerun repair: rc=$RC calls='$(calls)' link='$(readlink "$TMP/root/etc/resolv.conf")' dropin='$(dropin)': $OUT"
+fi
+
+host
+printf '[main]\nrc-manager=symlink\n' > "$TMP/state/nm-config-late"
+run hand_dns_to_resolved
+expect_fatal "a NetworkManager file read after the drop-in that sets rc-manager stops the install, naming it" \
+    "rc-manager=symlink" "rm -f $TMP/root/nm/90-spinifex-dns.conf"
+case "$(calls)" in *"restart NetworkManager"*) fail "rc-manager override: restarted NetworkManager first: $(calls)" ;; esac
+
+host
+stub ln ':'
+run hand_dns_to_resolved
+expect_fatal "a resolv.conf that will not stay linked to the stub stops the install" \
+    "is not a link to /run/systemd/resolve/stub-resolv.conf"
+
 host
 mkdir -p "$TMP/root/saved"
 echo "the original" > "$TMP/root/saved/resolv.conf.orig"
@@ -155,7 +219,8 @@ host
 printf 'Global:\nLink 2 (br-wan):\n' > "$TMP/state/resolvectl-dns"
 run hand_dns_to_resolved
 expect_fatal "no upstream servers stops the install, naming the check and printing the restore" \
-    "'resolvectl dns' lists no DNS servers" "systemctl disable --now systemd-resolved" "cp -a --remove-destination $TMP/root/saved/resolv.conf.orig"
+    "'resolvectl dns' lists no DNS servers" "systemctl disable --now systemd-resolved" "cp -a --remove-destination $TMP/root/saved/resolv.conf.orig" \
+    "rm -f $TMP/root/nm/90-spinifex-dns.conf"

 host
 : > "$TMP/state/getent-fail"
@@ -1,6 +1,6 @@
 #!/bin/sh
-# setup_sqs_store_image_test.sh — tests for shipping the SQS store and worker
-# microVM images: staging them into a release tree and installing with setup.sh.
+# setup_sqs_worker_image_test.sh — tests for shipping the SQS microVM
+# image: staging it into a release tree and installing with setup.sh.
 #
 # setup.sh is sourced with INSTALL_SPINIFEX_LIB_ONLY=1 so main never runs, and
 # installs into a temporary SPINIFEX_MICROVM_DIR with SUDO cleared.
@@ -16,7 +16,7 @@ fails=0
 pass() { echo "  ok   - $1"; }
 fail() { echo "  FAIL - $1"; fails=$((fails + 1)); }

-echo "-- SQS store and worker microVM image release and install"
+echo "-- SQS microVM image release and install"

 # image <dir> <tag>: a fake built image whose bytes name its tag.
 image() {
@@ -35,11 +35,11 @@ release() {
 }

 # install_image <extract-dir> <microvm-dir> [path-prefix]: runs
-# install_sqs_store_image and sets RC and OUT.
+# install_sqs_worker_image and sets RC and OUT.
 install_image() {
    RC=0
    OUT=$(PATH="${3:-}$PATH" EXTRACT_DIR="$1" SPINIFEX_MICROVM_DIR="$2" INSTALL_SPINIFEX_LIB_ONLY=1 \
-        bash -c '. "$1"; SUDO=""; install_sqs_store_image' _ "$SETUP" 2>&1) || RC=$?
+        bash -c '. "$1"; SUDO=""; install_sqs_worker_image' _ "$SETUP" 2>&1) || RC=$?
 }

