fix(ci): give the nightly's single-node service suites node SSH to their own VM

* feat(lambda): record SQS event-source mappings behind Create, Get, List and Delete

Create refuses every option this build does not implement, reads the
queue as the function's execution role and requires the queue's
visibility window to cover the function's run limit. Delete records
Deleting for the poller to finish. Get and Delete authorize against
the mapping ARN.

* feat(lambda): poll mapped SQS queues in test-link gateways

The poller receives and deletes as the function's execution role and
invokes through the internal executor. A batch is deleted only after
the function succeeds; an error, refusal or unknown outcome leaves the
messages to reappear.

* test(lambda): give the poller tests their in-package reason

* fix: derive the mapping UUID so duplicate creates collide atomically

Create no longer scans every mapping. The UUID is derived from the
account, Region, function and queue, so the keyed create refuses a
duplicate pair in one atomic write with no second record to orphan.

* fix: mark the poller in-process SQS request as received over TLS

The request never leaves the TLS-terminating gateway, so an execution
role that requires aws:SecureTransport must still authorize it.

* fix: renew the poller session per event and back off on failed batches

Each event now runs on a session that outlives its invocation and its
delete, so a later split event is never acknowledged on expired
credentials. A failed, refused or unknown invocation makes the worker
back off before the next receive.

* feat: page ListEventSourceMappings with Marker and NextMarker

* test: drive the poller through InvokeEventSource on invalid and busy outcomes

* fix: skip locally held management addresses when allocating

The shared record can lose an entry an instance on this node still holds. Allocate now treats every address in the local cache as taken, refuses cached or recorded hits that another local instance holds, and resets its results on each CAS attempt.

* fix: refuse ListEventSourceMappings until listing can be bounded

The mapping store can only be listed whole, so a list request would
wait on the whole inventory. Refuse it, and skip repeated keys when the
gateway lists mappings at startup.

.github/workflows/e2e-nightly.yml

@@ -578,6 +578,8 @@ jobs:
              echo \"::group::suite ${SUITE}\"; \
              date -u +%Y-%m-%dT%H:%M:%SZ > '${ARTIFACT_DIR}/test-${SUITE}.start'; \
              GITHUB_ACTIONS=true FORCE_COLOR=1 SPINIFEX_E2E=1 SPINIFEX_MODE=single \
+               SPINIFEX_NODES='${{ steps.vm.outputs.wan_ip }}' SPINIFEX_SSH_USER=tf-user \
+               SPINIFEX_SSH_KEY=\$HOME/.ssh/tf-user-ap-southeast-2 \
                ARTIFACT_DIR='${ARTIFACT_DIR}' \
                ./${SUITE}.test -test.v -test.timeout=${TMO} \
                2>&1 | tee '${ARTIFACT_DIR}/test-${SUITE}.log'; \

architecture/layering/allowlist_test.go

@@ -17,7 +17,6 @@ var allowlist = []allowance{
 	{From: "operator/accountlifecycle", To: "domains/ec2/awsapi/instance", Debt: "accountteardown (dependency)", Reason: "EC2 reapers call the awsapi action adapters; needs domain-owned cleanup capabilities"},
 	{From: "operator/accountlifecycle", To: "domains/ec2/awsapi/routetable", Debt: "accountteardown (dependency)", Reason: "EC2 reapers call the awsapi action adapters; needs domain-owned cleanup capabilities"},
 	{From: "operator/accountlifecycle", To: "domains/ec2/awsapi/snapshot", Debt: "accountteardown (dependency)", Reason: "EC2 reapers call the awsapi action adapters; needs domain-owned cleanup capabilities"},
-	{From: "operator/accountlifecycle", To: "domains/ec2/awsapi/volume", Debt: "accountteardown (dependency)", Reason: "EC2 reapers call the awsapi action adapters; needs domain-owned cleanup capabilities"},
 	{From: "operator/accountlifecycle", To: "domains/ec2/awsapi/vpc", Debt: "accountteardown (dependency)", Reason: "EC2 reapers call the awsapi action adapters; needs domain-owned cleanup capabilities"},
 
 	{From: "agents/ecs/agent", To: "domains/ecs", Debt: "agents/ecs/agent", Reason: "assignment/state wire types from the ECS implementation; needs a guest/controller contract"},

build/microvm/init.sh

@@ -32,6 +32,8 @@ load_mod qemu_fw_cfg || load_mod fw_cfg_sysfs || true
 load_mod failover || true
 load_mod net_failover || true
 load_mod virtio_net || true
+# lb-agent probes each data NIC's gateway on an AF_PACKET socket.
+load_mod af_packet || true
 
 NETCFG=/sys/firmware/qemu_fw_cfg/by_name/opt/spinifex/netcfg/raw
 i=0

build/microvm/lambda-runtime/init.sh

@@ -16,8 +16,8 @@ load_mod() {
     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
+# lambda-agent reads its identity from fw_cfg, or the configuration disk, and
+# seals it 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
@@ -26,8 +26,30 @@ load_mod virtio_net || echo "[init] virtio_net did not load" >/dev/console
 
 ip link set lo up
 
+# Without fw_cfg data the executor attaches the same entries as a tar archive
+# on the second disk; lambda-agent reads them here and erases the disk before
+# the function starts. qemu_fw_cfg loads without a device, so test the entry.
+CFG_ROOT=/spinifex-cfg
+FWCFG_ROOT=/sys/firmware/qemu_fw_cfg
+if [ ! -f "$FWCFG_ROOT/by_name/opt/spinifex/netcfg/raw" ]; then
+    load_mod virtio_blk || echo "[init] virtio_blk did not load" >/dev/console
+    i=0
+    while [ "$i" -lt 25 ] && [ ! -b /dev/vdb ]; do
+        sleep 0.04
+        i=$((i + 1))
+    done
+    if [ -b /dev/vdb ]; then
+        mkdir -m 0700 "$CFG_ROOT"
+        if ! mount -t tmpfs -o mode=0700,size=1m tmpfs "$CFG_ROOT" ||
+            ! tar -xf /dev/vdb -C "$CFG_ROOT"; then
+            echo "[init] configuration disk not readable" >/dev/console
+        fi
+        FWCFG_ROOT=$CFG_ROOT
+    fi
+fi
+
 # The guest's addresses come from the same netcfg entry as other system VMs.
-NETCFG=/sys/firmware/qemu_fw_cfg/by_name/opt/spinifex/netcfg/raw
+NETCFG=$FWCFG_ROOT/by_name/opt/spinifex/netcfg/raw
 i=0
 while [ "$i" -lt 25 ] && [ ! -f "$NETCFG" ]; do
     sleep 0.04

cmd/lambda-agent/connect.go

@@ -5,6 +5,7 @@ import (
 	"crypto/tls"
 	"errors"
 	"fmt"
+	"os"
 
 	lambdav1 "github.com/mulgadc/spinifex/contracts/lambda/v1"
 	"github.com/mulgadc/spinifex/spinifex/agents/lambda/hostlink"
@@ -25,9 +26,22 @@ type startup struct {
 }
 
 func defaultStartup() startup {
+	if _, err := os.Stat(hostlink.ConfigDiskRoot); err == nil {
+		return configDiskStartup(hostlink.ConfigDiskRoot, hostlink.ConfigDiskDevice)
+	}
 	return startup{fwCfgDir: hostlink.FwCfgDir, fwCfgRoot: hostlink.FwCfgRoot, unload: hostlink.UnloadFwCfg}
 }
 
+// configDiskStartup reads the entries init extracted from the configuration
+// disk, and seals by erasing the disk and removing the extracted tree.
+func configDiskStartup(root, device string) startup {
+	return startup{
+		fwCfgDir:  root + "/by_name/" + lambdav1.FwCfgPrefix,
+		fwCfgRoot: root,
+		unload:    func() error { return hostlink.SealConfigDisk(device, root) },
+	}
+}
+
 // connect loads the identity, seals fw_cfg whether or not the load succeeded,
 // then dials the host. Under TLS 1.3 a host refusal is seen only on first read.
 func (s startup) connect(ctx context.Context) (*tls.Conn, lambdav1.Scope, int, error) {

cmd/lambda-agent/connect_configdisk_test.go

@@ -0,0 +1,43 @@
+//test:in-package — package main cannot be imported by an external test package.
+
+package main
+
+import (
+	"bytes"
+	"os"
+	"path/filepath"
+	"testing"
+
+	"github.com/stretchr/testify/require"
+)
+
+// An identity that does not load still seals the configuration disk: the
+// guest key on it is erased before the agent refuses to start.
+func TestConfigDiskStartup_RefusedLoadErasesDisk(t *testing.T) {
+	dir := t.TempDir()
+	root := filepath.Join(dir, "cfg")
+	require.NoError(t, os.MkdirAll(filepath.Join(root, "by_name"), 0o700))
+	device := filepath.Join(dir, "vdb")
+	require.NoError(t, os.WriteFile(device, bytes.Repeat([]byte("guest-key"), 512), 0o600))
+
+	_, _, code, err := configDiskStartup(root, device).connect(t.Context())
+	require.Error(t, err)
+	require.Equal(t, exitRefused, code)
+	got, rerr := os.ReadFile(device)
+	require.NoError(t, rerr)
+	require.Len(t, got, 9*512)
+	require.Equal(t, make([]byte, len(got)), got, "configuration disk erased")
+}
+
+// A disk that cannot be erased fails closed: the agent refuses, and the
+// extracted tree is kept so the seal check sees it.
+func TestConfigDiskStartup_UnerasableDiskRefuses(t *testing.T) {
+	dir := t.TempDir()
+	root := filepath.Join(dir, "cfg")
+	require.NoError(t, os.MkdirAll(filepath.Join(root, "by_name"), 0o700))
+
+	_, _, code, err := configDiskStartup(root, filepath.Join(dir, "missing")).connect(t.Context())
+	require.ErrorContains(t, err, "erase configuration disk")
+	require.Equal(t, exitRefused, code)
+	require.DirExists(t, root)
+}

contracts/ec2/v1/commands.go

@@ -12,6 +12,11 @@ const (
 	// this contract family. It is the NATS permission/subscription pattern for
 	// the per-instance command route, rather than the broader descendant match.
 	InstanceCommandSubjectWildcard = InstanceCommandSubjectPrefix + "*"
+
+	// HandledHeader is set to "1" on every reply an owner sends to a
+	// modify-instance-metadata-options or set-disable-api-termination command.
+	// An owner that predates those commands cannot set it.
+	HandledHeader = "Spinifex-Ec2-Cmd-Handled"
 )
 
 // InstanceCommandSubject returns the targeted command subject for instanceID.
@@ -23,7 +28,9 @@ func InstanceCommandSubject(instanceID string) string {
 }
 
 // EC2InstanceCommand is the NATS wire format for EC2 instance commands
-// (stop, terminate, start, attach/detach-volume, attach/detach-eni).
+// (stop, terminate, start, reboot, attach/detach-volume, attach/detach-eni,
+// IAM profile, spot lineage, tags, monitoring, metadata options and
+// termination protection).
 type EC2InstanceCommand struct {
 	ID                        string                     `json:"id"`
 	Attributes                EC2CommandAttributes       `json:"attributes"`
@@ -35,6 +42,9 @@ type EC2InstanceCommand struct {
 	SpotLineageData           *SpotLineageData           `json:"spot_lineage_data,omitempty"`
 	InstanceTagsData          *InstanceTagsData          `json:"instance_tags_data,omitempty"`
 	InstanceMonitoringData    *InstanceMonitoringData    `json:"instance_monitoring_data,omitempty"`
+
+	InstanceMetadataOptionsData *InstanceMetadataOptionsData `json:"instance_metadata_options_data,omitempty"`
+	DisableApiTerminationData   *DisableApiTerminationData   `json:"disable_api_termination_data,omitempty"`
 }
 
 // EC2CommandAttributes indicates which action the daemon should perform.
@@ -52,6 +62,9 @@ type EC2CommandAttributes struct {
 	SetInstanceTags             bool `json:"set_instance_tags,omitempty"`
 	RemoveInstanceTags          bool `json:"remove_instance_tags,omitempty"`
 	SetInstanceMonitoring       bool `json:"set_instance_monitoring,omitempty"`
+
+	ModifyInstanceMetadataOptions bool `json:"modify_instance_metadata_options,omitempty"`
+	SetDisableApiTermination      bool `json:"set_disable_api_termination,omitempty"`
 }
 
 // AttachVolumeData carries parameters for an attach-volume command.
@@ -103,6 +116,23 @@ type InstanceMonitoringData struct {
 	Enabled bool `json:"enabled"`
 }
 
+// InstanceMetadataOptionsData carries the requested metadata options for a
+// modify-instance-metadata-options command. Unset fields are left unchanged.
+type InstanceMetadataOptionsData struct {
+	HttpTokens              string `json:"http_tokens,omitempty"`
+	HttpEndpoint            string `json:"http_endpoint,omitempty"`
+	HttpProtocolIpv6        string `json:"http_protocol_ipv6,omitempty"`
+	InstanceMetadataTags    string `json:"instance_metadata_tags,omitempty"`
+	HttpPutResponseHopLimit *int64 `json:"http_put_response_hop_limit,omitempty"`
+}
+
+// DisableApiTerminationData carries the desired termination-protection value
+// for a set-disable-api-termination command. Carried as data so "false" is an
+// explicit value, not the absence of one.
+type DisableApiTerminationData struct {
+	Value bool `json:"value"`
+}
+
 // SpotLineageData carries the SIR id stamped onto a spot-launched instance.
 // Lifecycle is always "spot" for this command, so only the SIR id travels.
 type SpotLineageData struct {

contracts/ec2/v1/commands_test.go

@@ -24,6 +24,9 @@ func TestEC2InstanceCommandJSONContract(t *testing.T) {
 			SetInstanceTags:             true,
 			RemoveInstanceTags:          true,
 			SetInstanceMonitoring:       true,
+
+			ModifyInstanceMetadataOptions: true,
+			SetDisableApiTermination:      true,
 		},
 		AttachVolumeData:          &AttachVolumeData{VolumeID: "vol-1", Device: "/dev/sdf"},
 		DetachVolumeData:          &DetachVolumeData{VolumeID: "vol-2", Device: "/dev/sdg", Force: true},
@@ -33,6 +36,15 @@ func TestEC2InstanceCommandJSONContract(t *testing.T) {
 		SpotLineageData:           &SpotLineageData{SpotInstanceRequestId: "sir-1"},
 		InstanceTagsData:          &InstanceTagsData{Tags: map[string]string{"Name": "example"}, TagKeys: []string{"old"}},
 		InstanceMonitoringData:    &InstanceMonitoringData{Enabled: true},
+
+		InstanceMetadataOptionsData: &InstanceMetadataOptionsData{
+			HttpTokens:              "required",
+			HttpEndpoint:            "enabled",
+			HttpProtocolIpv6:        "disabled",
+			InstanceMetadataTags:    "enabled",
+			HttpPutResponseHopLimit: int64Ptr(3),
+		},
+		DisableApiTerminationData: &DisableApiTerminationData{Value: false},
 	}
 
 	data, err := json.Marshal(command)
@@ -52,7 +64,9 @@ func TestEC2InstanceCommandJSONContract(t *testing.T) {
 			"set_spot_lineage": true,
 			"set_instance_tags": true,
 			"remove_instance_tags": true,
-			"set_instance_monitoring": true
+			"set_instance_monitoring": true,
+			"modify_instance_metadata_options": true,
+			"set_disable_api_termination": true
 		},
 		"attach_volume_data": {"volume_id": "vol-1", "device": "/dev/sdf"},
 		"detach_volume_data": {"volume_id": "vol-2", "device": "/dev/sdg", "force": true},
@@ -61,10 +75,64 @@ func TestEC2InstanceCommandJSONContract(t *testing.T) {
 		"iam_profile_association_data": {"instance_profile_arn": "arn:aws:iam::123456789012:instance-profile/example"},
 		"spot_lineage_data": {"spot_instance_request_id": "sir-1"},
 		"instance_tags_data": {"tags": {"Name": "example"}, "tag_keys": ["old"]},
-		"instance_monitoring_data": {"enabled": true}
+		"instance_monitoring_data": {"enabled": true},
+		"instance_metadata_options_data": {
+			"http_tokens": "required",
+			"http_endpoint": "enabled",
+			"http_protocol_ipv6": "disabled",
+			"instance_metadata_tags": "enabled",
+			"http_put_response_hop_limit": 3
+		},
+		"disable_api_termination_data": {"value": false}
 	}`, string(data))
 }
 
+func int64Ptr(v int64) *int64 { return &v }
+
+// An owner that predates the metadata and termination-protection commands
+// sends and receives payloads without the new fields.
+func TestEC2InstanceCommandNewFieldsAreOptionalOnTheWire(t *testing.T) {
+	var n1 EC2InstanceCommand
+	require.NoError(t, json.Unmarshal([]byte(`{
+		"id": "i-0123456789abcdef0",
+		"attributes": {"stop_instance": true},
+		"instance_monitoring_data": {"enabled": true}
+	}`), &n1))
+	require.Equal(t, "i-0123456789abcdef0", n1.ID)
+	require.True(t, n1.Attributes.StopInstance)
+	require.False(t, n1.Attributes.ModifyInstanceMetadataOptions)
+	require.False(t, n1.Attributes.SetDisableApiTermination)
+	require.Nil(t, n1.InstanceMetadataOptionsData)
+	require.Nil(t, n1.DisableApiTerminationData)
+
+	unset, err := json.Marshal(EC2InstanceCommand{ID: "i-1", Attributes: EC2CommandAttributes{StopInstance: true}})
+	require.NoError(t, err)
+	require.NotContains(t, string(unset), "modify_instance_metadata_options")
+	require.NotContains(t, string(unset), "set_disable_api_termination")
+	require.NotContains(t, string(unset), "instance_metadata_options_data")
+	require.NotContains(t, string(unset), "disable_api_termination_data")
+}
+
+func TestInstanceMetadataOptionsDataOmitsUnsetFields(t *testing.T) {
+	data, err := json.Marshal(InstanceMetadataOptionsData{HttpPutResponseHopLimit: int64Ptr(1)})
+	require.NoError(t, err)
+	require.JSONEq(t, `{"http_put_response_hop_limit": 1}`, string(data))
+
+	data, err = json.Marshal(InstanceMetadataOptionsData{})
+	require.NoError(t, err)
+	require.JSONEq(t, `{}`, string(data))
+}
+
+func TestDisableApiTerminationDataKeepsExplicitFalse(t *testing.T) {
+	data, err := json.Marshal(DisableApiTerminationData{Value: false})
+	require.NoError(t, err)
+	require.JSONEq(t, `{"value": false}`, string(data))
+}
+
+func TestHandledHeaderName(t *testing.T) {
+	require.Equal(t, "Spinifex-Ec2-Cmd-Handled", HandledHeader)
+}
+
 func TestInstanceCommandSubjects(t *testing.T) {
 	require.Equal(t, "ec2.cmd.i-0123456789abcdef0", InstanceCommandSubject("i-0123456789abcdef0"))
 	require.Equal(t, "ec2.cmd.*", InstanceCommandSubjectWildcard)

contracts/lambda/v1/README.md

@@ -28,6 +28,9 @@ Its encoded frame is at most the base64 length of 6 MiB plus 64 KiB of metadata
 The credential travels as its own frame of at most 8 KiB, and the reply has the same payload and frame bounds as the request.
 The daemon holds at most two connections at once, from accept to close, and leaves the rest unaccepted in the listen backlog.
 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.
+In the same builds, the function owner moves a created function with stored code from `Pending` to `Active` once this node's daemon serves the pinned image and the cluster configuration has exactly one node; a function with environment variables or a code copy over the test-link bound stays `Pending`, and `StateReason` names the unmet condition.
+Release builds never leave `Pending`.
+The Mulga deploy role installs a test-link build with `deploy_lambda_testlink=true`; this profile does not meet Q-199 fail-closed revocation, and its guests take management addresses whose record can expire (mulga-0hsub).
 
 **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.
 
@@ -41,6 +44,8 @@ Under TLS 1.3 the guest's handshake can complete before the host refuses its cer
 
 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.
+On a VMM without fw_cfg (the test-link Firecracker executor), the executor attaches the same entries as a tar archive on the second, writable virtio-blk disk, laid out as `by_name/<entry>/raw`.
+The runtime image's init extracts it to a root-only tmpfs at `/spinifex-cfg`, and the agent reads the entries there, then overwrites the whole disk with zeros and removes the tree; it refuses to start if the tree is still present.
 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.

docs/testing/e2e-coverage.md

@@ -108,6 +108,8 @@ Three entry points, in the order they run:
 
 **Its own suite and its own cell, for instancerecovery's reasons plus one.** It takes a node away, so anything sharing the environment reports that outage as its own failure; it needs its own VPC, public subnet and a backend placement no shared fixture describes; and a failure here has to read as "the load balancer did not survive its host" rather than as one more red subtest in a suite about load balancer features.
 
+**A dead data NIC is the other half, and there the state does change.** `LBSelfHealsADeadDataNIC` disables one data NIC of a running load balancer guest through a second QMP client and runs for an internet-facing NLB and an internal ALB. The backends and an unrelated customer instance in the same subnet are launched more than 21 minutes before the fault, so a repair that reached a customer instance would find one the platform has long stopped expecting to change. A state watch samples `DescribeLoadBalancers` every two seconds and requires the sequence `active`, `active_impaired`, `active` with nothing in between, while the internet-facing run also watches the name continuously and serves traffic through it within five minutes. Afterwards the guest, its interfaces and its address are unchanged but its process is not, the backends and the bystander keep their instance ID, launch time and process (ADR-0003:S7), and the daemon journals exactly one `ELBv2 data-path repair started` line carrying the fields an operator needs. The internal ALB has no client inside the VPC, so it asserts the state sequence and target health rather than traffic through its name.
+
 ### `iam` — everything IAM and STS
 
 Policy evaluation against live requests rather than unit-level: a deny applying to the next request without a new credential, resource-scoped grants naming one key pair, a user rename needing `iam:UpdateUser` on both the old and the new name, `aws:SourceIp` seeing the real client address, assumed-role sessions carrying role policies, and revocation stopping one key alone.
@@ -129,7 +131,7 @@ The client reports on its serial console because the runner has no route into th
 
 ### `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 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.
+`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. The private-endpoint NIC of the cluster's endpoint NLB is disabled through QMP and every worker must be Ready again within ten minutes while the cluster stays `ACTIVE` and the NLB keeps its guest instance. 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
 

scripts/build-microvm-image.sh

@@ -209,6 +209,7 @@ echo "[build-microvm-image] nginx: binary + stream module present"
 #   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)
+#   af_packet.ko*  — AF_PACKET sockets (lb-agent's per-NIC gateway ARP probe)
 # All other loadable modules are removed. Drivers compiled into the kernel
 # (CONFIG_VIRTIO_MMIO=y etc.) are unaffected — they live in vmlinuz, not here.
 echo "[build-microvm-image] stripping kernel modules to virtio+fw_cfg only..."
@@ -222,6 +223,7 @@ for kver_dir in "$CHROOT_DIR/lib/modules"/*/; do
     find "$kernel_dir" -name "virtio*.ko.gz" -o -name "virtio*.ko" \
                        -o -name "*fw_cfg*.ko.gz" -o -name "*fw_cfg*.ko" \
                        -o -name "*failover*.ko.gz" -o -name "*failover*.ko" \
+                       -o -name "af_packet.ko.gz" -o -name "af_packet.ko" \
         2>/dev/null | while read -r mod; do
         rel="${mod#"$kernel_dir/"}"
         dest_dir="$tmp_keep/$(dirname "$rel")"
@@ -240,6 +242,15 @@ for kver_dir in "$CHROOT_DIR/lib/modules"/*/; do
     done)
     rm -rf "$tmp_keep"
 
+    # lb-agent's data-path probe needs AF_PACKET; without it every LB would
+    # report a dead NIC. Accept a kernel with CONFIG_PACKET built in.
+    if ! find "$kernel_dir" -name "af_packet.ko*" | grep -q .; then
+        if ! grep -q "CONFIG_PACKET=y" "$KERNEL_CONFIG" 2>/dev/null; then
+            echo "ERROR: af_packet.ko missing from the kept modules — lb-agent's data-path probe cannot open its socket" >&2
+            exit 1
+        fi
+    fi
+
     # 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

spinifex/agents/elbv2/agent.go

@@ -21,6 +21,7 @@ import (
 	"path/filepath"
 	"strconv"
 	"strings"
+	"sync"
 	"syscall"
 	"time"
 
@@ -94,6 +95,18 @@ type Agent struct {
 	statsFn func(socketPath string) ([]ServerStatus, error)
 	// For testing: override the active health prober (nginx/NLB).
 	probeFn func(targets []HealthTarget) []ServerStatus
+
+	// Data-path probe: netcfg and sysfs locations, the ARP exchange, the
+	// instance ID lookup and per-NIC miss counters (overridable in tests).
+	netcfgPath   string
+	sysNetDir    string
+	probeWait    time.Duration
+	arpProbeFn   arpProbeFn
+	instanceIDFn func() (string, error)
+	instanceID   string
+	idSkip       int
+	probeMu      sync.Mutex
+	probeStates  map[string]*nicProbeState
 }
 
 // New creates a new LB agent for the given load balancer.
@@ -150,6 +163,12 @@ func New(lbID, gatewayURL, accessKey, secretKey, region string) (*Agent, error)
 		reloadNginxFn: reloadNginx,
 		statsFn:       queryHAProxyStats,
 		probeFn:       probeHealthTargets,
+		netcfgPath:    netcfgPath,
+		sysNetDir:     sysClassNet,
+		probeWait:     probeWait,
+		arpProbeFn:    arpProbe,
+		instanceIDFn:  fetchInstanceID,
+		probeStates:   map[string]*nicProbeState{},
 	}, nil
 }
 
@@ -207,7 +226,7 @@ func (a *Agent) tick() {
 		}
 	}
 
-	resp, err := a.sendHeartbeat(servers)
+	resp, err := a.sendHeartbeat(servers, a.dataPathEvidence())
 	if err != nil {
 		// Include the dial target so the serial console disambiguates a
 		// transport failure (never reached AWSGW — "send request: dial ...")
@@ -288,7 +307,7 @@ func (a *Agent) enginePaths(engine string) (configPath, pidPath, certDir string,
 }
 
 // sendHeartbeat sends a heartbeat with health report to the gateway.
-func (a *Agent) sendHeartbeat(servers []ServerStatus) (*heartbeatResponse, error) {
+func (a *Agent) sendHeartbeat(servers []ServerStatus, ev dataPathEvidence) (*heartbeatResponse, error) {
 	params := url.Values{}
 	params.Set("Action", "LBAgentHeartbeat")
 	params.Set("Version", "2015-12-01")
@@ -300,6 +319,7 @@ func (a *Agent) sendHeartbeat(servers []ServerStatus) (*heartbeatResponse, error
 		params.Set("Servers.member."+idx+".Server", s.Server)
 		params.Set("Servers.member."+idx+".Status", s.Status)
 	}
+	ev.addTo(params)
 
 	body, err := a.signedPost(params)
 	if err != nil {

spinifex/agents/elbv2/arp_linux.go

@@ -0,0 +1,62 @@
+//go:build linux
+
+package elbv2
+
+import (
+	"errors"
+	"fmt"
+	"net"
+	"net/netip"
+	"syscall"
+	"time"
+)
+
+// arpProbe sends one ARP request for gateway out of iface on an AF_PACKET
+// socket bound to it and waits for the gateway's reply on the same socket.
+func arpProbe(iface string, mac net.HardwareAddr, src, gateway netip.Addr, wait time.Duration) error {
+	ifi, err := net.InterfaceByName(iface)
+	if err != nil {
+		return fmt.Errorf("interface %s: %w", iface, err)
+	}
+
+	proto := htons(syscall.ETH_P_ARP)
+	fd, err := syscall.Socket(syscall.AF_PACKET, syscall.SOCK_RAW, int(proto))
+	if err != nil {
+		return fmt.Errorf("packet socket: %w", err)
+	}
+	defer syscall.Close(fd)
+
+	addr := &syscall.SockaddrLinklayer{Protocol: proto, Ifindex: ifi.Index}
+	if err := syscall.Bind(fd, addr); err != nil {
+		return fmt.Errorf("bind %s: %w", iface, err)
+	}
+
+	deadline := time.Now().Add(wait)
+	if err := syscall.Sendto(fd, buildARPRequest(mac, src, gateway), 0, addr); err != nil {
+		return fmt.Errorf("send ARP request: %w", err)
+	}
+
+	buf := make([]byte, 128)
+	for {
+		left := time.Until(deadline)
+		if left <= 0 {
+			return errARPNoReply
+		}
+		tv := syscall.NsecToTimeval(int64(left))
+		if err := syscall.SetsockoptTimeval(fd, syscall.SOL_SOCKET, syscall.SO_RCVTIMEO, &tv); err != nil {
+			return fmt.Errorf("set receive deadline: %w", err)
+		}
+		n, _, err := syscall.Recvfrom(fd, buf, 0)
+		if err != nil {
+			if errors.Is(err, syscall.EAGAIN) || errors.Is(err, syscall.EINTR) {
+				continue
+			}
+			return fmt.Errorf("receive ARP reply: %w", err)
+		}
+		if isARPReplyFrom(buf[:n], mac, gateway) {
+			return nil
+		}
+	}
+}
+
+func htons(v uint16) uint16 { return v<<8 | v>>8 }

spinifex/agents/elbv2/arp_other.go

@@ -0,0 +1,15 @@
+//go:build !linux
+
+package elbv2
+
+import (
+	"errors"
+	"net"
+	"net/netip"
+	"time"
+)
+
+// arpProbe is only meaningful inside the Linux guest.
+func arpProbe(string, net.HardwareAddr, netip.Addr, netip.Addr, time.Duration) error {
+	return errors.New("ARP probe is only supported on linux")
+}

spinifex/agents/elbv2/datapath.go

@@ -0,0 +1,303 @@
+package elbv2
+
+import (
+	"bufio"
+	"bytes"
+	"encoding/binary"
+	"errors"
+	"fmt"
+	"net"
+	"net/netip"
+	"os"
+	"path/filepath"
+	"sort"
+	"strconv"
+	"strings"
+	"sync"
+	"time"
+)
+
+const (
+	// netcfgPath is the fw_cfg blob the microVM init script reads to configure NICs.
+	netcfgPath = "/sys/firmware/qemu_fw_cfg/by_name/opt/spinifex/netcfg/raw"
+
+	sysClassNet = "/sys/class/net"
+
+	// probeFailureThreshold is the number of consecutive missed probes after
+	// which a NIC is reported as failing (about 15 s at the poll interval).
+	probeFailureThreshold = 3
+
+	// arpExchangeWait bounds one ARP exchange. probeWait bounds how long a tick
+	// waits for all probes, so a stuck probe never delays the heartbeat.
+	arpExchangeWait = time.Second
+	probeWait       = 1500 * time.Millisecond
+)
+
+// nicConfig is one NIC of the netcfg blob that has a gateway to probe.
+type nicConfig struct {
+	MAC     string
+	Addr    netip.Addr
+	Gateway netip.Addr
+}
+
+// NICReport is the per-NIC data-path evidence carried in the heartbeat.
+type NICReport struct {
+	MAC                 string
+	ProbeOK             bool
+	ConsecutiveFailures int64
+	RxPackets           int64
+	TxPackets           int64
+	RxDropped           int64
+	TxDropped           int64
+}
+
+// arpProbeFn sends one ARP request for gateway out of iface and returns nil
+// when the gateway's reply arrives within timeout.
+type arpProbeFn func(iface string, mac net.HardwareAddr, src, gateway netip.Addr, timeout time.Duration) error
+
+// parseNetcfg reads the KEY=value netcfg blob and returns the NICs that have a
+// gateway, in NIC order. The mgmt NIC carries no gateway and is skipped, as
+// is any NIC without a usable MAC, address or gateway.
+func parseNetcfg(blob []byte) []nicConfig {
+	type raw struct{ mac, cidr, gw string }
+	byIndex := map[int]*raw{}
+
+	sc := bufio.NewScanner(bytes.NewReader(blob))
+	for sc.Scan() {
+		line := strings.TrimSpace(sc.Text())
+		key, val, ok := strings.Cut(line, "=")
+		if !ok || !strings.HasPrefix(key, "NIC") {
+			continue
+		}
+		idxStr, field, ok := strings.Cut(strings.TrimPrefix(key, "NIC"), "_")
+		if !ok {
+			continue
+		}
+		idx, err := strconv.Atoi(idxStr)
+		if err != nil {
+			continue
+		}
+		r := byIndex[idx]
+		if r == nil {
+			r = &raw{}
+			byIndex[idx] = r
+		}
+		switch field {
+		case "MAC":
+			r.mac = strings.ToLower(val)
+		case "CIDR":
+			r.cidr = val
+		case "GW":
+			r.gw = val
+		}
+	}
+
+	idxs := make([]int, 0, len(byIndex))
+	for i := range byIndex {
+		idxs = append(idxs, i)
+	}
+	sort.Ints(idxs)
+
+	var out []nicConfig
+	for _, i := range idxs {
+		r := byIndex[i]
+		if r.mac == "" || r.cidr == "" || r.gw == "" {
+			continue
+		}
+		prefix, err := netip.ParsePrefix(r.cidr)
+		if err != nil || !prefix.Addr().Is4() {
+			continue
+		}
+		gw, err := netip.ParseAddr(r.gw)
+		if err != nil || !gw.Is4() {
+			continue
+		}
+		out = append(out, nicConfig{MAC: r.mac, Addr: prefix.Addr(), Gateway: gw})
+	}
+	return out
+}
+
+// ifaceByMAC returns the interface under sysNetDir whose address is mac.
+func ifaceByMAC(sysNetDir, mac string) (string, error) {
+	entries, err := os.ReadDir(sysNetDir)
+	if err != nil {
+		return "", err
+	}
+	for _, e := range entries {
+		b, err := os.ReadFile(filepath.Join(sysNetDir, e.Name(), "address"))
+		if err != nil {
+			continue
+		}
+		if strings.EqualFold(strings.TrimSpace(string(b)), mac) {
+			return e.Name(), nil
+		}
+	}
+	return "", fmt.Errorf("no interface with MAC %s", mac)
+}
+
+// readCounters fills the sysfs statistics of iface into r.
+func readCounters(sysNetDir, iface string, r *NICReport) {
+	read := func(name string) int64 {
+		b, err := os.ReadFile(filepath.Join(sysNetDir, iface, "statistics", name))
+		if err != nil {
+			return 0
+		}
+		v, _ := strconv.ParseInt(strings.TrimSpace(string(b)), 10, 64)
+		return v
+	}
+	r.RxPackets = read("rx_packets")
+	r.TxPackets = read("tx_packets")
+	r.RxDropped = read("rx_dropped")
+	r.TxDropped = read("tx_dropped")
+}
+
+var errARPNoReply = errors.New("no ARP reply from the gateway")
+
+// buildARPRequest returns the Ethernet frame asking who has target.
+func buildARPRequest(mac net.HardwareAddr, src, target netip.Addr) []byte {
+	f := make([]byte, 0, 42)
+	f = append(f, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff)
+	f = append(f, mac...)
+	f = binary.BigEndian.AppendUint16(f, 0x0806)
+	f = binary.BigEndian.AppendUint16(f, 1)      // Ethernet
+	f = binary.BigEndian.AppendUint16(f, 0x0800) // IPv4
+	f = append(f, 6, 4)
+	f = binary.BigEndian.AppendUint16(f, 1) // request
+	f = append(f, mac...)
+	s := src.As4()
+	f = append(f, s[:]...)
+	f = append(f, 0, 0, 0, 0, 0, 0)
+	t := target.As4()
+	f = append(f, t[:]...)
+	return f
+}
+
+// isARPReplyFrom reports whether frame is an ARP reply sent by sender to mac.
+func isARPReplyFrom(frame []byte, mac net.HardwareAddr, sender netip.Addr) bool {
+	if len(frame) < 42 || binary.BigEndian.Uint16(frame[12:14]) != 0x0806 {
+		return false
+	}
+	if binary.BigEndian.Uint16(frame[20:22]) != 2 || !bytes.Equal(frame[32:38], mac) {
+		return false
+	}
+	s := sender.As4()
+	return bytes.Equal(frame[28:32], s[:])
+}
+
+// nicProbeState carries the miss counter of one NIC across ticks. A probe that
+// has not returned is still in flight, and counts as a miss without starting
+// another goroutine behind it.
+type nicProbeState struct {
+	mu       sync.Mutex
+	failures int64
+	inflight bool
+}
+
+type probeResult struct {
+	ok  bool
+	set bool
+}
+
+// probeNICs probes every gateway-bearing data NIC once, in parallel, and
+// returns within probeWait however long a probe blocks.
+func (a *Agent) probeNICs() []NICReport {
+	blob, err := os.ReadFile(a.netcfgPath)
+	if err != nil {
+		return nil
+	}
+	nics := parseNetcfg(blob)
+	if len(nics) == 0 {
+		return nil
+	}
+
+	reports := make([]NICReport, len(nics))
+	results := make([]chan probeResult, len(nics))
+	for i, n := range nics {
+		results[i] = make(chan probeResult, 1)
+		st := a.probeState(n.MAC)
+
+		st.mu.Lock()
+		busy := st.inflight
+		if !busy {
+			st.inflight = true
+		}
+		st.mu.Unlock()
+		if busy {
+			results[i] <- probeResult{set: true}
+			continue
+		}
+
+		go func(n nicConfig, st *nicProbeState, out chan<- probeResult) {
+			ok := a.probeOne(n)
+			st.mu.Lock()
+			st.inflight = false
+			st.mu.Unlock()
+			out <- probeResult{ok: ok, set: true}
+		}(n, st, results[i])
+	}
+
+	deadline := time.NewTimer(a.probeWait)
+	defer deadline.Stop()
+	expired := false
+	for i, n := range nics {
+		var res probeResult
+		if expired {
+			select {
+			case res = <-results[i]:
+			default:
+			}
+		} else {
+			select {
+			case res = <-results[i]:
+			case <-deadline.C:
+				// Out of time: an unfinished probe is a miss for this tick.
+				expired = true
+			}
+		}
+
+		st := a.probeState(n.MAC)
+		st.mu.Lock()
+		if res.ok {
+			st.failures = 0
+		} else {
+			st.failures++
+		}
+		failures := st.failures
+		st.mu.Unlock()
+
+		r := NICReport{
+			MAC:                 n.MAC,
+			ProbeOK:             failures < probeFailureThreshold,
+			ConsecutiveFailures: failures,
+		}
+		if iface, err := ifaceByMAC(a.sysNetDir, n.MAC); err == nil {
+			readCounters(a.sysNetDir, iface, &r)
+		}
+		reports[i] = r
+	}
+	return reports
+}
+
+// probeOne runs one ARP exchange for n, false on any failure.
+func (a *Agent) probeOne(n nicConfig) bool {
+	iface, err := ifaceByMAC(a.sysNetDir, n.MAC)
+	if err != nil {
+		return false
+	}
+	hw, err := net.ParseMAC(n.MAC)
+	if err != nil {
+		return false
+	}
+	return a.arpProbeFn(iface, hw, n.Addr, n.Gateway, arpExchangeWait) == nil
+}
+
+func (a *Agent) probeState(mac string) *nicProbeState {
+	a.probeMu.Lock()
+	defer a.probeMu.Unlock()
+	st := a.probeStates[mac]
+	if st == nil {
+		st = &nicProbeState{}
+		a.probeStates[mac] = st
+	}
+	return st
+}

spinifex/agents/elbv2/datapath_test.go

@@ -0,0 +1,484 @@
+package elbv2
+
+import (
+	"errors"
+	"fmt"
+	"net"
+	"net/http"
+	"net/http/httptest"
+	"net/netip"
+	"net/url"
+	"os"
+	"path/filepath"
+	"sync"
+	"sync/atomic"
+	"testing"
+	"time"
+)
+
+const (
+	testDataMAC = "02:00:00:aa:00:01"
+	testMgmtMAC = "02:00:00:aa:00:02"
+	testNetcfg  = "NIC0_MAC=" + testDataMAC + "\nNIC0_CIDR=10.0.1.5/24\nNIC0_GW=10.0.1.1\nNIC0_DEFAULT=1\n" +
+		"NIC1_MAC=" + testMgmtMAC + "\nNIC1_CIDR=10.255.0.9/24\nNIC1_DEFAULT=0\n"
+)
+
+// datapathAgent returns a test agent whose netcfg and sysfs live in a temp dir.
+func datapathAgent(t *testing.T, gwURL, netcfg string) *Agent {
+	t.Helper()
+	a := newTestAgent(t, gwURL)
+	dir := t.TempDir()
+
+	a.netcfgPath = filepath.Join(dir, "netcfg")
+	if err := os.WriteFile(a.netcfgPath, []byte(netcfg), 0o600); err != nil {
+		t.Fatal(err)
+	}
+
+	a.sysNetDir = filepath.Join(dir, "net")
+	for name, mac := range map[string]string{"ens3": testDataMAC, "ens4": testMgmtMAC} {
+		stats := filepath.Join(a.sysNetDir, name, "statistics")
+		if err := os.MkdirAll(stats, 0o755); err != nil {
+			t.Fatal(err)
+		}
+		if err := os.WriteFile(filepath.Join(a.sysNetDir, name, "address"), []byte(mac+"\n"), 0o600); err != nil {
+			t.Fatal(err)
+		}
+		for file, v := range map[string]string{"rx_packets": "100", "tx_packets": "200", "rx_dropped": "3", "tx_dropped": "4"} {
+			if err := os.WriteFile(filepath.Join(stats, file), []byte(v+"\n"), 0o600); err != nil {
+				t.Fatal(err)
+			}
+		}
+	}
+
+	a.instanceIDFn = func() (string, error) { return "i-0123456789abcdef0", nil }
+	return a
+}
+
+func TestParseNetcfg_SkipsMgmtNICAndMissingGateway(t *testing.T) {
+	blob := testNetcfg +
+		"NIC2_MAC=02:00:00:aa:00:03\nNIC2_CIDR=10.0.2.5/24\nNIC2_GW=10.0.2.1\nNIC2_DEFAULT=0\n" +
+		"NIC3_MAC=02:00:00:aa:00:04\nNIC3_DHCP=1\nNIC3_DEFAULT=0\n" +
+		"NIC4_CIDR=10.0.4.5/24\nNIC4_GW=10.0.4.1\n" +
+		"NIC5_MAC=02:00:00:aa:00:06\nNIC5_CIDR=not-a-cidr\nNIC5_GW=10.0.5.1\n"
+
+	got := parseNetcfg([]byte(blob))
+	if len(got) != 2 {
+		t.Fatalf("got %d NICs, want the 2 with a gateway: %+v", len(got), got)
+	}
+	if got[0].MAC != testDataMAC || got[0].Gateway != netip.MustParseAddr("10.0.1.1") || got[0].Addr != netip.MustParseAddr("10.0.1.5") {
+		t.Errorf("first NIC = %+v", got[0])
+	}
+	if got[1].MAC != "02:00:00:aa:00:03" || got[1].Gateway != netip.MustParseAddr("10.0.2.1") {
+		t.Errorf("second NIC = %+v", got[1])
+	}
+}
+
+func TestParseNetcfg_OrdersByNICIndexAndLowercasesMAC(t *testing.T) {
+	blob := "NIC10_MAC=02:00:00:AA:00:0A\nNIC10_CIDR=10.0.9.5/24\nNIC10_GW=10.0.9.1\n" +
+		"NIC2_MAC=02:00:00:AA:00:02\nNIC2_CIDR=10.0.2.5/24\nNIC2_GW=10.0.2.1\n"
+	got := parseNetcfg([]byte(blob))
+	if len(got) != 2 || got[0].MAC != "02:00:00:aa:00:02" || got[1].MAC != "02:00:00:aa:00:0a" {
+		t.Fatalf("order or case wrong: %+v", got)
+	}
+}
+
+func TestIfaceByMACAndCounters(t *testing.T) {
+	a := datapathAgent(t, "http://unused", testNetcfg)
+
+	iface, err := ifaceByMAC(a.sysNetDir, "02:00:00:AA:00:01")
+	if err != nil || iface != "ens3" {
+		t.Fatalf("iface = %q, err = %v", iface, err)
+	}
+	if _, err := ifaceByMAC(a.sysNetDir, "02:00:00:ff:ff:ff"); err == nil {
+		t.Error("an unknown MAC must not resolve to an interface")
+	}
+
+	var r NICReport
+	readCounters(a.sysNetDir, iface, &r)
+	if r.RxPackets != 100 || r.TxPackets != 200 || r.RxDropped != 3 || r.TxDropped != 4 {
+		t.Errorf("counters = %+v", r)
+	}
+}
+
+func TestARPFrames_RequestShapeAndReplyMatching(t *testing.T) {
+	mac, _ := net.ParseMAC(testDataMAC)
+	src, gw := netip.MustParseAddr("10.0.1.5"), netip.MustParseAddr("10.0.1.1")
+
+	req := buildARPRequest(mac, src, gw)
+	if len(req) != 42 || req[12] != 0x08 || req[13] != 0x06 || req[21] != 1 {
+		t.Fatalf("request frame malformed: % x", req)
+	}
+	if string(req[38:42]) != string(gw.AsSlice()) {
+		t.Errorf("target IP not the gateway: % x", req[38:42])
+	}
+
+	reply := make([]byte, 42)
+	copy(reply[0:6], mac)
+	reply[12], reply[13] = 0x08, 0x06
+	reply[21] = 2
+	copy(reply[28:32], gw.AsSlice())
+	copy(reply[32:38], mac)
+	if !isARPReplyFrom(reply, mac, gw) {
+		t.Error("a reply from the gateway to our MAC must match")
+	}
+	if isARPReplyFrom(reply, mac, netip.MustParseAddr("10.0.1.99")) {
+		t.Error("a reply from another host must not match")
+	}
+	reply[21] = 1
+	if isARPReplyFrom(reply, mac, gw) {
+		t.Error("a request must not count as a reply")
+	}
+	if isARPReplyFrom(reply[:20], mac, gw) {
+		t.Error("a short frame must not match")
+	}
+}
+
+func TestProbeNICs_ProbesOnlyTheGatewayNIC(t *testing.T) {
+	a := datapathAgent(t, "http://unused", testNetcfg)
+	var mu sync.Mutex
+	var probed []string
+	a.arpProbeFn = func(iface string, _ net.HardwareAddr, src, gw netip.Addr, _ time.Duration) error {
+		mu.Lock()
+		defer mu.Unlock()
+		probed = append(probed, fmt.Sprintf("%s %s>%s", iface, src, gw))
+		return nil
+	}
+
+	reports := a.probeNICs()
+	if len(reports) != 1 || reports[0].MAC != testDataMAC || !reports[0].ProbeOK || reports[0].ConsecutiveFailures != 0 {
+		t.Fatalf("reports = %+v", reports)
+	}
+	if len(probed) != 1 || probed[0] != "ens3 10.0.1.5>10.0.1.1" {
+		t.Errorf("probed = %v, want only the data NIC", probed)
+	}
+	if reports[0].RxPackets != 100 || reports[0].TxDropped != 4 {
+		t.Errorf("counters missing from report: %+v", reports[0])
+	}
+}
+
+func TestProbeNICs_TwoMissesThenSuccessNeverReportsFailing(t *testing.T) {
+	a := datapathAgent(t, "http://unused", testNetcfg)
+	results := []error{errARPNoReply, errARPNoReply, nil, errARPNoReply, errARPNoReply, nil}
+	var n atomic.Int32
+	a.arpProbeFn = func(string, net.HardwareAddr, netip.Addr, netip.Addr, time.Duration) error {
+		return results[n.Add(1)-1]
+	}
+
+	for i, want := range []int64{1, 2, 0, 1, 2, 0} {
+		r := a.probeNICs()
+		if len(r) != 1 || !r[0].ProbeOK || r[0].ConsecutiveFailures != want {
+			t.Fatalf("tick %d: report = %+v, want ok with %d failures", i, r, want)
+		}
+	}
+}
+
+func TestProbeNICs_ThreeMissesReportFailingAndRecoveryClears(t *testing.T) {
+	a := datapathAgent(t, "http://unused", testNetcfg)
+	var fail atomic.Bool
+	fail.Store(true)
+	a.arpProbeFn = func(string, net.HardwareAddr, netip.Addr, netip.Addr, time.Duration) error {
+		if fail.Load() {
+			return errARPNoReply
+		}
+		return nil
+	}
+
+	for i := 1; i <= 4; i++ {
+		r := a.probeNICs()
+		wantOK := i < 3 // failing from the third consecutive miss
+		if r[0].ProbeOK != wantOK || r[0].ConsecutiveFailures != int64(i) {
+			t.Fatalf("miss %d: report = %+v, want ProbeOK=%v", i, r[0], wantOK)
+		}
+	}
+
+	fail.Store(false)
+	if r := a.probeNICs(); !r[0].ProbeOK || r[0].ConsecutiveFailures != 0 {
+		t.Errorf("a successful probe must clear the failure: %+v", r[0])
+	}
+}
+
+func TestProbeNICs_BlockedProbeIsBoundedAndCountsAsMiss(t *testing.T) {
+	a := datapathAgent(t, "http://unused", testNetcfg)
+	a.probeWait = 50 * time.Millisecond
+	release := make(chan struct{})
+	defer close(release)
+	var calls atomic.Int32
+	a.arpProbeFn = func(string, net.HardwareAddr, netip.Addr, netip.Addr, time.Duration) error {
+		calls.Add(1)
+		<-release
+		return nil
+	}
+
+	for i := 1; i <= 3; i++ {
+		start := time.Now()
+		r := a.probeNICs()
+		if d := time.Since(start); d > time.Second {
+			t.Fatalf("tick %d waited %s on a stuck probe", i, d)
+		}
+		if len(r) != 1 || r[0].ConsecutiveFailures != int64(i) {
+			t.Fatalf("tick %d: report = %+v, want %d misses", i, r, i)
+		}
+	}
+	if got := calls.Load(); got != 1 {
+		t.Errorf("a stuck probe was started %d times, want 1 (no goroutine pile-up)", got)
+	}
+}
+
+func TestTick_BlockedProbeDoesNotDelayHeartbeatBeyondOneTick(t *testing.T) {
+	var got atomic.Pointer[url.Values]
+	gw := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		_ = r.ParseForm()
+		if r.FormValue("Action") == "LBAgentHeartbeat" {
+			v := r.Form
+			got.Store(&v)
+		}
+		fmt.Fprint(w, `<LBAgentHeartbeatResponse><LBAgentHeartbeatResult><Status>active</Status><ConfigHash>h1</ConfigHash></LBAgentHeartbeatResult></LBAgentHeartbeatResponse>`)
+	}))
+	defer gw.Close()
+
+	a := datapathAgent(t, gw.URL, testNetcfg)
+	a.localConfigHash = "h1"
+	a.probeWait = 50 * time.Millisecond
+	release := make(chan struct{})
+	defer close(release)
+	a.arpProbeFn = func(string, net.HardwareAddr, netip.Addr, netip.Addr, time.Duration) error {
+		<-release
+		return nil
+	}
+
+	start := time.Now()
+	a.tick()
+	if d := time.Since(start); d >= pollInterval {
+		t.Fatalf("tick took %s with a stuck probe, want under one %s tick", d, pollInterval)
+	}
+	if got.Load() == nil {
+		t.Fatal("the heartbeat was never sent")
+	}
+}
+
+func TestHeartbeat_EncodesEvidenceWhenPresent(t *testing.T) {
+	var got atomic.Pointer[url.Values]
+	gw := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		_ = r.ParseForm()
+		v := r.Form
+		got.Store(&v)
+		fmt.Fprint(w, `<LBAgentHeartbeatResponse><LBAgentHeartbeatResult><Status>active</Status><ConfigHash>h1</ConfigHash></LBAgentHeartbeatResult></LBAgentHeartbeatResponse>`)
+	}))
+	defer gw.Close()
+
+	a := datapathAgent(t, gw.URL, testNetcfg)
+	a.localConfigHash = "h1"
+	a.arpProbeFn = func(string, net.HardwareAddr, netip.Addr, netip.Addr, time.Duration) error { return errARPNoReply }
+
+	a.tick()
+	v := got.Load()
+	if v == nil {
+		t.Fatal("no heartbeat received")
+	}
+	want := map[string]string{
+		"InstanceID":                        "i-0123456789abcdef0",
+		"NICs.member.1.MAC":                 testDataMAC,
+		"NICs.member.1.ProbeOK":             "true",
+		"NICs.member.1.ConsecutiveFailures": "1",
+		"NICs.member.1.RxPackets":           "100",
+		"NICs.member.1.TxPackets":           "200",
+		"NICs.member.1.RxDropped":           "3",
+		"NICs.member.1.TxDropped":           "4",
+	}
+	for k, w := range want {
+		if g := v.Get(k); g != w {
+			t.Errorf("%s = %q, want %q", k, g, w)
+		}
+	}
+	if v.Get("NICs.member.2.MAC") != "" {
+		t.Error("the mgmt NIC must not be reported")
+	}
+}
+
+func TestHeartbeat_OmitsEvidenceWithoutNetcfg(t *testing.T) {
+	var got atomic.Pointer[url.Values]
+	gw := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		_ = r.ParseForm()
+		v := r.Form
+		got.Store(&v)
+		fmt.Fprint(w, `<LBAgentHeartbeatResponse><LBAgentHeartbeatResult><Status>active</Status><ConfigHash>h1</ConfigHash></LBAgentHeartbeatResult></LBAgentHeartbeatResponse>`)
+	}))
+	defer gw.Close()
+
+	a := datapathAgent(t, gw.URL, testNetcfg)
+	a.localConfigHash = "h1"
+	a.netcfgPath = filepath.Join(t.TempDir(), "absent")
+	a.instanceIDFn = func() (string, error) { return "", errors.New("must not be asked without NICs") }
+
+	a.tick()
+	v := got.Load()
+	if v == nil {
+		t.Fatal("no heartbeat received")
+	}
+	for k := range *v {
+		if k == "InstanceID" || len(k) >= 4 && k[:4] == "NICs" {
+			t.Errorf("unexpected evidence parameter %s", k)
+		}
+	}
+}
+
+func TestDataPathEvidence_InstanceIDRetriedUntilKnown(t *testing.T) {
+	a := datapathAgent(t, "http://unused", testNetcfg)
+	a.arpProbeFn = func(string, net.HardwareAddr, netip.Addr, netip.Addr, time.Duration) error { return nil }
+	var calls atomic.Int32
+	a.instanceIDFn = func() (string, error) {
+		if calls.Add(1) < 3 {
+			return "", errors.New("metadata not ready")
+		}
+		return "i-late", nil
+	}
+
+	// A failure backs the lookup off for imdsRetryTicks-1 ticks.
+	ticks := func(n int) string {
+		var id string
+		for range n {
+			id = a.dataPathEvidence().InstanceID
+		}
+		return id
+	}
+	if got := ticks(imdsRetryTicks); got != "" || calls.Load() != 1 {
+		t.Fatalf("after %d ticks: id %q, calls %d; want no id and one lookup", imdsRetryTicks, got, calls.Load())
+	}
+	if got := ticks(imdsRetryTicks); got != "" || calls.Load() != 2 {
+		t.Fatalf("second window: id %q, calls %d; want no id and two lookups", got, calls.Load())
+	}
+	if got := ticks(imdsRetryTicks); got != "i-late" || calls.Load() != 3 {
+		t.Fatalf("third window: id %q, calls %d; want i-late and three lookups", got, calls.Load())
+	}
+	ticks(imdsRetryTicks)
+	if calls.Load() != 3 {
+		t.Errorf("metadata asked %d times, want 3 (stop once known)", calls.Load())
+	}
+}
+
+func TestTick_BlockedInstanceIDLookupDoesNotDelayHeartbeat(t *testing.T) {
+	var got atomic.Pointer[url.Values]
+	gw := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		_ = r.ParseForm()
+		if r.FormValue("Action") == "LBAgentHeartbeat" {
+			v := r.Form
+			got.Store(&v)
+		}
+		fmt.Fprint(w, `<LBAgentHeartbeatResponse><LBAgentHeartbeatResult><Status>active</Status><ConfigHash>h1</ConfigHash></LBAgentHeartbeatResult></LBAgentHeartbeatResponse>`)
+	}))
+	defer gw.Close()
+
+	a := datapathAgent(t, gw.URL, testNetcfg)
+	a.localConfigHash = "h1"
+	a.arpProbeFn = func(string, net.HardwareAddr, netip.Addr, netip.Addr, time.Duration) error { return nil }
+	release := make(chan struct{})
+	defer close(release)
+	var calls atomic.Int32
+	a.instanceIDFn = func() (string, error) {
+		calls.Add(1)
+		<-release
+		return "", errors.New("unreachable")
+	}
+
+	for i := range 3 {
+		start := time.Now()
+		a.tick()
+		if d := time.Since(start); d >= pollInterval {
+			t.Fatalf("tick %d took %s with IMDS blocked, want under one %s tick", i, d, pollInterval)
+		}
+	}
+	v := got.Load()
+	if v == nil {
+		t.Fatal("the heartbeat was never sent")
+	}
+	if v.Get("InstanceID") != "" {
+		t.Error("InstanceID must be omitted while the lookup fails")
+	}
+	if v.Get("NICs.member.1.MAC") == "" {
+		t.Error("NIC evidence must still be sent")
+	}
+	if calls.Load() != 1 {
+		t.Errorf("blocked lookup started %d times in 3 ticks, want 1 (backoff)", calls.Load())
+	}
+}
+
+func TestFetchInstanceID_SharesOneDeadline(t *testing.T) {
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		time.Sleep(3 * imdsWait)
+	}))
+	defer srv.Close()
+	old := imdsBase
+	imdsBase = srv.URL + "/latest"
+	defer func() { imdsBase = old }()
+
+	start := time.Now()
+	if _, err := fetchInstanceID(); err == nil {
+		t.Fatal("a hung metadata service must fail the lookup")
+	}
+	if d := time.Since(start); d > imdsWait+imdsWait/2 {
+		t.Fatalf("lookup took %s, want about one %s deadline", d, imdsWait)
+	}
+}
+
+func TestARPProbe_NoResponderFailsClosed(t *testing.T) {
+	mac, _ := net.ParseMAC(testDataMAC)
+	// Nobody answers an ARP request sent on loopback, so a probe that returned
+	// nil here would mask a dead data path. It fails either on the socket (no
+	// CAP_NET_RAW) or on the missing reply.
+	err := arpProbe("lo", mac, netip.MustParseAddr("10.9.9.2"), netip.MustParseAddr("10.9.9.1"), 100*time.Millisecond)
+	if err == nil {
+		t.Fatal("a probe with no responder must fail")
+	}
+	if err := arpProbe("no-such-iface0", mac, netip.MustParseAddr("10.9.9.2"), netip.MustParseAddr("10.9.9.1"), 100*time.Millisecond); err == nil {
+		t.Fatal("a probe on a missing interface must fail")
+	}
+}
+
+func TestFetchInstanceID(t *testing.T) {
+	serve := func(tokenStatus, idStatus int, wantToken string) *httptest.Server {
+		return httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+			switch {
+			case r.Method == http.MethodPut && r.URL.Path == "/latest/api/token":
+				w.WriteHeader(tokenStatus)
+				fmt.Fprint(w, "tok123")
+			case r.URL.Path == "/latest/meta-data/instance-id":
+				if got := r.Header.Get("X-Aws-Ec2-Metadata-Token"); got != wantToken {
+					t.Errorf("token header = %q, want %q", got, wantToken)
+				}
+				w.WriteHeader(idStatus)
+				fmt.Fprint(w, " i-abc123 \n")
+			default:
+				http.NotFound(w, r)
+			}
+		}))
+	}
+	saved := imdsBase
+	defer func() { imdsBase = saved }()
+
+	for _, tc := range []struct {
+		name                  string
+		tokenStatus, idStatus int
+		wantToken             string
+		wantID                string
+		wantErr               bool
+	}{
+		{"token then id", http.StatusOK, http.StatusOK, "tok123", "i-abc123", false},
+		{"token refused falls back to v1", http.StatusNotFound, http.StatusOK, "", "i-abc123", false},
+		{"id endpoint error", http.StatusOK, http.StatusServiceUnavailable, "tok123", "", true},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			srv := serve(tc.tokenStatus, tc.idStatus, tc.wantToken)
+			defer srv.Close()
+			imdsBase = srv.URL + "/latest"
+			id, err := fetchInstanceID()
+			if (err != nil) != tc.wantErr || id != tc.wantID {
+				t.Fatalf("id = %q, err = %v", id, err)
+			}
+		})
+	}
+
+	imdsBase = "http://127.0.0.1:1/latest"
+	if _, err := fetchInstanceID(); err == nil {
+		t.Error("an unreachable metadata service must fail")
+	}
+}

spinifex/agents/elbv2/evidence.go

@@ -0,0 +1,133 @@
+package elbv2
+
+import (
+	"context"
+	"fmt"
+	"io"
+	"net/http"
+	"net/url"
+	"strconv"
+	"strings"
+	"time"
+)
+
+const (
+	imdsWait       = time.Second
+	imdsRetryTicks = 6
+)
+
+// imdsBase is the in-VM metadata endpoint (a variable so tests can point it at a fake).
+var imdsBase = "http://169.254.169.254/latest"
+
+// dataPathEvidence is the optional part of a heartbeat: which VM the agent runs
+// in and what each data NIC's gateway probe saw. An empty value adds nothing,
+// which is what an agent without a netcfg blob sends.
+type dataPathEvidence struct {
+	InstanceID string
+	NICs       []NICReport
+}
+
+// addTo encodes the evidence as additional heartbeat query parameters.
+func (e dataPathEvidence) addTo(params url.Values) {
+	if e.InstanceID != "" {
+		params.Set("InstanceID", e.InstanceID)
+	}
+	for i, n := range e.NICs {
+		p := "NICs.member." + strconv.Itoa(i+1) + "."
+		params.Set(p+"MAC", n.MAC)
+		params.Set(p+"ProbeOK", strconv.FormatBool(n.ProbeOK))
+		params.Set(p+"ConsecutiveFailures", strconv.FormatInt(n.ConsecutiveFailures, 10))
+		params.Set(p+"RxPackets", strconv.FormatInt(n.RxPackets, 10))
+		params.Set(p+"TxPackets", strconv.FormatInt(n.TxPackets, 10))
+		params.Set(p+"RxDropped", strconv.FormatInt(n.RxDropped, 10))
+		params.Set(p+"TxDropped", strconv.FormatInt(n.TxDropped, 10))
+	}
+}
+
+// dataPathEvidence gathers the probe results and instance ID for one heartbeat.
+// Neither step can fail the heartbeat: missing evidence is simply omitted.
+func (a *Agent) dataPathEvidence() dataPathEvidence {
+	nics := a.probeNICs()
+	if len(nics) == 0 {
+		return dataPathEvidence{}
+	}
+	if a.instanceID == "" {
+		a.lookupInstanceID()
+	}
+	return dataPathEvidence{InstanceID: a.instanceID, NICs: nics}
+}
+
+// lookupInstanceID asks for the instance ID without ever holding up a tick: the
+// call is bounded by imdsWait and, after a failure, retried only every
+// imdsRetryTicks ticks.
+func (a *Agent) lookupInstanceID() {
+	if a.idSkip > 0 {
+		a.idSkip--
+		return
+	}
+	type result struct {
+		id  string
+		err error
+	}
+	ch := make(chan result, 1)
+	go func() {
+		id, err := a.instanceIDFn()
+		ch <- result{id, err}
+	}()
+	timer := time.NewTimer(imdsWait)
+	defer timer.Stop()
+	select {
+	case r := <-ch:
+		if r.err == nil && r.id != "" {
+			a.instanceID = r.id
+			return
+		}
+	case <-timer.C:
+	}
+	a.idSkip = imdsRetryTicks - 1
+}
+
+// fetchInstanceID reads the instance ID from the in-VM metadata service. The
+// token request and the read share one deadline.
+func fetchInstanceID() (string, error) {
+	ctx, cancel := context.WithTimeout(context.Background(), imdsWait)
+	defer cancel()
+	c := &http.Client{}
+
+	var token string
+	if req, err := http.NewRequestWithContext(ctx, http.MethodPut, imdsBase+"/api/token", nil); err == nil {
+		req.Header.Set("X-Aws-Ec2-Metadata-Token-Ttl-Seconds", "60")
+		if resp, err := c.Do(req); err == nil {
+			b, _ := io.ReadAll(io.LimitReader(resp.Body, 256))
+			resp.Body.Close()
+			if resp.StatusCode == http.StatusOK {
+				token = strings.TrimSpace(string(b))
+			}
+		}
+	}
+
+	req, err := http.NewRequestWithContext(ctx, http.MethodGet, imdsBase+"/meta-data/instance-id", nil)
+	if err != nil {
+		return "", err
+	}
+	if token != "" {
+		req.Header.Set("X-Aws-Ec2-Metadata-Token", token)
+	}
+	resp, err := c.Do(req)
+	if err != nil {
+		return "", err
+	}
+	defer resp.Body.Close()
+	if resp.StatusCode != http.StatusOK {
+		return "", fmt.Errorf("instance-id: status %d", resp.StatusCode)
+	}
+	b, err := io.ReadAll(io.LimitReader(resp.Body, 256))
+	if err != nil {
+		return "", err
+	}
+	id := strings.TrimSpace(string(b))
+	if id == "" {
+		return "", fmt.Errorf("instance-id: empty")
+	}
+	return id, nil
+}

spinifex/agents/lambda/hostlink/configdisk.go

@@ -0,0 +1,71 @@
+package hostlink
+
+import (
+	"errors"
+	"fmt"
+	"io"
+	"os"
+
+	lambdav1 "github.com/mulgadc/spinifex/contracts/lambda/v1"
+	"golang.org/x/sys/unix"
+)
+
+// On a VMM without fw_cfg the executor attaches the same entries as a tar
+// archive on a second disk, which the runtime image's init extracts here in
+// the fw_cfg sysfs layout, so the fw_cfg readers work unchanged.
+const (
+	ConfigDiskRoot   = "/spinifex-cfg"
+	ConfigDiskDir    = ConfigDiskRoot + "/by_name/" + lambdav1.FwCfgPrefix
+	ConfigDiskDevice = "/dev/vdb"
+)
+
+// zeroChunk bounds each write while erasing the configuration disk.
+const zeroChunk = 64 << 10
+
+// unmount is a var so tests without privilege can stand in for the tmpfs.
+var unmount = unix.Unmount
+
+// SealConfigDisk overwrites the whole configuration disk with zeros, then
+// removes the extracted tree. The tree stays if the disk cannot be erased, so
+// SealFwCfg's check on root fails closed.
+func SealConfigDisk(device, root string) error {
+	if err := zeroDevice(device); err != nil {
+		return fmt.Errorf("hostlink: erase configuration disk: %w", err)
+	}
+	if err := unmount(root, 0); err != nil && !errors.Is(err, unix.EINVAL) {
+		return fmt.Errorf("hostlink: unmount %s: %w", root, err)
+	}
+	if err := os.RemoveAll(root); err != nil {
+		return fmt.Errorf("hostlink: remove %s: %w", root, err)
+	}
+	return nil
+}
+
+func zeroDevice(device string) error {
+	f, err := os.OpenFile(device, os.O_WRONLY, 0)
+	if err != nil {
+		return err
+	}
+	size, err := f.Seek(0, io.SeekEnd)
+	if err == nil {
+		_, err = f.Seek(0, io.SeekStart)
+	}
+	if err != nil {
+		_ = f.Close()
+		return err
+	}
+	zeros := make([]byte, zeroChunk)
+	for left := size; left > 0; {
+		n := min(left, int64(len(zeros)))
+		if _, err := f.Write(zeros[:n]); err != nil {
+			_ = f.Close()
+			return err
+		}
+		left -= n
+	}
+	if err := f.Sync(); err != nil {
+		_ = f.Close()
+		return err
+	}
+	return f.Close()
+}

spinifex/agents/lambda/hostlink/configdisk_test.go

@@ -0,0 +1,63 @@
+//test:in-package reaches the unexported configDisk and fwCfgTree readers to seal and read a real archive
+
+package hostlink
+
+import (
+	"bytes"
+	"os"
+	"path/filepath"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"golang.org/x/sys/unix"
+)
+
+// configDisk writes a fake disk holding secret and an extracted tree; the
+// stand-in unmount reports the tree is not a mount, as on a plain directory.
+func configDisk(t *testing.T, secret []byte) (device, root string) {
+	t.Helper()
+	dir := t.TempDir()
+	device = filepath.Join(dir, "vdb")
+	require.NoError(t, os.WriteFile(device, append(bytes.Repeat([]byte{0xa5}, 3*zeroChunk+7), secret...), 0o600))
+	root = fwCfgTree(t)
+	require.NoError(t, os.WriteFile(filepath.Join(root, "by_name", "key"), secret, 0o600))
+	old := unmount
+	unmount = func(string, int) error { return unix.EINVAL }
+	t.Cleanup(func() { unmount = old })
+	return device, root
+}
+
+func TestSealConfigDisk(t *testing.T) {
+	secret := []byte("guest-key-material")
+
+	t.Run("erases the disk and removes the tree", func(t *testing.T) {
+		device, root := configDisk(t, secret)
+		before, err := os.Stat(device)
+		require.NoError(t, err)
+		require.NoError(t, SealFwCfg(root, func() error { return SealConfigDisk(device, root) }))
+
+		disk, err := os.ReadFile(device)
+		require.NoError(t, err)
+		assert.Len(t, disk, int(before.Size()))
+		assert.Equal(t, make([]byte, len(disk)), disk, "configuration disk not fully erased")
+		_, err = os.Stat(root)
+		assert.ErrorIs(t, err, os.ErrNotExist)
+	})
+	t.Run("an unwritable disk keeps the tree and fails closed", func(t *testing.T) {
+		device, root := configDisk(t, secret)
+		require.NoError(t, os.Chmod(device, 0o400))
+		err := SealFwCfg(root, func() error { return SealConfigDisk(device, root) })
+		require.Error(t, err)
+		assert.Contains(t, err.Error(), "still readable")
+		_, err = os.Stat(root)
+		assert.NoError(t, err)
+	})
+	t.Run("a failed unmount keeps the tree", func(t *testing.T) {
+		device, root := configDisk(t, secret)
+		unmount = func(string, int) error { return unix.EBUSY }
+		require.Error(t, SealFwCfg(root, func() error { return SealConfigDisk(device, root) }))
+		_, err := os.Stat(root)
+		assert.NoError(t, err)
+	})
+}

spinifex/domains/ec2/accountcleanup/remainder.go

@@ -0,0 +1,155 @@
+package accountcleanup
+
+import (
+	"context"
+	"log/slog"
+	"slices"
+
+	"github.com/aws/aws-sdk-go/aws"
+	"github.com/aws/aws-sdk-go/service/ec2"
+	awserrors "github.com/mulgadc/spinifex/spinifex/foundation/aws/errors"
+	awsidentifiers "github.com/mulgadc/spinifex/spinifex/foundation/aws/identifiers"
+)
+
+// ReasonRecordRemains marks a resource whose record the owner still holds.
+const ReasonRecordRemains = "record-remains"
+
+// ComputeRemainder reads what compute the account still holds, changing nothing.
+type ComputeRemainder interface {
+	ComputeRemaining(ctx context.Context, accountID string) (*Outcome, error)
+}
+
+// StorageRemainder reads what volumes and tag residue the account still holds,
+// changing nothing.
+type StorageRemainder interface {
+	StorageRemaining(ctx context.Context, accountID string) (*Outcome, error)
+}
+
+var (
+	_ ComputeRemainder = (*ComputeCleanup)(nil)
+	_ StorageRemainder = (*StorageCleanup)(nil)
+)
+
+func validRemainderAccount(accountID string) error {
+	if accountID == "" {
+		return awserrors.Errorf(awserrors.ErrorMissingParameter, "The request must specify an account ID.")
+	}
+	if !awsidentifiers.IsAccountID(accountID) {
+		return awserrors.Errorf(awserrors.ErrorInvalidParameterValue, "The account ID must be 12 digits.")
+	}
+	return nil
+}
+
+// ComputeRemaining lists every instance the account still holds and every spot
+// request that can still launch. A read that does not complete is an incomplete
+// reason, never a shorter list. The result is a snapshot, not evidence of absence.
+func (c *ComputeCleanup) ComputeRemaining(ctx context.Context, accountID string) (*Outcome, error) {
+	if err := validRemainderAccount(accountID); err != nil {
+		return nil, err
+	}
+	if c == nil || c.spot == nil || c.instances == nil || c.teardown == nil {
+		return nil, awserrors.Errorf(awserrors.ErrorServerInternal, "Compute cleanup is not fully wired.")
+	}
+	res := &Outcome{AccountID: accountID}
+	defer res.finish()
+
+	scan, err := c.spot.ScanAccount(ctx, accountID)
+	if err != nil {
+		slog.WarnContext(ctx, "ComputeRemaining: spot scan failed", "account_id", accountID, "err", err)
+		res.incomplete(ReasonSpotInventoryIncomplete)
+	} else {
+		for _, r := range slices.Concat(scan.Active, scan.Terminal) {
+			if !terminalSpotState(r.State) {
+				res.Items = append(res.Items, Item{ResourceID: r.RequestID, Kind: KindSpotRequest, State: r.State, Disposition: DispositionResidue, Reason: ReasonRecordRemains})
+			}
+		}
+	}
+
+	inventory, err := c.instances.AccountInstances(ctx, accountID)
+	if err != nil || !inventoryUsable(inventory) {
+		slog.WarnContext(ctx, "ComputeRemaining: instance inventory incomplete", "account_id", accountID, "err", err)
+		res.incomplete(ReasonInstanceInventoryIncomplete)
+	} else {
+		for _, reservation := range inventory.Reservations {
+			for _, in := range reservation.Instances {
+				state := ""
+				if in.State != nil {
+					state = aws.StringValue(in.State.Name)
+				}
+				res.Items = append(res.Items, Item{ResourceID: aws.StringValue(in.InstanceId), Kind: KindInstance, State: state, Disposition: DispositionResidue, Reason: ReasonRecordRemains})
+			}
+		}
+	}
+
+	outstanding, err := c.teardown.AccountTeardownOutstanding(accountID)
+	if err != nil {
+		slog.WarnContext(ctx, "ComputeRemaining: teardown unread", "account_id", accountID, "err", err)
+		res.incomplete(ReasonTeardownUnread)
+	} else {
+		for _, o := range outstanding {
+			if !res.hasInstance(o.InstanceID) {
+				res.Items = append(res.Items, Item{ResourceID: o.InstanceID, Kind: KindInstance, State: o.State, Disposition: DispositionResidue, Reason: ReasonTerminating})
+			}
+		}
+	}
+	return res, nil
+}
+
+// StorageRemaining lists every volume record the account still holds, deleting
+// ones included, and the tag entries of resources whose owner holds no record.
+func (c *StorageCleanup) StorageRemaining(ctx context.Context, accountID string) (*Outcome, error) {
+	if err := validRemainderAccount(accountID); err != nil {
+		return nil, err
+	}
+	if c == nil || c.volumes == nil || c.instances == nil || c.teardown == nil || c.spot == nil || c.tags == nil {
+		return nil, awserrors.Errorf(awserrors.ErrorServerInternal, "Storage cleanup is not fully wired.")
+	}
+	res := &Outcome{AccountID: accountID}
+	defer res.finish()
+
+	volumes, err := c.volumes.VolumeInventory(ctx, accountID)
+	if err != nil {
+		slog.WarnContext(ctx, "StorageRemaining: volume inventory incomplete", "account_id", accountID, "err", err)
+		res.incomplete(ReasonVolumeInventoryIncomplete)
+	}
+	for _, v := range volumes {
+		reason := ReasonRecordRemains
+		if v.State == ec2.VolumeStateDeleting {
+			reason = ReasonVolumeDeleting
+		}
+		res.Items = append(res.Items, Item{ResourceID: v.VolumeID, Kind: KindVolume, State: v.State, Disposition: DispositionResidue, Reason: reason})
+	}
+
+	live, err := c.instances.AccountInstances(ctx, accountID)
+	if err != nil || !inventoryUsable(live) {
+		slog.WarnContext(ctx, "StorageRemaining: no complete instance view", "account_id", accountID, "err", err)
+		res.incomplete(ReasonHolderProofIncomplete)
+		return res, nil
+	}
+	outstanding, err := c.teardown.AccountTeardownOutstanding(accountID)
+	if err != nil {
+		slog.WarnContext(ctx, "StorageRemaining: teardown unread", "account_id", accountID, "err", err)
+		res.incomplete(ReasonHolderProofIncomplete)
+		return res, nil
+	}
+
+	tagged, err := c.tags.ScanAccountResourceIDs(ctx, accountID, storageTagPrefixes...)
+	if err != nil {
+		slog.WarnContext(ctx, "StorageRemaining: tag index read failed", "account_id", accountID, "err", err)
+		res.incomplete(ReasonTagIndexUnread)
+		return res, nil
+	}
+	owned, resolved := c.ownedResourceIDs(ctx, accountID, live, outstanding)
+	for _, id := range tagged {
+		switch {
+		case !resolved[tagPrefix(id)]:
+			res.incomplete(ReasonTagOwnerUnresolved)
+			res.Items = append(res.Items, Item{ResourceID: id, Kind: KindTag, Disposition: DispositionResidue, Reason: ReasonOwnerRecordsUnresolved})
+		case owned[id]:
+			// The resource itself is reported; its tags are not residue.
+		default:
+			res.Items = append(res.Items, Item{ResourceID: id, Kind: KindTag, Disposition: DispositionResidue, Reason: ReasonRecordRemains})
+		}
+	}
+	return res, nil
+}

spinifex/domains/ec2/accountcleanup/remainder_test.go

@@ -0,0 +1,97 @@
+package accountcleanup
+
+import (
+	"errors"
+	"testing"
+
+	"github.com/aws/aws-sdk-go/service/ec2"
+	"github.com/mulgadc/spinifex/spinifex/domains/ec2/accountinventory"
+	ec2instance "github.com/mulgadc/spinifex/spinifex/domains/ec2/instance"
+	ec2spotinstance "github.com/mulgadc/spinifex/spinifex/domains/ec2/spotinstance"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestComputeRemainingReportsEveryHeldInstanceAndOpenSpot(t *testing.T) {
+	r := newComputeRig()
+	r.spot.scans = []*ec2spotinstance.AccountScan{{
+		Active:   []ec2spotinstance.AccountSpotRequest{{RequestID: "sir-0aaaaaaaaaaaaaaa1", State: ec2.SpotInstanceStateOpen}},
+		Terminal: []ec2spotinstance.AccountSpotRequest{{RequestID: "sir-0aaaaaaaaaaaaaaa2", State: ec2.SpotInstanceStateClosed}},
+	}}
+	r.teardown.outstanding = []ec2instance.OutstandingInstance{{InstanceID: "i-0aaaaaaaaaaaaaa99", State: ec2.InstanceStateNameTerminated}}
+
+	res, err := r.cleanup().ComputeRemaining(t.Context(), testAccount)
+	require.NoError(t, err)
+
+	assert.Equal(t, StatusIncomplete, res.Status)
+	assert.NotNil(t, itemFor(res, runningID))
+	assert.NotNil(t, itemFor(res, stoppedID))
+	assert.NotNil(t, itemFor(res, "i-0aaaaaaaaaaaaaa99"))
+	assert.NotNil(t, itemFor(res, "sir-0aaaaaaaaaaaaaaa1"))
+	assert.Nil(t, itemFor(res, "sir-0aaaaaaaaaaaaaaa2"), "a closed request launches nothing")
+	assert.Empty(t, r.terminator.terminated, "a re-read changes nothing")
+}
+
+func TestComputeRemainingCleanWhenNothingHeld(t *testing.T) {
+	r := newComputeRig()
+	r.inventory.out = listing(nil)
+	res, err := r.cleanup().ComputeRemaining(t.Context(), testAccount)
+	require.NoError(t, err)
+	assert.Equal(t, StatusClean, res.Status)
+}
+
+func TestComputeRemainingUnreadStoreIsIncompleteNeverClean(t *testing.T) {
+	for name, mutate := range map[string]func(*computeRig){
+		"instances": func(r *computeRig) { r.inventory.out, r.inventory.err = nil, errors.New("node silent") },
+		"spot":      func(r *computeRig) { r.spot.scanErr = errors.New("spot unreadable") },
+		"teardown":  func(r *computeRig) { r.teardown.err = errors.New("no responder") },
+	} {
+		t.Run(name, func(t *testing.T) {
+			r := newComputeRig()
+			r.inventory.out = listing(nil)
+			mutate(r)
+			res, err := r.cleanup().ComputeRemaining(t.Context(), testAccount)
+			require.NoError(t, err)
+			assert.Equal(t, StatusIncomplete, res.Status)
+			assert.NotEmpty(t, res.IncompleteReasons)
+		})
+	}
+}
+
+func TestStorageRemainingReportsVolumeRecordsAndTagResidue(t *testing.T) {
+	r := newStorageRig()
+	r.volumes.vols = []accountinventory.VolumeItem{
+		volume(freeVol, ec2.VolumeStateAvailable, ""),
+		volume("vol-0aaaaaaaaaaaaaaa2", ec2.VolumeStateDeleting, ""),
+	}
+	r.tags.ids = []string{"i-0aaaaaaaaaaaaaaaf", freeVol}
+
+	res, err := r.cleanup().StorageRemaining(t.Context(), testAccount)
+	require.NoError(t, err)
+
+	assert.Equal(t, StatusIncomplete, res.Status)
+	assert.Equal(t, ReasonRecordRemains, itemFor(res, freeVol).Reason)
+	assert.Equal(t, ReasonVolumeDeleting, itemFor(res, "vol-0aaaaaaaaaaaaaaa2").Reason)
+	assert.Equal(t, KindTag, itemFor(res, "i-0aaaaaaaaaaaaaaaf").Kind, "tags of a resource with no record are residue")
+	assert.Empty(t, r.deleter.deleted, "a re-read deletes nothing")
+	assert.Empty(t, r.tags.deleted, "a re-read clears no tag")
+}
+
+func TestStorageRemainingUnreadStoreIsIncompleteNeverClean(t *testing.T) {
+	for name, mutate := range map[string]func(*storageRig){
+		"volumes":   func(r *storageRig) { r.volumes.failAt = -1 },
+		"instances": func(r *storageRig) { r.instances.out, r.instances.err = nil, errors.New("node silent") },
+		"teardown":  func(r *storageRig) { r.teardown.err = errors.New("no responder") },
+		"tags":      func(r *storageRig) { r.tags.scanErr = errors.New("tag index unreadable") },
+	} {
+		t.Run(name, func(t *testing.T) {
+			r := newStorageRig()
+			r.volumes.vols = nil
+			mutate(r)
+			res, err := r.cleanup().StorageRemaining(t.Context(), testAccount)
+			require.NoError(t, err)
+			assert.Equal(t, StatusIncomplete, res.Status)
+			assert.NotEmpty(t, res.IncompleteReasons)
+		})
+	}
+}