 same() { cmp -s "$1" "$2"; }
@@ -48,13 +48,13 @@ same() { cmp -s "$1" "$2"; }
 image "$TMP/built" v1
 release "$TMP/rel"
 sh "$STAGE" "$TMP/built" "$TMP/rel" > "$TMP/stage.out"
-if same "$TMP/built/vmlinuz" "$TMP/rel/microvm-sqs-store/vmlinuz" \
-    && same "$TMP/built/initramfs.cpio.gz" "$TMP/rel/microvm-sqs-store/initramfs.cpio.gz"; then
+if same "$TMP/built/vmlinuz" "$TMP/rel/microvm-sqs-worker/vmlinuz" \
+    && same "$TMP/built/initramfs.cpio.gz" "$TMP/rel/microvm-sqs-worker/initramfs.cpio.gz"; then
    pass "stages the built image beside the load balancer's"
 else
    fail "staged image differs from the build"
 fi
-if [ ! -e "$TMP/rel/microvm/sqs-store" ] && [ "$(ls "$TMP/rel/microvm" | wc -l)" -eq 3 ]; then
+if [ ! -e "$TMP/rel/microvm/sqs-worker" ] && [ "$(ls "$TMP/rel/microvm" | wc -l)" -eq 3 ]; then
    pass "leaves the load balancer's microvm directory as it was"
 else
    fail "load balancer directory changed: $(ls "$TMP/rel/microvm")"
@@ -63,21 +63,21 @@ fi
 release "$TMP/rel-none"
 mkdir -p "$TMP/half"; printf 'k\n' > "$TMP/half/vmlinuz"
 OUT=$(sh "$STAGE" "$TMP/half" "$TMP/rel-none")
-if [ ! -e "$TMP/rel-none/microvm-sqs-store" ] && echo "$OUT" | grep -q 'WARNING: SQS store microVM image missing'; then
+if [ ! -e "$TMP/rel-none/microvm-sqs-worker" ] && echo "$OUT" | grep -q 'WARNING: SQS worker microVM image missing'; then
    pass "stages nothing from an incomplete build, and says so"
 else
    fail "incomplete build staged or unannounced: $OUT"
 fi

-# --- an older installer meets a release that carries the store image ---
+# --- an older installer meets a release that carries the SQS image ---
 # It copies microvm/* without -r; the sibling directory must not reach it.
 mkdir -p "$TMP/old-dest"
-if cp "$TMP/rel/microvm/"* "$TMP/old-dest/" 2> "$TMP/old.err" && [ ! -e "$TMP/old-dest/sqs-store" ]; then
-    pass "an older installer's copy still succeeds and ignores the store image"
+if cp "$TMP/rel/microvm/"* "$TMP/old-dest/" 2> "$TMP/old.err" && [ ! -e "$TMP/old-dest/sqs-worker" ]; then
+    pass "an older installer's copy still succeeds and ignores the SQS image"
 else
    fail "older installer copy failed: $(cat "$TMP/old.err")"
 fi
-mkdir -p "$TMP/nested/microvm/sqs-store" "$TMP/nested-dest"
+mkdir -p "$TMP/nested/microvm/sqs-worker" "$TMP/nested-dest"
 printf 'x\n' > "$TMP/nested/microvm/vmlinuz"
 if cp "$TMP/nested/microvm/"* "$TMP/nested-dest/" 2> /dev/null; then
    fail "control: a nested directory did not break the older copy, so the test proves nothing"
@@ -87,58 +87,58 @@ fi

 # --- install ---
 install_image "$TMP/rel" "$TMP/mv"
-if [ "$RC" = 0 ] && same "$TMP/built/vmlinuz" "$TMP/mv/sqs-store/vmlinuz" \
-    && same "$TMP/built/initramfs.cpio.gz" "$TMP/mv/sqs-store/initramfs.cpio.gz"; then
+if [ "$RC" = 0 ] && same "$TMP/built/vmlinuz" "$TMP/mv/sqs-worker/vmlinuz" \
+    && same "$TMP/built/initramfs.cpio.gz" "$TMP/mv/sqs-worker/initramfs.cpio.gz"; then
    pass "installs the staged image where the daemon looks"
 else
    fail "install rc=$RC: $OUT"
 fi
-mode=$(stat -c '%a' "$TMP/mv/sqs-store/initramfs.cpio.gz" 2> /dev/null || echo none)
+mode=$(stat -c '%a' "$TMP/mv/sqs-worker/initramfs.cpio.gz" 2> /dev/null || echo none)
 if [ "$mode" = 644 ]; then pass "installs the image 0644"; else fail "initramfs mode $mode"; fi

 image "$TMP/built2" v2
 release "$TMP/rel2"
 sh "$STAGE" "$TMP/built2" "$TMP/rel2" > /dev/null
 install_image "$TMP/rel2" "$TMP/mv"
-if [ "$RC" = 0 ] && same "$TMP/built2/vmlinuz" "$TMP/mv/sqs-store/vmlinuz" \
-    && same "$TMP/built2/initramfs.cpio.gz" "$TMP/mv/sqs-store/initramfs.cpio.gz"; then
+if [ "$RC" = 0 ] && same "$TMP/built2/vmlinuz" "$TMP/mv/sqs-worker/vmlinuz" \
+    && same "$TMP/built2/initramfs.cpio.gz" "$TMP/mv/sqs-worker/initramfs.cpio.gz"; then
    pass "an upgrade replaces both files"
 else
    fail "upgrade rc=$RC: $OUT"
 fi