spinifex/domains/ec2/awsapi/instance/AccountTeardownOutstanding.go

@@ -0,0 +1,28 @@
+package instance
+
+import (
+	"context"
+	"fmt"
+	"time"
+
+	ec2instance "github.com/mulgadc/spinifex/spinifex/domains/ec2/instance"
+	natsmsg "github.com/mulgadc/spinifex/spinifex/foundation/messaging/nats"
+	"github.com/nats-io/nats.go"
+)
+
+// AccountTeardownOutstandingSubject serves the instances of an account whose
+// teardown is not complete, from the shared record store.
+const AccountTeardownOutstandingSubject = "ec2.AccountTeardownOutstanding"
+
+var accountTeardownOutstandingTimeout = 15 * time.Second
+
+// AccountTeardownOutstanding reads the account's unfinished instance teardowns.
+// No responder, a timeout or an error reply is an error, never an empty list.
+func AccountTeardownOutstanding(ctx context.Context, natsConn *nats.Conn, accountID string) ([]ec2instance.OutstandingInstance, error) {
+	out, err := natsmsg.NATSRequest[ec2instance.TeardownOutstandingOutput](ctx, natsConn, AccountTeardownOutstandingSubject,
+		&ec2instance.TeardownOutstandingInput{}, accountTeardownOutstandingTimeout, accountID)
+	if err != nil {
+		return nil, fmt.Errorf("read outstanding teardown: %w", err)
+	}
+	return out.Instances, nil
+}

spinifex/domains/ec2/awsapi/instance/AccountTeardownOutstanding_test.go

@@ -0,0 +1,88 @@
+package instance
+
+import (
+	"context"
+	"encoding/json"
+	"testing"
+	"time"
+
+	ec2instance "github.com/mulgadc/spinifex/spinifex/domains/ec2/instance"
+	awserrors "github.com/mulgadc/spinifex/spinifex/foundation/aws/errors"
+	"github.com/nats-io/nats.go"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func serveTeardownOutstanding(t *testing.T, nc *nats.Conn, reply []byte) {
+	t.Helper()
+	_, err := nc.Subscribe(AccountTeardownOutstandingSubject, func(msg *nats.Msg) { _ = msg.Respond(reply) })
+	require.NoError(t, err)
+	require.NoError(t, nc.Flush())
+}
+
+func TestAccountTeardownOutstanding_ReturnsTheServedInstances(t *testing.T) {
+	t.Parallel()
+	_, nc := startTestNATSServer(t)
+	reply, err := json.Marshal(ec2instance.TeardownOutstandingOutput{Instances: []ec2instance.OutstandingInstance{{InstanceID: "i-1"}}})
+	require.NoError(t, err)
+	serveTeardownOutstanding(t, nc, reply)
+
+	got, err := AccountTeardownOutstanding(context.Background(), nc, "123456789012")
+
+	require.NoError(t, err)
+	require.Len(t, got, 1)
+	assert.Equal(t, "i-1", got[0].InstanceID)
+}
+
+// A read that cannot be answered must never look like an empty list, because an
+// empty list lets the storage cleanup believe no QEMU still holds a volume.
+func TestAccountTeardownOutstanding_UnansweredReadIsAnError(t *testing.T) {
+	t.Parallel()
+
+	t.Run("no responder", func(t *testing.T) {
+		t.Parallel()
+		_, nc := startTestNATSServer(t)
+
+		got, err := AccountTeardownOutstanding(context.Background(), nc, "123456789012")
+
+		require.Error(t, err)
+		assert.Nil(t, got)
+		assert.ErrorIs(t, err, nats.ErrNoResponders)
+	})
+	t.Run("error reply", func(t *testing.T) {
+		t.Parallel()
+		_, nc := startTestNATSServer(t)
+		serveTeardownOutstanding(t, nc, awserrors.GenerateErrorPayload(awserrors.ErrorServerInternal))
+
+		got, err := AccountTeardownOutstanding(context.Background(), nc, "123456789012")
+
+		require.Error(t, err)
+		assert.Nil(t, got)
+	})
+	t.Run("unreadable reply", func(t *testing.T) {
+		t.Parallel()
+		_, nc := startTestNATSServer(t)
+		serveTeardownOutstanding(t, nc, []byte("not json"))
+
+		got, err := AccountTeardownOutstanding(context.Background(), nc, "123456789012")
+
+		require.Error(t, err)
+		assert.Nil(t, got)
+	})
+}
+
+// Not parallel: it shortens the package timeout.
+func TestAccountTeardownOutstanding_TimeoutIsAnError(t *testing.T) {
+	_, nc := startTestNATSServer(t)
+	_, err := nc.Subscribe(AccountTeardownOutstandingSubject, func(*nats.Msg) {})
+	require.NoError(t, err)
+	require.NoError(t, nc.Flush())
+	previous := accountTeardownOutstandingTimeout
+	accountTeardownOutstandingTimeout = 200 * time.Millisecond
+	t.Cleanup(func() { accountTeardownOutstandingTimeout = previous })
+
+	got, err := AccountTeardownOutstanding(context.Background(), nc, "123456789012")
+
+	require.Error(t, err)
+	assert.Nil(t, got)
+}

spinifex/domains/ec2/awsapi/instance/ModifyInstanceAttribute.go

@@ -11,6 +11,7 @@ import (
 	"time"
 
 	"github.com/aws/aws-sdk-go/service/ec2"
+	"github.com/mulgadc/spinifex/contracts/ec2/v1"
 	"github.com/mulgadc/spinifex/spinifex/foundation/aws/errors"
 	"github.com/nats-io/nats.go"
 )
@@ -65,6 +66,20 @@ func ModifyInstanceAttribute(ctx context.Context, input *ec2.ModifyInstanceAttri
 
 	slog.InfoContext(ctx, "ModifyInstanceAttribute: Processing request", "instance_id", *input.InstanceId)
 
+	// Termination protection on a running instance lives on its owner, so it
+	// goes there first. A nil value is not a value and keeps the queue path.
+	if input.DisableApiTermination != nil && input.DisableApiTermination.Value != nil {
+		cmd := ec2v1.EC2InstanceCommand{
+			ID:                        *input.InstanceId,
+			Attributes:                ec2v1.EC2CommandAttributes{SetDisableApiTermination: true},
+			DisableApiTerminationData: &ec2v1.DisableApiTerminationData{Value: *input.DisableApiTermination.Value},
+		}
+		if _, err := routeInstanceMutation[ec2.ModifyInstanceAttributeOutput](ctx, natsConn, *input.InstanceId, accountID, cmd, "ec2.ModifyInstanceAttribute", input); err != nil {
+			return ec2.ModifyInstanceAttributeOutput{}, err
+		}
+		return ec2.ModifyInstanceAttributeOutput{}, nil
+	}
+
 	jsonData, err := json.Marshal(input)
 	if err != nil {
 		slog.ErrorContext(ctx, "ModifyInstanceAttribute: Failed to marshal request", "instance_id", *input.InstanceId, "err", err)

spinifex/domains/ec2/awsapi/instance/ModifyInstanceAttribute_test.go

@@ -3,10 +3,13 @@ package instance
 import (
 	"context"
 	"encoding/json"
+	"sync"
+	"sync/atomic"
 	"testing"
 
 	"github.com/aws/aws-sdk-go/aws"
 	"github.com/aws/aws-sdk-go/service/ec2"
+	"github.com/mulgadc/spinifex/contracts/ec2/v1"
 	"github.com/mulgadc/spinifex/spinifex/foundation/aws/errors"
 	"github.com/nats-io/nats.go"
 	"github.com/stretchr/testify/assert"
@@ -226,3 +229,53 @@ func TestModifyInstanceAttribute_ValidationFailure(t *testing.T) {
 	require.Error(t, err)
 	assert.Equal(t, awserrors.ErrorInvalidParameterValue, err.Error())
 }
+
+func TestModifyInstanceAttribute_DisableApiTerminationRouting(t *testing.T) {
+	const id = "i-0123456789abcdef0"
+	_, nc := startTestNATSServer(t)
+
+	var gotValues []bool
+	var mu sync.Mutex
+	_, err := nc.Subscribe(ec2v1.InstanceCommandSubject(id), func(msg *nats.Msg) {
+		var cmd ec2v1.EC2InstanceCommand
+		if json.Unmarshal(msg.Data, &cmd) == nil && cmd.Attributes.SetDisableApiTermination && cmd.DisableApiTerminationData != nil {
+			mu.Lock()
+			gotValues = append(gotValues, cmd.DisableApiTerminationData.Value)
+			mu.Unlock()
+		}
+		reply := nats.NewMsg(msg.Reply)
+		reply.Data = []byte(`{}`)
+		reply.Header.Set(ec2v1.HandledHeader, "1")
+		_ = msg.RespondMsg(reply)
+	})
+	require.NoError(t, err)
+	var queue atomic.Int32
+	_, err = nc.Subscribe("ec2.ModifyInstanceAttribute", func(msg *nats.Msg) {
+		queue.Add(1)
+		_ = msg.Respond([]byte(`{}`))
+	})
+	require.NoError(t, err)
+
+	for _, v := range []bool{true, false} {
+		_, err := ModifyInstanceAttribute(context.Background(), &ec2.ModifyInstanceAttributeInput{
+			InstanceId:            aws.String(id),
+			DisableApiTermination: &ec2.AttributeBooleanValue{Value: aws.Bool(v)},
+		}, nc, "acct")
+		require.NoError(t, err)
+	}
+	mu.Lock()
+	assert.Equal(t, []bool{true, false}, gotValues, "explicit true and false both reach the owner")
+	mu.Unlock()
+	assert.Zero(t, queue.Load())
+
+	// A nil value is not a value: queue path, no owner command.
+	_, err = ModifyInstanceAttribute(context.Background(), &ec2.ModifyInstanceAttributeInput{
+		InstanceId:            aws.String(id),
+		DisableApiTermination: &ec2.AttributeBooleanValue{},
+	}, nc, "acct")
+	require.NoError(t, err)
+	assert.Equal(t, int32(1), queue.Load())
+	mu.Lock()
+	assert.Len(t, gotValues, 2, "no owner command for a nil value")
+	mu.Unlock()
+}

spinifex/domains/ec2/awsapi/instance/ModifyInstanceMetadataOptions.go

@@ -3,11 +3,11 @@ package instance
 import (
 	"context"
 	"errors"
-	"github.com/mulgadc/spinifex/spinifex/foundation/messaging/nats"
 	"strings"
-	"time"
 
+	"github.com/aws/aws-sdk-go/aws"
 	"github.com/aws/aws-sdk-go/service/ec2"
+	"github.com/mulgadc/spinifex/contracts/ec2/v1"
 	"github.com/mulgadc/spinifex/spinifex/foundation/aws/errors"
 	"github.com/nats-io/nats.go"
 )
@@ -22,5 +22,16 @@ func ModifyInstanceMetadataOptions(ctx context.Context, input *ec2.ModifyInstanc
 	if !strings.HasPrefix(*input.InstanceId, "i-") {
 		return nil, errors.New(awserrors.ErrorInvalidInstanceIDMalformed)
 	}
-	return natsmsg.NATSRequest[ec2.ModifyInstanceMetadataOptionsOutput](ctx, natsConn, "ec2.ModifyInstanceMetadataOptions", input, 30*time.Second, accountID)
+	cmd := ec2v1.EC2InstanceCommand{
+		ID:         *input.InstanceId,
+		Attributes: ec2v1.EC2CommandAttributes{ModifyInstanceMetadataOptions: true},
+		InstanceMetadataOptionsData: &ec2v1.InstanceMetadataOptionsData{
+			HttpTokens:              aws.StringValue(input.HttpTokens),
+			HttpEndpoint:            aws.StringValue(input.HttpEndpoint),
+			HttpProtocolIpv6:        aws.StringValue(input.HttpProtocolIpv6),
+			InstanceMetadataTags:    aws.StringValue(input.InstanceMetadataTags),
+			HttpPutResponseHopLimit: input.HttpPutResponseHopLimit,
+		},
+	}
+	return routeInstanceMutation[ec2.ModifyInstanceMetadataOptionsOutput](ctx, natsConn, *input.InstanceId, accountID, cmd, "ec2.ModifyInstanceMetadataOptions", input)
 }

spinifex/domains/ec2/awsapi/instance/ModifyInstanceMetadataOptions_test.go

@@ -2,11 +2,16 @@ package instance
 
 import (
 	"context"
+	"encoding/json"
+	"sync/atomic"
 	"testing"
+	"time"
 
 	"github.com/aws/aws-sdk-go/aws"
 	"github.com/aws/aws-sdk-go/service/ec2"
+	"github.com/mulgadc/spinifex/contracts/ec2/v1"
 	"github.com/mulgadc/spinifex/spinifex/foundation/aws/errors"
+	"github.com/nats-io/nats.go"
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
 )
@@ -30,3 +35,179 @@ func TestModifyInstanceMetadataOptions_RejectsMalformedID(t *testing.T) {
 		})
 	}
 }
+
+const metaTestID = "i-0123456789abcdef0"
+
+func metaInput(hop int64) *ec2.ModifyInstanceMetadataOptionsInput {
+	return &ec2.ModifyInstanceMetadataOptionsInput{
+		InstanceId:              aws.String(metaTestID),
+		HttpPutResponseHopLimit: aws.Int64(hop),
+	}
+}
+
+// ownerReply answers one ec2.cmd request, optionally carrying the handled marker.
+func ownerReply(t *testing.T, nc *nats.Conn, handled bool, body []byte, calls *atomic.Int32) {
+	t.Helper()
+	_, err := nc.Subscribe(ec2v1.InstanceCommandSubject(metaTestID), func(msg *nats.Msg) {
+		calls.Add(1)
+		reply := nats.NewMsg(msg.Reply)
+		reply.Data = body
+		if handled {
+			reply.Header.Set(ec2v1.HandledHeader, "1")
+		}
+		_ = msg.RespondMsg(reply)
+	})
+	require.NoError(t, err)
+}
+
+func queueReply(t *testing.T, nc *nats.Conn, body []byte, calls *atomic.Int32) {
+	t.Helper()
+	_, err := nc.QueueSubscribe("ec2.ModifyInstanceMetadataOptions", "spinifex-workers", func(msg *nats.Msg) {
+		calls.Add(1)
+		_ = msg.Respond(body)
+	})
+	require.NoError(t, err)
+}
+
+func metaOutputJSON(t *testing.T, hop int64) []byte {
+	t.Helper()
+	b, err := json.Marshal(&ec2.ModifyInstanceMetadataOptionsOutput{
+		InstanceId:              aws.String(metaTestID),
+		InstanceMetadataOptions: &ec2.InstanceMetadataOptionsResponse{HttpPutResponseHopLimit: aws.Int64(hop)},
+	})
+	require.NoError(t, err)
+	return b
+}
+
+func TestModifyInstanceMetadataOptions_OwnerAnswerReturnedAsIs(t *testing.T) {
+	_, nc := startTestNATSServer(t)
+	var owner, queue atomic.Int32
+	ownerReply(t, nc, true, metaOutputJSON(t, 3), &owner)
+	queueReply(t, nc, []byte(`{}`), &queue)
+
+	out, err := ModifyInstanceMetadataOptions(context.Background(), metaInput(3), nc, "acct")
+	require.NoError(t, err)
+	assert.Equal(t, int64(3), aws.Int64Value(out.InstanceMetadataOptions.HttpPutResponseHopLimit))
+	assert.Equal(t, int32(1), owner.Load())
+	assert.Zero(t, queue.Load())
+}
+
+func TestModifyInstanceMetadataOptions_OwnerErrorWithHeaderIsNotRetried(t *testing.T) {
+	_, nc := startTestNATSServer(t)
+	var owner, queue atomic.Int32
+	ownerReply(t, nc, true, awserrors.GenerateErrorPayload(awserrors.ErrorServerInternal), &owner)
+	queueReply(t, nc, metaOutputJSON(t, 3), &queue)
+
+	_, err := ModifyInstanceMetadataOptions(context.Background(), metaInput(3), nc, "acct")
+	require.Error(t, err)
+	assert.Equal(t, awserrors.ErrorServerInternal, err.Error())
+	assert.Zero(t, queue.Load())
+}
+
+func TestModifyInstanceMetadataOptions_OwnerValidationErrorRelayed(t *testing.T) {
+	_, nc := startTestNATSServer(t)
+	var owner, queue atomic.Int32
+	ownerReply(t, nc, true, awserrors.GenerateErrorPayload(awserrors.ErrorUnsupportedOperation), &owner)
+	queueReply(t, nc, metaOutputJSON(t, 3), &queue)
+
+	_, err := ModifyInstanceMetadataOptions(context.Background(), metaInput(3), nc, "acct")
+	require.Error(t, err)
+	assert.Equal(t, awserrors.ErrorUnsupportedOperation, err.Error())
+	assert.Zero(t, queue.Load())
+}
+
+func TestModifyInstanceMetadataOptions_HeaderlessOtherErrorReturnedAsIs(t *testing.T) {
+	_, nc := startTestNATSServer(t)
+	var owner, queue atomic.Int32
+	ownerReply(t, nc, false, awserrors.GenerateErrorPayload(awserrors.ErrorInvalidInstanceIDNotFound), &owner)
+	queueReply(t, nc, metaOutputJSON(t, 3), &queue)
+
+	_, err := ModifyInstanceMetadataOptions(context.Background(), metaInput(3), nc, "acct")
+	require.Error(t, err)
+	assert.Equal(t, awserrors.ErrorInvalidInstanceIDNotFound, err.Error())
+	assert.Zero(t, queue.Load())
+}
+
+func TestS9_N1OwnerRejectionFallsBackOnceToQueueGroup(t *testing.T) {
+	_, nc := startTestNATSServer(t)
+	var owner, queue atomic.Int32
+	ownerReply(t, nc, false, awserrors.GenerateErrorPayload(awserrors.ErrorServerInternal), &owner)
+	queueReply(t, nc, metaOutputJSON(t, 3), &queue)
+
+	out, err := ModifyInstanceMetadataOptions(context.Background(), metaInput(3), nc, "acct")
+	require.NoError(t, err)
+	assert.Equal(t, int64(3), aws.Int64Value(out.InstanceMetadataOptions.HttpPutResponseHopLimit))
+	assert.Equal(t, int32(1), owner.Load(), "exactly one owner request")
+	assert.Equal(t, int32(1), queue.Load(), "exactly one queue-group request")
+}
+
+func TestModifyInstanceMetadataOptions_NoOwnerUsesQueueGroupAfterThreeAttempts(t *testing.T) {
+	_, nc := startTestNATSServer(t)
+	var sleeps atomic.Int32
+	prev := cmdRetrySleep
+	cmdRetrySleep = func(time.Duration) { sleeps.Add(1) }
+	t.Cleanup(func() { cmdRetrySleep = prev })
+
+	var queue atomic.Int32
+	queueReply(t, nc, metaOutputJSON(t, 2), &queue)
+
+	out, err := ModifyInstanceMetadataOptions(context.Background(), metaInput(2), nc, "acct")
+	require.NoError(t, err)
+	assert.Equal(t, int64(2), aws.Int64Value(out.InstanceMetadataOptions.HttpPutResponseHopLimit))
+	assert.Equal(t, int32(2), sleeps.Load(), "three attempts means two backoffs")
+	assert.Equal(t, int32(1), queue.Load())
+}
+
+func TestModifyInstanceMetadataOptions_AbsentIsNotFound(t *testing.T) {
+	_, nc := startTestNATSServer(t)
+	var queue atomic.Int32
+	queueReply(t, nc, awserrors.GenerateErrorPayload(awserrors.ErrorInvalidInstanceIDNotFound), &queue)
+
+	_, err := ModifyInstanceMetadataOptions(context.Background(), metaInput(2), nc, "acct")
+	require.Error(t, err)
+	assert.Equal(t, awserrors.ErrorInvalidInstanceIDNotFound, err.Error())
+}
+
+func TestModifyInstanceMetadataOptions_OwnerTimeoutIsServerInternalWithoutFallback(t *testing.T) {
+	_, nc := startTestNATSServer(t)
+	prev := instanceCmdTimeout
+	instanceCmdTimeout = 50 * time.Millisecond
+	t.Cleanup(func() { instanceCmdTimeout = prev })
+
+	// The owner applies the change, then withholds its reply.
+	var applied, queue atomic.Int32
+	_, err := nc.Subscribe(ec2v1.InstanceCommandSubject(metaTestID), func(*nats.Msg) { applied.Add(1) })
+	require.NoError(t, err)
+	queueReply(t, nc, metaOutputJSON(t, 3), &queue)
+
+	_, err = ModifyInstanceMetadataOptions(context.Background(), metaInput(3), nc, "acct")
+	require.Error(t, err)
+	assert.Equal(t, awserrors.ErrorServerInternal, err.Error())
+	assert.Equal(t, int32(1), applied.Load(), "the owner saw the command, so the outcome is unknown")
+	assert.Zero(t, queue.Load(), "a timeout never falls back to the queue group")
+}
+
+func TestModifyInstanceMetadataOptions_TimeoutAfterApplyIsUnknownNotRolledBack(t *testing.T) {
+	_, nc := startTestNATSServer(t)
+	prev := instanceCmdTimeout
+	instanceCmdTimeout = 50 * time.Millisecond
+	t.Cleanup(func() { instanceCmdTimeout = prev })
+
+	// The owner persists the change, then the reply is lost.
+	var persisted atomic.Int64
+	_, err := nc.Subscribe(ec2v1.InstanceCommandSubject(metaTestID), func(msg *nats.Msg) {
+		var cmd ec2v1.EC2InstanceCommand
+		if json.Unmarshal(msg.Data, &cmd) == nil && cmd.InstanceMetadataOptionsData != nil {
+			persisted.Store(aws.Int64Value(cmd.InstanceMetadataOptionsData.HttpPutResponseHopLimit))
+		}
+	})
+	require.NoError(t, err)
+	var queue atomic.Int32
+	queueReply(t, nc, metaOutputJSON(t, 1), &queue)
+
+	_, err = ModifyInstanceMetadataOptions(context.Background(), metaInput(3), nc, "acct")
+	require.Error(t, err)
+	assert.Equal(t, awserrors.ErrorServerInternal, err.Error())
+	assert.Zero(t, queue.Load(), "no fallback that could race a live owner")
+	assert.Equal(t, int64(3), persisted.Load(), "the gateway error does not undo what the owner stored")
+}

spinifex/domains/ec2/awsapi/instance/instanceid.go

@@ -99,3 +99,46 @@ func classifyUnownedInstance(ctx context.Context, natsConn *nats.Conn, instanceI
 	}
 	return bucketStateAbsent, allOK
 }
+
+// routeInstanceMutation sends cmd to the instance's owner first and falls back
+// to the queue-group subject only when that is safe: no daemon runs the
+// instance (ErrNoResponders after every retry), or the owner is N-1 and
+// rejected the command without writing (ServerInternal with no HandledHeader).
+// A timeout or other transport error is an unknown outcome and never falls back.
+func routeInstanceMutation[Out any](ctx context.Context, natsConn *nats.Conn, instanceID, accountID string, cmd ec2v1.EC2InstanceCommand, queueSubject string, queueInput any) (*Out, error) {
+	queue := func() (*Out, error) {
+		return natsmsg.NATSRequest[Out](ctx, natsConn, queueSubject, queueInput, 30*time.Second, accountID)
+	}
+
+	data, err := json.Marshal(cmd)
+	if err != nil {
+		return nil, errors.New(awserrors.ErrorServerInternal)
+	}
+	msg, err := requestInstanceCmd(ctx, natsConn, instanceID, accountID, data)
+	if err != nil {
+		if errors.Is(err, nats.ErrNoResponders) {
+			return queue()
+		}
+		slog.ErrorContext(ctx, "Owner request failed, outcome unknown", "instance_id", instanceID, "subject", queueSubject, "err", err)
+		return nil, errors.New(awserrors.ErrorServerInternal)
+	}
+
+	responseError, parseErr := awserrors.ValidateErrorPayload(msg.Data)
+	if parseErr != nil {
+		handled := msg.Header.Get(ec2v1.HandledHeader) == "1"
+		if !handled && *responseError.Code == awserrors.ErrorServerInternal {
+			slog.WarnContext(ctx, "Owner did not recognise the command, using the queue group", "instance_id", instanceID, "subject", queueSubject)
+			return queue()
+		}
+		if responseError.Message != nil && *responseError.Message != "" {
+			return nil, awserrors.Errorf(*responseError.Code, "%s", *responseError.Message)
+		}
+		return nil, errors.New(*responseError.Code)
+	}
+
+	var out Out
+	if err := json.Unmarshal(msg.Data, &out); err != nil {
+		return nil, errors.New(awserrors.ErrorServerInternal)
+	}
+	return &out, nil
+}

spinifex/domains/ec2/instance/teardown_outstanding.go

@@ -1,12 +1,33 @@
 package instance
 
+import "context"
+
 // OutstandingInstance is an instance of an account that is not yet fully torn
 // down, with the state its record carries.
 type OutstandingInstance struct {
-	InstanceID string
+	InstanceID string `json:"instance_id"`
 	// State is the record's lifecycle state, such as running or terminated.
-	State string
+	State string `json:"state"`
 	// Teardown is the per-dependent progress (done, pending or failed) the
 	// record carries. It is empty when no dependent has been recorded.
-	Teardown map[string]string
+	Teardown map[string]string `json:"teardown,omitempty"`
+}
+
+// TeardownOutstandingInput is the empty request of the outstanding-teardown
+// read; the account comes from the message header.
+type TeardownOutstandingInput struct{}
+
+// TeardownOutstandingOutput carries every instance not yet fully torn down.
+type TeardownOutstandingOutput struct {
+	Instances []OutstandingInstance `json:"instances"`
+}
+
+// AccountTeardownOutstandingRPC serves AccountTeardownOutstanding over the
+// message bus, so a process without the instance record store can read it.
+func (s *InstanceServiceImpl) AccountTeardownOutstandingRPC(_ context.Context, _ *TeardownOutstandingInput, accountID string) (*TeardownOutstandingOutput, error) {
+	outstanding, err := s.AccountTeardownOutstanding(accountID)
+	if err != nil {
+		return nil, err
+	}
+	return &TeardownOutstandingOutput{Instances: outstanding}, nil
 }

spinifex/domains/ec2/systeminstance/launcher.go

@@ -57,7 +57,7 @@ const (
 	BootAMI BootMode = "ami"
 )
 
-// SystemInstanceLauncher launches and terminates system-managed VMs.
+// SystemInstanceLauncher launches, reboots and terminates system-managed VMs.
 // Defined here to avoid a circular import between elbv2/eks and daemon.
 type SystemInstanceLauncher interface {
 	// LaunchSystemInstance creates and starts a system-managed VM, returning its ID and private IP.
@@ -65,6 +65,10 @@ type SystemInstanceLauncher interface {
 
 	// TerminateSystemInstance stops and cleans up a system-managed VM.
 	TerminateSystemInstance(instanceID string) error
+
+	// RebootSystemInstance restarts a system-managed load balancer VM into a
+	// fresh QEMU on its owning node, keeping its ID, ENIs and addresses.
+	RebootSystemInstance(instanceID string) error
 }
 
 // SystemInstanceInput describes a system VM to launch.

spinifex/domains/elbv2/agent_heartbeat_wire_test.go

@@ -0,0 +1,73 @@
+package elbv2
+
+import (
+	"testing"
+
+	"github.com/mulgadc/spinifex/spinifex/ingress/aws/query"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestLBAgentHeartbeatInput_OldAgentDecodesUnchanged(t *testing.T) {
+	t.Parallel()
+	// Exactly the parameters an agent without the data-path probe sends.
+	params := map[string]string{
+		"Action":                   "LBAgentHeartbeat",
+		"Version":                  "2015-12-01",
+		"LBID":                     "lb-abc123",
+		"Servers.member.1.Backend": "bk_tg1",
+		"Servers.member.1.Server":  "srv_i-web1",
+		"Servers.member.1.Status":  "UP",
+		"Servers.member.2.Backend": "bk_tg1",
+		"Servers.member.2.Server":  "srv_i-web2",
+		"Servers.member.2.Status":  "DOWN",
+	}
+
+	var in LBAgentHeartbeatInput
+	require.NoError(t, query.QueryParamsToStruct(params, &in))
+
+	require.NotNil(t, in.LBID)
+	assert.Equal(t, "lb-abc123", *in.LBID)
+	require.Len(t, in.Servers, 2)
+	assert.Equal(t, "DOWN", *in.Servers[1].Status)
+	assert.Nil(t, in.InstanceID, "an old agent reports no instance ID")
+	assert.Empty(t, in.NICs, "an old agent reports no NICs")
+	assert.Len(t, in.toHealthReport().Servers, 2)
+}
+
+func TestLBAgentHeartbeatInput_DecodesDataPathEvidence(t *testing.T) {
+	t.Parallel()
+	params := map[string]string{
+		"LBID":                              "lb-abc123",
+		"InstanceID":                        "i-0123456789abcdef0",
+		"NICs.member.1.MAC":                 "02:00:00:aa:00:01",
+		"NICs.member.1.ProbeOK":             "false",
+		"NICs.member.1.ConsecutiveFailures": "4",
+		"NICs.member.1.RxPackets":           "100",
+		"NICs.member.1.TxPackets":           "200",
+		"NICs.member.1.RxDropped":           "3",
+		"NICs.member.1.TxDropped":           "9",
+		"NICs.member.2.MAC":                 "02:00:00:aa:00:03",
+		"NICs.member.2.ProbeOK":             "true",
+		"NICs.member.2.ConsecutiveFailures": "0",
+	}
+
+	var in LBAgentHeartbeatInput
+	require.NoError(t, query.QueryParamsToStruct(params, &in))
+
+	require.NotNil(t, in.InstanceID)
+	assert.Equal(t, "i-0123456789abcdef0", *in.InstanceID)
+	require.Len(t, in.NICs, 2)
+
+	n := in.NICs[0]
+	assert.Equal(t, "02:00:00:aa:00:01", *n.MAC)
+	assert.False(t, *n.ProbeOK)
+	assert.EqualValues(t, 4, *n.ConsecutiveFailures)
+	assert.EqualValues(t, 100, *n.RxPackets)
+	assert.EqualValues(t, 200, *n.TxPackets)
+	assert.EqualValues(t, 3, *n.RxDropped)
+	assert.EqualValues(t, 9, *n.TxDropped)
+
+	assert.True(t, *in.NICs[1].ProbeOK)
+	assert.Nil(t, in.NICs[1].RxPackets, "an omitted counter stays unset")
+}

spinifex/domains/elbv2/agent_types.go

@@ -7,9 +7,25 @@ import (
 
 // LBAgentHeartbeatInput is sent by the LB agent on each heartbeat tick,
 // including its health report so the daemon can update target health.
+//
+// InstanceID and NICs are optional: an older agent sends neither.
 type LBAgentHeartbeatInput struct {
-	LBID    *string                `locationName:"LBID" type:"string"`
-	Servers []*LBAgentServerStatus `locationName:"Servers" type:"list"`
+	LBID       *string                `locationName:"LBID" type:"string"`
+	Servers    []*LBAgentServerStatus `locationName:"Servers" type:"list"`
+	InstanceID *string                `locationName:"InstanceID" type:"string"`
+	NICs       []*LBAgentNICStatus    `locationName:"NICs" type:"list"`
+}
+
+// LBAgentNICStatus is the gateway probe result and interface counters of one
+// data NIC, as reported by the LB agent.
+type LBAgentNICStatus struct {
+	MAC                 *string `locationName:"MAC" type:"string"`
+	ProbeOK             *bool   `locationName:"ProbeOK" type:"boolean"`
+	ConsecutiveFailures *int64  `locationName:"ConsecutiveFailures" type:"long"`
+	RxPackets           *int64  `locationName:"RxPackets" type:"long"`
+	TxPackets           *int64  `locationName:"TxPackets" type:"long"`
+	RxDropped           *int64  `locationName:"RxDropped" type:"long"`
+	TxDropped           *int64  `locationName:"TxDropped" type:"long"`
 }
 
 // LBAgentServerStatus represents a single backend server's health status as reported by the LB agent.

spinifex/domains/elbv2/launcher_test.go

@@ -14,6 +14,9 @@ type mockSystemInstanceLauncher struct {
 	launchErr      error
 	terminateErr   error
 	terminateDone  chan struct{} // closed after TerminateSystemInstance completes
+	onTerminate    func(instanceID string)
+	rebootCalls    []string
+	rebootErr      error
 }
 
 func (m *mockSystemInstanceLauncher) LaunchSystemInstance(input *SystemInstanceInput) (*SystemInstanceOutput, error) {
@@ -25,12 +28,22 @@ func (m *mockSystemInstanceLauncher) TerminateSystemInstance(instanceID string)
 	m.mu.Lock()
 	m.terminateCalls = append(m.terminateCalls, instanceID)
 	m.mu.Unlock()
+	if m.onTerminate != nil {
+		m.onTerminate(instanceID)
+	}
 	if m.terminateDone != nil {
 		close(m.terminateDone)
 	}
 	return m.terminateErr
 }
 
+func (m *mockSystemInstanceLauncher) RebootSystemInstance(instanceID string) error {
+	m.mu.Lock()
+	m.rebootCalls = append(m.rebootCalls, instanceID)
+	m.mu.Unlock()
+	return m.rebootErr
+}
+
 // waitTerminate blocks until TerminateSystemInstance has been called.
 func (m *mockSystemInstanceLauncher) waitTerminate() {
 	if m.terminateDone != nil {

spinifex/domains/elbv2/lb_converge_test.go

@@ -0,0 +1,192 @@
+package elbv2
+
+import (
+	"context"
+	"testing"
+	"time"
+
+	"github.com/aws/aws-sdk-go/aws"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestLBRepair_AnotherReservationWinsTheFirstCAS(t *testing.T) {
+	t.Parallel()
+	e := newRepairEnv(t)
+	lb := e.seed("lbr20", true)
+	cands := e.candidates()
+	require.Len(t, cands, 1)
+	require.True(t, e.rec.eligible(context.Background(), cands[0], e.clock))
+
+	_, err := e.svc.store.UpdateLBHealth(context.Background(), lb.LoadBalancerID, func(rec *LBDataPathRecord) (bool, error) {
+		rec.Repairs = append(rec.Repairs, LBRepair{At: e.clock, InstanceID: lb.InstanceID, Outcome: RepairReserved})
+		return true, nil
+	})
+	require.NoError(t, err)
+
+	assert.False(t, e.rec.repair(context.Background(), cands[0]))
+	assert.Empty(t, e.reboots())
+	assert.Len(t, e.health(lb.LoadBalancerID).Repairs, 1, "the losing reservation must not add an entry")
+}
+
+func TestLBRepair_LeaseLostAfterReservationKeepsItCountedAndDoesNotDispatch(t *testing.T) {
+	t.Parallel()
+	e := newRepairEnv(t)
+	e.seed("lbr21", true)
+	cands := e.candidates()
+	require.Len(t, cands, 1)
+
+	calls := 0
+	e.rec.isLeader = func() bool {
+		calls++
+		return calls == 1
+	}
+	assert.True(t, e.rec.repair(context.Background(), cands[0]), "a written reservation is reported as spent")
+	assert.Empty(t, e.reboots())
+
+	repairs := e.health("lbr21").Repairs
+	require.Len(t, repairs, 1)
+	assert.Equal(t, RepairReserved, repairs[0].Outcome, "the entry stays counted with no outcome")
+}
+
+func TestLBRepair_NonLeaderNeverReserves(t *testing.T) {
+	t.Parallel()
+	e := newRepairEnv(t)
+	e.seed("lbr22", true)
+	cands := e.candidates()
+	require.Len(t, cands, 1)
+
+	e.rec.isLeader = func() bool { return false }
+	assert.False(t, e.rec.repair(context.Background(), cands[0]))
+	assert.Empty(t, e.health("lbr22").Repairs)
+}
+
+func TestLBHealth_FailedResumeThenNewVMHeartbeatsConverge(t *testing.T) {
+	t.Parallel()
+	svc := setupTestService(t)
+	lb := seedHealthLB(t, svc)
+	ctx := context.Background()
+
+	// The old VM's evidence was suspended by SetSubnets, whose resume write then failed.
+	_, err := svc.store.UpdateLBHealth(ctx, lb.LoadBalancerID, func(rec *LBDataPathRecord) (bool, error) {
+		rec.InstanceID = "i-oldvm"
+		rec.Suspended = true
+		rec.SuspendedAt = time.Now().UTC()
+		return true, nil
+	})
+	require.NoError(t, err)
+
+	healthBeat(t, svc, aws.String(healthInst), nicStatus(healthMACData, true, 0, 10))
+	rec, err := svc.store.GetLBHealth(ctx, lb.LoadBalancerID)
+	require.NoError(t, err)
+	assert.False(t, rec.Suspended, "a heartbeat from the VM the LB names lifts the stale suspension")
+	assert.Equal(t, healthInst, rec.InstanceID)
+
+	healthBeat(t, svc, aws.String(healthInst), nicStatus(healthMACData, false, 3, 11))
+	rec, err = svc.store.GetLBHealth(ctx, lb.LoadBalancerID)
+	require.NoError(t, err)
+	require.Contains(t, rec.NICs, healthMACData, "a later NIC failure is recorded and repairable")
+}
+
+func TestLBHealth_SuspensionIsNotLiftedByAHeartbeatFromAVMTheLBDoesNotName(t *testing.T) {
+	t.Parallel()
+	svc := setupTestService(t)
+	lb := seedHealthLB(t, svc)
+	ctx := context.Background()
+
+	_, err := svc.store.UpdateLBHealth(ctx, lb.LoadBalancerID, func(rec *LBDataPathRecord) (bool, error) {
+		rec.InstanceID = "i-oldvm"
+		rec.Suspended = true
+		rec.SuspendedAt = time.Now().UTC()
+		return true, nil
+	})
+	require.NoError(t, err)
+
+	healthBeat(t, svc, aws.String("i-oldvm"), nicStatus(healthMACData, false, 3, 1))
+	rec, err := svc.store.GetLBHealth(ctx, lb.LoadBalancerID)
+	require.NoError(t, err)
+	assert.True(t, rec.Suspended)
+	assert.Equal(t, "i-oldvm", rec.InstanceID)
+}
+
+func TestLBHealth_SuspensionBackstopLiftsOnlyAfterThirtyMinutes(t *testing.T) {
+	t.Parallel()
+	svc := setupTestService(t)
+	lb := seedHealthLB(t, svc)
+	ctx := context.Background()
+
+	suspend := func(age time.Duration) {
+		_, err := svc.store.UpdateLBHealth(ctx, lb.LoadBalancerID, func(rec *LBDataPathRecord) (bool, error) {
+			rec.InstanceID = healthInst
+			rec.Suspended = true
+			rec.SuspendedAt = time.Now().UTC().Add(-age)
+			return true, nil
+		})
+		require.NoError(t, err)
+	}
+
+	suspend(29 * time.Minute)
+	healthBeat(t, svc, aws.String(healthInst), nicStatus(healthMACData, false, 3, 1))
+	rec, err := svc.store.GetLBHealth(ctx, lb.LoadBalancerID)
+	require.NoError(t, err)
+	assert.True(t, rec.Suspended, "inside the backstop the suspension holds")
+
+	suspend(31 * time.Minute)
+	healthBeat(t, svc, aws.String(healthInst), nicStatus(healthMACData, false, 3, 2))
+	rec, err = svc.store.GetLBHealth(ctx, lb.LoadBalancerID)
+	require.NoError(t, err)
+	assert.False(t, rec.Suspended, "past the backstop the current VM's heartbeat lifts it")
+	assert.Contains(t, rec.NICs, healthMACData)
+}
+
+func TestLBHealth_EstablishedFailureSurvivesAnAgentRestart(t *testing.T) {
+	t.Parallel()
+	svc := setupTestService(t)
+	lb := seedHealthLB(t, svc)
+	ctx := context.Background()
+
+	healthBeat(t, svc, aws.String(healthInst), nicStatus(healthMACData, false, 3, 1))
+	first, err := svc.store.GetLBHealth(ctx, lb.LoadBalancerID)
+	require.NoError(t, err)
+	since := first.NICs[healthMACData].FailingSince
+
+	// A restarted agent reports healthy until it has missed three probes again.
+	healthBeat(t, svc, aws.String(healthInst), nicStatus(healthMACData, true, 1, 2))
+	healthBeat(t, svc, aws.String(healthInst), nicStatus(healthMACData, true, 2, 3))
+	rec, err := svc.store.GetLBHealth(ctx, lb.LoadBalancerID)
+	require.NoError(t, err)
+	require.Contains(t, rec.NICs, healthMACData)
+	assert.True(t, since.Equal(rec.NICs[healthMACData].FailingSince), "FailingSince is kept")
+
+	healthBeat(t, svc, aws.String(healthInst), nicStatus(healthMACData, true, 0, 4))
+	rec, err = svc.store.GetLBHealth(ctx, lb.LoadBalancerID)
+	require.NoError(t, err)
+	assert.Empty(t, rec.NICs, "only a clean probe clears the failure")
+}
+
+func TestLBHealth_OldVMHeartbeatPausedAcrossPublicationLeavesTheNewBinding(t *testing.T) {
+	t.Parallel()
+	svc := setupTestService(t)
+	lb := seedHealthLB(t, svc)
+	ctx := context.Background()
+
+	// SetSubnets has already published the new VM and rebound the evidence.
+	lb.InstanceID = "i-newvm"
+	require.NoError(t, svc.store.PutLoadBalancer(ctx, lb))
+	require.NoError(t, svc.store.ResumeLBHealth(ctx, lb.LoadBalancerID, "i-newvm"))
+
+	// The old VM's heartbeat read the LB before that and runs now.
+	stale := *lb
+	stale.InstanceID = healthInst
+	svc.recordDataPathEvidence(ctx, &stale, &LBAgentHeartbeatInput{
+		LBID:       aws.String(lb.LoadBalancerID),
+		InstanceID: aws.String(healthInst),
+		NICs:       []*LBAgentNICStatus{nicStatus(healthMACData, false, 5, 1)},
+	})
+
+	rec, err := svc.store.GetLBHealth(ctx, lb.LoadBalancerID)
+	require.NoError(t, err)
+	require.NotNil(t, rec)
+	assert.Equal(t, "i-newvm", rec.InstanceID)
+	assert.Empty(t, rec.NICs, "the old VM's failure must not land on the new binding")
+}

spinifex/domains/elbv2/lb_health.go

@@ -0,0 +1,333 @@
+package elbv2
+
+import (
+	"context"
+	"encoding/json"
+	"errors"
+	"fmt"
+	"log/slog"
+	"strings"
+	"time"
+
+	"github.com/aws/aws-sdk-go/aws"
+	"github.com/mulgadc/spinifex/spinifex/foundation/state/kvutil"
+	"github.com/nats-io/nats.go/jetstream"
+)
+
+const (
+	// KeyPrefixLBHealth holds the data-path evidence of one load balancer. It is
+	// a separate key from the LB record so whole-record writes cannot drop it.
+	KeyPrefixLBHealth = "lbhealth."
+
+	// StateActiveImpaired is what DescribeLoadBalancers reports while a data NIC
+	// of the LB's current VM is failing its gateway probe.
+	StateActiveImpaired = "active_impaired"
+
+	// lbSuspendBackstop is how long a suspension may outlive its mutation before
+	// a heartbeat from the current VM lifts it.
+	lbSuspendBackstop = 30 * time.Minute
+)
+
+// Outcomes of a recorded repair. Reserved is written before the repair is
+// dispatched, so an entry whose outcome was never updated still counts.
+const (
+	RepairReserved    = "reserved"
+	RepairDispatched  = "dispatched"
+	RepairRefused     = "refused"
+	RepairUnreachable = "unreachable"
+)
+
+// LBNICFailure is one data NIC that was failing its gateway probe when the
+// transition was recorded.
+type LBNICFailure struct {
+	ENIID               string    `json:"eni_id,omitempty"`
+	FailingSince        time.Time `json:"failing_since"`
+	ConsecutiveFailures int64     `json:"consecutive_failures"`
+	RxPackets           int64     `json:"rx_packets"`
+	TxPackets           int64     `json:"tx_packets"`
+	RxDropped           int64     `json:"rx_dropped"`
+	TxDropped           int64     `json:"tx_dropped"`
+}
+
+// LBRepair is one repair of an LB's VM and how it ended.
+type LBRepair struct {
+	At         time.Time `json:"at"`
+	InstanceID string    `json:"instance_id"`
+	ENIID      string    `json:"eni_id,omitempty"`
+	Reason     string    `json:"reason,omitempty"`
+	Outcome    string    `json:"outcome"`
+}
+
+// LBDataPathRecord is the data-path evidence of one load balancer VM. NICs holds
+// only failing NICs, keyed by lower-case MAC; an absent NIC is healthy.
+type LBDataPathRecord struct {
+	InstanceID  string                   `json:"instance_id"`
+	Suspended   bool                     `json:"suspended,omitempty"`
+	SuspendedAt time.Time                `json:"suspended_at,omitzero"`
+	BoundAt     time.Time                `json:"bound_at,omitzero"`
+	NICs        map[string]*LBNICFailure `json:"nics,omitempty"`
+	Repairs     []LBRepair               `json:"repairs,omitempty"`
+}
+
+// failing reports whether any NIC of the record is failing.
+func (r *LBDataPathRecord) failing() bool {
+	return r != nil && len(r.NICs) > 0
+}
+
+// GetLBHealth returns the evidence record of a load balancer, or nil when none
+// has been written.
+func (s *Store) GetLBHealth(ctx context.Context, lbID string) (*LBDataPathRecord, error) {
+	entry, err := s.kv.Get(ctx, KeyPrefixLBHealth+lbID)
+	if err != nil {
+		if errors.Is(err, jetstream.ErrKeyNotFound) {
+			return nil, nil
+		}
+		return nil, err
+	}
+	var rec LBDataPathRecord
+	if err := json.Unmarshal(entry.Value(), &rec); err != nil {
+		return nil, fmt.Errorf("unmarshal lb health: %w", err)
+	}
+	return &rec, nil
+}
+
+// UpdateLBHealth applies mutate to the evidence record under optimistic
+// concurrency, creating it when absent. mutate returns false to write nothing.
+func (s *Store) UpdateLBHealth(ctx context.Context, lbID string, mutate func(*LBDataPathRecord) (bool, error)) (*LBDataPathRecord, error) {
+	return kvutil.Update(ctx, s.kv, KeyPrefixLBHealth+lbID, kvutil.CASConfig{
+		Attempts:       targetGroupCASAttempts,
+		CreateIfAbsent: true,
+		Exhausted: func(_ string, attempts int) error {
+			return fmt.Errorf("elbv2: lb health %s contended, CAS exhausted after %d attempts", lbID, attempts)
+		},
+	}, mutate)
+}
+
+// DeleteLBHealth removes the evidence record of a load balancer.
+func (s *Store) DeleteLBHealth(ctx context.Context, lbID string) error {
+	err := s.kv.Delete(ctx, KeyPrefixLBHealth+lbID)
+	if err != nil && !errors.Is(err, jetstream.ErrKeyNotFound) {
+		return err
+	}
+	return nil
+}
+
+// SuspendLBHealth marks the load balancer as being torn down or reshaped, so
+// no repair is admitted against its VM until the mutation finishes.
+func (s *Store) SuspendLBHealth(ctx context.Context, lbID string) error {
+	_, err := s.UpdateLBHealth(ctx, lbID, func(rec *LBDataPathRecord) (bool, error) {
+		if rec.Suspended {
+			return false, nil
+		}
+		rec.Suspended = true
+		rec.SuspendedAt = time.Now().UTC()
+		return true, nil
+	})
+	return err
+}
+
+// ResumeLBHealth lifts a suspension. A non-empty instanceID also binds the
+// record to that VM and drops the old VM's failing NICs.
+func (s *Store) ResumeLBHealth(ctx context.Context, lbID, instanceID string) error {
+	_, err := s.UpdateLBHealth(ctx, lbID, func(rec *LBDataPathRecord) (bool, error) {
+		changed := rec.Suspended
+		rec.Suspended = false
+		rec.SuspendedAt = time.Time{}
+		if instanceID != "" && rec.InstanceID != instanceID {
+			rec.InstanceID = instanceID
+			rec.BoundAt = time.Now().UTC()
+			rec.NICs = nil
+			changed = true
+		}
+		return changed, nil
+	})
+	return err
+}
+
+// failingNICMACs returns the lower-case MACs that are failing. A probe that
+// failed starts a failure. An established failure ends only when the latest
+// probe succeeded, because an agent restarted by a repair reports healthy for
+// its first misses.
+func failingNICMACs(nics []*LBAgentNICStatus, established map[string]*LBNICFailure) map[string]*LBAgentNICStatus {
+	out := make(map[string]*LBAgentNICStatus)
+	for _, n := range nics {
+		if n == nil || n.MAC == nil || n.ProbeOK == nil {
+			continue
+		}
+		mac := strings.ToLower(*n.MAC)
+		if !*n.ProbeOK || (established[mac] != nil && aws.Int64Value(n.ConsecutiveFailures) != 0) {
+			out[mac] = n
+		}
+	}
+	return out
+}
+
+// recordDataPathEvidence stores the data-path evidence of one heartbeat. It
+// writes only when the set of failing NICs changes, and ignores evidence that
+// does not come from the VM the LB record currently names.
+func (s *ELBv2ServiceImpl) recordDataPathEvidence(ctx context.Context, lb *LoadBalancerRecord, input *LBAgentHeartbeatInput) {
+	if lb.InstanceID == "" || input.InstanceID == nil || *input.InstanceID != lb.InstanceID || len(input.NICs) == 0 {
+		return
+	}
+
+	rec, err := s.store.GetLBHealth(ctx, lb.LoadBalancerID)
+	if err != nil {
+		slog.WarnContext(ctx, "LBAgentHeartbeat: failed to read data-path evidence", "lbId", lb.LoadBalancerID, "err", err)
+		return
+	}
+	converge := false
+	if rec != nil && rec.Suspended {
+		if !s.suspensionConverges(ctx, lb, rec, *input.InstanceID) {
+			return
+		}
+		converge = true
+	}
+	var known map[string]*LBNICFailure
+	if rec != nil && rec.InstanceID == lb.InstanceID {
+		known = rec.NICs
+	}
+	failing := failingNICMACs(input.NICs, known)
+	if !converge && sameMACs(known, failing) && (len(failing) == 0 || rec != nil) {
+		return
+	}
+
+	eniByMAC := s.resolveLBENIs(ctx, lb, failing)
+	now := time.Now().UTC()
+	_, err = s.store.UpdateLBHealth(ctx, lb.LoadBalancerID, func(cur *LBDataPathRecord) (bool, error) {
+		changed := false
+		if cur.Suspended {
+			if !converge || !suspensionLifts(cur, lb.InstanceID, now) {
+				return false, nil
+			}
+			cur.Suspended = false
+			cur.SuspendedAt = time.Time{}
+			changed = true
+			if cur.InstanceID != lb.InstanceID {
+				cur.InstanceID = lb.InstanceID
+				cur.BoundAt = now
+				cur.NICs = nil
+			}
+		}
+		if cur.InstanceID != "" && cur.InstanceID != lb.InstanceID {
+			return changed, nil
+		}
+		prev := cur.NICs
+		failing := failingNICMACs(input.NICs, prev)
+		if cur.InstanceID == lb.InstanceID && sameMACs(prev, failing) {
+			return changed, nil
+		}
+		next := make(map[string]*LBNICFailure, len(failing))
+		for mac, n := range failing {
+			f := &LBNICFailure{
+				ENIID:               eniByMAC[mac],
+				FailingSince:        now,
+				ConsecutiveFailures: aws.Int64Value(n.ConsecutiveFailures),
+				RxPackets:           aws.Int64Value(n.RxPackets),
+				TxPackets:           aws.Int64Value(n.TxPackets),
+				RxDropped:           aws.Int64Value(n.RxDropped),
+				TxDropped:           aws.Int64Value(n.TxDropped),
+			}
+			if old, ok := prev[mac]; ok {
+				f.FailingSince = old.FailingSince
+			}
+			next[mac] = f
+		}
+		cur.InstanceID = lb.InstanceID
+		cur.NICs = next
+		return true, nil
+	})
+	if err != nil {
+		slog.WarnContext(ctx, "LBAgentHeartbeat: failed to store data-path evidence", "lbId", lb.LoadBalancerID, "err", err)
+		return
+	}
+	for mac, n := range failing {
+		if _, ok := known[mac]; !ok {
+			slog.WarnContext(ctx, "ELBv2 data-path probe failing",
+				"lbArn", lb.LoadBalancerArn, "lbId", lb.LoadBalancerID, "instanceId", lb.InstanceID,
+				"eniId", eniByMAC[mac], "mac", mac, "consecutive_failures", aws.Int64Value(n.ConsecutiveFailures),
+				"rx_packets", aws.Int64Value(n.RxPackets), "tx_packets", aws.Int64Value(n.TxPackets),
+				"rx_dropped", aws.Int64Value(n.RxDropped), "tx_dropped", aws.Int64Value(n.TxDropped))
+		}
+	}
+	for mac := range known {
+		if _, ok := failing[mac]; !ok {
+			slog.InfoContext(ctx, "ELBv2 data-path probe recovered",
+				"lbArn", lb.LoadBalancerArn, "lbId", lb.LoadBalancerID, "instanceId", lb.InstanceID, "mac", mac)
+		}
+	}
+}
+
+// suspensionLifts reports whether a suspended record may be lifted by a
+// heartbeat from instanceID: it is bound to an older VM, or it has outlived the
+// backstop while bound to the current one.
+func suspensionLifts(rec *LBDataPathRecord, instanceID string, now time.Time) bool {
+	if rec.InstanceID != "" && rec.InstanceID != instanceID {
+		return true
+	}
+	return !rec.SuspendedAt.IsZero() && now.Sub(rec.SuspendedAt) >= lbSuspendBackstop
+}
+
+// suspensionConverges confirms against a fresh read of the LB record that the
+// VM the heartbeat reports is the one the LB currently names.
+func (s *ELBv2ServiceImpl) suspensionConverges(ctx context.Context, lb *LoadBalancerRecord, rec *LBDataPathRecord, heartbeatInstance string) bool {
+	if !suspensionLifts(rec, heartbeatInstance, time.Now().UTC()) {
+		return false
+	}
+	fresh, err := s.store.GetLoadBalancer(ctx, lb.LoadBalancerID)
+	return err == nil && fresh != nil && fresh.InstanceID == heartbeatInstance
+}
+
+// sameMACs reports whether two failing sets name the same NICs.
+func sameMACs[A, B any](a map[string]A, b map[string]B) bool {
+	if len(a) != len(b) {
+		return false
+	}
+	for mac := range a {
+		if _, ok := b[mac]; !ok {
+			return false
+		}
+	}
+	return true
+}
+
+// resolveLBENIs maps failing MACs to ENI IDs from the LB's cross-account ENIs
+// and one describe of its own ENIs. A MAC it cannot resolve maps to nothing.
+func (s *ELBv2ServiceImpl) resolveLBENIs(ctx context.Context, lb *LoadBalancerRecord, failing map[string]*LBAgentNICStatus) map[string]string {
+	out := make(map[string]string, len(failing))
+	for _, x := range lb.CrossAccountENIs {
+		mac := strings.ToLower(x.ENIMac)
+		if _, ok := failing[mac]; ok {
+			out[mac] = x.ENIID
+		}
+	}
+	if len(out) == len(failing) {
+		return out
+	}
+	for id, eni := range s.describeENIs(ctx, lb.ENIs, lb.AccountID) {
+		mac := strings.ToLower(aws.StringValue(eni.MacAddress))
+		if _, ok := failing[mac]; ok && out[mac] == "" {
+			out[mac] = id
+		}
+	}
+	return out
+}
+
+// impairedByEvidence reports whether the LB's current VM has a failing data NIC.
+func (s *ELBv2ServiceImpl) impairedByEvidence(ctx context.Context, r *LoadBalancerRecord) bool {
+	if r.State != StateActive || r.InstanceID == "" {
+		return false
+	}
+	rec, err := s.store.GetLBHealth(ctx, r.LoadBalancerID)
+	if err != nil {
+		slog.DebugContext(ctx, "lbRecordToSDK: failed to read data-path evidence", "lbId", r.LoadBalancerID, "err", err)
+		return false
+	}
+	return rec != nil && !rec.Suspended && rec.InstanceID == r.InstanceID && rec.failing()
+}
+
+// resumeLBHealth lifts a suspension after a failure that left the VM untouched.
+func (s *ELBv2ServiceImpl) resumeLBHealth(ctx context.Context, lbID string) {
+	if err := s.store.ResumeLBHealth(ctx, lbID, ""); err != nil {
+		slog.WarnContext(ctx, "failed to lift data-path suspension", "lbId", lbID, "err", err)
+	}
+}

spinifex/domains/elbv2/lb_health_test.go

@@ -0,0 +1,370 @@
+package elbv2
+
+import (
+	"context"
+	"errors"
+	"sync/atomic"
+	"testing"
+	"time"
+
+	"github.com/aws/aws-sdk-go/aws"
+	"github.com/aws/aws-sdk-go/service/elbv2"
+	"github.com/nats-io/nats.go/jetstream"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+const (
+	healthLBID    = "lbhealth01"
+	healthInst    = "i-lbvm01"
+	healthMACData = "02:00:00:aa:00:01"
+	healthMACX    = "02:00:00:aa:00:02"
+)
+
+func seedHealthLB(t *testing.T, svc *ELBv2ServiceImpl) *LoadBalancerRecord {
+	t.Helper()
+	lb := &LoadBalancerRecord{
+		LoadBalancerArn: "arn:aws:elasticloadbalancing:us-east-1:" + testAccountID + ":loadbalancer/app/health-lb/" + healthLBID,
+		LoadBalancerID:  healthLBID,
+		Name:            "health-lb",
+		State:           StateActive,
+		InstanceID:      healthInst,
+		AccountID:       testAccountID,
+		CrossAccountENIs: []ExtraENIInput{
+			{ENIID: "eni-cross01", ENIMac: healthMACX},
+		},
+		CreatedAt: time.Now().UTC(),
+	}
+	require.NoError(t, svc.store.PutLoadBalancer(context.Background(), lb))
+	return lb
+}
+
+func nicStatus(mac string, ok bool, failures, rx int64) *LBAgentNICStatus {
+	return &LBAgentNICStatus{
+		MAC:                 aws.String(mac),
+		ProbeOK:             aws.Bool(ok),
+		ConsecutiveFailures: aws.Int64(failures),
+		RxPackets:           aws.Int64(rx),
+		TxPackets:           aws.Int64(rx + 5),
+		RxDropped:           aws.Int64(0),
+		TxDropped:           aws.Int64(2),
+	}
+}
+
+func healthBeat(t *testing.T, svc *ELBv2ServiceImpl, instanceID *string, nics ...*LBAgentNICStatus) {
+	t.Helper()
+	_, err := svc.LBAgentHeartbeat(context.Background(), &LBAgentHeartbeatInput{
+		LBID:       aws.String(healthLBID),
+		InstanceID: instanceID,
+		NICs:       nics,
+	}, testAccountID)
+	require.NoError(t, err)
+}
+
+func healthRevision(t *testing.T, svc *ELBv2ServiceImpl) uint64 {
+	t.Helper()
+	entry, err := svc.store.kv.Get(context.Background(), KeyPrefixLBHealth+healthLBID)
+	require.NoError(t, err)
+	return entry.Revision()
+}
+
+func TestLBHealth_SteadyHeartbeatsWriteNothing(t *testing.T) {
+	t.Parallel()
+	svc := setupTestService(t)
+	seedHealthLB(t, svc)
+
+	for range 5 {
+		healthBeat(t, svc, aws.String(healthInst), nicStatus(healthMACData, true, 0, 100))
+	}
+	rec, err := svc.store.GetLBHealth(context.Background(), healthLBID)
+	require.NoError(t, err)
+	assert.Nil(t, rec, "healthy heartbeats must not create an evidence record")
+
+	healthBeat(t, svc, aws.String(healthInst), nicStatus(healthMACData, false, 3, 100))
+	rev := healthRevision(t, svc)
+	for i := range 5 {
+		healthBeat(t, svc, aws.String(healthInst), nicStatus(healthMACData, false, int64(4+i), int64(100+i)))
+	}
+	assert.Equal(t, rev, healthRevision(t, svc), "a NIC that stays failing must not rewrite the record")
+}
+
+func TestLBHealth_FailingNICWritesFailingSinceOnceAndRecoveryClears(t *testing.T) {
+	t.Parallel()
+	svc := setupTestService(t)
+	seedHealthLB(t, svc)
+
+	before := time.Now().UTC().Add(-time.Second)
+	healthBeat(t, svc, aws.String(healthInst), nicStatus(healthMACData, false, 3, 777), nicStatus(healthMACX, true, 0, 1))
+
+	rec, err := svc.store.GetLBHealth(context.Background(), healthLBID)
+	require.NoError(t, err)
+	require.NotNil(t, rec)
+	assert.Equal(t, healthInst, rec.InstanceID)
+	require.Len(t, rec.NICs, 1)
+	f := rec.NICs[healthMACData]
+	require.NotNil(t, f)
+	assert.True(t, f.FailingSince.After(before))
+	assert.EqualValues(t, 777, f.RxPackets)
+	assert.EqualValues(t, 782, f.TxPackets)
+	assert.EqualValues(t, 2, f.TxDropped)
+	assert.EqualValues(t, 3, f.ConsecutiveFailures)
+
+	healthBeat(t, svc, aws.String(healthInst), nicStatus(healthMACData, true, 0, 900))
+	rec, err = svc.store.GetLBHealth(context.Background(), healthLBID)
+	require.NoError(t, err)
+	require.NotNil(t, rec)
+	assert.Empty(t, rec.NICs, "recovery clears the failing NIC")
+}
+
+func TestLBHealth_SecondNICFailingKeepsFirstFailingSince(t *testing.T) {
+	t.Parallel()
+	svc := setupTestService(t)
+	seedHealthLB(t, svc)
+
+	healthBeat(t, svc, aws.String(healthInst), nicStatus(healthMACData, false, 3, 1))
+	first, err := svc.store.GetLBHealth(context.Background(), healthLBID)
+	require.NoError(t, err)
+	since := first.NICs[healthMACData].FailingSince
+
+	healthBeat(t, svc, aws.String(healthInst), nicStatus(healthMACData, false, 9, 2), nicStatus(healthMACX, false, 3, 2))
+	rec, err := svc.store.GetLBHealth(context.Background(), healthLBID)
+	require.NoError(t, err)
+	require.Len(t, rec.NICs, 2)
+	assert.True(t, since.Equal(rec.NICs[healthMACData].FailingSince))
+	assert.Equal(t, "eni-cross01", rec.NICs[healthMACX].ENIID, "a cross-account NIC resolves to its ENI without a describe")
+}
+
+func TestLBHealth_WrongOrMissingInstanceIDIsIgnored(t *testing.T) {
+	t.Parallel()
+	svc := setupTestService(t)
+	seedHealthLB(t, svc)
+
+	failing := nicStatus(healthMACData, false, 5, 1)
+	healthBeat(t, svc, aws.String("i-someoneelse"), failing)
+	healthBeat(t, svc, aws.String(""), failing)
+	healthBeat(t, svc, nil, failing)
+
+	rec, err := svc.store.GetLBHealth(context.Background(), healthLBID)
+	require.NoError(t, err)
+	assert.Nil(t, rec, "evidence from another VM, or with no instance ID, must be ignored")
+}
+
+func TestLBHealth_WholeRecordPutLeavesEvidenceIntact(t *testing.T) {
+	t.Parallel()
+	svc := setupTestService(t)
+	lb := seedHealthLB(t, svc)
+
+	healthBeat(t, svc, aws.String(healthInst), nicStatus(healthMACData, false, 3, 1))
+	want, err := svc.store.GetLBHealth(context.Background(), healthLBID)
+	require.NoError(t, err)
+	rev := healthRevision(t, svc)
+
+	lb.StateReason = "rewritten by an older daemon"
+	require.NoError(t, svc.store.PutLoadBalancer(context.Background(), lb))
+
+	got, err := svc.store.GetLBHealth(context.Background(), healthLBID)
+	require.NoError(t, err)
+	assert.Equal(t, want, got)
+	assert.Equal(t, rev, healthRevision(t, svc))
+}
+
+func TestLBHealth_DeleteLoadBalancerRemovesTheKey(t *testing.T) {
+	t.Parallel()
+	svc := setupTestService(t)
+	lb := seedHealthLB(t, svc)
+	healthBeat(t, svc, aws.String(healthInst), nicStatus(healthMACData, false, 3, 1))
+
+	_, err := svc.DeleteLoadBalancer(context.Background(), &elbv2.DeleteLoadBalancerInput{
+		LoadBalancerArn: aws.String(lb.LoadBalancerArn),
+	}, testAccountID)
+	require.NoError(t, err)
+
+	rec, err := svc.store.GetLBHealth(context.Background(), healthLBID)
+	require.NoError(t, err)
+	assert.Nil(t, rec)
+}
+
+func TestLBHealth_DeleteLoadBalancerSuspendsBeforeTerminatingTheVM(t *testing.T) {
+	t.Parallel()
+	svc := setupTestService(t)
+	lb := seedHealthLB(t, svc)
+	seen := make(chan bool, 1)
+	svc.InstanceLauncher = &mockSystemInstanceLauncher{
+		onTerminate: func(string) {
+			rec, err := svc.store.GetLBHealth(context.Background(), healthLBID)
+			seen <- err == nil && rec != nil && rec.Suspended
+		},
+	}
+
+	_, err := svc.DeleteLoadBalancer(context.Background(), &elbv2.DeleteLoadBalancerInput{
+		LoadBalancerArn: aws.String(lb.LoadBalancerArn),
+	}, testAccountID)
+	require.NoError(t, err)
+
+	select {
+	case suspended := <-seen:
+		assert.True(t, suspended, "repair must be excluded before the VM is terminated")
+	case <-time.After(5 * time.Second):
+		t.Fatal("the LB VM was never terminated")
+	}
+}
+
+func TestLBHealth_SetSubnetsSuspendsFirstAndResumesOnTheNewVM(t *testing.T) {
+	t.Parallel()
+	svc, vpcSvc, mock := setupSubnetTestService(t)
+	vid := vpcID(t, vpcSvc)
+	sub1 := getTestSubnetID(t, vpcSvc, vid, "10.0.40.0/24", "us-east-1a")
+	sub2 := getTestSubnetID(t, vpcSvc, vid, "10.0.41.0/24", "us-east-1b")
+
+	out, err := svc.CreateLoadBalancer(context.Background(), &elbv2.CreateLoadBalancerInput{
+		Name:    aws.String("health-sub-lb"),
+		Subnets: []*string{aws.String(sub1)},
+	}, testAccountID)
+	require.NoError(t, err)
+	svc.WaitLaunches()
+	lbID := aws.StringValue(out.LoadBalancers[0].LoadBalancerArn)
+	rec0, err := svc.store.GetLoadBalancerByArn(context.Background(), lbID)
+	require.NoError(t, err)
+	require.Equal(t, "i-lb", rec0.InstanceID)
+
+	_, err = svc.LBAgentHeartbeat(context.Background(), &LBAgentHeartbeatInput{
+		LBID:       aws.String(rec0.LoadBalancerID),
+		InstanceID: aws.String("i-lb"),
+		NICs:       []*LBAgentNICStatus{nicStatus(healthMACData, false, 3, 1)},
+	}, testAccountID)
+	require.NoError(t, err)
+
+	suspendedAtTerminate := false
+	mock.onTerminate = func(string) {
+		rec, gerr := svc.store.GetLBHealth(context.Background(), rec0.LoadBalancerID)
+		suspendedAtTerminate = gerr == nil && rec != nil && rec.Suspended
+	}
+	mock.launchResult = &SystemInstanceOutput{InstanceID: "i-new", PrivateIP: "10.0.1.6"}
+
+	_, err = svc.SetSubnets(context.Background(), &elbv2.SetSubnetsInput{
+		LoadBalancerArn: aws.String(lbID),
+		Subnets:         []*string{aws.String(sub1), aws.String(sub2)},
+	}, testAccountID)
+	require.NoError(t, err)
+	assert.True(t, suspendedAtTerminate, "repair must be excluded before the VM is terminated")
+
+	rec, err := svc.store.GetLBHealth(context.Background(), rec0.LoadBalancerID)
+	require.NoError(t, err)
+	require.NotNil(t, rec)
+	assert.False(t, rec.Suspended, "the suspension is lifted once the new VM is published")
+	assert.Equal(t, "i-new", rec.InstanceID)
+	assert.Empty(t, rec.NICs, "the old VM's failing NIC does not carry over")
+}
+
+func TestLBHealth_SetSubnetsFailureBeforeTheVMIsTouchedResumes(t *testing.T) {
+	t.Parallel()
+	svc, vpcSvc, mock := setupSubnetTestService(t)
+	vid := vpcID(t, vpcSvc)
+	sub1 := getTestSubnetID(t, vpcSvc, vid, "10.0.42.0/24", "us-east-1a")
+	sub2 := getTestSubnetID(t, vpcSvc, vid, "10.0.43.0/24", "us-east-1b")
+
+	out, err := svc.CreateLoadBalancer(context.Background(), &elbv2.CreateLoadBalancerInput{
+		Name:    aws.String("health-resume-lb"),
+		Subnets: []*string{aws.String(sub1)},
+	}, testAccountID)
+	require.NoError(t, err)
+	svc.WaitLaunches()
+	arn := aws.StringValue(out.LoadBalancers[0].LoadBalancerArn)
+	rec0, err := svc.store.GetLoadBalancerByArn(context.Background(), arn)
+	require.NoError(t, err)
+
+	mock.terminateErr = assert.AnError
+	_, err = svc.SetSubnets(context.Background(), &elbv2.SetSubnetsInput{
+		LoadBalancerArn: aws.String(arn),
+		Subnets:         []*string{aws.String(sub1), aws.String(sub2)},
+	}, testAccountID)
+	require.Error(t, err)
+
+	rec, err := svc.store.GetLBHealth(context.Background(), rec0.LoadBalancerID)
+	require.NoError(t, err)
+	require.NotNil(t, rec)
+	assert.False(t, rec.Suspended, "a failed reshape that left the VM running must not leave repair excluded")
+}
+
+func TestLBHealth_DescribeShowsImpairedOnlyForFailingNICOfCurrentVM(t *testing.T) {
+	t.Parallel()
+	svc := setupTestService(t)
+	lb := seedHealthLB(t, svc)
+	code := func() string {
+		out, err := svc.DescribeLoadBalancers(context.Background(), &elbv2.DescribeLoadBalancersInput{
+			LoadBalancerArns: []*string{aws.String(lb.LoadBalancerArn)},
+		}, testAccountID)
+		require.NoError(t, err)
+		require.Len(t, out.LoadBalancers, 1)
+		return aws.StringValue(out.LoadBalancers[0].State.Code)
+	}
+
+	assert.Equal(t, StateActive, code())
+
+	healthBeat(t, svc, aws.String(healthInst), nicStatus(healthMACData, false, 3, 1))
+	assert.Equal(t, StateActiveImpaired, code())
+
+	healthBeat(t, svc, aws.String(healthInst), nicStatus(healthMACData, true, 0, 2))
+	assert.Equal(t, StateActive, code(), "active again once the probe recovers")
+
+	// Evidence recorded against a VM the LB no longer names never impairs it.
+	_, err := svc.store.UpdateLBHealth(context.Background(), healthLBID, func(r *LBDataPathRecord) (bool, error) {
+		r.InstanceID = "i-oldvm"
+		r.NICs = map[string]*LBNICFailure{healthMACData: {FailingSince: time.Now().UTC()}}
+		return true, nil
+	})
+	require.NoError(t, err)
+	assert.Equal(t, StateActive, code())
+
+	// A LB that is not active is reported as it is.
+	_, err = svc.store.UpdateLBHealth(context.Background(), healthLBID, func(r *LBDataPathRecord) (bool, error) {
+		r.InstanceID = healthInst
+		return true, nil
+	})
+	require.NoError(t, err)
+	require.Equal(t, StateActiveImpaired, code())
+	_, err = svc.store.UpdateLoadBalancer(context.Background(), healthLBID, func(r *LoadBalancerRecord) (bool, error) {
+		r.State = StateFailed
+		return true, nil
+	})
+	require.NoError(t, err)
+	assert.Equal(t, StateFailed, code())
+}
+
+// failKeysKV fails key listing once armed, which is what the listener cascade
+// of DeleteLoadBalancer reads first.
+type failKeysKV struct {
+	jetstream.KeyValue
+
+	armed atomic.Bool
+}
+
+func (k *failKeysKV) Keys(ctx context.Context, opts ...jetstream.WatchOpt) ([]string, error) {
+	if k.armed.Load() {
+		return nil, errors.New("injected key listing failure")
+	}
+	return k.KeyValue.Keys(ctx, opts...)
+}
+
+func TestLBHealth_DeleteLoadBalancerListenerCascadeFailureResumes(t *testing.T) {
+	t.Parallel()
+	svc := setupTestService(t)
+	lb := seedHealthLB(t, svc)
+	kv := &failKeysKV{KeyValue: svc.store.kv}
+	svc.store.kv = kv
+	mock := &mockSystemInstanceLauncher{}
+	svc.InstanceLauncher = mock
+	kv.armed.Store(true)
+
+	_, err := svc.DeleteLoadBalancer(context.Background(), &elbv2.DeleteLoadBalancerInput{
+		LoadBalancerArn: aws.String(lb.LoadBalancerArn),
+	}, testAccountID)
+	require.Error(t, err)
+	kv.armed.Store(false)
+
+	rec, err := svc.store.GetLBHealth(context.Background(), healthLBID)
+	require.NoError(t, err)
+	require.NotNil(t, rec)
+	assert.False(t, rec.Suspended, "a delete that failed before touching the VM must not leave repair excluded")
+	assert.Empty(t, mock.terminateCalls)
+}