-# A release without the store image: nothing installed, and a clear warning.
+# A release without the SQS image: nothing installed, and a clear warning.
 install_image "$TMP/rel-none" "$TMP/mv-none"
-if [ "$RC" = 0 ] && [ ! -e "$TMP/mv-none/sqs-store" ] && echo "$OUT" | grep -q 'no SQS store microVM image; SQS store launches will be refused'; then
+if [ "$RC" = 0 ] && [ ! -e "$TMP/mv-none/sqs-worker" ] && echo "$OUT" | grep -q 'no SQS worker microVM image; SQS worker launches will be refused'; then
    pass "a release without the image installs nothing and warns"
 else
-    fail "release without image: rc=$RC exists=$([ -e "$TMP/mv-none/sqs-store" ] && echo yes || echo no) $OUT"
+    fail "release without image: rc=$RC exists=$([ -e "$TMP/mv-none/sqs-worker" ] && echo yes || echo no) $OUT"
 fi

-# installed <dir>: a store image from an earlier install, plus an unrelated file.
+# installed <dir>: an SQS image from an earlier install, plus an unrelated file.
 installed() {
-    image "$1/sqs-store" old
-    printf 'keep\n' > "$1/sqs-store/operator-note"
+    image "$1/sqs-worker" old
+    printf 'keep\n' > "$1/sqs-worker/operator-note"
 }
-gone() { [ ! -e "$1/sqs-store/vmlinuz" ] && [ ! -e "$1/sqs-store/initramfs.cpio.gz" ]; }
-kept() { [ "$(cat "$1/sqs-store/operator-note" 2> /dev/null)" = keep ]; }
+gone() { [ ! -e "$1/sqs-worker/vmlinuz" ] && [ ! -e "$1/sqs-worker/initramfs.cpio.gz" ]; }
+kept() { [ "$(cat "$1/sqs-worker/operator-note" 2> /dev/null)" = keep ]; }

 # Upgrading to a release without the image must not leave the old one usable.
 installed "$TMP/mv-up"
 install_image "$TMP/rel-none" "$TMP/mv-up"
 if [ "$RC" = 0 ] && gone "$TMP/mv-up" && kept "$TMP/mv-up" \
-    && echo "$OUT" | grep -q 'Removed the SQS store microVM image left by an earlier install' \
-    && echo "$OUT" | grep -q 'SQS store launches will be refused'; then
+    && echo "$OUT" | grep -q 'Removed the SQS worker microVM image left by an earlier install' \
+    && echo "$OUT" | grep -q 'SQS worker launches will be refused'; then
    pass "a release without the image removes the earlier one, keeps other files, and warns"
 else
    fail "release without image over an installed one: rc=$RC $OUT"
 fi