spinifex/domains/elbv2/lb_repair.go

@@ -0,0 +1,377 @@
+package elbv2
+
+import (
+	"context"
+	"errors"
+	"log/slog"
+	"sort"
+	"strings"
+	"sync"
+	"time"
+
+	"github.com/mulgadc/spinifex/spinifex/domains/ec2/systeminstance"
+	"github.com/mulgadc/spinifex/spinifex/foundation/state/kvlease"
+	"github.com/mulgadc/spinifex/spinifex/foundation/telemetry"
+	"github.com/nats-io/nats.go"
+	"github.com/nats-io/nats.go/jetstream"
+)
+
+const (
+	// KVBucketELBv2Leader holds the lease key of the data-path reconciler.
+	KVBucketELBv2Leader        = "spinifex-elbv2-leader"
+	KVBucketELBv2LeaderVersion = 1
+	KVBucketELBv2LeaderTTL     = 60 * time.Second
+
+	lbRepairLeaderKey     = "datapath"
+	lbRepairLeaseRenew    = 20 * time.Second
+	lbRepairPassInterval  = 30 * time.Second
+	lbRepairMinFailing    = 60 * time.Second
+	lbRepairMinVMAge      = 5 * time.Minute
+	lbRepairCooldown      = 10 * time.Minute
+	lbRepairBudgetWindow  = 24 * time.Hour
+	lbRepairBudget        = 3
+	lbRepairFleetMinCount = 3
+
+	// A heartbeat that persists only every heartbeatPersistInterval is still
+	// fresh for twice that long.
+	lbRepairHeartbeatMaxAge = 2 * heartbeatPersistInterval
+)
+
+// SystemInstanceRebooter is the only capability the repair path has over a VM.
+type SystemInstanceRebooter interface {
+	RebootSystemInstance(instanceID string) error
+}
+
+var _ SystemInstanceRebooter = SystemInstanceLauncher(nil)
+
+// counts reports whether a repair entry uses up cooldown and budget. Only a
+// repair no node received is free.
+func (r LBRepair) counts() bool {
+	return r.Outcome != RepairUnreachable
+}
+
+// repairWindow counts the entries inside window ending at now, and returns the
+// newest of them.
+func repairWindow(repairs []LBRepair, now time.Time, window time.Duration) (n int, newest time.Time) {
+	for _, r := range repairs {
+		if !r.counts() || now.Sub(r.At) >= window {
+			continue
+		}
+		n++
+		if r.At.After(newest) {
+			newest = r.At
+		}
+	}
+	return n, newest
+}
+
+// lbRepairReconciler restarts the VM of a load balancer whose data NIC has been
+// failing its gateway probe. Only the lease holder acts.
+type lbRepairReconciler struct {
+	svc      *ELBv2ServiceImpl
+	rebooter SystemInstanceRebooter
+	isLeader func() bool
+	now      func() time.Time
+
+	mu             sync.Mutex
+	budgetReported map[string]bool
+}
+
+func newLBRepairReconciler(svc *ELBv2ServiceImpl, rebooter SystemInstanceRebooter, isLeader func() bool) *lbRepairReconciler {
+	return &lbRepairReconciler{
+		svc:            svc,
+		rebooter:       rebooter,
+		isLeader:       isLeader,
+		now:            func() time.Time { return time.Now().UTC() },
+		budgetReported: make(map[string]bool),
+	}
+}
+
+// StartDataPathReconciler elects a leader among the daemons and runs the repair
+// pass on it every lbRepairPassInterval until ctx or the service is cancelled.
+func (s *ELBv2ServiceImpl) StartDataPathReconciler(ctx context.Context, holder string) {
+	if s.InstanceLauncher == nil || s.store == nil || s.store.js == nil {
+		return
+	}
+	lease, err := kvlease.New(kvlease.Config{
+		Name:   "elbv2/datapath",
+		Bucket: s.leaderBucket,
+		Key:    lbRepairLeaderKey,
+		Holder: holder,
+		TTL:    KVBucketELBv2LeaderTTL,
+		Renew:  lbRepairLeaseRenew,
+		Retry:  lbRepairLeaseRenew,
+	})
+	if err != nil {
+		slog.ErrorContext(ctx, "ELBv2 data-path reconciler: lease config invalid", "err", err)
+		return
+	}
+	rec := newLBRepairReconciler(s, s.InstanceLauncher, lease.Held)
+	go lease.Run(ctx)
+	go func() {
+		ticker := time.NewTicker(lbRepairPassInterval)
+		defer ticker.Stop()
+		for {
+			select {
+			case <-s.ctx.Done():
+				return
+			case <-ctx.Done():
+				return
+			case <-ticker.C:
+				rec.pass(ctx)
+			}
+		}
+	}()
+}
+
+func (s *ELBv2ServiceImpl) leaderBucket(ctx context.Context) (jetstream.KeyValue, error) {
+	return kvlease.OpenBucket(ctx, s.store.js, kvlease.BucketConfig{
+		Name:    KVBucketELBv2Leader,
+		TTL:     KVBucketELBv2LeaderTTL,
+		Version: KVBucketELBv2LeaderVersion,
+	})
+}
+
+// candidate is a load balancer that is failing and may be repaired.
+type lbRepairCandidate struct {
+	lb    *LoadBalancerRecord
+	rec   *LBDataPathRecord
+	mac   string
+	nic   *LBNICFailure
+	since time.Time
+}
+
+// pass runs one repair pass. It starts at most one repair.
+func (r *lbRepairReconciler) pass(ctx context.Context) {
+	if !r.isLeader() {
+		return
+	}
+	lbs, err := r.svc.store.ListLoadBalancersStrict(ctx)
+	if err != nil {
+		slog.WarnContext(ctx, "ELBv2 data-path reconciler: failed to list load balancers", "err", err)
+		return
+	}
+	now := r.now()
+
+	var heartbeating, failing int
+	var cands []lbRepairCandidate
+	live := make(map[string]bool, len(lbs))
+	for _, lb := range lbs {
+		live[lb.LoadBalancerID] = true
+		if lb.State != StateActive || lb.InstanceID == "" || now.Sub(lb.LastHeartbeat) >= lbRepairHeartbeatMaxAge {
+			continue
+		}
+		heartbeating++
+		rec, err := r.svc.store.GetLBHealth(ctx, lb.LoadBalancerID)
+		if err != nil || rec == nil || rec.Suspended || rec.InstanceID != lb.InstanceID || !rec.failing() {
+			continue
+		}
+		failing++
+		c := lbRepairCandidate{lb: lb, rec: rec}
+		for mac, nic := range rec.NICs {
+			if c.nic == nil || nic.FailingSince.Before(c.since) {
+				c.mac, c.nic, c.since = mac, nic, nic.FailingSince
+			}
+		}
+		cands = append(cands, c)
+	}
+	r.forgetBudgetReports(live)
+
+	if failing >= lbRepairFleetMinCount && failing*2 > heartbeating {
+		slog.WarnContext(ctx, "ELBv2 data-path repair suppressed: fleet-wide probe failure",
+			"failing_lbs", failing, "heartbeating_lbs", heartbeating)
+		return
+	}
+
+	sort.Slice(cands, func(i, j int) bool { return cands[i].since.Before(cands[j].since) })
+	for _, c := range cands {
+		if r.eligible(ctx, c, now) && r.repair(ctx, c) {
+			return
+		}
+	}
+}
+
+// eligible applies the per-LB conditions that need no write.
+func (r *lbRepairReconciler) eligible(ctx context.Context, c lbRepairCandidate, now time.Time) bool {
+	if now.Sub(c.since) < lbRepairMinFailing {
+		return false
+	}
+	if now.Sub(vmRunningSince(c.lb, c.rec)) < lbRepairMinVMAge {
+		return false
+	}
+	n, newest := repairWindow(c.rec.Repairs, now, lbRepairBudgetWindow)
+	if n >= lbRepairBudget {
+		r.reportBudgetSpent(ctx, c, n)
+		return false
+	}
+	r.clearBudgetReport(c.lb.LoadBalancerID)
+	return now.Sub(newest) >= lbRepairCooldown || newest.IsZero()
+}
+
+// vmRunningSince is when the LB's current VM last came up as far as the control
+// plane can tell: its activation, a relaunch, or the last repair.
+func vmRunningSince(lb *LoadBalancerRecord, rec *LBDataPathRecord) time.Time {
+	since := lb.ActiveSince
+	if since.IsZero() {
+		since = lb.CreatedAt
+	}
+	if rec.BoundAt.After(since) {
+		since = rec.BoundAt
+	}
+	for _, e := range rec.Repairs {
+		if e.counts() && e.At.After(since) {
+			since = e.At
+		}
+	}
+	return since
+}
+
+// repair reserves the attempt durably, then dispatches it. It reports whether a
+// repair was started, which ends the pass.
+func (r *lbRepairReconciler) repair(ctx context.Context, c lbRepairCandidate) bool {
+	if !r.isLeader() {
+		return false
+	}
+	lb, err := r.svc.store.GetLoadBalancer(ctx, c.lb.LoadBalancerID)
+	if err != nil || lb == nil || lb.State != StateActive || lb.InstanceID != c.lb.InstanceID {
+		return false
+	}
+	now := r.now()
+	reason := "data NIC failing gateway probe"
+	var attempt int
+	reserved := false
+	_, err = r.svc.store.UpdateLBHealth(ctx, lb.LoadBalancerID, func(cur *LBDataPathRecord) (bool, error) {
+		// The callback reruns after a conflict, so admission is judged by the run
+		// that wrote the reservation and nothing earlier.
+		reserved, attempt = false, 0
+		nic := cur.NICs[c.mac]
+		if cur.Suspended || cur.InstanceID != lb.InstanceID || nic == nil || now.Sub(nic.FailingSince) < lbRepairMinFailing {
+			return false, nil
+		}
+		n, newest := repairWindow(cur.Repairs, now, lbRepairBudgetWindow)
+		if n >= lbRepairBudget || (!newest.IsZero() && now.Sub(newest) < lbRepairCooldown) {
+			return false, nil
+		}
+		cur.Repairs = append(pruneRepairs(cur.Repairs, now), LBRepair{
+			At: now, InstanceID: lb.InstanceID, ENIID: nic.ENIID, Reason: reason, Outcome: RepairReserved,
+		})
+		attempt = n + 1
+		reserved = true
+		return true, nil
+	})
+	if err != nil {
+		slog.WarnContext(ctx, "ELBv2 data-path repair: reservation failed", "lbId", lb.LoadBalancerID, "err", err)
+		return false
+	}
+	if !reserved {
+		return false
+	}
+
+	attrs := repairLogAttrs(lb, c, attempt, reason, now)
+	if !r.isLeader() {
+		slog.WarnContext(ctx, "ELBv2 data-path repair not dispatched: leadership lost after reservation", attrs...)
+		return true
+	}
+	slog.WarnContext(ctx, "ELBv2 data-path repair started", attrs...)
+	outcome := r.dispatch(lb.InstanceID)
+	if outcome != "" {
+		r.recordOutcome(ctx, lb.LoadBalancerID, now, lb.InstanceID, outcome)
+	}
+	switch outcome {
+	case RepairDispatched:
+		slog.InfoContext(ctx, "ELBv2 data-path repair dispatched", attrs...)
+	case RepairRefused:
+		slog.ErrorContext(ctx, "ELBv2 data-path repair refused by the owning node", attrs...)
+	case RepairUnreachable:
+		slog.WarnContext(ctx, "ELBv2 data-path repair: owning node unreachable, will retry", attrs...)
+	default:
+		slog.WarnContext(ctx, "ELBv2 data-path repair: outcome unknown, counted", attrs...)
+	}
+	return true
+}
+
+// dispatch asks the owning node to restart exactly the LB's own VM. It returns
+// an empty outcome when the reply was lost, which leaves the entry reserved.
+func (r *lbRepairReconciler) dispatch(instanceID string) string {
+	err := r.rebooter.RebootSystemInstance(instanceID)
+	switch {
+	case err == nil:
+		return RepairDispatched
+	case errors.Is(err, nats.ErrNoResponders), strings.Contains(err.Error(), "no responders"):
+		return RepairUnreachable
+	case errors.Is(err, systeminstance.ErrSystemInstanceNotFound),
+		errors.Is(err, systeminstance.ErrSystemInstanceNotSystem),
+		errors.Is(err, systeminstance.ErrSystemInstanceOwnershipUnknown):
+		return RepairRefused
+	default:
+		return ""
+	}
+}
+
+func (r *lbRepairReconciler) recordOutcome(ctx context.Context, lbID string, at time.Time, instanceID, outcome string) {
+	_, err := r.svc.store.UpdateLBHealth(ctx, lbID, func(cur *LBDataPathRecord) (bool, error) {
+		for i := range cur.Repairs {
+			e := &cur.Repairs[i]
+			if e.At.Equal(at) && e.InstanceID == instanceID && e.Outcome == RepairReserved {
+				e.Outcome = outcome
+				return true, nil
+			}
+		}
+		return false, nil
+	})
+	if err != nil {
+		slog.WarnContext(ctx, "ELBv2 data-path repair: failed to record outcome", "lbId", lbID, "outcome", outcome, "err", err)
+	}
+}
+
+// pruneRepairs drops entries older than the budget window.
+func pruneRepairs(repairs []LBRepair, now time.Time) []LBRepair {
+	kept := repairs[:0:0]
+	for _, e := range repairs {
+		if now.Sub(e.At) < lbRepairBudgetWindow {
+			kept = append(kept, e)
+		}
+	}
+	return kept
+}
+
+func repairLogAttrs(lb *LoadBalancerRecord, c lbRepairCandidate, attempt int, reason string, now time.Time) []any {
+	return []any{
+		"lbArn", lb.LoadBalancerArn, "lbId", lb.LoadBalancerID, "instanceId", lb.InstanceID, "nodeId", lb.NodeID,
+		"eniId", c.nic.ENIID, "mac", c.mac, "failing_since", c.nic.FailingSince,
+		"failing_ms", otelsetup.Millis(now.Sub(c.nic.FailingSince)),
+		"consecutive_failures", c.nic.ConsecutiveFailures,
+		"rx_packets", c.nic.RxPackets, "tx_packets", c.nic.TxPackets,
+		"rx_dropped", c.nic.RxDropped, "tx_dropped", c.nic.TxDropped,
+		"repair_attempt", attempt, "reason", reason,
+	}
+}
+
+func (r *lbRepairReconciler) reportBudgetSpent(ctx context.Context, c lbRepairCandidate, n int) {
+	r.mu.Lock()
+	already := r.budgetReported[c.lb.LoadBalancerID]
+	r.budgetReported[c.lb.LoadBalancerID] = true
+	r.mu.Unlock()
+	if already {
+		return
+	}
+	slog.ErrorContext(ctx, "ELBv2 data-path repair budget exhausted",
+		"lbArn", c.lb.LoadBalancerArn, "lbId", c.lb.LoadBalancerID, "instanceId", c.lb.InstanceID,
+		"eniId", c.nic.ENIID, "mac", c.mac, "repairs_24h", n)
+}
+
+func (r *lbRepairReconciler) clearBudgetReport(lbID string) {
+	r.mu.Lock()
+	delete(r.budgetReported, lbID)
+	r.mu.Unlock()
+}
+
+func (r *lbRepairReconciler) forgetBudgetReports(live map[string]bool) {
+	r.mu.Lock()
+	defer r.mu.Unlock()
+	for id := range r.budgetReported {
+		if !live[id] {
+			delete(r.budgetReported, id)
+		}
+	}
+}

spinifex/domains/elbv2/lb_repair_test.go

@@ -0,0 +1,418 @@
+package elbv2
+
+import (
+	"bytes"
+	"context"
+	"errors"
+	"fmt"
+	"log/slog"
+	"strings"
+	"testing"
+	"time"
+
+	"github.com/mulgadc/spinifex/spinifex/domains/ec2/systeminstance"
+	"github.com/nats-io/nats.go"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+const repairMAC = "02:00:00:bb:00:01"
+
+type repairEnv struct {
+	t        *testing.T
+	svc      *ELBv2ServiceImpl
+	rec      *lbRepairReconciler
+	launcher *mockSystemInstanceLauncher
+	clock    time.Time
+	leader   bool
+}
+
+func newRepairEnv(t *testing.T) *repairEnv {
+	t.Helper()
+	e := &repairEnv{
+		t:        t,
+		svc:      setupTestService(t),
+		launcher: &mockSystemInstanceLauncher{},
+		clock:    time.Date(2026, 11, 5, 12, 0, 0, 0, time.UTC),
+		leader:   true,
+	}
+	e.rec = newLBRepairReconciler(e.svc, e.launcher, func() bool { return e.leader })
+	e.rec.now = func() time.Time { return e.clock }
+	return e
+}
+
+// seed stores an active LB whose VM has run for ten minutes and whose data NIC
+// has been failing for two, with a heartbeat from this instant.
+func (e *repairEnv) seed(id string, failing bool) *LoadBalancerRecord {
+	e.t.Helper()
+	lb := &LoadBalancerRecord{
+		LoadBalancerArn: "arn:aws:elasticloadbalancing:us-east-1:" + testAccountID + ":loadbalancer/app/" + id + "/" + id,
+		LoadBalancerID:  id,
+		Name:            id,
+		State:           StateActive,
+		InstanceID:      "i-" + id,
+		AccountID:       testAccountID,
+		NodeID:          "node-1",
+		CreatedAt:       e.clock.Add(-time.Hour),
+		ActiveSince:     e.clock.Add(-10 * time.Minute),
+		LastHeartbeat:   e.clock,
+	}
+	require.NoError(e.t, e.svc.store.PutLoadBalancer(context.Background(), lb))
+	if failing {
+		e.fail(id, 2*time.Minute)
+	}
+	return lb
+}
+
+func (e *repairEnv) fail(id string, age time.Duration) {
+	e.t.Helper()
+	_, err := e.svc.store.UpdateLoadBalancer(context.Background(), id, func(lb *LoadBalancerRecord) (bool, error) {
+		lb.LastHeartbeat = e.clock
+		return true, nil
+	})
+	require.NoError(e.t, err)
+	_, err = e.svc.store.UpdateLBHealth(context.Background(), id, func(rec *LBDataPathRecord) (bool, error) {
+		rec.InstanceID = "i-" + id
+		rec.NICs = map[string]*LBNICFailure{
+			repairMAC: {ENIID: "eni-" + id, FailingSince: e.clock.Add(-age), ConsecutiveFailures: 9, RxPackets: 4, TxPackets: 8},
+		}
+		return true, nil
+	})
+	require.NoError(e.t, err)
+}
+
+func (e *repairEnv) health(id string) *LBDataPathRecord {
+	e.t.Helper()
+	rec, err := e.svc.store.GetLBHealth(context.Background(), id)
+	require.NoError(e.t, err)
+	require.NotNil(e.t, rec)
+	return rec
+}
+
+func (e *repairEnv) pass() { e.rec.pass(context.Background()) }
+
+func (e *repairEnv) reboots() []string {
+	e.launcher.mu.Lock()
+	defer e.launcher.mu.Unlock()
+	return append([]string(nil), e.launcher.rebootCalls...)
+}
+
+func captureLogs(t *testing.T) *bytes.Buffer {
+	t.Helper()
+	var buf bytes.Buffer
+	prev := slog.Default()
+	slog.SetDefault(slog.New(slog.NewTextHandler(&buf, &slog.HandlerOptions{Level: slog.LevelDebug})))
+	t.Cleanup(func() { slog.SetDefault(prev) })
+	return &buf
+}
+
+func TestLBRepair_RepairsOnlyAfterSixtySecondsOfFailure(t *testing.T) {
+	t.Parallel()
+	e := newRepairEnv(t)
+	e.seed("lbr1", false)
+	e.fail("lbr1", 59*time.Second)
+	e.pass()
+	assert.Empty(t, e.reboots())
+
+	e.clock = e.clock.Add(2 * time.Second)
+	e.pass()
+	assert.Equal(t, []string{"i-lbr1"}, e.reboots())
+	h := e.health("lbr1")
+	require.Len(t, h.Repairs, 1)
+	assert.Equal(t, RepairDispatched, h.Repairs[0].Outcome)
+	assert.Equal(t, "eni-lbr1", h.Repairs[0].ENIID)
+}
+
+func TestLBRepair_ContinuouslyHeartbeatingLBIsNeverStale(t *testing.T) {
+	t.Parallel()
+	// A healthy agent persists its heartbeat only every minute, so the stored
+	// timestamp can be any age below that plus one pass at the reconciler.
+	for age := time.Duration(0); age < lbRepairHeartbeatMaxAge; age += 15 * time.Second {
+		e := newRepairEnv(t)
+		lb := e.seed("lbr2", true)
+		lb.LastHeartbeat = e.clock.Add(-age)
+		require.NoError(t, e.svc.store.PutLoadBalancer(context.Background(), lb))
+		e.pass()
+		assert.Len(t, e.reboots(), 1, "heartbeat %s old", age)
+	}
+
+	e := newRepairEnv(t)
+	lb := e.seed("lbr2", true)
+	lb.LastHeartbeat = e.clock.Add(-lbRepairHeartbeatMaxAge)
+	require.NoError(t, e.svc.store.PutLoadBalancer(context.Background(), lb))
+	e.pass()
+	assert.Empty(t, e.reboots(), "a heartbeat older than the bound is stale")
+}
+
+func TestLBRepair_NeverWithoutCurrentEvidence(t *testing.T) {
+	t.Parallel()
+	mutate := map[string]func(*LoadBalancerRecord){
+		"non-active LB":      func(lb *LoadBalancerRecord) { lb.State = StateProvisioning },
+		"failed LB":          func(lb *LoadBalancerRecord) { lb.State = StateFailed },
+		"empty instance ID":  func(lb *LoadBalancerRecord) { lb.InstanceID = "" },
+		"changed instance":   func(lb *LoadBalancerRecord) { lb.InstanceID = "i-other" },
+		"stale heartbeat":    func(lb *LoadBalancerRecord) { lb.LastHeartbeat = lb.LastHeartbeat.Add(-time.Hour) },
+		"VM under 5 minutes": func(lb *LoadBalancerRecord) { lb.ActiveSince = lb.LastHeartbeat.Add(-4 * time.Minute) },
+	}
+	for name, fn := range mutate {
+		t.Run(name, func(t *testing.T) {
+			t.Parallel()
+			e := newRepairEnv(t)
+			lb := e.seed("lbr3", true)
+			fn(lb)
+			require.NoError(t, e.svc.store.PutLoadBalancer(context.Background(), lb))
+			e.pass()
+			assert.Empty(t, e.reboots())
+		})
+	}
+}
+
+func TestLBRepair_NonHolderNeverRepairs(t *testing.T) {
+	t.Parallel()
+	e := newRepairEnv(t)
+	e.seed("lbr4", true)
+	e.leader = false
+	e.pass()
+	assert.Empty(t, e.reboots())
+	e.leader = true
+	e.pass()
+	assert.Len(t, e.reboots(), 1)
+}
+
+func TestLBRepair_OneRepairPerPass(t *testing.T) {
+	t.Parallel()
+	e := newRepairEnv(t)
+	e.seed("lbr5a", true)
+	e.seed("lbr5b", true)
+	for i := range 3 {
+		e.seed(fmt.Sprintf("lbr5h%d", i), false)
+	}
+	e.fail("lbr5b", 3*time.Minute)
+
+	e.pass()
+	assert.Equal(t, []string{"i-lbr5b"}, e.reboots(), "the longest-failing LB goes first")
+	e.pass()
+	assert.Equal(t, []string{"i-lbr5b", "i-lbr5a"}, e.reboots())
+}
+
+func TestLBRepair_FleetWideFailureRepairsNothing(t *testing.T) {
+	e := newRepairEnv(t)
+	logs := captureLogs(t)
+	e.seed("lbr6a", true)
+	e.seed("lbr6b", true)
+	e.seed("lbr6c", true)
+	e.seed("lbr6d", false)
+	e.pass()
+	assert.Empty(t, e.reboots())
+	assert.Contains(t, logs.String(), "ELBv2 data-path repair suppressed: fleet-wide probe failure")
+
+	// Three failing out of seven is not more than half.
+	for i := range 3 {
+		e.seed(fmt.Sprintf("lbr6h%d", i), false)
+	}
+	e.pass()
+	assert.Len(t, e.reboots(), 1)
+}
+
+func TestLBRepair_CooldownAndDailyBudget(t *testing.T) {
+	e := newRepairEnv(t)
+	logs := captureLogs(t)
+	e.seed("lbr7", true)
+
+	e.pass()
+	require.Len(t, e.reboots(), 1)
+
+	e.clock = e.clock.Add(9 * time.Minute)
+	e.fail("lbr7", 2*time.Minute)
+	e.pass()
+	assert.Len(t, e.reboots(), 1, "inside the 10 minute cooldown")
+
+	for want := 2; want <= 3; want++ {
+		e.clock = e.clock.Add(11 * time.Minute)
+		e.fail("lbr7", 2*time.Minute)
+		e.pass()
+		assert.Len(t, e.reboots(), want)
+	}
+
+	e.clock = e.clock.Add(11 * time.Minute)
+	e.fail("lbr7", 2*time.Minute)
+	e.pass()
+	e.pass()
+	assert.Len(t, e.reboots(), 3, "the fourth repair in 24 hours is refused")
+	assert.Equal(t, 1, strings.Count(logs.String(), "ELBv2 data-path repair budget exhausted"))
+
+	e.clock = e.clock.Add(24 * time.Hour)
+	e.fail("lbr7", 2*time.Minute)
+	e.pass()
+	assert.Len(t, e.reboots(), 4, "the budget frees after 24 hours")
+}
+
+func TestLBRepair_UnreachableOwnerIsRetriedAndNotCounted(t *testing.T) {
+	t.Parallel()
+	e := newRepairEnv(t)
+	e.launcher.rebootErr = fmt.Errorf("nats request: %w", nats.ErrNoResponders)
+	e.seed("lbr8", true)
+	for range 4 {
+		e.pass()
+		e.clock = e.clock.Add(30 * time.Second)
+		e.fail("lbr8", 2*time.Minute)
+	}
+	assert.Len(t, e.reboots(), 4, "no node received it, so cooldown and budget do not apply")
+	for _, r := range e.health("lbr8").Repairs {
+		assert.Equal(t, RepairUnreachable, r.Outcome)
+	}
+
+	e.launcher.rebootErr = nil
+	e.pass()
+	assert.Len(t, e.reboots(), 5)
+}
+
+func TestLBRepair_RefusedAndAmbiguousRepairsStillCount(t *testing.T) {
+	t.Parallel()
+	for name, err := range map[string]error{
+		"refused": fmt.Errorf("%w: i-lbr9", systeminstance.ErrSystemInstanceNotSystem),
+		"lost":    errors.New("timeout waiting for reply"),
+	} {
+		t.Run(name, func(t *testing.T) {
+			t.Parallel()
+			e := newRepairEnv(t)
+			e.launcher.rebootErr = err
+			e.seed("lbr9", true)
+			e.pass()
+			e.clock = e.clock.Add(30 * time.Second)
+			e.fail("lbr9", 2*time.Minute)
+			e.pass()
+			assert.Len(t, e.reboots(), 1, "inside the cooldown")
+			want := RepairRefused
+			if name == "lost" {
+				want = RepairReserved
+			}
+			assert.Equal(t, want, e.health("lbr9").Repairs[0].Outcome)
+		})
+	}
+}
+
+func TestLBRepair_ReservationWithoutOutcomeCountsAfterCrashOrLostLease(t *testing.T) {
+	t.Parallel()
+	e := newRepairEnv(t)
+	e.seed("lbr10", true)
+	// A leader reserved and died before dispatching.
+	_, err := e.svc.store.UpdateLBHealth(context.Background(), "lbr10", func(rec *LBDataPathRecord) (bool, error) {
+		rec.Repairs = append(rec.Repairs, LBRepair{At: e.clock.Add(-time.Minute), InstanceID: "i-lbr10", Outcome: RepairReserved})
+		return true, nil
+	})
+	require.NoError(t, err)
+
+	e.pass()
+	assert.Empty(t, e.reboots(), "the new leader sees the counted reservation")
+
+	e.leader = false
+	e.pass()
+	e.leader = true
+	assert.Empty(t, e.reboots())
+}
+
+func TestLBRepair_PausedBetweenEligibilityAndReservationRepairsNothing(t *testing.T) {
+	t.Parallel()
+	cases := map[string]func(e *repairEnv, lb *LoadBalancerRecord){
+		"DeleteLoadBalancer starts": func(e *repairEnv, lb *LoadBalancerRecord) {
+			require.NoError(t, e.svc.store.SuspendLBHealth(context.Background(), lb.LoadBalancerID))
+		},
+		"SetSubnets relaunches the VM": func(e *repairEnv, lb *LoadBalancerRecord) {
+			lb.InstanceID = "i-new"
+			require.NoError(t, e.svc.store.PutLoadBalancer(context.Background(), lb))
+		},
+		"SetSubnets rebinds the evidence": func(e *repairEnv, lb *LoadBalancerRecord) {
+			require.NoError(t, e.svc.store.ResumeLBHealth(context.Background(), lb.LoadBalancerID, "i-new"))
+		},
+	}
+	for name, interleave := range cases {
+		t.Run(name, func(t *testing.T) {
+			t.Parallel()
+			e := newRepairEnv(t)
+			lb := e.seed("lbr11", true)
+			cands := e.candidates()
+			require.Len(t, cands, 1)
+			require.True(t, e.rec.eligible(context.Background(), cands[0], e.clock))
+
+			interleave(e, lb)
+
+			assert.False(t, e.rec.repair(context.Background(), cands[0]))
+			assert.Empty(t, e.reboots())
+		})
+	}
+}
+
+func (e *repairEnv) candidates() []lbRepairCandidate {
+	e.t.Helper()
+	var out []lbRepairCandidate
+	lbs, err := e.svc.store.ListLoadBalancersStrict(context.Background())
+	require.NoError(e.t, err)
+	for _, lb := range lbs {
+		rec, err := e.svc.store.GetLBHealth(context.Background(), lb.LoadBalancerID)
+		require.NoError(e.t, err)
+		if rec == nil {
+			continue
+		}
+		c := lbRepairCandidate{lb: lb, rec: rec}
+		for mac, nic := range rec.NICs {
+			c.mac, c.nic, c.since = mac, nic, nic.FailingSince
+		}
+		out = append(out, c)
+	}
+	return out
+}
+
+// Verifies: ADR-0003:S7.
+func TestS7_LBRepairActsOnlyOnTheLBsOwnVM(t *testing.T) {
+	t.Parallel()
+	e := newRepairEnv(t)
+	e.seed("lbr12", true)
+	e.seed("lbr12ok", false)
+	e.pass()
+	e.pass()
+	assert.Equal(t, []string{"i-lbr12"}, e.reboots(), "only the failing LB's own instance ID reaches the launcher")
+	e.launcher.mu.Lock()
+	defer e.launcher.mu.Unlock()
+	assert.Empty(t, e.launcher.terminateCalls)
+	assert.Empty(t, e.launcher.launchCalls)
+}
+
+func TestLBRepair_TeardownNeverTriggersRepair(t *testing.T) {
+	t.Parallel()
+	ctx := context.Background()
+	t.Run("DeleteLoadBalancer", func(t *testing.T) {
+		t.Parallel()
+		e := newRepairEnv(t)
+		lb := e.seed("lbr13a", true)
+		require.NoError(t, e.svc.store.SuspendLBHealth(ctx, lb.LoadBalancerID))
+		e.pass()
+		assert.Empty(t, e.reboots())
+	})
+	t.Run("SetSubnets window", func(t *testing.T) {
+		t.Parallel()
+		e := newRepairEnv(t)
+		lb := e.seed("lbr13b", true)
+		require.NoError(t, e.svc.store.SuspendLBHealth(ctx, lb.LoadBalancerID))
+		lb.InstanceID = "i-relaunched"
+		require.NoError(t, e.svc.store.PutLoadBalancer(ctx, lb))
+		e.pass()
+		assert.Empty(t, e.reboots(), "suspended, and the old VM's evidence names another instance")
+	})
+	t.Run("transient probe failure on the relaunched VM", func(t *testing.T) {
+		t.Parallel()
+		e := newRepairEnv(t)
+		lb := e.seed("lbr13c", false)
+		lb.InstanceID = "i-relaunched"
+		require.NoError(t, e.svc.store.PutLoadBalancer(ctx, lb))
+		require.NoError(t, e.svc.store.ResumeLBHealth(ctx, lb.LoadBalancerID, "i-relaunched"))
+		_, err := e.svc.store.UpdateLBHealth(ctx, lb.LoadBalancerID, func(rec *LBDataPathRecord) (bool, error) {
+			rec.BoundAt = e.clock.Add(-90 * time.Second)
+			rec.NICs = map[string]*LBNICFailure{repairMAC: {FailingSince: e.clock.Add(-80 * time.Second)}}
+			return true, nil
+		})
+		require.NoError(t, err)
+		e.pass()
+		assert.Empty(t, e.reboots(), "a freshly relaunched VM gets five minutes")
+	})
+}

spinifex/domains/elbv2/service_impl.go

@@ -1138,6 +1138,7 @@ func (s *ELBv2ServiceImpl) LBAgentHeartbeat(ctx context.Context, input *LBAgentH
 			if rec.State == StateProvisioning {
 				rec.State = StateActive
 				rec.StateReason = ""
+				rec.ActiveSince = now
 			}
 			rec.LastHeartbeat = now
 			return true, nil
@@ -1162,6 +1163,8 @@ func (s *ELBv2ServiceImpl) LBAgentHeartbeat(ctx context.Context, input *LBAgentH
 		}
 	}
 
+	s.recordDataPathEvidence(ctx, lb, input)
+
 	// Process health report directly — no JSON round-trip needed.
 	if len(input.Servers) > 0 {
 		s.hc.handleHealthReportDirect(ctx, input.toHealthReport())
@@ -1853,15 +1856,23 @@ func (s *ELBv2ServiceImpl) DeleteLoadBalancer(ctx context.Context, input *elbv2.
 		return &elbv2.DeleteLoadBalancerOutput{}, nil
 	}
 
+	// Exclude data-path repair before anything touches the VM or its ENIs.
+	if err := s.store.SuspendLBHealth(ctx, lb.LoadBalancerID); err != nil {
+		slog.ErrorContext(ctx, "DeleteLoadBalancer: failed to suspend data-path repair", "lbId", lb.LoadBalancerID, "err", err)
+		return nil, errors.New(awserrors.ErrorServerInternal)
+	}
+
 	// Cascade-delete all listeners and their rules so no orphan pins a TG as ResourceInUse.
 	listeners, err := s.store.ListListenersByLB(ctx, lb.LoadBalancerArn)
 	if err != nil {
 		slog.ErrorContext(ctx, "DeleteLoadBalancer: failed to list listeners for cascade", "lbArn", lb.LoadBalancerArn, "err", err)
+		s.resumeLBHealth(ctx, lb.LoadBalancerID)
 		return nil, errors.New(awserrors.ErrorServerInternal)
 	}
 	for _, l := range listeners {
 		if err := s.deleteListenerCascade(ctx, l); err != nil {
 			slog.ErrorContext(ctx, "DeleteLoadBalancer: failed to cascade listener delete", "listenerID", l.ListenerID, "err", err)
+			s.resumeLBHealth(ctx, lb.LoadBalancerID)
 			return nil, errors.New(awserrors.ErrorServerInternal)
 		}
 	}
@@ -1910,6 +1921,9 @@ func (s *ELBv2ServiceImpl) DeleteLoadBalancer(ctx context.Context, input *elbv2.
 		slog.ErrorContext(ctx, "DeleteLoadBalancer: failed to delete record", "lbId", lb.LoadBalancerID, "err", err)
 		return nil, errors.New(awserrors.ErrorServerInternal)
 	}
+	if err := s.store.DeleteLBHealth(ctx, lb.LoadBalancerID); err != nil {
+		slog.WarnContext(ctx, "DeleteLoadBalancer: failed to delete data-path evidence", "lbId", lb.LoadBalancerID, "err", err)
+	}
 
 	// Release the name claim so the name is reusable. Idempotent on a missing
 	// key, so a delete that races the record removal still converges.
@@ -3456,6 +3470,9 @@ func (s *ELBv2ServiceImpl) lbRecordToSDK(ctx context.Context, r *LoadBalancerRec
 	if r.StateReason != "" {
 		lb.State.Reason = aws.String(r.StateReason)
 	}
+	if s.impairedByEvidence(ctx, r) {
+		lb.State.Code = aws.String(StateActiveImpaired)
+	}
 
 	for _, sg := range r.SecurityGroups {
 		lb.SecurityGroups = append(lb.SecurityGroups, aws.String(sg))

spinifex/domains/elbv2/service_impl_lb_network.go

@@ -172,6 +172,18 @@ func (s *ELBv2ServiceImpl) SetSubnets(ctx context.Context, input *elbv2.SetSubne
 		return s.setSubnetsOutput(ctx, lb), nil
 	}
 
+	// Exclude data-path repair before any ENI or VM call. It stays excluded once the
+	// VM is terminated and is lifted when the new VM is published.
+	if err := s.store.SuspendLBHealth(ctx, lb.LoadBalancerID); err != nil {
+		slog.ErrorContext(ctx, "SetSubnets: failed to suspend data-path repair", "arn", *input.LoadBalancerArn, "err", err)
+		return nil, errors.New(awserrors.ErrorServerInternal)
+	}
+	resume := func() {
+		if err := s.store.ResumeLBHealth(ctx, lb.LoadBalancerID, ""); err != nil {
+			slog.WarnContext(ctx, "SetSubnets: failed to lift data-path suspension", "lbId", lb.LoadBalancerID, "err", err)
+		}
+	}
+
 	// Create ENIs for added subnets; roll back on failure to avoid leaks.
 	newENIBySubnet := make(map[string]string, len(toAdd))
 	newAZBySubnet := make(map[string]string, len(toAdd))
@@ -188,6 +200,7 @@ func (s *ELBv2ServiceImpl) SetSubnets(ctx context.Context, input *elbv2.SetSubne
 		eniID, az, eniErr := s.createLBENI(ctx, subnetID, lb, accountID)
 		if eniErr != nil {
 			rollbackNewENIs()
+			resume()
 			slog.ErrorContext(ctx, "SetSubnets: failed to create ENI", "subnet", subnetID, "err", eniErr)
 			return nil, errors.New(awserrors.ErrorELBv2SubnetNotFound)
 		}
@@ -209,6 +222,7 @@ func (s *ELBv2ServiceImpl) SetSubnets(ctx context.Context, input *elbv2.SetSubne
 	if lb.InstanceID != "" && s.InstanceLauncher != nil {
 		if err := s.InstanceLauncher.TerminateSystemInstance(lb.InstanceID); err != nil {
 			rollbackNewENIs()
+			resume()
 			slog.ErrorContext(ctx, "SetSubnets: failed to terminate LB VM for relaunch", "arn", *input.LoadBalancerArn, "instanceId", lb.InstanceID, "err", err)
 			return nil, errors.New(awserrors.ErrorServerInternal)
 		}
@@ -268,6 +282,9 @@ func (s *ELBv2ServiceImpl) SetSubnets(ctx context.Context, input *elbv2.SetSubne
 		slog.ErrorContext(ctx, "SetSubnets: failed to persist LB", "arn", *input.LoadBalancerArn, "err", err)
 		return nil, errors.New(awserrors.ErrorServerInternal)
 	}
+	if err := s.store.ResumeLBHealth(ctx, lb.LoadBalancerID, lb.InstanceID); err != nil {
+		slog.WarnContext(ctx, "SetSubnets: failed to lift data-path suspension", "lbId", lb.LoadBalancerID, "err", err)
+	}
 
 	slog.InfoContext(ctx, "SetSubnets completed", "arn", *input.LoadBalancerArn, "subnets", len(desired), "added", len(toAdd), "removed", len(toRemove), "state", lb.State)
 	return s.setSubnetsOutput(ctx, lb), nil

spinifex/domains/elbv2/service_impl_test.go

@@ -2108,6 +2108,8 @@ func (m *mockTerminateLauncher) TerminateSystemInstance(instanceID string) error
 	return nil
 }
 
+func (m *mockTerminateLauncher) RebootSystemInstance(_ string) error { return nil }
+
 // --- Scheme unit tests ---
 
 func TestCreateLoadBalancer_InternetFacingScheme_DNSName(t *testing.T) {

spinifex/domains/elbv2/system_instance_launcher_nats.go

@@ -96,4 +96,25 @@ func (l *natsSystemInstanceLauncher) TerminateSystemInstance(instanceID string)
 	return nil
 }
 
+func (l *natsSystemInstanceLauncher) RebootSystemInstance(instanceID string) error {
+	if instanceID == "" {
+		return fmt.Errorf("system instance reboot: empty instance ID")
+	}
+
+	subject := fmt.Sprintf("system.RebootInstance.%s", instanceID)
+	reply, err := l.nc.Request(subject, nil, l.timeout)
+	if err != nil {
+		return fmt.Errorf("nats request %s: %w", subject, err)
+	}
+
+	var env natsSystemInstanceTerminateEnvelope
+	if err := json.Unmarshal(reply.Data, &env); err != nil {
+		return fmt.Errorf("decode reboot reply: %w", err)
+	}
+	if env.Error != "" {
+		return fmt.Errorf("%s", env.Error)
+	}
+	return nil
+}
+
 var _ SystemInstanceLauncher = (*natsSystemInstanceLauncher)(nil)

spinifex/domains/elbv2/system_instance_launcher_nats_test.go

@@ -4,6 +4,7 @@ import (
 	"encoding/json"
 	"errors"
 	"sync"
+	"sync/atomic"
 	"testing"
 	"time"
 
@@ -311,3 +312,69 @@ func TestNATSSystemInstanceLauncher_RefusesNonLoadBalancerKindBeforeSending(t *t
 	defer mu.Unlock()
 	assert.Equal(t, 1, received, "only the load balancer launch may be sent")
 }
+
+func TestNATSSystemInstanceLauncher_RebootEmptyID(t *testing.T) {
+	t.Parallel()
+	_, nc, _ := testutil.StartTestJetStream(t)
+	defer nc.Close()
+
+	err := NewNATSSystemInstanceLauncher(nc, time.Second).RebootSystemInstance("")
+	require.Error(t, err)
+	assert.Contains(t, err.Error(), "empty instance ID")
+}
+
+func TestNATSSystemInstanceLauncher_RebootRoundTrip(t *testing.T) {
+	t.Parallel()
+	_, nc, _ := testutil.StartTestJetStream(t)
+	defer nc.Close()
+
+	var asked atomic.Int32
+	sub, err := nc.Subscribe("system.RebootInstance.i-lb", func(msg *nats.Msg) {
+		asked.Add(1)
+		_ = msg.Respond([]byte(`{}`))
+	})
+	require.NoError(t, err)
+	defer func() { _ = sub.Unsubscribe() }()
+
+	require.NoError(t, NewNATSSystemInstanceLauncher(nc, 2*time.Second).RebootSystemInstance("i-lb"))
+	assert.EqualValues(t, 1, asked.Load())
+}
+
+func TestNATSSystemInstanceLauncher_RebootNoResponderKeepsTransportError(t *testing.T) {
+	t.Parallel()
+	_, nc, _ := testutil.StartTestJetStream(t)
+	defer nc.Close()
+
+	err := NewNATSSystemInstanceLauncher(nc, 250*time.Millisecond).RebootSystemInstance("i-no-one-listens")
+	require.ErrorIs(t, err, nats.ErrNoResponders)
+}
+
+func TestNATSSystemInstanceLauncher_RebootDecodeFailureAndRemoteError(t *testing.T) {
+	t.Parallel()
+	_, nc, _ := testutil.StartTestJetStream(t)
+	defer nc.Close()
+
+	bad, err := nc.Subscribe("system.RebootInstance.i-decode", func(msg *nats.Msg) {
+		_ = msg.Respond([]byte("{not valid json"))
+	})
+	require.NoError(t, err)
+	defer func() { _ = bad.Unsubscribe() }()
+
+	refused, err := nc.Subscribe("system.RebootInstance.i-refused", func(msg *nats.Msg) {
+		payload, _ := json.Marshal(struct {
+			Error string `json:"error,omitempty"`
+		}{Error: "not a system instance"})
+		_ = msg.Respond(payload)
+	})
+	require.NoError(t, err)
+	defer func() { _ = refused.Unsubscribe() }()
+
+	l := NewNATSSystemInstanceLauncher(nc, 2*time.Second)
+	err = l.RebootSystemInstance("i-decode")
+	require.Error(t, err)
+	assert.Contains(t, err.Error(), "decode reboot reply")
+
+	err = l.RebootSystemInstance("i-refused")
+	require.Error(t, err)
+	assert.Contains(t, err.Error(), "not a system instance")
+}

spinifex/domains/elbv2/types.go

@@ -155,6 +155,7 @@ type LoadBalancerRecord struct {
 	Tags             map[string]string  `json:"tags,omitempty"`
 	AccountID        string             `json:"account_id"`
 	CreatedAt        time.Time          `json:"created_at"`
+	ActiveSince      time.Time          `json:"active_since,omitzero"` // First heartbeat of the current VM generation
 }
 