-# An incomplete store image refuses and removes the earlier pair, because the
+# An incomplete SQS image refuses and removes the earlier pair, because the
 # binaries this release installed before it would otherwise boot the old one.
 installed "$TMP/mv-bad"
 cp -r "$TMP/rel2" "$TMP/rel-bad"
-rm "$TMP/rel-bad/microvm-sqs-store/initramfs.cpio.gz"
+rm "$TMP/rel-bad/microvm-sqs-worker/initramfs.cpio.gz"
 install_image "$TMP/rel-bad" "$TMP/mv-bad"
 if [ "$RC" != 0 ] && echo "$OUT" | grep -q 'incomplete: initramfs.cpio.gz is missing' \
    && gone "$TMP/mv-bad" && kept "$TMP/mv-bad"; then
@@ -154,17 +154,17 @@ for case in missing incomplete; do
    sh "$STAGE" "$TMP/built" "$TMP/rel-$case" > /dev/null
    if [ "$case" = missing ]; then rm -r "$TMP/built-$case"; else rm "$TMP/built-$case/initramfs.cpio.gz"; fi
    OUT=$(sh "$STAGE" "$TMP/built-$case" "$TMP/rel-$case")
-    if [ ! -e "$TMP/rel-$case/microvm-sqs-store" ] && echo "$OUT" | grep -q 'tarball will not include it'; then
+    if [ ! -e "$TMP/rel-$case/microvm-sqs-worker" ] && echo "$OUT" | grep -q 'tarball will not include it'; then
        pass "restaging from a $case build ships no earlier image"
    else
-        fail "restaging from a $case build left: $(ls -R "$TMP/rel-$case/microvm-sqs-store" 2> /dev/null)"
+        fail "restaging from a $case build left: $(ls -R "$TMP/rel-$case/microvm-sqs-worker" 2> /dev/null)"
    fi
 done
-mkdir -p "$TMP/rel-other/microvm-sqs-store"
-image "$TMP/rel-other/microvm-sqs-store" old
-printf 'keep\n' > "$TMP/rel-other/microvm-sqs-store/note"
+mkdir -p "$TMP/rel-other/microvm-sqs-worker"
+image "$TMP/rel-other/microvm-sqs-worker" old
+printf 'keep\n' > "$TMP/rel-other/microvm-sqs-worker/note"
 sh "$STAGE" "$TMP/no-such-build" "$TMP/rel-other" > /dev/null
-if [ ! -e "$TMP/rel-other/microvm-sqs-store/vmlinuz" ] && [ "$(cat "$TMP/rel-other/microvm-sqs-store/note")" = keep ]; then
+if [ ! -e "$TMP/rel-other/microvm-sqs-worker/vmlinuz" ] && [ "$(cat "$TMP/rel-other/microvm-sqs-worker/note")" = keep ]; then
    pass "restaging removes only the image files"
 else
    fail "restaging touched more than the image files"
@@ -180,65 +180,25 @@ exec /usr/bin/install "$@"
 EOF
 chmod +x "$TMP/bin/install"
 install_image "$TMP/rel" "$TMP/mv" "$TMP/bin:"
-if [ "$RC" != 0 ] && same "$TMP/built/vmlinuz" "$TMP/mv/sqs-store/vmlinuz" && [ ! -e "$TMP/mv/sqs-store/initramfs.cpio.gz" ]; then
+if [ "$RC" != 0 ] && same "$TMP/built/vmlinuz" "$TMP/mv/sqs-worker/vmlinuz" && [ ! -e "$TMP/mv/sqs-worker/initramfs.cpio.gz" ]; then
    pass "an interrupted install leaves the initramfs missing, not mismatched"
 else
-    fail "interrupted install: rc=$RC initramfs=$([ -e "$TMP/mv/sqs-store/initramfs.cpio.gz" ] && echo present || echo missing)"
+    fail "interrupted install: rc=$RC initramfs=$([ -e "$TMP/mv/sqs-worker/initramfs.cpio.gz" ] && echo present || echo missing)"
 fi

-# --- SQS worker image: same staging and install path, its own directory ---
-install_worker() {
-    RC=0
-    OUT=$(EXTRACT_DIR="$1" SPINIFEX_MICROVM_DIR="$2" INSTALL_SPINIFEX_LIB_ONLY=1 \
-        bash -c '. "$1"; SUDO=""; install_sqs_worker_image' _ "$SETUP" 2>&1) || RC=$?
-}
-image "$TMP/wbuilt" w1
-release "$TMP/wrel"
-sh "$STAGE" "$TMP/built" "$TMP/wrel" > /dev/null
-sh "$STAGE" "$TMP/wbuilt" "$TMP/wrel" sqs-worker > /dev/null
-if same "$TMP/wbuilt/vmlinuz" "$TMP/wrel/microvm-sqs-worker/vmlinuz" \
-    && same "$TMP/wbuilt/initramfs.cpio.gz" "$TMP/wrel/microvm-sqs-worker/initramfs.cpio.gz" \
-    && same "$TMP/built/vmlinuz" "$TMP/wrel/microvm-sqs-store/vmlinuz"; then
-    pass "the worker image stages beside the store image"
+image "$TMP/keep/sqs-store" old-store
+install_image "$TMP/rel" "$TMP/keep"
+if [ "$RC" = 0 ] && [ "$(cat "$TMP/keep/sqs-store/vmlinuz")" = kernel-old-store ] \
+    && [ "$(cat "$TMP/keep/sqs-store/initramfs.cpio.gz")" = initramfs-old-store ]; then
+    pass "an install leaves a separate store image from an earlier install untouched"
 else
-    fail "worker staging: $(ls -R "$TMP/wrel")"
+    fail "earlier store image: rc=$RC $(ls -R "$TMP/keep")"
 fi
-install_worker "$TMP/wrel" "$TMP/wmv"
-if [ "$RC" = 0 ] && same "$TMP/wbuilt/initramfs.cpio.gz" "$TMP/wmv/sqs-worker/initramfs.cpio.gz" \
-    && [ ! -e "$TMP/wmv/sqs-store" ]; then
-    pass "the worker image installs into its own directory only"
+if OUT=$(sh "$STAGE" "$TMP/built" "$TMP/rel" sqs-store 2>&1); then
+    fail "the separate store image was staged"
 else
-    fail "worker install: rc=$RC $OUT"
-fi
-release "$TMP/wrel-none"
-OUT=$(sh "$STAGE" "$TMP/no-such-build" "$TMP/wrel-none" sqs-worker)
-if [ ! -e "$TMP/wrel-none/microvm-sqs-worker" ] && echo "$OUT" | grep -q 'SQS worker microVM image missing'; then
-    pass "a missing worker build stages nothing and says so"
-else
-    fail "missing worker build: $OUT"
-fi
-install_worker "$TMP/wrel-none" "$TMP/wmv"
-if [ "$RC" = 0 ] && [ ! -e "$TMP/wmv/sqs-worker/vmlinuz" ] && [ ! -e "$TMP/wmv/sqs-worker/initramfs.cpio.gz" ] \
-    && echo "$OUT" | grep -q 'no SQS worker microVM image; SQS worker launches will be refused'; then
-    pass "a release without the worker image removes the earlier one and warns"
-else
-    fail "release without worker image: rc=$RC $OUT"
-fi
-cp -r "$TMP/wrel" "$TMP/wrel-bad"
-rm "$TMP/wrel-bad/microvm-sqs-worker/vmlinuz"
-install_worker "$TMP/wrel" "$TMP/wmv"
-install_worker "$TMP/wrel-bad" "$TMP/wmv"
-if [ "$RC" != 0 ] && echo "$OUT" | grep -q 'SQS worker microVM image in this release is incomplete: vmlinuz is missing' \
-    && [ ! -e "$TMP/wmv/sqs-worker/initramfs.cpio.gz" ]; then
-    pass "an incomplete worker image is refused and the earlier pair is removed"
-else
-    fail "incomplete worker image: rc=$RC $OUT"
-fi
-if OUT=$(sh "$STAGE" "$TMP/wbuilt" "$TMP/wrel" lb 2>&1); then
-    fail "an unknown image name was staged"
-else
-    pass "an unknown image name is refused"
+    pass "the separate store image is no longer staged"
 fi

 [ "$fails" = 0 ] || { echo "FAIL: $fails"; exit 1; }
-echo "  setup_sqs_store_image_test ok"
+echo "  setup_sqs_worker_image_test ok"
@@ -29,6 +29,7 @@ MODELS=(
    rds/2014-10-31
    route-53/2013-04-01
    s3/2006-03-01
+    sqs/2012-11-05
    sts/2011-06-15
 )

@@ -4,8 +4,9 @@

 ```bash
 export VULTR_API_KEY=...            # the key's IP allowlist must admit this host and every node
-./deploy-vultr.sh up --channel dev                                   # one VX VM
-./deploy-vultr.sh up --channel dev --nodes 3 --plan vbm-8c-64gb-amd  # three bare-metal servers
+./deploy-vultr.sh up                                                 # one VX VM, latest release
+./deploy-vultr.sh up --nodes 3 --plan vbm-8c-64gb-amd                # three bare-metal servers
+./deploy-vultr.sh up --channel dev                                   # the newest dev build
 ./deploy-vultr.sh up --name v1 --distro spinifex-<version>.tar.gz    # a named deployment from a local tarball
 ./deploy-vultr.sh status
 ./deploy-vultr.sh credentials
--- a/Show More
+++ b/Show More