 // HealthTargetSpec is a backend the nginx agent actively health-checks, computed

spinifex/domains/lambda/awsapi/eventsource.go

@@ -0,0 +1,277 @@
+package awsapi
+
+import (
+	"bytes"
+	"context"
+	"crypto/sha256"
+	"encoding/json"
+	"errors"
+	"fmt"
+	"net/http"
+	"regexp"
+	"strings"
+	"time"
+
+	"github.com/aws/aws-sdk-go/aws"
+	"github.com/aws/aws-sdk-go/service/lambda"
+	"github.com/mulgadc/spinifex/spinifex/domains/lambda/eventsource"
+	"github.com/mulgadc/spinifex/spinifex/domains/lambda/function"
+	"github.com/mulgadc/spinifex/spinifex/foundation/aws/arn"
+	awserrors "github.com/mulgadc/spinifex/spinifex/foundation/aws/errors"
+)
+
+// MappingStore is what the handlers need of the event-source mapping store.
+type MappingStore interface {
+	Create(ctx context.Context, m eventsource.Mapping) error
+	Get(ctx context.Context, uuid string) (eventsource.Mapping, uint64, error)
+	List(ctx context.Context) ([]eventsource.Entry, error)
+	Update(ctx context.Context, m eventsource.Mapping, revision uint64) (uint64, error)
+}
+
+var _ MappingStore = (*eventsource.Store)(nil)
+
+// SourceChecker reads an SQS queue's visibility timeout as the function's
+// execution role, so a queue the role cannot read is refused at create.
+type SourceChecker interface {
+	QueueVisibilityTimeout(ctx context.Context, rec function.Record, queueARN string) (int64, error)
+}
+
+// Batch size bounds for an SQS standard queue without a batching window.
+const (
+	defaultBatchSize = 10
+	maxBatchSize     = 10
+)
+
+// mappingCreateFields are the CreateEventSourceMapping members this build
+// accepts; every other member is refused rather than ignored.
+var mappingCreateFields = map[string]bool{
+	"FunctionName": true, "EventSourceArn": true, "BatchSize": true, "Enabled": true,
+	"MaximumBatchingWindowInSeconds": true,
+}
+
+var mappingUUIDRE = regexp.MustCompile(`^[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}$`)
+
+func mappingNotFound() error {
+	return awserrors.Errorf(awserrors.ErrorResourceNotFoundException, "The resource you requested does not exist.")
+}
+
+// mappingARN is the IAM resource of one mapping, or "*" for a malformed UUID.
+func mappingARN(region, accountID, uuid string) string {
+	if region == "" || accountID == "" || !mappingUUIDRE.MatchString(uuid) {
+		return "*"
+	}
+	return fmt.Sprintf("arn:aws:lambda:%s:%s:event-source-mapping:%s", region, accountID, uuid)
+}
+
+type mappingCreate struct {
+	FunctionName, QueueARN string
+	BatchSize              int64
+}
+
+// decodeMappingCreate reads the body strictly and refuses every option this
+// build does not implement.
+func decodeMappingCreate(body []byte) (*mappingCreate, error) {
+	var raw map[string]json.RawMessage
+	dec := json.NewDecoder(bytes.NewReader(body))
+	if err := dec.Decode(&raw); err != nil {
+		return nil, invalid("The request body could not be parsed as JSON.")
+	}
+	for k := range raw {
+		if !mappingCreateFields[k] {
+			return nil, unsupported(k)
+		}
+	}
+	var in struct {
+		FunctionName                   *string `json:"FunctionName"`
+		EventSourceArn                 *string `json:"EventSourceArn"`
+		BatchSize                      *int64  `json:"BatchSize"`
+		Enabled                        *bool   `json:"Enabled"`
+		MaximumBatchingWindowInSeconds *int64  `json:"MaximumBatchingWindowInSeconds"`
+	}
+	if err := json.Unmarshal(body, &in); err != nil {
+		return nil, invalid("The request body could not be parsed as JSON.")
+	}
+	if in.FunctionName == nil || *in.FunctionName == "" {
+		return nil, invalid("1 validation error detected: Value null at 'functionName' failed to satisfy constraint: Member must not be null")
+	}
+	if in.EventSourceArn == nil || *in.EventSourceArn == "" {
+		return nil, invalid("Unrecognized event source, must be kinesis, dynamodb stream or sqs.")
+	}
+	if in.Enabled != nil && !*in.Enabled {
+		return nil, unsupported("Enabled false")
+	}
+	if in.MaximumBatchingWindowInSeconds != nil && *in.MaximumBatchingWindowInSeconds != 0 {
+		return nil, unsupported("MaximumBatchingWindowInSeconds")
+	}
+	m := &mappingCreate{FunctionName: *in.FunctionName, QueueARN: *in.EventSourceArn, BatchSize: defaultBatchSize}
+	if in.BatchSize != nil {
+		if *in.BatchSize < 1 || *in.BatchSize > maxBatchSize {
+			return nil, invalid("BatchSize must be between 1 and %d for an SQS queue without a batching window.", maxBatchSize)
+		}
+		m.BatchSize = *in.BatchSize
+	}
+	return m, nil
+}
+
+// checkQueueARN refuses anything but a standard SQS queue in the caller's
+// own account and Region.
+func checkQueueARN(region, accountID, queueARN string) error {
+	parts := strings.Split(queueARN, ":")
+	if len(parts) != 6 || parts[0] != "arn" || parts[1] != "aws" {
+		return invalid("Unrecognized event source, must be kinesis, dynamodb stream or sqs.")
+	}
+	if parts[2] != "sqs" {
+		return unsupported("an event source other than SQS")
+	}
+	if parts[3] != region || parts[4] != accountID {
+		return unsupported("a queue in another account or Region")
+	}
+	if parts[5] == "" {
+		return invalid("Unrecognized event source, must be kinesis, dynamodb stream or sqs.")
+	}
+	if strings.HasSuffix(parts[5], ".fifo") {
+		return unsupported("a FIFO queue")
+	}
+	return nil
+}
+
+// mappingUUID derives one UUID per account, Region, function and queue, so the
+// keyed create itself refuses a duplicate with no inventory scan or second record.
+func mappingUUID(accountID, region, function, queueARN string) string {
+	b := sha256.Sum256([]byte(accountID + "\x00" + region + "\x00" + function + "\x00" + queueARN))
+	b[6] = b[6]&0x0f | 0x50
+	b[8] = b[8]&0x3f | 0x80
+	return fmt.Sprintf("%x-%x-%x-%x-%x", b[0:4], b[4:6], b[6:8], b[8:10], b[10:16])
+}
+
+func mappingShape(m eventsource.Mapping, state string) *lambda.EventSourceMappingConfiguration {
+	return &lambda.EventSourceMappingConfiguration{
+		UUID:                           aws.String(m.UUID),
+		BatchSize:                      aws.Int64(m.BatchSize),
+		MaximumBatchingWindowInSeconds: aws.Int64(0),
+		EventSourceArn:                 aws.String(m.EventSourceARN),
+		FunctionArn:                    aws.String(m.FunctionARN),
+		LastModified:                   aws.Time(m.LastModified),
+		State:                          aws.String(state),
+		StateTransitionReason:          aws.String(m.StateTransitionReason),
+	}
+}
+
+func (d Deps) mappings(ctx context.Context) (MappingStore, error) {
+	if d.Mappings == nil {
+		return nil, serviceError("event-source mappings are unavailable")
+	}
+	return d.Mappings(ctx)
+}
+
+// createEventSourceMapping resolves the function, reads the queue as its
+// execution role, and records an Enabled mapping its poller picks up.
+func (d Deps) createEventSourceMapping(ctx context.Context, c *call) (reply, error) {
+	in, err := decodeMappingCreate(c.Body)
+	if err != nil {
+		return reply{}, err
+	}
+	if err := checkQueueARN(c.Region, c.AccountID, in.QueueARN); err != nil {
+		return reply{}, err
+	}
+	name, q, err := resolveFunctionName(c.Region, c.AccountID, in.FunctionName)
+	if err != nil {
+		return reply{}, err
+	}
+	if q != "" {
+		return reply{}, unsupported("a qualified FunctionName")
+	}
+	rec, _, err := d.Meta.Get(ctx, c.AccountID, c.Region, name)
+	if errors.Is(err, function.ErrNotFound) {
+		return reply{}, notFound(c.Region, c.AccountID, name)
+	}
+	if err != nil {
+		return reply{}, err
+	}
+	if d.Sources == nil {
+		return reply{}, serviceError("event-source mappings are unavailable")
+	}
+	visibility, err := d.Sources.QueueVisibilityTimeout(ctx, rec, in.QueueARN)
+	if err != nil {
+		return reply{}, err
+	}
+	if rec.Timeout > visibility {
+		return reply{}, invalid("Queue visibility timeout: %d seconds is less than Function timeout: %d seconds", visibility, rec.Timeout)
+	}
+	store, err := d.mappings(ctx)
+	if err != nil {
+		return reply{}, err
+	}
+	uuid := mappingUUID(c.AccountID, c.Region, name, in.QueueARN)
+	m := eventsource.Mapping{
+		FormatVersion: eventsource.FormatVersion, UUID: uuid, AccountID: c.AccountID, Region: c.Region,
+		FunctionName: name, FunctionARN: arn.FormatLambdaFunction(c.Region, c.AccountID, name),
+		EventSourceARN: in.QueueARN, BatchSize: in.BatchSize, State: eventsource.StateEnabled,
+		StateTransitionReason: eventsource.ReasonUserInitiated, LastModified: time.Now().UTC().Truncate(time.Millisecond),
+	}
+	if err := store.Create(ctx, m); err != nil {
+		if errors.Is(err, eventsource.ErrConflict) {
+			return reply{}, awserrors.Errorf(awserrors.ErrorResourceConflictException,
+				"An event source mapping with SQS arn (\" %s \") and function (\" %s \") already exists. Please update or delete the existing mapping with UUID %s",
+				in.QueueARN, name, uuid)
+		}
+		return reply{}, err
+	}
+	return reply{Status: http.StatusAccepted, Body: mappingShape(m, "Creating")}, nil
+}
+
+// ownMapping reads a mapping of the caller's account and Region.
+func (d Deps) ownMapping(ctx context.Context, c *call) (MappingStore, eventsource.Mapping, uint64, error) {
+	if !mappingUUIDRE.MatchString(c.Params[0]) {
+		return nil, eventsource.Mapping{}, 0, mappingNotFound()
+	}
+	store, err := d.mappings(ctx)
+	if err != nil {
+		return nil, eventsource.Mapping{}, 0, err
+	}
+	m, rev, err := store.Get(ctx, c.Params[0])
+	if errors.Is(err, eventsource.ErrNotFound) || (err == nil && (m.AccountID != c.AccountID || m.Region != c.Region)) {
+		return nil, eventsource.Mapping{}, 0, mappingNotFound()
+	}
+	return store, m, rev, err
+}
+
+func (d Deps) getEventSourceMapping(ctx context.Context, c *call) (reply, error) {
+	_, m, _, err := d.ownMapping(ctx, c)
+	if err != nil {
+		return reply{}, err
+	}
+	return reply{Status: http.StatusOK, Body: mappingShape(m, m.State)}, nil
+}
+
+// deleteEventSourceMapping records Deleting and returns; the poller stops its
+// worker and only then removes the record.
+func (d Deps) deleteEventSourceMapping(ctx context.Context, c *call) (reply, error) {
+	for range 3 {
+		store, m, rev, err := d.ownMapping(ctx, c)
+		if err != nil {
+			return reply{}, err
+		}
+		if m.State == eventsource.StateDeleting {
+			return reply{Status: http.StatusAccepted, Body: mappingShape(m, m.State)}, nil
+		}
+		m.State = eventsource.StateDeleting
+		m.StateTransitionReason = eventsource.ReasonUserInitiated
+		m.LastModified = time.Now().UTC().Truncate(time.Millisecond)
+		_, err = store.Update(ctx, m, rev)
+		if errors.Is(err, eventsource.ErrConflict) {
+			continue
+		}
+		if err != nil {
+			return reply{}, err
+		}
+		return reply{Status: http.StatusAccepted, Body: mappingShape(m, m.State)}, nil
+	}
+	return reply{}, serviceError("the mapping changed concurrently; retry the request")
+}
+
+// listEventSourceMappings is refused: the mapping store can only be listed
+// whole, and a request must not wait on the whole inventory.
+func (d Deps) listEventSourceMappings(_ context.Context, _ *call) (reply, error) {
+	return reply{}, invalid("ListEventSourceMappings is not supported in this build; use GetEventSourceMapping with the mapping UUID")
+}

spinifex/domains/lambda/awsapi/eventsource_test.go

@@ -0,0 +1,230 @@
+package awsapi_test
+
+import (
+	"context"
+	"encoding/json"
+	"net/http"
+	"sync"
+	"testing"
+
+	testutil "github.com/mulgadc/spinifex/internal/testkit"
+	"github.com/mulgadc/spinifex/internal/testkit/gwtest"
+	iamdomain "github.com/mulgadc/spinifex/spinifex/domains/iam"
+	"github.com/mulgadc/spinifex/spinifex/domains/lambda/awsapi"
+	"github.com/mulgadc/spinifex/spinifex/domains/lambda/eventsource"
+	"github.com/mulgadc/spinifex/spinifex/domains/lambda/function"
+	awserrors "github.com/mulgadc/spinifex/spinifex/foundation/aws/errors"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+type fakeSources struct {
+	mu         sync.Mutex
+	visibility int64
+	err        error
+	asRole     []string
+}
+
+func (f *fakeSources) QueueVisibilityTimeout(_ context.Context, rec function.Record, _ string) (int64, error) {
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	f.asRole = append(f.asRole, rec.Role)
+	return f.visibility, f.err
+}
+
+const mappingsPath = "/2015-03-31/event-source-mappings/"
+
+var queueARN = "arn:aws:sqs:" + gwtest.Region + ":" + gwtest.AccountID + ":orders"
+
+func newMappingEnv(t *testing.T, statements ...iamdomain.Statement) (*env, *eventsource.Store, *fakeSources) {
+	t.Helper()
+	e := newEnv(t, statements...)
+	_, err := e.do(http.MethodPost, "/2015-03-31/functions", createBody("hello", func(b map[string]any) { b["Timeout"] = 30 }))
+	require.NoError(t, err)
+	_, _, js := testutil.StartTestJetStream(t)
+	store, err := eventsource.Open(t.Context(), js)
+	require.NoError(t, err)
+	src := &fakeSources{visibility: 30}
+	e.deps.Mappings = func(context.Context) (awsapi.MappingStore, error) { return store, nil }
+	e.deps.Sources = src
+	return e, store, src
+}
+
+func mappingBody(mutate func(map[string]any)) map[string]any {
+	b := map[string]any{"FunctionName": "hello", "EventSourceArn": queueARN}
+	if mutate != nil {
+		mutate(b)
+	}
+	return b
+}
+
+func decodeMapping(t *testing.T, raw []byte) map[string]any {
+	t.Helper()
+	var out map[string]any
+	require.NoError(t, json.Unmarshal(raw, &out))
+	return out
+}
+
+func TestEventSourceMapping_CreateGetDelete(t *testing.T) {
+	e, store, src := newMappingEnv(t)
+	rec, err := e.do(http.MethodPost, mappingsPath, mappingBody(func(b map[string]any) { b["BatchSize"] = 4 }))
+	require.NoError(t, err)
+	require.Equal(t, http.StatusAccepted, rec.Code)
+	created := decodeMapping(t, rec.Body.Bytes())
+	assert.Equal(t, "Creating", created["State"])
+	assert.EqualValues(t, 4, created["BatchSize"])
+	assert.Equal(t, queueARN, created["EventSourceArn"])
+	assert.Equal(t, []string{roleARN}, src.asRole, "the queue is read as the function's execution role")
+	uuid, _ := created["UUID"].(string)
+
+	stored, _, err := store.Get(t.Context(), uuid)
+	require.NoError(t, err)
+	assert.Equal(t, eventsource.StateEnabled, stored.State)
+
+	rec, err = e.do(http.MethodGet, mappingsPath+uuid, nil)
+	require.NoError(t, err)
+	assert.Equal(t, "Enabled", decodeMapping(t, rec.Body.Bytes())["State"])
+
+	_, err = e.do(http.MethodPost, mappingsPath, mappingBody(nil))
+	testutil.RequireAWSCode(t, err, awserrors.ErrorResourceConflictException)
+
+	for range 2 {
+		rec, err = e.do(http.MethodDelete, mappingsPath+uuid, nil)
+		require.NoError(t, err)
+		require.Equal(t, http.StatusAccepted, rec.Code)
+		assert.Equal(t, "Deleting", decodeMapping(t, rec.Body.Bytes())["State"])
+	}
+	stored, _, err = store.Get(t.Context(), uuid)
+	require.NoError(t, err, "delete leaves the record for the poller to remove")
+	assert.Equal(t, eventsource.StateDeleting, stored.State)
+}
+
+func TestEventSourceMapping_ConcurrentDuplicateCreatesYieldOneMapping(t *testing.T) {
+	first, store, src := newMappingEnv(t)
+	const n = 8
+	gateways := []*env{first}
+	for len(gateways) < n {
+		e := newEnv(t)
+		_, err := e.do(http.MethodPost, "/2015-03-31/functions", createBody("hello", func(b map[string]any) { b["Timeout"] = 30 }))
+		require.NoError(t, err)
+		e.deps.Mappings = first.deps.Mappings
+		e.deps.Sources = src
+		gateways = append(gateways, e)
+	}
+	codes := make(chan error, n)
+	var wg sync.WaitGroup
+	for _, e := range gateways {
+		wg.Go(func() {
+			_, err := e.do(http.MethodPost, mappingsPath, mappingBody(nil))
+			codes <- err
+		})
+	}
+	wg.Wait()
+	close(codes)
+	created := 0
+	for err := range codes {
+		if err == nil {
+			created++
+			continue
+		}
+		testutil.RequireAWSCode(t, err, awserrors.ErrorResourceConflictException)
+	}
+	assert.Equal(t, 1, created)
+	entries, err := store.List(t.Context())
+	require.NoError(t, err)
+	assert.Len(t, entries, 1)
+}
+
+func TestEventSourceMapping_ListIsRefusedWithoutReadingTheStore(t *testing.T) {
+	e, store, _ := newMappingEnv(t)
+	_, err := e.do(http.MethodPost, mappingsPath, mappingBody(nil))
+	require.NoError(t, err)
+	for _, path := range []string{mappingsPath, mappingsPath + "?FunctionName=hello", mappingsPath + "?MaxItems=2"} {
+		_, err = e.do(http.MethodGet, path, nil)
+		testutil.RequireAWSCode(t, err, awserrors.ErrorInvalidParameterValueException)
+	}
+	entries, err := store.List(t.Context())
+	require.NoError(t, err)
+	assert.Len(t, entries, 1, "the refusal leaves the mapping in place")
+}
+
+func TestEventSourceMapping_RefusesWhatThisBuildDoesNotImplement(t *testing.T) {
+	e, store, _ := newMappingEnv(t)
+	cases := map[string]func(map[string]any){
+		"FIFO queue":      func(b map[string]any) { b["EventSourceArn"] = queueARN + ".fifo" },
+		"another account": func(b map[string]any) { b["EventSourceArn"] = "arn:aws:sqs:" + gwtest.Region + ":210987654321:orders" },
+		"another Region":  func(b map[string]any) { b["EventSourceArn"] = "arn:aws:sqs:us-east-1:" + gwtest.AccountID + ":orders" },
+		"not SQS": func(b map[string]any) {
+			b["EventSourceArn"] = "arn:aws:kinesis:" + gwtest.Region + ":" + gwtest.AccountID + ":stream/s"
+		},
+		"filter criteria":   func(b map[string]any) { b["FilterCriteria"] = map[string]any{"Filters": []any{}} },
+		"disabled":          func(b map[string]any) { b["Enabled"] = false },
+		"batching window":   func(b map[string]any) { b["MaximumBatchingWindowInSeconds"] = 5 },
+		"batch over 10":     func(b map[string]any) { b["BatchSize"] = 11 },
+		"batch zero":        func(b map[string]any) { b["BatchSize"] = 0 },
+		"qualified name":    func(b map[string]any) { b["FunctionName"] = "hello:1" },
+		"missing source":    func(b map[string]any) { delete(b, "EventSourceArn") },
+		"missing function":  func(b map[string]any) { delete(b, "FunctionName") },
+		"starting position": func(b map[string]any) { b["StartingPosition"] = "LATEST" },
+	}
+	for name, mutate := range cases {
+		t.Run(name, func(t *testing.T) {
+			_, err := e.do(http.MethodPost, mappingsPath, mappingBody(mutate))
+			testutil.RequireAWSCode(t, err, awserrors.ErrorInvalidParameterValueException)
+		})
+	}
+	entries, err := store.List(t.Context())
+	require.NoError(t, err)
+	assert.Empty(t, entries)
+}
+
+func TestEventSourceMapping_VisibilityTimeoutMustCoverTheFunctionTimeout(t *testing.T) {
+	e, store, src := newMappingEnv(t)
+	src.visibility = 29
+	_, err := e.do(http.MethodPost, mappingsPath, mappingBody(nil))
+	testutil.RequireAWSCode(t, err, awserrors.ErrorInvalidParameterValueException)
+
+	src.visibility, src.err = 0, awserrors.Errorf(awserrors.ErrorAccessDeniedException, "denied")
+	_, err = e.do(http.MethodPost, mappingsPath, mappingBody(nil))
+	testutil.RequireAWSCode(t, err, awserrors.ErrorAccessDeniedException)
+
+	entries, err := store.List(t.Context())
+	require.NoError(t, err)
+	assert.Empty(t, entries)
+
+	_, err = e.do(http.MethodPost, mappingsPath, mappingBody(func(b map[string]any) { b["FunctionName"] = "absent" }))
+	testutil.RequireAWSCode(t, err, awserrors.ErrorResourceNotFoundException)
+}
+
+func TestEventSourceMapping_AnotherAccountsMappingIsNotFound(t *testing.T) {
+	e, store, _ := newMappingEnv(t)
+	m := eventsource.Mapping{
+		FormatVersion: eventsource.FormatVersion, UUID: "0a1b2c3d-4e5f-4a6b-8c7d-9e0f1a2b3c4d", AccountID: "210987654321",
+		Region: gwtest.Region, FunctionName: "hello", EventSourceARN: "arn:aws:sqs:" + gwtest.Region + ":210987654321:orders",
+		BatchSize: 1, State: eventsource.StateEnabled,
+	}
+	require.NoError(t, store.Create(t.Context(), m))
+	_, err := e.do(http.MethodGet, mappingsPath+m.UUID, nil)
+	testutil.RequireAWSCode(t, err, awserrors.ErrorResourceNotFoundException)
+	_, err = e.do(http.MethodDelete, mappingsPath+m.UUID, nil)
+	testutil.RequireAWSCode(t, err, awserrors.ErrorResourceNotFoundException)
+	_, err = e.do(http.MethodGet, mappingsPath+"not-a-uuid", nil)
+	testutil.RequireAWSCode(t, err, awserrors.ErrorResourceNotFoundException)
+}
+
+func TestEventSourceMapping_GetAndDeleteAuthorizeAgainstTheMappingARN(t *testing.T) {
+	const uuid = "0a1b2c3d-4e5f-4a6b-8c7d-9e0f1a2b3c4d"
+	mapping := "arn:aws:lambda:" + gwtest.Region + ":" + gwtest.AccountID + ":event-source-mapping:" + uuid
+	e, store, _ := newMappingEnv(t, gwtest.Statement("Allow", "*", "*"), gwtest.Statement("Deny", "lambda:*", mapping))
+	require.NoError(t, store.Create(t.Context(), eventsource.Mapping{
+		FormatVersion: eventsource.FormatVersion, UUID: uuid, AccountID: gwtest.AccountID, Region: gwtest.Region,
+		FunctionName: "hello", EventSourceARN: queueARN, BatchSize: 1, State: eventsource.StateEnabled,
+	}))
+	_, err := e.do(http.MethodGet, mappingsPath+uuid, nil)
+	gwtest.AssertDenied(t, err)
+	_, err = e.do(http.MethodDelete, mappingsPath+uuid, nil)
+	gwtest.AssertDenied(t, err)
+	m, _, err := store.Get(t.Context(), uuid)
+	require.NoError(t, err)
+	assert.Equal(t, eventsource.StateEnabled, m.State)
+}

spinifex/domains/lambda/awsapi/handler_test.go

@@ -765,8 +765,9 @@ func TestAuthorization(t *testing.T) {
 
 func TestRegistration_ResolvesActionsAndInventory(t *testing.T) {
 	reg := awsapi.NewRegistration(awsapi.Deps{})
-	assert.Equal(t, []string{"CreateFunction", "GetFunction", "GetFunctionCodeSigningConfig", "GetFunctionConfiguration", "Invoke", "ListVersionsByFunction"}, reg.Inventory.Registered)
+	assert.Equal(t, []string{"CreateEventSourceMapping", "CreateFunction", "DeleteEventSourceMapping", "GetEventSourceMapping", "GetFunction", "GetFunctionCodeSigningConfig", "GetFunctionConfiguration", "Invoke", "ListEventSourceMappings", "ListVersionsByFunction"}, reg.Inventory.Registered)
 	assert.Equal(t, "ListVersionsByFunction", reg.ResolveAction(httptest.NewRequest(http.MethodGet, "/2015-03-31/functions/x/versions", nil)))
+	assert.Equal(t, "DeleteEventSourceMapping", reg.ResolveAction(httptest.NewRequest(http.MethodDelete, "/2015-03-31/event-source-mappings/x", nil)))
 	assert.Equal(t, "GetFunctionCodeSigningConfig", reg.ResolveAction(httptest.NewRequest(http.MethodGet, "/2020-06-30/functions/x/code-signing-config", nil)))
 	assert.Equal(t, "CreateFunction", reg.ResolveAction(httptest.NewRequest(http.MethodPost, "/2015-03-31/functions", nil)))
 	assert.Equal(t, "GetFunction", reg.ResolveAction(httptest.NewRequest(http.MethodGet, "/2015-03-31/functions/x", nil)))

spinifex/domains/lambda/awsapi/invoke.go

@@ -114,16 +114,32 @@ func (d Deps) invoke(ctx context.Context, c *call) (reply, error) {
 	if err != nil {
 		return reply{}, err
 	}
+	out, err := d.runSync(ctx, rec, revision, c.Invoke.Qualifier, c.Body)
+	if err != nil {
+		return reply{}, err
+	}
+	hdr := http.Header{}
+	hdr.Set("X-Amz-Executed-Version", out.ExecutedVersion)
+	if out.FunctionError != "" {
+		hdr.Set("X-Amz-Function-Error", out.FunctionError)
+	}
+	return reply{Status: http.StatusOK, Body: rawJSON(out.Payload), Header: hdr}, nil
+}
+
+// runSync runs one synchronous attempt on a resolved record and classifies
+// its result; every error is the AWS error Invoke returns. A nil error means
+// the result is valid, with or without a function error.
+func (d Deps) runSync(ctx context.Context, rec function.Record, revision uint64, qualifier string, payload []byte) (SyncResult, error) {
 	if rec.State != function.StateActive {
-		return reply{}, awserrors.Errorf(awserrors.ErrorResourceConflictException,
+		return SyncResult{}, awserrors.Errorf(awserrors.ErrorResourceConflictException,
 			"The operation cannot be performed at this time. The function is currently in the following state: %s", rec.State)
 	}
 	if d.Exec == nil || rec.Code.State != function.CodeStored {
-		return reply{}, serviceError("function execution is unavailable")
+		return SyncResult{}, serviceError("function execution is unavailable")
 	}
 
 	if err := ctx.Err(); err != nil {
-		return reply{}, err
+		return SyncResult{}, err
 	}
 	// Once accepted, a disconnecting caller does not cancel the attempt; only
 	// the wait is bounded, and a late result is discarded.
@@ -133,31 +149,37 @@ func (d Deps) invoke(ctx context.Context, c *call) (reply, error) {
 	out, err := d.Exec.InvokeSync(actx, SyncInvocation{
 		Function:  rec,
 		Revision:  revision,
-		Qualifier: c.Invoke.Qualifier,
-		Payload:   c.Body,
+		Qualifier: qualifier,
+		Payload:   payload,
 		WaitUntil: waitUntil,
 	})
 	if refusal, ok := errors.AsType[*RefusalError](err); ok {
 		if _, coded := awserrors.ResolveErrorCode(refusal.Err); coded {
-			return reply{}, refusal.Err
+			return SyncResult{}, refusal.Err
 		}
 		err = refusal
 	}
 	if err != nil {
 		slog.ErrorContext(ctx, "Lambda: synchronous attempt outcome unknown", "function_id", rec.FunctionID, "err", err)
-		return reply{}, serviceError("the invocation outcome is unknown")
+		return SyncResult{}, serviceError("the invocation outcome is unknown")
 	}
 	if out.ExecutedVersion == "" || len(out.Payload) > maxSyncPayload {
 		slog.ErrorContext(ctx, "Lambda: executor returned an invalid result", "function_id", rec.FunctionID,
 			"payload_bytes", len(out.Payload))
-		return reply{}, serviceError("the invocation outcome is unknown")
+		return SyncResult{}, serviceError("the invocation outcome is unknown")
 	}
-	hdr := http.Header{}
-	hdr.Set("X-Amz-Executed-Version", out.ExecutedVersion)
-	if out.FunctionError != "" {
-		hdr.Set("X-Amz-Function-Error", out.FunctionError)
+	return out, nil
+}
+
+// InvokeEventSource runs one event-source batch on $LATEST through the same
+// state, code and result checks as Invoke, without Invoke's IAM check: the
+// mapping was authorized when it was created.
+func (d Deps) InvokeEventSource(ctx context.Context, accountID, region, name string, payload []byte) (SyncResult, error) {
+	rec, revision, err := d.Meta.Get(ctx, accountID, region, name)
+	if err != nil {
+		return SyncResult{}, err
 	}
-	return reply{Status: http.StatusOK, Body: rawJSON(out.Payload), Header: hdr}, nil
+	return d.runSync(ctx, rec, revision, "", payload)
 }
 
 // checkInvokeOptions refuses every request option this build does not

spinifex/domains/lambda/awsapi/registration.go

@@ -54,6 +54,12 @@ type Deps struct {
 	AEAD cipher.AEAD
 	// Exec runs synchronous attempts; nil refuses every Invoke.
 	Exec Executor
+	// Mappings opens the event-source mapping store; nil refuses every
+	// mapping action.
+	Mappings func(ctx context.Context) (MappingStore, error)
+	// Sources reads a queue as a function's execution role; nil refuses
+	// CreateEventSourceMapping.
+	Sources SourceChecker
 }
 
 // NewRegistration returns the Lambda REST-JSON registration.
@@ -136,6 +142,8 @@ func (d Deps) serve(w http.ResponseWriter, inv dispatch.Invocation) error {
 		if name, _, err := resolveFunctionName(inv.Region, inv.AccountID, params[0]); err == nil {
 			resource = functionARN(inv.Region, inv.AccountID, name)
 		}
+	case "GetEventSourceMapping", "DeleteEventSourceMapping":
+		resource = mappingARN(inv.Region, inv.AccountID, params[0])
 	case "Invoke":
 		// Invoke validates deterministically before authorizing, and its
 		// resource comes from that same validated target. URL.Query drops
@@ -160,8 +168,7 @@ func (d Deps) serve(w http.ResponseWriter, inv dispatch.Invocation) error {
 		}
 	}
 	// Only CreateFunction and GetFunction touch the code store.
-	needsCodes := action != "GetFunctionConfiguration" && action != "GetFunctionCodeSigningConfig" &&
-		action != "ListVersionsByFunction" && action != "Invoke"
+	needsCodes := action == "CreateFunction" || action == "GetFunction"
 	if d.Meta == nil || (d.Codes == nil && needsCodes) {
 		return errors.New(awserrors.ErrorServiceException)
 	}

spinifex/domains/lambda/awsapi/routes.go

@@ -66,6 +66,30 @@ var routes = []route{
 		Handler: func(ctx context.Context, d Deps, c *call) (reply, error) {
 			return d.invoke(ctx, c)
 		}},
+	{Method: "POST", Pattern: "/2015-03-31/event-source-mappings/", Action: "CreateEventSourceMapping",
+		Handler: func(ctx context.Context, d Deps, c *call) (reply, error) {
+			return d.createEventSourceMapping(ctx, c)
+		}},
+	{Method: "POST", Pattern: "/2015-03-31/event-source-mappings", Action: "CreateEventSourceMapping",
+		Handler: func(ctx context.Context, d Deps, c *call) (reply, error) {
+			return d.createEventSourceMapping(ctx, c)
+		}},
+	{Method: "GET", Pattern: "/2015-03-31/event-source-mappings/", Action: "ListEventSourceMappings",
+		Handler: func(ctx context.Context, d Deps, c *call) (reply, error) {
+			return d.listEventSourceMappings(ctx, c)
+		}},
+	{Method: "GET", Pattern: "/2015-03-31/event-source-mappings", Action: "ListEventSourceMappings",
+		Handler: func(ctx context.Context, d Deps, c *call) (reply, error) {
+			return d.listEventSourceMappings(ctx, c)
+		}},
+	{Method: "GET", Pattern: "/2015-03-31/event-source-mappings/{UUID}", Action: "GetEventSourceMapping",
+		Handler: func(ctx context.Context, d Deps, c *call) (reply, error) {
+			return d.getEventSourceMapping(ctx, c)
+		}},
+	{Method: "DELETE", Pattern: "/2015-03-31/event-source-mappings/{UUID}", Action: "DeleteEventSourceMapping",
+		Handler: func(ctx context.Context, d Deps, c *call) (reply, error) {
+			return d.deleteEventSourceMapping(ctx, c)
+		}},
 }
 
 var router = rest.NewRouter("lambda", routes)

spinifex/domains/lambda/eventsource/mapping.go

@@ -0,0 +1,170 @@
+// Package eventsource holds the Lambda event-source mapping record and its
+// JetStream KV store.
+package eventsource
+
+import (
+	"context"
+	"encoding/json"
+	"errors"
+	"fmt"
+	"time"
+
+	"github.com/mulgadc/spinifex/spinifex/foundation/state/kvutil"
+	"github.com/nats-io/nats.go/jetstream"
+)
+
+// FormatVersion is the only record layout this build reads or writes.
+const FormatVersion = 1
+
+// KVBucket holds every mapping, keyed by UUID; the UUID is global in AWS.
+const KVBucket = "lambda-event-source-mappings"
+
+// Mapping states this build records. A mapping is Enabled from creation and
+// Deleting from the delete request until its poller has stopped.
+const (
+	StateEnabled  = "Enabled"
+	StateDeleting = "Deleting"
+)
+
+// ReasonUserInitiated is the transition reason of a state the caller asked for.
+const ReasonUserInitiated = "USER_INITIATED"
+
+// ErrNotFound is returned for a UUID with no mapping.
+var ErrNotFound = errors.New("eventsource: mapping not found")
+
+// ErrConflict is returned when the record changed since it was read.
+var ErrConflict = errors.New("eventsource: mapping changed concurrently")
+
+// Mapping is one SQS event-source mapping.
+type Mapping struct {
+	FormatVersion  int    `json:"formatVersion"`
+	UUID           string `json:"uuid"`
+	AccountID      string `json:"accountId"`
+	Region         string `json:"region"`
+	FunctionName   string `json:"functionName"`
+	FunctionARN    string `json:"functionArn"`
+	EventSourceARN string `json:"eventSourceArn"`
+	BatchSize      int64  `json:"batchSize"`
+	State          string `json:"state"`
+	// StateTransitionReason is USER_INITIATED, or the last receive failure
+	// prefixed "PROBLEM: " while polling cannot proceed.
+	StateTransitionReason string    `json:"stateTransitionReason"`
+	LastModified          time.Time `json:"lastModified"`
+}
+
+// Store reads and writes mappings in KVBucket.
+type Store struct{ kv jetstream.KeyValue }
+
+// Open returns the store, creating its bucket on first use.
+func Open(ctx context.Context, js jetstream.KeyValueManager) (*Store, error) {
+	kv, err := kvutil.GetOrCreateBucket(ctx, js, KVBucket, 1)
+	if err != nil {
+		return nil, fmt.Errorf("eventsource: open bucket: %w", err)
+	}
+	return &Store{kv: kv}, nil
+}
+
+func key(uuid string) string { return "esm." + uuid }
+
+// Create writes a new mapping; a UUID already present is a conflict.
+func (s *Store) Create(ctx context.Context, m Mapping) error {
+	b, err := json.Marshal(m)
+	if err != nil {
+		return err
+	}
+	if _, err := s.kv.Create(ctx, key(m.UUID), b); err != nil {
+		if errors.Is(err, jetstream.ErrKeyExists) {
+			return ErrConflict
+		}
+		return fmt.Errorf("eventsource: create: %w", err)
+	}
+	return nil
+}
+
+// Get returns the mapping and its KV revision.
+func (s *Store) Get(ctx context.Context, uuid string) (Mapping, uint64, error) {
+	e, err := s.kv.Get(ctx, key(uuid))
+	if errors.Is(err, jetstream.ErrKeyNotFound) {
+		return Mapping{}, 0, ErrNotFound
+	}
+	if err != nil {
+		return Mapping{}, 0, fmt.Errorf("eventsource: get: %w", err)
+	}
+	m, err := decode(e.Value())
+	return m, e.Revision(), err
+}
+
+// Entry is one listed mapping with the revision it was read at.
+type Entry struct {
+	Mapping  Mapping
+	Revision uint64
+}
+
+// List returns every mapping in the bucket once.
+func (s *Store) List(ctx context.Context) ([]Entry, error) {
+	lister, err := s.kv.ListKeys(ctx)
+	if err != nil {
+		return nil, fmt.Errorf("eventsource: list: %w", err)
+	}
+	defer func() { _ = lister.Stop() }()
+	var out []Entry
+	// A KV listing may repeat a key that changes while it is being listed.
+	seen := map[string]bool{}
+	for k := range lister.Keys() {
+		if seen[k] {
+			continue
+		}
+		seen[k] = true
+		e, err := s.kv.Get(ctx, k)
+		if errors.Is(err, jetstream.ErrKeyNotFound) {
+			continue
+		}
+		if err != nil {
+			return nil, fmt.Errorf("eventsource: list get: %w", err)
+		}
+		m, err := decode(e.Value())
+		if err != nil {
+			return nil, err
+		}
+		out = append(out, Entry{Mapping: m, Revision: e.Revision()})
+	}
+	return out, nil
+}
+
+// Update replaces the mapping only if it is still at revision.
+func (s *Store) Update(ctx context.Context, m Mapping, revision uint64) (uint64, error) {
+	b, err := json.Marshal(m)
+	if err != nil {
+		return 0, err
+	}
+	rev, err := s.kv.Update(ctx, key(m.UUID), b, revision)
+	if err != nil {
+		if errors.Is(err, jetstream.ErrKeyExists) {
+			return 0, ErrConflict
+		}
+		return 0, fmt.Errorf("eventsource: update: %w", err)
+	}
+	return rev, nil
+}
+
+// Remove deletes the mapping only if it is still at revision.
+func (s *Store) Remove(ctx context.Context, uuid string, revision uint64) error {
+	if err := s.kv.Delete(ctx, key(uuid), jetstream.LastRevision(revision)); err != nil {
+		if errors.Is(err, jetstream.ErrKeyExists) {
+			return ErrConflict
+		}
+		return fmt.Errorf("eventsource: remove: %w", err)
+	}
+	return nil
+}
+
+func decode(b []byte) (Mapping, error) {
+	var m Mapping
+	if err := json.Unmarshal(b, &m); err != nil {
+		return Mapping{}, fmt.Errorf("eventsource: decode: %w", err)
+	}
+	if m.FormatVersion != FormatVersion {
+		return Mapping{}, fmt.Errorf("eventsource: record format %d is not %d", m.FormatVersion, FormatVersion)
+	}
+	return m, nil
+}

spinifex/domains/lambda/eventsource/mapping_test.go

@@ -0,0 +1,42 @@
+package eventsource_test
+
+import (
+	"testing"
+
+	testutil "github.com/mulgadc/spinifex/internal/testkit"
+	"github.com/mulgadc/spinifex/spinifex/domains/lambda/eventsource"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestStore_RevisionsGuardEveryChange(t *testing.T) {
+	_, _, js := testutil.StartTestJetStream(t)
+	s, err := eventsource.Open(t.Context(), js)
+	require.NoError(t, err)
+
+	entries, err := s.List(t.Context())
+	require.NoError(t, err)
+	assert.Empty(t, entries)
+
+	m := eventsource.Mapping{FormatVersion: eventsource.FormatVersion, UUID: "u1", State: eventsource.StateEnabled}
+	require.NoError(t, s.Create(t.Context(), m))
+	require.ErrorIs(t, s.Create(t.Context(), m), eventsource.ErrConflict)
+
+	got, rev, err := s.Get(t.Context(), "u1")
+	require.NoError(t, err)
+	assert.Equal(t, m, got)
+
+	m.State = eventsource.StateDeleting
+	newRev, err := s.Update(t.Context(), m, rev)
+	require.NoError(t, err)
+	_, err = s.Update(t.Context(), m, rev)
+	require.ErrorIs(t, err, eventsource.ErrConflict, "a stale revision never overwrites")
+	require.ErrorIs(t, s.Remove(t.Context(), "u1", rev), eventsource.ErrConflict, "a stale revision never removes")
+
+	require.NoError(t, s.Remove(t.Context(), "u1", newRev))
+	_, _, err = s.Get(t.Context(), "u1")
+	require.ErrorIs(t, err, eventsource.ErrNotFound)
+	entries, err = s.List(t.Context())
+	require.NoError(t, err)
+	assert.Empty(t, entries)
+}

spinifex/domains/lambda/function/activate.go

@@ -0,0 +1,155 @@
+package function
+
+import (
+	"context"
+	"encoding/json"
+	"errors"
+	"fmt"
+	"log/slog"
+	"strings"
+	"time"
+
+	"github.com/nats-io/nats.go/jetstream"
+)
+
+// Readiness reports why a function whose code is stored cannot run yet, or nil
+// when it can. Its error text becomes the function's StateReason.
+type Readiness func(ctx context.Context, rec Record) error
+
+// SetReadiness enables activation. Without it no function ever leaves Pending.
+func (s *MetaService) SetReadiness(r Readiness) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	s.ready = r
+	if s.nudge == nil {
+		s.nudge = make(chan struct{}, 1)
+	}
+}
+
+func (s *MetaService) readiness() (Readiness, chan struct{}) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	return s.ready, s.nudge
+}
+
+// wake asks the activation loop for an early pass; it never blocks.
+func (s *MetaService) wake() {
+	if _, nudge := s.readiness(); nudge != nil {
+		select {
+		case nudge <- struct{}{}:
+		default:
+		}
+	}
+}
+
+// RunActivation activates pending functions until ctx ends: once at start, on
+// every code-stored completion, and every interval, so a crash between storing
+// the code and activating it is converged by the next pass.
+func (s *MetaService) RunActivation(ctx context.Context, interval time.Duration) {
+	ready, nudge := s.readiness()
+	if ready == nil {
+		return
+	}
+	t := time.NewTicker(interval)
+	defer t.Stop()
+	for {
+		if err := s.ActivatePending(ctx); err != nil && ctx.Err() == nil {
+			slog.Warn("Lambda: activation pass incomplete; retrying", "err", err)
+		}
+		select {
+		case <-ctx.Done():
+			return
+		case <-nudge:
+		case <-t.C:
+		}
+	}
+}
+
+// ActivatePending makes one pass over every account's functions, moving each
+// Pending function with stored code to Active when readiness allows it.
+func (s *MetaService) ActivatePending(ctx context.Context) error {
+	ready, _ := s.readiness()
+	if ready == nil {
+		return nil
+	}
+	var accounts []string
+	names := s.js.KeyValueStoreNames(ctx)
+	for name := range names.Name() {
+		if id, ok := strings.CutPrefix(name, KVBucketAccountPrefix); ok {
+			accounts = append(accounts, id)
+		}
+	}
+	if err := names.Error(); err != nil {
+		return fmt.Errorf("lambda: list account buckets: %w", err)
+	}
+	var errs []error
+	for _, accountID := range accounts {
+		if err := s.activateAccount(ctx, accountID, ready); err != nil {
+			errs = append(errs, err)
+		}
+	}
+	return errors.Join(errs...)
+}
+
+func (s *MetaService) activateAccount(ctx context.Context, accountID string, ready Readiness) error {
+	kv, err := s.bucket(ctx, accountID)
+	if err != nil {
+		return err
+	}
+	keys, err := kv.ListKeysFiltered(ctx, kvFunctionKeyPrefix+">")
+	if errors.Is(err, jetstream.ErrNoKeysFound) {
+		return nil
+	}
+	if err != nil {
+		return fmt.Errorf("lambda: list functions of account %s: %w", accountID, err)
+	}
+	var errs []error
+	for key := range keys.Keys() {
+		if err := s.activateOne(ctx, kv, key, ready); err != nil {
+			errs = append(errs, err)
+		}
+	}
+	return errors.Join(errs...)
+}
+
+// activateOne writes the readiness outcome by compare-and-swap at the revision
+// it read, so a concurrent update or delete wins and is never overwritten.
+func (s *MetaService) activateOne(ctx context.Context, kv jetstream.KeyValue, key string, ready Readiness) error {
+	entry, err := kv.Get(ctx, key)
+	if errors.Is(err, jetstream.ErrKeyNotFound) || errors.Is(err, jetstream.ErrKeyDeleted) {
+		return nil
+	}
+	if err != nil {
+		return fmt.Errorf("lambda: read %s: %w", key, err)
+	}
+	rec, err := decode(entry.Value())
+	if err != nil {
+		return err
+	}
+	if rec.State != StatePending || rec.Code.State != CodeStored {
+		return nil
+	}
+	if notReady := ready(ctx, rec); notReady != nil {
+		if rec.StateReason == notReady.Error() {
+			return nil
+		}
+		rec.StateReasonCode, rec.StateReason = ReasonCodeCreating, notReady.Error()
+	} else {
+		rec.State, rec.StateReasonCode, rec.StateReason = StateActive, "", ""
+	}
+	data, err := json.Marshal(rec)
+	if err != nil {
+		return err
+	}
+	if _, err := kv.Update(ctx, key, data, entry.Revision()); err != nil {
+		if errors.Is(err, jetstream.ErrKeyExists) {
+			return nil
+		}
+		return fmt.Errorf("lambda: write readiness of %s: %w", key, err)
+	}
+	if rec.State == StateActive {
+		slog.Info("Lambda: function active", "account_id", rec.AccountID, "region", rec.Region,
+			"function_name", rec.FunctionName, "function_id", rec.FunctionID)
+	}
+	return nil
+}

spinifex/domains/lambda/function/activate_test.go

@@ -0,0 +1,120 @@
+package function_test
+
+import (
+	"context"
+	"errors"
+	"testing"
+	"time"
+
+	"github.com/mulgadc/spinifex/internal/testkit"
+	"github.com/mulgadc/spinifex/spinifex/domains/lambda/function"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+const (
+	actAccount = "000000000001"
+	actRegion  = "ap-southeast-2"
+	actDigest  = "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa"
+)
+
+func newOwner(t *testing.T) *function.MetaService {
+	t.Helper()
+	_, nc, _ := testutil.StartTestJetStream(t)
+	return function.NewMetaService(testutil.NewJetStream(t, nc))
+}
+
+func create(t *testing.T, svc *function.MetaService, name string) uint64 {
+	t.Helper()
+	resp, err := svc.Create(t.Context(), &function.CreateRequest{Region: actRegion, Record: function.Record{
+		FunctionID: "id-" + name, FunctionName: name, Role: "arn:aws:iam::000000000001:role/r",
+		Runtime: "provided.al2023", Handler: "bootstrap", Timeout: 3, MemorySize: 128,
+		Code: function.Code{RevisionID: "rev-" + name},
+	}}, actAccount)
+	require.NoError(t, err)
+	return resp.Revision
+}
+
+func get(t *testing.T, svc *function.MetaService, name string) *function.GetResponse {
+	t.Helper()
+	resp, err := svc.Get(t.Context(), &function.GetRequest{Region: actRegion, FunctionName: name}, actAccount)
+	require.NoError(t, err)
+	return resp
+}
+
+func storeCode(t *testing.T, svc *function.MetaService, name string, rev uint64) {
+	t.Helper()
+	_, err := svc.CodeStored(t.Context(), &function.CodeStoredRequest{Region: actRegion, FunctionName: name, Revision: rev, SHA256: actDigest, Size: 9}, actAccount)
+	require.NoError(t, err)
+}
+
+func TestActivation_WithoutReadinessNothingLeavesPending(t *testing.T) {
+	svc := newOwner(t)
+	storeCode(t, svc, "hello", create(t, svc, "hello"))
+
+	require.NoError(t, svc.ActivatePending(t.Context()))
+	ctx, cancel := context.WithCancel(t.Context())
+	cancel()
+	svc.RunActivation(ctx, time.Millisecond)
+
+	got := get(t, svc, "hello").Record
+	assert.Equal(t, function.StatePending, got.State)
+	assert.Equal(t, function.ReasonCreating, got.StateReason)
+}
+
+func TestActivation_UnmetConditionStaysPendingWithItsReason(t *testing.T) {
+	svc := newOwner(t)
+	storeCode(t, svc, "hello", create(t, svc, "hello"))
+	svc.SetReadiness(func(context.Context, function.Record) error { return errors.New("The node is not serving.") })
+
+	require.NoError(t, svc.ActivatePending(t.Context()))
+	first := get(t, svc, "hello")
+	assert.Equal(t, function.StatePending, first.Record.State)
+	assert.Equal(t, function.ReasonCodeCreating, first.Record.StateReasonCode)
+	assert.Equal(t, "The node is not serving.", first.Record.StateReason)
+
+	require.NoError(t, svc.ActivatePending(t.Context()))
+	assert.Equal(t, first.Revision, get(t, svc, "hello").Revision, "an unchanged reason is not rewritten")
+}
+
+func TestActivation_ReadyStoredFunctionBecomesActive(t *testing.T) {
+	svc := newOwner(t)
+	storeCode(t, svc, "hello", create(t, svc, "hello"))
+	storingRev := create(t, svc, "storing")
+	var seen []string
+	svc.SetReadiness(func(_ context.Context, rec function.Record) error {
+		seen = append(seen, rec.FunctionName)
+		return nil
+	})
+
+	require.NoError(t, svc.ActivatePending(t.Context()))
+
+	got := get(t, svc, "hello").Record
+	assert.Equal(t, function.StateActive, got.State)
+	assert.Empty(t, got.StateReason)
+	assert.Empty(t, got.StateReasonCode)
+	assert.Equal(t, function.CodeStored, got.Code.State)
+	storing := get(t, svc, "storing")
+	assert.Equal(t, function.StatePending, storing.Record.State, "a function still storing its code is never activated")
+	assert.Equal(t, storingRev, storing.Revision)
+	assert.Equal(t, []string{"hello"}, seen)
+}
+
+func TestActivation_CodeStoredWakesTheLoop(t *testing.T) {
+	svc := newOwner(t)
+	storeCode(t, svc, "first", create(t, svc, "first"))
+	// Readiness is enabled after first is stored, so no wake is buffered and
+	// only the start pass can activate first.
+	svc.SetReadiness(func(context.Context, function.Record) error { return nil })
+	ctx, cancel := context.WithCancel(t.Context())
+	done := make(chan struct{})
+	go func() { defer close(done); svc.RunActivation(ctx, time.Hour) }()
+	t.Cleanup(func() { cancel(); <-done })
+	require.Eventually(t, func() bool { return get(t, svc, "first").Record.State == function.StateActive },
+		10*time.Second, 20*time.Millisecond, "the start pass activates what was already stored")
+
+	storeCode(t, svc, "hello", create(t, svc, "hello"))
+
+	require.Eventually(t, func() bool { return get(t, svc, "hello").Record.State == function.StateActive },
+		10*time.Second, 20*time.Millisecond, "after the start pass, only the code-stored wake can activate within the hour")
+}

spinifex/domains/lambda/function/meta.go

@@ -44,6 +44,8 @@ type MetaService struct {
 
 	mu      sync.Mutex
 	buckets map[string]jetstream.KeyValue
+	ready   Readiness
+	nudge   chan struct{}
 }
 
 // NewMetaService builds the service over a JetStream handle.
@@ -306,6 +308,7 @@ func (s *MetaService) CodeStored(ctx context.Context, req *CodeStoredRequest, ac
 		return nil, fmt.Errorf("lambda: update function record: %w", err)
 	}
 	attach(ctx, kv, up, upRev, rev)
+	s.wake()
 	return &CodeStoredResponse{Record: rec, Revision: rev}, nil
 }
 

spinifex/domains/ochre/fakes_test.go

@@ -360,6 +360,8 @@ func (f *fakeInstanceLauncher) TerminateSystemInstance(instanceID string) error
 	return nil
 }
 
+func (f *fakeInstanceLauncher) RebootSystemInstance(string) error { return nil }
+
 // fakeAMIResolver always resolves the same serving AMI.
 type fakeAMIResolver struct{}
 

spinifex/domains/rds/instance/delete.go

@@ -66,6 +66,12 @@ func (s *Service) DeleteDBInstance(ctx context.Context, input *rds.DeleteDBInsta
 			"DB instance %s is already being deleted with %s; a retry must repeat that choice", id, snapshotChoice(rec.FinalSnapshotIdentifier))
 	}
 
+	// The handler that accepted the delete tears it down, and the reconciler
+	// resumes one whose handler is gone, so a repeat only reports the state.
+	if rec.Status == StatusDeleting {
+		return &rds.DeleteDBInstanceOutput{DBInstance: s.projectDBInstance(rec)}, nil
+	}
+
 	// Reserve the final snapshot before the instance enters deleting. A collision
 	// therefore fails while the VM and data volume are still intact, and a crash
 	// afterwards leaves a creating record the reconciler can resolve.
@@ -74,29 +80,29 @@ func (s *Service) DeleteDBInstance(ctx context.Context, input *rds.DeleteDBInsta
 		return nil, err
 	}
 
-	// A repeat of a delete already under way re-runs the teardown rather than
-	// being rejected: the first call may have died partway through it.
-	if rec.Status != StatusDeleting {
-		now := time.Now().UTC()
-		rec.Status = StatusDeleting
-		rec.FailureReason = ""
-		rec.FinalSnapshotIdentifier = finalSnapshot
-		rec.TransitionStartedAt = &now
-		rec.UpdatedAt = now
-		if err := updateJSON(ctx, kv, DBInstanceKey(id), rev, rec); err != nil {
-			if reservation.creator {
-				s.rollbackFinalSnapshotReservation(ctx, kv, finalSnapshot, reservation.revision)
-			}
-			if errors.Is(err, kvstore.ErrConflict) {
-				return nil, awserrors.Errorf(awserrors.ErrorDBInstanceInvalidState,
-					"DB instance %s changed state concurrently; retry the delete", id)
-			}
-			return nil, err
+	now := time.Now().UTC()
+	rec.Status = StatusDeleting
+	rec.FailureReason = ""
+	rec.FinalSnapshotIdentifier = finalSnapshot
+	rec.TransitionStartedAt = &now
+	rec.UpdatedAt = now
+	if err := updateJSON(ctx, kv, DBInstanceKey(id), rev, rec); err != nil {
+		if reservation.creator {
+			s.rollbackFinalSnapshotReservation(ctx, kv, finalSnapshot, reservation.revision)
 		}
-		s.RecordEvent(ctx, accountID, EventSourceTypeDBInstance, id, "DB instance deleted.", EventCategoryDeletion)
+		if errors.Is(err, kvstore.ErrConflict) {
+			return nil, awserrors.Errorf(awserrors.ErrorDBInstanceInvalidState,
+				"DB instance %s changed state concurrently; retry the delete", id)
+		}
+		return nil, err
 	}
+	s.RecordEvent(ctx, accountID, EventSourceTypeDBInstance, id, "DB instance deleted.", EventCategoryDeletion)
 
-	if err := s.teardownDBInstance(ctx, kv, accountID, rec, reservation.creator); err != nil {
+	// Ends exactly when the reconciler's guard does, so the teardown is this
+	// handler's alone until the reconciler may resume it.
+	teardownCtx, cancel := context.WithDeadline(ctx, deleteClaimEnds(rec))
+	defer cancel()
+	if err := s.teardownDBInstance(teardownCtx, kv, accountID, rec, reservation.creator); err != nil {
 		// The reconciler finishes the delete once the modify lets go.
 		if errors.Is(err, errModifyInProgress) {
 			return &rds.DeleteDBInstanceOutput{DBInstance: s.projectDBInstance(rec)}, nil
@@ -202,6 +208,11 @@ func (s *Service) teardownDBInstance(ctx context.Context, kv *kvstore.Bucket, ac
 		s.stopEngineOrRecordFallback(ctx, accountID, rec, "deleting")
 	}
 
+	// The engine stop swallows a deadline, and a handler past its claim must not
+	// start a terminate the reconciler may already be running.
+	if err := ctx.Err(); err != nil {
+		return err
+	}
 	if err := s.terminateInstanceVM(ctx, rec.InstanceID, rec.DataVolumeID); err != nil {
 		return fmt.Errorf("rds: terminate the VM behind %s: %w", rec.DBInstanceIdentifier, err)
 	}

spinifex/domains/rds/instance/delete_test.go

@@ -1,9 +1,11 @@
 package instance
 
 import (
+	"context"
 	"errors"
 	"fmt"
 	"testing"
+	"time"
 
 	"github.com/aws/aws-sdk-go/aws"
 	"github.com/aws/aws-sdk-go/service/rds"
@@ -176,13 +178,12 @@ func TestDeleteDBInstance_ReservesTheFinalSnapshotBeforeCuttingIt(t *testing.T)
 }
 
 // A dead worker can leave the EC2 snapshot cut while its RDS record still says
-// creating. A retry adopts that exact account-scoped snapshot instead of
-// cutting another one and permanently pinning the source volume twice.
-func TestDeleteDBInstance_AdoptsAnInterruptedFinalSnapshot(t *testing.T) {
+// creating. The resumed teardown adopts that exact account-scoped snapshot
+// instead of cutting another one and permanently pinning the source volume twice.
+func TestReconciler_AdoptsAnInterruptedFinalSnapshot(t *testing.T) {
 	t.Parallel()
 	h := newLifecycleHarness(t, false)
-	rec := availableRecord()
-	rec.Status = StatusDeleting
+	rec := abandonedDelete(availableRecord())
 	rec.FinalSnapshotIdentifier = "orders-db-final"
 	seedInstance(t, h.svc, rec)
 
@@ -201,13 +202,9 @@ func TestDeleteDBInstance_AdoptsAnInterruptedFinalSnapshot(t *testing.T) {
 		},
 	}
 
-	_, err = h.svc.DeleteDBInstance(t.Context(), &rds.DeleteDBInstanceInput{
-		DBInstanceIdentifier:      aws.String(testDBID),
-		FinalDBSnapshotIdentifier: aws.String(rec.FinalSnapshotIdentifier),
-	}, testAccountID)
-	require.NoError(t, err)
+	require.NoError(t, onePass(t, NewReconciler(h.svc, "node-a")))
 
-	assert.Empty(t, h.snaps.created, "the retry must adopt the cut snapshot, not create another")
+	assert.Empty(t, h.snaps.created, "the resumed teardown must adopt the cut snapshot, not create another")
 	stored, found := h.snapshotRecord(t, rec.FinalSnapshotIdentifier)
 	require.True(t, found)
 	assert.Equal(t, SnapshotStatusAvailable, stored.Status)
@@ -325,26 +322,18 @@ func TestDeleteDBInstance_RetainsTheVolumeWhenTheVolumeStoreReportsItInUse(t *te
 
 // A teardown that died partway through is replayed, so every step has to treat
 // a resource that is already gone as work it has done.
-func TestDeleteDBInstance_IsIdempotentAcrossARetriedTeardown(t *testing.T) {
+func TestReconciler_IsIdempotentAcrossARetriedTeardown(t *testing.T) {
 	t.Parallel()
 	h := newLifecycleHarness(t, false)
-	rec := availableRecord()
-	// The shape a delete leaves behind when its caller died mid-teardown.
-	rec.Status = StatusDeleting
+	rec := abandonedDelete(availableRecord())
 	rec.FinalSnapshotIdentifier = "orders-db-final"
 	seedInstance(t, h.svc, rec)
-
-	input := &rds.DeleteDBInstanceInput{
-		DBInstanceIdentifier:      aws.String(testDBID),
-		FinalDBSnapshotIdentifier: aws.String("orders-db-final"),
-	}
-	_, err := h.svc.DeleteDBInstance(t.Context(), input, testAccountID)
-	require.NoError(t, err)
+	reconciler := NewReconciler(h.svc, "node-a")
+	require.NoError(t, onePass(t, reconciler))
 
 	// The record left behind by the interrupted attempt, replayed.
 	seedInstance(t, h.svc, rec)
-	_, err = h.svc.DeleteDBInstance(t.Context(), input, testAccountID)
-	require.NoError(t, err)
+	require.NoError(t, onePass(t, reconciler))
 
 	assert.Len(t, h.snaps.created, 1, "a replayed teardown must not take a second final snapshot")
 	assert.False(t, h.recordExists(t, testDBID))
@@ -355,9 +344,7 @@ func TestDeleteDBInstance_IsIdempotentAcrossARetriedTeardown(t *testing.T) {
 func TestReconciler_ResumesAnInterruptedDelete(t *testing.T) {
 	t.Parallel()
 	h := newLifecycleHarness(t, false)
-	rec := availableRecord()
-	rec.Status = StatusDeleting
-	seedInstance(t, h.svc, rec)
+	seedInstance(t, h.svc, abandonedDelete(availableRecord()))
 
 	reconciler := NewReconciler(h.svc, "node-a")
 	require.NoError(t, onePass(t, reconciler))
@@ -366,6 +353,87 @@ func TestReconciler_ResumesAnInterruptedDelete(t *testing.T) {
 	assert.False(t, h.recordExists(t, testDBID))
 }
 
+// A second teardown racing the handler's sees the VM mid-shutdown as gone and
+// fails on the still-attached volume, so the handler's delete is left to it.
+func TestReconciler_LeavesADeleteItsHandlerMayStillBeRunning(t *testing.T) {
+	t.Parallel()
+	h := newLifecycleHarness(t, false)
+	rec := availableRecord()
+	started := time.Now().UTC()
+	rec.Status = StatusDeleting
+	rec.TransitionStartedAt = &started
+	seedInstance(t, h.svc, rec)
+
+	require.NoError(t, onePass(t, NewReconciler(h.svc, "node-a")))
+
+	assert.Empty(t, h.launcher.terminated)
+	assert.True(t, h.recordExists(t, testDBID))
+}
+
+// A repeat must not start a teardown of its own alongside the one under way;
+// the reconciler is what resumes an abandoned one.
+func TestDeleteDBInstance_ARepeatStartsNoTeardown(t *testing.T) {
+	t.Parallel()
+	h := newLifecycleHarness(t, false)
+	seedInstance(t, h.svc, abandonedDelete(availableRecord()))
+
+	out, err := h.svc.DeleteDBInstance(t.Context(), skipFinalSnapshot(), testAccountID)
+	require.NoError(t, err)
+
+	assert.Equal(t, string(StatusDeleting), aws.StringValue(out.DBInstance.DBInstanceStatus))
+	assert.Empty(t, h.launcher.terminated)
+	assert.True(t, h.recordExists(t, testDBID))
+}
+
+// The reconciler resumes a delete from the persisted stamp, so the handler's
+// teardown must stop at that same instant, however long the acceptance took.
+func TestDeleteDBInstance_TeardownEndsWhenTheClaimDoes(t *testing.T) {
+	t.Parallel()
+	h := newLifecycleHarness(t, false)
+	seedInstance(t, h.svc, availableRecord())
+	var claimEnds time.Time
+	h.snaps.beforeCreate = func() { stored := h.record(t); claimEnds = deleteClaimEnds(&stored) }
+
+	_, err := h.svc.DeleteDBInstance(t.Context(), &rds.DeleteDBInstanceInput{
+		DBInstanceIdentifier:      aws.String(testDBID),
+		FinalDBSnapshotIdentifier: aws.String("orders-db-final"),
+	}, testAccountID)
+	require.NoError(t, err)
+
+	require.False(t, claimEnds.IsZero())
+	assert.True(t, h.snaps.deadline.Equal(claimEnds), "deadline %s, claim ends %s", h.snaps.deadline, claimEnds)
+}
+
+// A graceful stop that runs into the claim's end is swallowed, so the teardown
+// itself has to notice and leave the terminate to the reconciler.
+func TestTeardownDBInstance_StartsNoTerminatePastItsDeadline(t *testing.T) {
+	t.Parallel()
+	h := newLifecycleHarness(t, false)
+	h.agent.silenceType(CommandStopEngine)
+	rec := availableRecord()
+	rec.Status = StatusDeleting
+	rec.FinalSnapshotIdentifier = "orders-db-final"
+	seedInstance(t, h.svc, rec)
+	kv, err := h.svc.bucket(t.Context(), testAccountID)
+	require.NoError(t, err)
+
+	ctx, cancel := context.WithTimeout(t.Context(), 200*time.Millisecond)
+	defer cancel()
+	err = h.svc.teardownDBInstance(ctx, kv, testAccountID, &rec, true)
+
+	require.ErrorIs(t, err, context.DeadlineExceeded)
+	assert.Empty(t, h.launcher.terminated)
+	assert.True(t, h.recordExists(t, testDBID))
+}
+
+// The shape a delete leaves once its handler's request budget has run out.
+func abandonedDelete(rec DBInstanceRecord) DBInstanceRecord {
+	started := time.Now().UTC().Add(-deleteTimeout - time.Minute)
+	rec.Status = StatusDeleting
+	rec.TransitionStartedAt = &started
+	return rec
+}
+
 // A stop whose caller died leaves the VM possibly still running, so the stop is
 // re-issued rather than assumed.
 func TestReconciler_ResumesAnInterruptedStop(t *testing.T) {

spinifex/domains/rds/instance/lifecycle_test.go

@@ -117,7 +117,9 @@ func (f *fakeInstanceCommander) DetachVolumeFromStopped(_ context.Context, insta
 // DescribeSnapshots reports against the data volume, which is what decides
 // between deleting the volume and retaining it.
 type fakeSnapshots struct {
-	created       []*ec2.CreateSnapshotInput
+	created []*ec2.CreateSnapshotInput
+	// The context deadline the last CreateSnapshot ran under.
+	deadline      time.Time
 	holding       []string
 	deleted       []string
 	createErr     error
@@ -132,7 +134,8 @@ type fakeSnapshots struct {
 
 var _ snapshotProvider = (*fakeSnapshots)(nil)
 
-func (f *fakeSnapshots) CreateSnapshot(_ context.Context, input *ec2.CreateSnapshotInput, _ string) (*ec2.Snapshot, error) {
+func (f *fakeSnapshots) CreateSnapshot(ctx context.Context, input *ec2.CreateSnapshotInput, _ string) (*ec2.Snapshot, error) {
+	f.deadline, _ = ctx.Deadline()
 	if f.beforeCreate != nil {
 		f.beforeCreate()
 	}

spinifex/domains/rds/instance/reconciler.go

@@ -610,6 +610,11 @@ func (r *Reconciler) reconcileStopping(ctx context.Context, kv *kvstore.Bucket,
 // did; only a teardown still stuck past the bound is called failed, which is
 // what lets the customer retry the delete.
 func (r *Reconciler) reconcileDeleting(ctx context.Context, kv *kvstore.Bucket, rev uint64, accountID string, rec *DBInstanceRecord) error {
+	// The handler that accepted the delete tears it down until its claim ends; a
+	// second teardown alongside it sees the VM mid-shutdown as gone.
+	if time.Now().Before(deleteClaimEnds(rec)) {
+		return nil
+	}
 	err := r.svc.teardownDBInstance(ctx, kv, accountID, rec, false)
 	if err == nil || errors.Is(err, errFinalSnapshotInProgress) || errors.Is(err, errModifyInProgress) {
 		return nil
@@ -795,6 +800,12 @@ func (s *Service) findEC2SnapshotFor(ctx context.Context, accountID, dbSnapshotI
 // When the transition began. A record written by an older control plane carries
 // no stamp, so its last write stands in — it is never earlier than the
 // transition, so the bound cannot be under-counted.
+// When the handler that accepted a delete gives up its teardown, measured from
+// the persisted stamp so the handler and the reconciler agree on it.
+func deleteClaimEnds(rec *DBInstanceRecord) time.Time {
+	return transitionStarted(rec).Add(deleteTimeout)
+}
+
 func transitionStarted(rec *DBInstanceRecord) time.Time {
 	if rec.TransitionStartedAt != nil {
 		return *rec.TransitionStartedAt

spinifex/operator/accountlifecycle/absence_proof_silent_node_test.go

@@ -1,89 +0,0 @@
-package accountlifecycle
-
-//test:in-package — the volume reaper and its fake cluster are unexported, and
-// these drive its force path against a node that never answers.
-
-// Every test here waits out the fixed describe fan-out deadline a silent node
-// costs, two or three times over. This file sorts first so they take parallel
-// slots before the package's short tests queue ahead of them.
-
-import (
-	"testing"
-
-	"github.com/mulgadc/spinifex/spinifex/foundation/aws/errors"
-	"github.com/stretchr/testify/assert"
-	"github.com/stretchr/testify/require"
-)
-
-// A lone node's NotFound says nothing about the node that stayed silent, so
-// the holder is not proven gone and the volume keeps its attachment.
-func TestVolumeReaperForceRetainsAVolumeWhenALoneNodeAnswersNotFoundWhileAnotherIsSilent(t *testing.T) {
-	t.Parallel()
-	assertForceRetainsTheHeldVolume(t, func(cluster *fakeCluster) {
-		cluster.fail("ec2.DescribeInstances", awserrors.ErrorInvalidInstanceIDNotFound)
-	}, []string{fakeClusterNode, "node-2"})
-}
-
-// A terminated copy is not proof either while the node that may still run the
-// holder has not answered.
-func TestVolumeReaperForceRetainsAVolumeWhenATerminatedCopyIsFoundWhileTheHolderNodeIsSilent(t *testing.T) {
-	t.Parallel()
-	assertForceRetainsTheHeldVolume(t, func(cluster *fakeCluster) {
-		cluster.reply("ec2.DescribeTerminatedInstances", instancesIn("terminated", "i-held"))
-	}, []string{fakeClusterNode, "node-2"})
-}
-
-// The holder lives on node-2, which never answers, while node-1 answers every
-// fan-out twice. A count of replies reads that as complete and the holder as
-// gone; only judging by responder identity sees that node-2 is missing.
-func TestVolumeReaperForceRetainsAVolumeWhenOneNodeAnswersTwiceAndTheHolderNodeIsSilent(t *testing.T) {
-	t.Parallel()
-	cluster := newFakeCluster(t)
-	volume := newHeldVolume(cluster, "vol-held", "i-on-silent-node")
-	cluster.echoAsNode(t, "ec2.DescribeInstances", fakeClusterNode)
-
-	reaper := newVolumeReaper(cluster.nc, fakeClusterNode, "node-2")
-	err := reaper.Delete(silentClusterCtx(t), "000000000042", Resource{Kind: "volume", ID: "vol-held"}, true)
-
-	require.Error(t, err)
-	assert.Contains(t, err.Error(), "vol-held")
-	assert.Contains(t, err.Error(), "i-on-silent-node")
-	deleted, holder, detaches := volume.state()
-	assert.False(t, deleted, "a holder on a silent node is not proven gone")
-	assert.Equal(t, "i-on-silent-node", holder)
-	assert.Zero(t, detaches)
-	assert.NotContains(t, cluster.called(), forceDetachSubject)
-}
-
-func TestForcedTeardownKeepsTheAccountWhenOneNodeAnswersTwiceAndTheHolderNodeIsSilent(t *testing.T) {
-	t.Parallel()
-	cluster := newFakeCluster(t)
-	volume := newHeldVolume(cluster, "vol-held", "i-on-silent-node")
-	cluster.echoAsNode(t, "ec2.DescribeInstances", fakeClusterNode)
-
-	accounts := newFakeAccounts()
-	engine := quietEngine(accounts, newVolumeReaper(cluster.nc, fakeClusterNode, "node-2"))
-	req := request()
-	req.Force = true
-
-	result, err := engine.Teardown(silentClusterCtx(t), req)
-
-	require.ErrorIs(t, err, ErrResourcesStuck)
-	require.Equal(t, 1, result.StuckCount())
-	var stuck Stuck
-	for _, stage := range result.Stages {
-		if len(stage.Stuck) > 0 {
-			stuck = stage.Stuck[0]
-		}
-	}
-	assert.Equal(t, "vol-held", stuck.Resource.ID)
-	assert.Contains(t, stuck.Reason, "vol-held")
-	assert.Contains(t, stuck.Reason, "i-on-silent-node")
-	assert.False(t, result.AccountDeleted)
-	assert.False(t, accounts.deleted)
-	assert.Equal(t, AccountStatusTerminating, accounts.account.Status)
-	deleted, holder, detaches := volume.state()
-	assert.False(t, deleted)
-	assert.Equal(t, "i-on-silent-node", holder)
-	assert.Zero(t, detaches)
-}

spinifex/operator/accountlifecycle/cluster.go

@@ -35,6 +35,7 @@ func NewClusterEngine(
 	svc IAMBackend,
 	buckets BucketStore,
 	cleaner PlatformCleaner,
+	options ...ClusterOption,
 ) (*Engine, error) {
 	// Refused rather than skipped. A teardown missing its bucket reaper still
 	// reports every stage drained and still deletes the account record, leaving
@@ -95,5 +96,20 @@ func NewClusterEngine(
 	// May be nil: a dry run needs no cleanup, and Teardown and Exercise refuse
 	// to start without it.
 	engine.Cleanup = cleaner
+	for _, option := range options {
+		option(engine)
+	}
 	return engine, nil
 }
+
+// ClusterOption binds an optional capability to the engine NewClusterEngine builds.
+type ClusterOption func(*Engine)
+
+// WithComputeCleanup binds the EC2 compute and storage cleanup. Teardown
+// refuses to start without both, so an engine built without them can still
+// answer a dry run but never deletes an account.
+func WithComputeCleanup(compute ComputeStage, storage StorageStage) ClusterOption {
+	return func(e *Engine) {
+		e.Compute, e.Storage = compute, storage
+	}
+}

spinifex/operator/accountlifecycle/converge.go

@@ -0,0 +1,317 @@
+package accountlifecycle
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"slices"
+	"strings"
+	"time"
+
+	operatorv1 "github.com/mulgadc/spinifex/contracts/operator/v1"
+	"github.com/mulgadc/spinifex/spinifex/domains/ec2/accountcleanup"
+	otelsetup "github.com/mulgadc/spinifex/spinifex/foundation/telemetry"
+)
+
+// ComputeStage is the EC2 compute cleanup and its read-only remainder.
+type ComputeStage interface {
+	accountcleanup.ComputeCleaner
+	accountcleanup.ComputeRemainder
+}
+
+// StorageStage is the EC2 storage cleanup and its read-only remainder.
+type StorageStage interface {
+	accountcleanup.StorageCleaner
+	accountcleanup.StorageRemainder
+}
+
+var errComputeCleanupMissing = errors.New("teardown needs the EC2 compute and storage cleanup")
+
+// Converge bounds the loop that repeats the compute and storage cleanup inside
+// one teardown. Each cleanup call returns without waiting for teardown, so the
+// loop is what turns a terminate into a deleted volume.
+type Converge struct {
+	// Budget is the longest the loop runs, well under the job's own budget.
+	Budget time.Duration
+	// MinBackoff and MaxBackoff bound the wait between passes, which doubles.
+	MinBackoff time.Duration
+	MaxBackoff time.Duration
+	// IncompletePasses is how many passes in a row may end with a read that did
+	// not complete before the loop gives up.
+	IncompletePasses int
+}
+
+// DefaultConverge is sized under the 2 h job budget and over the 10 min
+// stuck-terminate backstop, so a wedged terminate still ends the loop clean.
+func DefaultConverge() Converge {
+	return Converge{Budget: 30 * time.Minute, MinBackoff: 5 * time.Second, MaxBackoff: time.Minute, IncompletePasses: 3}
+}
+
+func (e *Engine) convergeConfig() Converge {
+	cfg := DefaultConverge()
+	if e.Converge.Budget > 0 {
+		cfg.Budget = e.Converge.Budget
+	}
+	if e.Converge.MinBackoff > 0 {
+		cfg.MinBackoff = e.Converge.MinBackoff
+	}
+	if e.Converge.MaxBackoff > 0 {
+		cfg.MaxBackoff = e.Converge.MaxBackoff
+	}
+	if e.Converge.IncompletePasses > 0 {
+		cfg.IncompletePasses = e.Converge.IncompletePasses
+	}
+	return cfg
+}
+
+// holdsFence stops a holder that has been superseded or cancelled before it
+// issues another terminate or delete. It reads the account back and requires
+// it to still be TERMINATING under this operation and fence.
+func (e *Engine) holdsFence(ctx context.Context, req Request) error {
+	if err := ctx.Err(); err != nil {
+		return err
+	}
+	account, err := e.Accounts.GetAccount(req.AccountID)
+	if err != nil {
+		return fmt.Errorf("read account to confirm the fence: %w", err)
+	}
+	op := account.TerminationOp
+	if account.Status != AccountStatusTerminating || op == nil || op.OperationID != req.OperationID || op.Fence != req.Fence {
+		return errFenceLost
+	}
+	return nil
+}
+
+var errFenceLost = errors.New("the account is no longer held under this operation and fence")
+
+func waitContext(ctx context.Context, d time.Duration) error {
+	timer := time.NewTimer(d)
+	defer timer.Stop()
+	select {
+	case <-ctx.Done():
+		return ctx.Err()
+	case <-timer.C:
+		return nil
+	}
+}
+
+// convergent reports whether time alone can resolve an item.
+func convergent(it accountcleanup.Item) bool {
+	if it.Disposition == accountcleanup.DispositionPending {
+		return true
+	}
+	switch it.Reason {
+	case accountcleanup.ReasonTerminating, accountcleanup.ReasonVolumeDeleting, accountcleanup.ReasonTeardownUnconfirmed:
+		return true
+	}
+	return false
+}
+
+// passSignature identifies what a pass left, so two identical passes in a row
+// with nothing convergent show the loop is not going anywhere.
+func passSignature(legacy []Stuck, outcomes ...*accountcleanup.Outcome) string {
+	var lines []string
+	for _, s := range legacy {
+		lines = append(lines, "legacy/"+s.Resource.Kind+"/"+s.Resource.ID+"/"+s.Reason)
+	}
+	for _, out := range outcomes {
+		for _, it := range out.Items {
+			lines = append(lines, it.Kind+"/"+it.ResourceID+"/"+it.Disposition+"/"+it.Reason+"/"+it.State)
+		}
+		lines = append(lines, "incomplete:"+strings.Join(out.IncompleteReasons, ","))
+	}
+	slices.Sort(lines)
+	return strings.Join(lines, "\n")
+}
+
+// convergeCompute repeats the compute cleanup, the storage-position legacy
+// reapers, the storage cleanup and the remainder read until the account holds
+// no instance, spot request or volume, or the loop proves it cannot make
+// progress. It ends the teardown with the stuck items named, never with a
+// shorter list: a lost fence or a cancelled context is an error.
+func (e *Engine) convergeCompute(ctx context.Context, req Request, result *Result) error {
+	cfg := e.convergeConfig()
+	started := time.Now()
+	deadline := started.Add(cfg.Budget)
+	backoff := cfg.MinBackoff
+	wait := e.wait
+	if wait == nil {
+		wait = waitContext
+	}
+
+	attachments := StageResult{Stage: StageAttachments}
+	storage := StageResult{Stage: StageStorage}
+	var lastSignature string
+	var idlePasses, incompletePasses int
+
+	for pass := 1; ; pass++ {
+		if err := e.holdsFence(ctx, req); err != nil {
+			return err
+		}
+		computeOut, err := e.Compute.CleanupCompute(ctx, &accountcleanup.ComputeRequest{
+			AccountID: req.AccountID, CorrelationRef: req.OperationID, Force: req.Force,
+		})
+		if err != nil {
+			return fmt.Errorf("compute cleanup: %w", err)
+		}
+		legacy, err := e.runStage(ctx, StageStorage, req)
+		if err != nil {
+			return err
+		}
+		storageOut, err := e.Storage.CleanupStorage(ctx, &accountcleanup.StorageRequest{
+			AccountID: req.AccountID, CorrelationRef: req.OperationID, Force: req.Force,
+		})
+		if err != nil {
+			return fmt.Errorf("storage cleanup: %w", err)
+		}
+		computeLeft, err := e.Compute.ComputeRemaining(ctx, req.AccountID)
+		if err != nil {
+			return fmt.Errorf("compute remainder: %w", err)
+		}
+		storageLeft, err := e.Storage.StorageRemaining(ctx, req.AccountID)
+		if err != nil {
+			return fmt.Errorf("storage remainder: %w", err)
+		}
+
+		attachments.Deleted = appendDeleted(attachments.Deleted, computeOut)
+		storage.Deleted = appendDeleted(storage.Deleted, storageOut)
+		for _, r := range legacy.Deleted {
+			if !containsResource(storage.Deleted, r) {
+				storage.Deleted = append(storage.Deleted, r)
+			}
+		}
+		storage.Stuck = legacy.Stuck
+		if e.OnPass != nil {
+			e.OnPass(ctx, pass)
+		}
+		e.logger().InfoContext(ctx, "Teardown convergence pass",
+			"account_id", req.AccountID, "pass", pass,
+			"compute_remaining", len(computeLeft.Items), "storage_remaining", len(storageLeft.Items),
+			"legacy_stuck", len(legacy.Stuck), "elapsed_ms", otelsetup.Millis(time.Since(started)))
+
+		if len(computeLeft.Items) == 0 && len(storageLeft.Items) == 0 &&
+			computeLeft.Status == accountcleanup.StatusClean && storageLeft.Status == accountcleanup.StatusClean &&
+			len(legacy.Stuck) == 0 {
+			return e.finishConverge(ctx, result, attachments, storage, started)
+		}
+
+		var moving bool
+		for _, out := range []*accountcleanup.Outcome{computeOut, storageOut} {
+			moving = moving || slices.ContainsFunc(out.Items, convergent)
+		}
+		if len(computeOut.IncompleteReasons)+len(storageOut.IncompleteReasons)+len(computeLeft.IncompleteReasons)+len(storageLeft.IncompleteReasons) > 0 {
+			incompletePasses++
+		} else {
+			incompletePasses = 0
+		}
+		signature := passSignature(legacy.Stuck, computeOut, storageOut, computeLeft, storageLeft)
+		if signature == lastSignature && !moving {
+			idlePasses++
+		} else {
+			idlePasses = 0
+		}
+		lastSignature = signature
+
+		reason := ""
+		switch {
+		case idlePasses >= 1:
+			reason = "only items time cannot resolve remain"
+		case incompletePasses >= cfg.IncompletePasses:
+			reason = "a read stayed incomplete"
+		case time.Now().Add(backoff).After(deadline):
+			reason = "the convergence budget ended"
+		}
+		if reason != "" {
+			attachments.Stuck = stuckFrom(attachments.Stuck[:0], computeOut, computeLeft)
+			storage.Stuck = append(legacy.Stuck, stuckFrom(nil, storageOut, storageLeft)...)
+			if err := e.finishConverge(ctx, result, attachments, storage, started); err != nil {
+				return err
+			}
+			e.logger().WarnContext(ctx, "Account teardown stopped converging",
+				"accountID", req.AccountID, "reason", reason, "passes", pass)
+			return fmt.Errorf("%d resources left (%s): %w", result.StuckCount(), reason, ErrResourcesStuck)
+		}
+
+		if err := wait(ctx, backoff); err != nil {
+			return err
+		}
+		backoff = min(backoff*2, cfg.MaxBackoff)
+	}
+}
+
+func (e *Engine) finishConverge(ctx context.Context, result *Result, attachments, storage StageResult, started time.Time) error {
+	elapsed := time.Since(started).String()
+	attachments.Elapsed, storage.Elapsed = elapsed, elapsed
+	for _, stage := range []StageResult{attachments, storage} {
+		result.Stages = append(result.Stages, stage)
+		if e.OnStage != nil {
+			e.OnStage(ctx, stage)
+		}
+	}
+	return nil
+}
+
+func appendDeleted(into []Resource, out *accountcleanup.Outcome) []Resource {
+	for _, it := range out.Items {
+		if it.Disposition == accountcleanup.DispositionDeleted {
+			r := Resource{Kind: it.Kind, ID: it.ResourceID}
+			if !containsResource(into, r) {
+				into = append(into, r)
+			}
+		}
+	}
+	return into
+}
+
+// stuckFrom names every item the account still holds with the reason the
+// cleanup gave for keeping it, or that the record remains. An incomplete read
+// is named too, because it proves nothing about what is left.
+func stuckFrom(into []Stuck, cleanup, remainder *accountcleanup.Outcome) []Stuck {
+	reasons := map[string]string{}
+	for _, it := range cleanup.Items {
+		if it.Reason != "" && it.Disposition != accountcleanup.DispositionDeleted {
+			reasons[it.Kind+"/"+it.ResourceID] = it.Reason
+		}
+	}
+	for _, it := range remainder.Items {
+		reason := reasons[it.Kind+"/"+it.ResourceID]
+		if reason == "" {
+			reason = it.Reason
+		}
+		into = append(into, Stuck{Resource: Resource{Kind: it.Kind, ID: it.ResourceID, Detail: it.State}, Reason: reason})
+	}
+	for _, reason := range slices.Concat(cleanup.IncompleteReasons, remainder.IncompleteReasons) {
+		into = append(into, Stuck{Resource: Resource{Kind: "inventory", ID: reason}, Reason: reason})
+	}
+	return into
+}
+
+// rereadComputeStorage reads the instances, spot requests and volumes the
+// account still holds immediately before the record is removed, so anything
+// that appeared after the convergence loop ended keeps the account. A read
+// that fails or is incomplete is named, never taken as empty.
+func (e *Engine) rereadComputeStorage(ctx context.Context, accountID string, result *Result) {
+	compute, cerr := e.Compute.ComputeRemaining(ctx, accountID)
+	storage, serr := e.Storage.StorageRemaining(ctx, accountID)
+	e.addFinalStuck(result, StageAttachments, accountID, compute, cerr)
+	e.addFinalStuck(result, StageStorage, accountID, storage, serr)
+}
+
+func (e *Engine) addFinalStuck(result *Result, stage Stage, accountID string, out *accountcleanup.Outcome, err error) {
+	var stuck []Stuck
+	if err != nil || out == nil {
+		stuck = []Stuck{{Resource: Resource{Kind: "inventory", ID: stage.String()}, Reason: operatorv1.AccountingReasonRereadIncomplete}}
+	} else {
+		stuck = stuckFrom(nil, &accountcleanup.Outcome{}, out)
+	}
+	if len(stuck) == 0 {
+		return
+	}
+	for i := range result.Stages {
+		if result.Stages[i].Stage == stage {
+			result.Stages[i].Stuck = append(result.Stages[i].Stuck, stuck...)
+			return
+		}
+	}
+	result.Stages = append(result.Stages, StageResult{Stage: stage, Elapsed: "0s", Stuck: stuck})
+}

spinifex/operator/accountlifecycle/converge_test.go

@@ -0,0 +1,256 @@
+package accountlifecycle
+
+//test:in-package — the convergence loop drives unexported engine fields and
+// the unexported backoff hook.
+
+import (
+	"context"
+	"errors"
+	"testing"
+	"time"
+
+	"github.com/mulgadc/spinifex/spinifex/domains/ec2/accountcleanup"
+)
+
+// scriptedStage answers each pass from a script, repeating the last answer.
+type scriptedStage struct {
+	cleanups   []*accountcleanup.Outcome
+	remainders []*accountcleanup.Outcome
+	cleanupN   int
+	remainN    int
+	forces     []bool
+}
+
+func pick(list []*accountcleanup.Outcome, n int) *accountcleanup.Outcome {
+	if n >= len(list) {
+		n = len(list) - 1
+	}
+	return list[n]
+}
+
+func (s *scriptedStage) run(forceIn bool) *accountcleanup.Outcome {
+	s.forces = append(s.forces, forceIn)
+	out := pick(s.cleanups, s.cleanupN)
+	s.cleanupN++
+	return out
+}
+
+func (s *scriptedStage) remaining() *accountcleanup.Outcome {
+	out := pick(s.remainders, s.remainN)
+	s.remainN++
+	return out
+}
+
+type scriptedCompute struct{ scriptedStage }
+
+func (s *scriptedCompute) CleanupCompute(_ context.Context, req *accountcleanup.ComputeRequest) (*accountcleanup.ComputeResult, error) {
+	return s.run(req.Force), nil
+}
+
+func (s *scriptedCompute) ComputeRemaining(context.Context, string) (*accountcleanup.Outcome, error) {
+	return s.remaining(), nil
+}
+
+type scriptedStorage struct{ scriptedStage }
+
+func (s *scriptedStorage) CleanupStorage(_ context.Context, req *accountcleanup.StorageRequest) (*accountcleanup.StorageResult, error) {
+	return s.run(req.Force), nil
+}
+
+func (s *scriptedStorage) StorageRemaining(context.Context, string) (*accountcleanup.Outcome, error) {
+	return s.remaining(), nil
+}
+
+func outcome(items ...accountcleanup.Item) *accountcleanup.Outcome {
+	out := &accountcleanup.Outcome{AccountID: testAccountID, Status: accountcleanup.StatusClean, Items: items}
+	return out
+}
+
+func incomplete(reason string) *accountcleanup.Outcome {
+	return &accountcleanup.Outcome{AccountID: testAccountID, Status: accountcleanup.StatusIncomplete, IncompleteReasons: []string{reason}}
+}
+
+func item(kind, id, state, disposition, reason string) accountcleanup.Item {
+	return accountcleanup.Item{Kind: kind, ResourceID: id, State: state, Disposition: disposition, Reason: reason}
+}
+
+// convergeEngine runs a teardown with no real delay between passes.
+func convergeEngine(accounts AccountStore, compute *scriptedCompute, storage *scriptedStorage) (*Engine, *int) {
+	engine := quietEngine(accounts)
+	engine.Compute, engine.Storage = compute, storage
+	waits := 0
+	engine.wait = func(ctx context.Context, _ time.Duration) error {
+		waits++
+		return ctx.Err()
+	}
+	return engine, &waits
+}
+
+var (
+	runningInstance     = item(accountcleanup.KindInstance, "i-1", "running", accountcleanup.DispositionPending, accountcleanup.ReasonTerminating)
+	termInstance        = item(accountcleanup.KindInstance, "i-1", "terminated", accountcleanup.DispositionDeleted, "")
+	blockedHolderVolume = item(accountcleanup.KindVolume, "vol-1", "in-use", accountcleanup.DispositionBlocked, accountcleanup.ReasonHolderPresent)
+)
+
+func TestConvergeRunningInstanceFinishesInOneJob(t *testing.T) {
+	compute := &scriptedCompute{scriptedStage{
+		cleanups:   []*accountcleanup.Outcome{outcome(runningInstance), outcome(termInstance)},
+		remainders: []*accountcleanup.Outcome{outcome(item(accountcleanup.KindInstance, "i-1", "shutting-down", accountcleanup.DispositionResidue, accountcleanup.ReasonRecordRemains)), outcome()},
+	}}
+	storage := &scriptedStorage{scriptedStage{
+		cleanups:   []*accountcleanup.Outcome{outcome(blockedHolderVolume), outcome(item(accountcleanup.KindVolume, "vol-1", "available", accountcleanup.DispositionDeleted, ""))},
+		remainders: []*accountcleanup.Outcome{outcome(item(accountcleanup.KindVolume, "vol-1", "in-use", accountcleanup.DispositionResidue, accountcleanup.ReasonRecordRemains)), outcome()},
+	}}
+	accounts := newFakeAccounts()
+	engine, waits := convergeEngine(accounts, compute, storage)
+
+	result, err := engine.Teardown(context.Background(), request())
+	if err != nil {
+		t.Fatalf("a running instance must converge in one job, got %v", err)
+	}
+	if !result.AccountDeleted || !accounts.deleted {
+		t.Fatalf("account must be deleted once the loop ends clean: %+v", result)
+	}
+	if *waits != 1 || compute.cleanupN != 2 || storage.cleanupN != 2 {
+		t.Fatalf("expected two passes and one backoff, got passes=%d/%d waits=%d", compute.cleanupN, storage.cleanupN, *waits)
+	}
+}
+
+func TestConvergeStopsOnBlockerTimeCannotResolve(t *testing.T) {
+	managed := item(accountcleanup.KindInstance, "i-managed", "running", accountcleanup.DispositionBlocked, accountcleanup.ReasonManagedServiceOwner)
+	remain := item(accountcleanup.KindInstance, "i-managed", "running", accountcleanup.DispositionResidue, accountcleanup.ReasonRecordRemains)
+	compute := &scriptedCompute{scriptedStage{cleanups: []*accountcleanup.Outcome{outcome(managed)}, remainders: []*accountcleanup.Outcome{outcome(remain)}}}
+	storage := &scriptedStorage{scriptedStage{cleanups: []*accountcleanup.Outcome{outcome()}, remainders: []*accountcleanup.Outcome{outcome()}}}
+	accounts := newFakeAccounts()
+	engine, _ := convergeEngine(accounts, compute, storage)
+	engine.Converge = Converge{Budget: time.Hour}
+
+	result, err := engine.Teardown(context.Background(), request())
+	if !errors.Is(err, ErrResourcesStuck) {
+		t.Fatalf("a non-converging blocker must end the job stuck, got %v", err)
+	}
+	if accounts.deleted {
+		t.Fatal("account must not be deleted while an instance remains")
+	}
+	if compute.cleanupN > 2 {
+		t.Fatalf("loop must end before the budget on a blocker time cannot resolve, ran %d passes", compute.cleanupN)
+	}
+	var named bool
+	for _, stage := range result.Stages {
+		for _, s := range stage.Stuck {
+			named = named || (s.Resource.ID == "i-managed" && s.Reason == accountcleanup.ReasonManagedServiceOwner)
+		}
+	}
+	if !named {
+		t.Fatalf("stuck item must be named with the cleanup's reason: %+v", result.Stages)
+	}
+}
+
+// supersededAccounts reports a different holder once the first cleanup ran.
+type supersededAccounts struct {
+	*fakeAccounts
+
+	after *scriptedCompute
+}
+
+func (s *supersededAccounts) GetAccount(id string) (*Account, error) {
+	account, err := s.fakeAccounts.GetAccount(id)
+	if err != nil || s.after.cleanupN == 0 {
+		return account, err
+	}
+	taken := *account
+	taken.TerminationOp = &TerminationOp{OperationID: "other-op", Fence: account.TerminationOp.Fence + 1}
+	return &taken, nil
+}
+
+func TestConvergeLostFenceStopsBeforeNextPass(t *testing.T) {
+	compute := &scriptedCompute{scriptedStage{
+		cleanups:   []*accountcleanup.Outcome{outcome(runningInstance)},
+		remainders: []*accountcleanup.Outcome{outcome(item(accountcleanup.KindInstance, "i-1", "shutting-down", accountcleanup.DispositionResidue, accountcleanup.ReasonRecordRemains))},
+	}}
+	storage := &scriptedStorage{scriptedStage{cleanups: []*accountcleanup.Outcome{outcome()}, remainders: []*accountcleanup.Outcome{outcome()}}}
+	accounts := &supersededAccounts{fakeAccounts: newFakeAccounts(), after: compute}
+	engine, _ := convergeEngine(accounts, compute, storage)
+
+	// The deadline lets a loop that ignores the lost fence end, so it fails
+	// the pass count below instead of hanging.
+	ctx, cancel := context.WithTimeout(context.Background(), 500*time.Millisecond)
+	defer cancel()
+	_, err := engine.Teardown(ctx, request())
+	if err == nil || errors.Is(err, ErrResourcesStuck) {
+		t.Fatalf("a lost fence must stop the loop with an error, got %v", err)
+	}
+	if compute.cleanupN != 1 {
+		t.Fatalf("no terminate may be issued after the fence is lost, ran %d passes", compute.cleanupN)
+	}
+	if accounts.deleted {
+		t.Fatal("a superseded holder must not delete the account")
+	}
+}
+
+func TestConvergeCancelledContextStopsLoop(t *testing.T) {
+	compute := &scriptedCompute{scriptedStage{
+		cleanups:   []*accountcleanup.Outcome{outcome(runningInstance)},
+		remainders: []*accountcleanup.Outcome{outcome(item(accountcleanup.KindInstance, "i-1", "shutting-down", accountcleanup.DispositionResidue, accountcleanup.ReasonRecordRemains))},
+	}}
+	storage := &scriptedStorage{scriptedStage{cleanups: []*accountcleanup.Outcome{outcome()}, remainders: []*accountcleanup.Outcome{outcome()}}}
+	engine, _ := convergeEngine(newFakeAccounts(), compute, storage)
+	ctx, cancel := context.WithCancel(context.Background())
+	engine.OnPass = func(context.Context, int) { cancel() }
+
+	if _, err := engine.Teardown(ctx, request()); !errors.Is(err, context.Canceled) {
+		t.Fatalf("a cancelled job must stop the loop, got %v", err)
+	}
+	if compute.cleanupN != 1 {
+		t.Fatalf("no pass may start after cancel, ran %d", compute.cleanupN)
+	}
+}
+
+func TestConvergeIncompleteReadsEndStuckNotClean(t *testing.T) {
+	compute := &scriptedCompute{scriptedStage{cleanups: []*accountcleanup.Outcome{incomplete(accountcleanup.ReasonInstanceInventoryIncomplete)}, remainders: []*accountcleanup.Outcome{incomplete(accountcleanup.ReasonInstanceInventoryIncomplete)}}}
+	storage := &scriptedStorage{scriptedStage{cleanups: []*accountcleanup.Outcome{outcome()}, remainders: []*accountcleanup.Outcome{outcome()}}}
+	accounts := newFakeAccounts()
+	engine, _ := convergeEngine(accounts, compute, storage)
+
+	_, err := engine.Teardown(context.Background(), request())
+	if !errors.Is(err, ErrResourcesStuck) || accounts.deleted {
+		t.Fatalf("an unreadable inventory must never read as an empty account: err=%v deleted=%v", err, accounts.deleted)
+	}
+}
+
+// The loop keeps the account while the storage owner reports a volume still
+// recorded attached. That the owner refuses such a volume is the owner's own
+// invariant test in accountcleanup, against its real cleanup.
+func TestConvergeKeepsTheAccountWhileAnAttachedVolumeRemains(t *testing.T) {
+	compute := &scriptedCompute{scriptedStage{cleanups: []*accountcleanup.Outcome{outcome()}, remainders: []*accountcleanup.Outcome{outcome()}}}
+	storage := &scriptedStorage{scriptedStage{
+		cleanups:   []*accountcleanup.Outcome{outcome(blockedHolderVolume)},
+		remainders: []*accountcleanup.Outcome{outcome(item(accountcleanup.KindVolume, "vol-1", "in-use", accountcleanup.DispositionResidue, accountcleanup.ReasonRecordRemains))},
+	}}
+	for _, force := range []bool{false, true} {
+		accounts := newFakeAccounts()
+		engine, _ := convergeEngine(accounts, compute, storage)
+		req := request()
+		req.Force = force
+		storage.cleanupN, storage.remainN = 0, 0
+
+		_, err := engine.Teardown(context.Background(), req)
+		if !errors.Is(err, ErrResourcesStuck) || accounts.deleted {
+			t.Fatalf("force=%v: a volume still recorded attached must keep the account, err=%v deleted=%v", force, err, accounts.deleted)
+		}
+	}
+}
+
+func TestConvergeMissingCleanerFailsClosed(t *testing.T) {
+	engine := quietEngine(newFakeAccounts())
+	engine.Storage = nil
+	if _, err := engine.Teardown(context.Background(), request()); !errors.Is(err, errComputeCleanupMissing) {
+		t.Fatalf("compute without storage must refuse to start, got %v", err)
+	}
+}
+
+// cleanStages are compute and storage cleanups of an account that holds nothing.
+func cleanStages() (*scriptedCompute, *scriptedStorage) {
+	return &scriptedCompute{scriptedStage{cleanups: []*accountcleanup.Outcome{outcome()}, remainders: []*accountcleanup.Outcome{outcome()}}},
+		&scriptedStorage{scriptedStage{cleanups: []*accountcleanup.Outcome{outcome()}, remainders: []*accountcleanup.Outcome{outcome()}}}
+}

spinifex/operator/accountlifecycle/cutover_test.go

@@ -0,0 +1,169 @@
+package accountlifecycle
+
+//test:in-package — these drive the engine's unexported stage wiring with
+// scripted compute and storage owners, so each can count what it was asked.
+
+import (
+	"context"
+	"errors"
+	"testing"
+
+	"github.com/mulgadc/spinifex/spinifex/domains/ec2/accountcleanup"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+// failingRemainder answers the compute remainder read until its nth call.
+type failingRemainder struct {
+	ComputeStage
+
+	failOn int
+	calls  int
+}
+
+func (f *failingRemainder) ComputeRemaining(ctx context.Context, accountID string) (*accountcleanup.Outcome, error) {
+	f.calls++
+	if f.calls == f.failOn {
+		return nil, errors.New("owner unreachable")
+	}
+	return f.ComputeStage.ComputeRemaining(ctx, accountID)
+}
+
+func stuckIDs(result *Result) []string {
+	var ids []string
+	for _, stage := range result.Stages {
+		for _, s := range stage.Stuck {
+			ids = append(ids, s.Resource.ID)
+		}
+	}
+	return ids
+}
+
+// An account deletion that was never accepted must reach no terminate and no
+// volume delete: the quiesce refuses first, whatever Force says.
+func TestS7_NoTerminateOrVolumeDeleteWithoutAnAcceptedDeletion(t *testing.T) {
+	for _, force := range []bool{false, true} {
+		accounts := newFakeAccounts()
+		accounts.statusErr = errors.New("account is owned by a newer fence")
+		// The record already reads as ours, so only the quiesce itself stands
+		// between this holder and the terminate.
+		accounts.account.Status = AccountStatusTerminating
+		accounts.account.TerminationOp = &TerminationOp{OperationID: "op-1", Fence: 1}
+		engine := quietEngine(accounts)
+		compute, storage := cleanStages()
+		engine.Compute, engine.Storage = compute, storage
+
+		req := request()
+		req.OperationID, req.Fence, req.Force = "op-1", 1, force
+		_, err := engine.Teardown(context.Background(), req)
+
+		require.Error(t, err)
+		assert.Zero(t, compute.cleanupN, "force=%v: no terminate", force)
+		assert.Zero(t, storage.cleanupN, "force=%v: no volume delete", force)
+		assert.False(t, accounts.deleted)
+	}
+}
+
+// With the compute and storage owners missing the teardown refuses before it
+// quiesces anything, so a mis-wired gateway cannot half-delete an account.
+func TestS7_MissingComputeOwnerRefusesBeforeAnyStageRuns(t *testing.T) {
+	accounts := newFakeAccounts()
+	engine := quietEngine(accounts)
+	engine.Compute = nil
+
+	_, err := engine.Teardown(context.Background(), request())
+
+	require.ErrorIs(t, err, errComputeCleanupMissing)
+	assert.Empty(t, accounts.quiesces)
+	assert.False(t, accounts.deleted)
+}
+
+// A resource that appears after the loop ended, while the later stages ran,
+// still keeps the account, and it is named.
+func TestFinalRereadKeepsTheAccountWhenAnInstanceArrivesLate(t *testing.T) {
+	accounts := newFakeAccounts()
+	engine := quietEngine(accounts)
+	engine.Compute = &scriptedCompute{scriptedStage{
+		cleanups:   []*accountcleanup.Outcome{outcome()},
+		remainders: []*accountcleanup.Outcome{outcome(), outcome(item(accountcleanup.KindInstance, "i-late", "running", accountcleanup.DispositionResidue, accountcleanup.ReasonRecordRemains))},
+	}}
+
+	result, err := engine.Teardown(context.Background(), request())
+
+	require.ErrorIs(t, err, ErrResourcesStuck)
+	assert.False(t, accounts.deleted)
+	assert.Contains(t, stuckIDs(result), "i-late")
+}
+
+// A final read that cannot be answered proves nothing, so the account stays.
+func TestFinalRereadKeepsTheAccountWhenTheReadFails(t *testing.T) {
+	accounts := newFakeAccounts()
+	engine := quietEngine(accounts)
+	compute, _ := cleanStages()
+	engine.Compute = &failingRemainder{ComputeStage: compute, failOn: 2}
+
+	result, err := engine.Teardown(context.Background(), request())
+
+	require.ErrorIs(t, err, ErrResourcesStuck)
+	assert.False(t, accounts.deleted)
+	assert.Contains(t, stuckIDs(result), StageAttachments.String())
+}
+
+// An ECS, EKS or RDS owner that cannot finish deleting may be running on these
+// instances, so none is terminated under it.
+func TestStuckComputeOwnerLeavesTheInstancesAlone(t *testing.T) {
+	accounts := newFakeAccounts()
+	cluster := &fakeReaper{
+		kind: "ecs-cluster", stage: StageCompute, blockedUntilForce: true,
+		resources: []Resource{{Kind: "ecs-cluster", ID: "web"}},
+	}
+	engine := quietEngine(accounts, cluster)
+	compute, _ := cleanStages()
+	engine.Compute = compute
+
+	result, err := engine.Teardown(context.Background(), request())
+
+	require.ErrorIs(t, err, ErrResourcesStuck)
+	assert.Zero(t, compute.cleanupN, "no instance is terminated while the owner holds resources")
+	assert.False(t, accounts.deleted)
+	assert.Contains(t, stuckIDs(result), "web")
+}
+
+// The dry run reports what the actual deletion would destroy.
+func TestDryRunListsInstancesSpotRequestsAndVolumes(t *testing.T) {
+	accounts := newFakeAccounts()
+	engine := quietEngine(accounts)
+	engine.Compute = &scriptedCompute{scriptedStage{
+		cleanups: []*accountcleanup.Outcome{outcome()},
+		remainders: []*accountcleanup.Outcome{outcome(
+			item(accountcleanup.KindInstance, "i-1", "running", accountcleanup.DispositionPending, ""),
+			item(accountcleanup.KindSpotRequest, "sir-1", "open", accountcleanup.DispositionPending, ""),
+		)},
+	}}
+	engine.Storage = &scriptedStorage{scriptedStage{
+		cleanups:   []*accountcleanup.Outcome{outcome()},
+		remainders: []*accountcleanup.Outcome{outcome(item(accountcleanup.KindVolume, "vol-1", "available", accountcleanup.DispositionPending, ""))},
+	}}
+
+	result, err := engine.DryRun(context.Background(), testAccountID)
+
+	require.NoError(t, err)
+	var ids []string
+	for _, stage := range result.Stages {
+		for _, r := range stage.Deleted {
+			ids = append(ids, r.ID)
+		}
+	}
+	assert.ElementsMatch(t, []string{"i-1", "sir-1", "vol-1"}, ids)
+}
+
+// A dry run that cannot read the owners fails rather than listing less.
+func TestDryRunFailsWhenTheComputeOwnerCannotBeRead(t *testing.T) {
+	engine := quietEngine(newFakeAccounts())
+	compute, _ := cleanStages()
+	engine.Compute = &failingRemainder{ComputeStage: compute, failOn: 1}
+
+	_, err := engine.DryRun(context.Background(), testAccountID)
+
+	require.Error(t, err)
+}

spinifex/operator/accountlifecycle/ec2_platform_inventory_test.go

@@ -16,6 +16,7 @@ import (
 	"github.com/aws/aws-sdk-go/service/s3"
 	ec2v1 "github.com/mulgadc/spinifex/contracts/ec2/v1"
 	"github.com/mulgadc/spinifex/internal/testkit"
+	"github.com/mulgadc/spinifex/spinifex/domains/ec2/accountcleanup"
 	"github.com/mulgadc/spinifex/spinifex/domains/ec2/accountinventory"
 	iamdomain "github.com/mulgadc/spinifex/spinifex/domains/iam"
 	awserrors "github.com/mulgadc/spinifex/spinifex/foundation/aws/errors"
@@ -209,6 +210,21 @@ func TestNoLegacyReaperOwnsAnL09Kind(t *testing.T) {
 	}
 }
 
+// Instances, spot requests and volumes are the EC2 compute and storage
+// owners', so no legacy reaper may list or delete them either.
+func TestNoLegacyReaperOwnsAnL03L05Kind(t *testing.T) {
+	var reapers []Reaper
+	reapers = append(reapers, ECSReapers(nil)...)
+	reapers = append(reapers, EKSReapers(nil)...)
+	reapers = append(reapers, RDSReapers(nil)...)
+	reapers = append(reapers, ELBv2Reapers(nil)...)
+	reapers = append(reapers, EC2ExtraReapers(nil)...)
+	reapers = append(reapers, EC2Reapers(nil, 1, nil)...)
+	for _, reaper := range reapers {
+		assert.NotContains(t, []string{accountcleanup.KindInstance, accountcleanup.KindSpotRequest, accountcleanup.KindVolume}, reaper.Kind())
+	}
+}
+
 // A caller deadline shorter than the adapter's own timeout ends the request
 // at that deadline, and a context already done sends nothing.
 func TestEC2PlatformInventoryClientHonoursTheCallersDeadline(t *testing.T) {

spinifex/operator/accountlifecycle/engine.go

@@ -7,9 +7,11 @@ import (
 	"log/slog"
 	"slices"
 	"sort"
+	"strings"
 	"time"
 
 	operatorv1 "github.com/mulgadc/spinifex/contracts/operator/v1"
+	"github.com/mulgadc/spinifex/spinifex/domains/ec2/accountcleanup"
 )
 
 // AccountStatusTerminating mirrors iamdomain.AccountStatusTerminating.
@@ -41,6 +43,21 @@ type Engine struct {
 	// OnExercised receives what a non-finalizing run did after quiescence.
 	OnExercised func(ctx context.Context, result operatorv1.ExerciseResult)
 
+	// Compute and Storage are EC2's instance, spot request and volume cleanup.
+	// When set, Teardown repeats them within Converge's bounds in place of the
+	// legacy attachments and storage stages.
+	Compute ComputeStage
+	Storage StorageStage
+
+	// Converge bounds the repeat loop. The zero value uses DefaultConverge.
+	Converge Converge
+
+	// OnPass reports each finished pass of the loop so the caller can persist progress.
+	OnPass func(ctx context.Context, pass int)
+
+	// wait replaces the backoff delay in tests.
+	wait func(ctx context.Context, d time.Duration) error
+
 	// ec2Inventory is EC2's read-only platform inventory: Inventory lists the
 	// L09 kinds from it, and the quiescence evaluation and final re-read use it.
 	ec2Inventory ec2PlatformInventory
@@ -104,11 +121,44 @@ func (e *Engine) Inventory(ctx context.Context, accountID string) (*Result, erro
 			}
 			stageResult.Deleted = append(stageResult.Deleted, listed...)
 		}
+		if stage == StageAttachments || stage == StageStorage {
+			listed, err := e.listComputeStorage(ctx, stage, accountID)
+			if err != nil {
+				return nil, err
+			}
+			stageResult.Deleted = append(stageResult.Deleted, listed...)
+		}
 		result.Stages = append(result.Stages, stageResult)
 	}
 	return result, nil
 }
 
+// listComputeStorage reads what the EC2 compute and storage owners still hold.
+// An unanswered or incomplete read is an error, never a shorter list.
+func (e *Engine) listComputeStorage(ctx context.Context, stage Stage, accountID string) ([]Resource, error) {
+	if e.Compute == nil || e.Storage == nil {
+		return nil, errComputeCleanupMissing
+	}
+	var out *accountcleanup.Outcome
+	var err error
+	if stage == StageAttachments {
+		out, err = e.Compute.ComputeRemaining(ctx, accountID)
+	} else {
+		out, err = e.Storage.StorageRemaining(ctx, accountID)
+	}
+	if err != nil {
+		return nil, fmt.Errorf("inventory %s: %w", stage, err)
+	}
+	if len(out.IncompleteReasons) > 0 {
+		return nil, fmt.Errorf("inventory %s incomplete: %s", stage, strings.Join(out.IncompleteReasons, ","))
+	}
+	var found []Resource
+	for _, it := range out.Items {
+		found = append(found, Resource{Kind: it.Kind, ID: it.ResourceID, Detail: it.State})
+	}
+	return found, nil
+}
+
 // DryRun is the explicit operator dry run: exactly what Inventory returns.
 func (e *Engine) DryRun(ctx context.Context, accountID string) (*Result, error) {
 	return e.Inventory(ctx, accountID)
@@ -130,6 +180,9 @@ func (e *Engine) Teardown(ctx context.Context, req Request) (*Result, error) {
 	if e.Cleanup == nil || e.ec2Inventory == nil {
 		return nil, errPlatformCleanupMissing
 	}
+	if e.Compute == nil || e.Storage == nil {
+		return nil, errComputeCleanupMissing
+	}
 
 	generation, err := e.Accounts.QuiesceAccount(req.AccountID, req.OperationID, req.Fence)
 	if err != nil {
@@ -148,6 +201,9 @@ func (e *Engine) Teardown(ctx context.Context, req Request) (*Result, error) {
 	}
 	var platform *platformRun
 	for _, stage := range Stages() {
+		if stage == StageAttachments || stage == StageStorage {
+			continue
+		}
 		if stage == StagePlatform {
 			var err error
 			if platform, err = e.cleanPlatform(ctx, req, generation); err != nil {
@@ -165,8 +221,21 @@ func (e *Engine) Teardown(ctx context.Context, req Request) (*Result, error) {
 		if e.OnStage != nil {
 			e.OnStage(ctx, stageResult)
 		}
+		if stage == StageCompute {
+			if len(stageResult.Stuck) > 0 {
+				// An ECS, EKS or RDS owner that still holds resources may be
+				// running on these instances, so none is terminated under it.
+				e.logger().WarnContext(ctx, "Account teardown: compute owners still hold resources, instances left alone",
+					"accountID", account.AccountID, "stuck", len(stageResult.Stuck))
+				continue
+			}
+			if err := e.convergeCompute(ctx, req, result); err != nil {
+				return result, err
+			}
+		}
 	}
 	e.rereadPlatform(ctx, req, platform, result)
+	e.rereadComputeStorage(ctx, req.AccountID, result)
 
 	if stuck := result.StuckCount(); stuck > 0 {
 		// The account stays TERMINATING on purpose. Removing the record now

spinifex/operator/accountlifecycle/engine_test.go

@@ -130,6 +130,7 @@ func quietEngine(accounts AccountStore, reapers ...Reaper) *Engine {
 	// stage need not stand up the EC2 owners.
 	engine.Cleanup = &stubCleaner{result: cleanResult()}
 	engine.ec2Inventory = &stubInventory{reply: *completeInventory()}
+	engine.Compute, engine.Storage = cleanStages()
 	return engine
 }
 

spinifex/operator/accountlifecycle/exercise.go

@@ -157,6 +157,10 @@ type InstanceReferenceReader struct {
 
 var _ accountcleanup.InstanceReferenceReader = (*InstanceReferenceReader)(nil)
 
+// terminalInstanceStates are the states an instance no longer holds anything
+// in. Anything else is still occupying a volume, an address and a host.
+var terminalInstanceStates = map[string]bool{"terminated": true, "shutting-down": true}
+
 // NewInstanceReferenceReader builds the reader over the configured node set.
 func NewInstanceReferenceReader(nc *nats.Conn, nodeIDs []string) *InstanceReferenceReader {
 	return &InstanceReferenceReader{nc: nc, nodeIDs: nodeIDs, budget: ec2PlatformInventoryBudget}
@@ -165,7 +169,7 @@ func NewInstanceReferenceReader(nc *nats.Conn, nodeIDs []string) *InstanceRefere
 func (r *InstanceReferenceReader) AccountInstanceReferences(ctx context.Context, accountID string) (*accountcleanup.References, error) {
 	ctx, cancel := context.WithTimeout(ctx, r.budget)
 	defer cancel()
-	out, err := ec2instanceapi.DescribeAccountInstancesComplete(ctx, r.nc, r.nodeIDs, accountID)
+	out, err := ec2instanceapi.DescribeAccountInstancesStrict(ctx, r.nc, r.nodeIDs, accountID)
 	if err != nil {
 		return nil, err
 	}

spinifex/operator/accountlifecycle/quiescence_test.go

@@ -241,9 +241,9 @@ func TestTeardownReportsQuiescenceAfterTheQuiesceAndBeforeTheFirstStage(t *testi
 	assert.True(t, accounts.deleted)
 }
 
-// A fence the read-back does not confirm must not spend an inventory call, and a
-// failed read-back must not stop the teardown: this step records and gates nothing.
-func TestTeardownRecordsAFailedReadbackAndStillRunsEveryStage(t *testing.T) {
+// A fence the read-back does not confirm must not spend an inventory call, and
+// the convergence loop must not run a pass behind a fence it cannot read.
+func TestTeardownRecordsAFailedReadbackAndRunsNoPassBehindIt(t *testing.T) {
 	accounts := newFakeAccounts()
 	engine := quietEngine(accounts)
 	inventory := &stubInventory{reply: *completeInventory()}
@@ -257,11 +257,11 @@ func TestTeardownRecordsAFailedReadbackAndStillRunsEveryStage(t *testing.T) {
 	req.OperationID, req.Fence = "op-1", 1
 	_, err := engine.Teardown(context.Background(), req)
 
-	require.NoError(t, err)
+	require.Error(t, err)
 	require.Len(t, got.Unresolved, 1)
 	assert.Equal(t, operatorv1.ReasonIAMReadbackFailed, got.Unresolved[0].Reason)
-	assert.Equal(t, 1, inventory.calls, "only the final re-read: no evaluation reads inventory behind an unconfirmed fence")
-	assert.True(t, accounts.deleted)
+	assert.Equal(t, 0, inventory.calls, "no evaluation reads inventory behind an unconfirmed fence")
+	assert.False(t, accounts.deleted)
 }
 
 // readbackFails serves the pre-quiesce read and fails the one after it.

spinifex/operator/accountlifecycle/realcleanup_test.go

@@ -0,0 +1,104 @@
+package accountlifecycle
+
+import (
+	"context"
+	"errors"
+	"sync/atomic"
+	"testing"
+	"time"
+
+	"github.com/aws/aws-sdk-go/service/ec2"
+	"github.com/mulgadc/spinifex/spinifex/domains/ec2/accountcleanup"
+	"github.com/mulgadc/spinifex/spinifex/domains/ec2/accountinventory"
+	ec2instance "github.com/mulgadc/spinifex/spinifex/domains/ec2/instance"
+	ec2spotinstance "github.com/mulgadc/spinifex/spinifex/domains/ec2/spotinstance"
+)
+
+// Stores that answer as an account does once its instance record has expired:
+// no instance, no outstanding teardown, no spot request, no tag residue.
+type emptyComputeStores struct{}
+
+func (emptyComputeStores) ScanAccount(context.Context, string) (*ec2spotinstance.AccountScan, error) {
+	return &ec2spotinstance.AccountScan{}, nil
+}
+
+func (emptyComputeStores) CancelSpotInstanceRequests(context.Context, *ec2.CancelSpotInstanceRequestsInput, string) (*ec2.CancelSpotInstanceRequestsOutput, error) {
+	return &ec2.CancelSpotInstanceRequestsOutput{}, nil
+}
+
+func (emptyComputeStores) AccountInstances(context.Context, string) (*ec2.DescribeInstancesOutput, error) {
+	return &ec2.DescribeInstancesOutput{}, nil
+}
+
+func (emptyComputeStores) AccountTeardownOutstanding(string) ([]ec2instance.OutstandingInstance, error) {
+	return nil, nil
+}
+
+func (emptyComputeStores) AccountManagedInstances(context.Context, string) (map[string]bool, error) {
+	return map[string]bool{}, nil
+}
+
+func (emptyComputeStores) ScanAccountResourceIDs(context.Context, string, ...string) ([]string, error) {
+	return nil, nil
+}
+
+func (emptyComputeStores) DeleteAllTags(context.Context, string, string) error { return nil }
+
+type heldVolumes struct{ items []accountinventory.VolumeItem }
+
+func (h heldVolumes) VolumeInventory(context.Context, string) ([]accountinventory.VolumeItem, error) {
+	return h.items, nil
+}
+
+type countingEC2 struct{ terminates, deletes atomic.Int32 }
+
+func (c *countingEC2) TerminateInstance(context.Context, string, string) error {
+	c.terminates.Add(1)
+	return nil
+}
+
+func (c *countingEC2) DeleteVolume(context.Context, string, string) error {
+	c.deletes.Add(1)
+	return nil
+}
+
+// ADR-0003:S7. The instance's terminated record has expired and the instance is
+// absent from the inventory, but its volume is still recorded in-use and
+// attached. With the real compute and storage cleanup bound to these stores,
+// nothing is deleted whatever Force says, the account is not removed, and the
+// volume is named.
+func TestS7_ExpiredTerminatedRecordAndAttachedVolumeDeleteNothingAndKeepTheAccount(t *testing.T) {
+	for _, force := range []bool{false, true} {
+		stores, ec2ops := emptyComputeStores{}, &countingEC2{}
+		accounts := newFakeAccounts()
+		engine := quietEngine(accounts)
+		engine.Compute = accountcleanup.NewCompute(stores, stores, ec2ops, stores, stores)
+		engine.Storage = accountcleanup.NewStorage(
+			heldVolumes{items: []accountinventory.VolumeItem{{VolumeID: "vol-held", State: "in-use", AttachedInstance: "i-gone", DeleteOnTermination: true}}},
+			ec2ops, stores, stores, stores, stores)
+		engine.wait = func(ctx context.Context, _ time.Duration) error { return ctx.Err() }
+
+		req := request()
+		req.Force = force
+		result, err := engine.Teardown(context.Background(), req)
+
+		if !errors.Is(err, ErrResourcesStuck) {
+			t.Fatalf("force=%v: a held volume must end the teardown stuck, got %v", force, err)
+		}
+		if ec2ops.deletes.Load() != 0 || ec2ops.terminates.Load() != 0 {
+			t.Fatalf("force=%v: nothing may be deleted or terminated, deletes=%d terminates=%d", force, ec2ops.deletes.Load(), ec2ops.terminates.Load())
+		}
+		if accounts.deleted {
+			t.Fatalf("force=%v: the account must not be removed", force)
+		}
+		named := false
+		for _, stage := range result.Stages {
+			for _, s := range stage.Stuck {
+				named = named || s.Resource.ID == "vol-held"
+			}
+		}
+		if !named {
+			t.Fatalf("force=%v: the stuck volume must be named in the result", force)
+		}
+	}
+}

spinifex/operator/accountlifecycle/reapers_ec2.go

@@ -2,11 +2,8 @@ package accountlifecycle
 
 import (
 	"context"
-	"fmt"
-	"github.com/mulgadc/spinifex/spinifex/foundation/messaging/nats"
 	"slices"
 	"strings"
-	"time"
 
 	"github.com/aws/aws-sdk-go/aws"
 	"github.com/aws/aws-sdk-go/service/ec2"
@@ -14,10 +11,8 @@ import (
 	ec2eipapi "github.com/mulgadc/spinifex/spinifex/domains/ec2/awsapi/eip"
 	ec2igwapi "github.com/mulgadc/spinifex/spinifex/domains/ec2/awsapi/igw"
 	ec2imageapi "github.com/mulgadc/spinifex/spinifex/domains/ec2/awsapi/image"
-	ec2instanceapi "github.com/mulgadc/spinifex/spinifex/domains/ec2/awsapi/instance"
 	ec2routetableapi "github.com/mulgadc/spinifex/spinifex/domains/ec2/awsapi/routetable"
 	ec2snapshotapi "github.com/mulgadc/spinifex/spinifex/domains/ec2/awsapi/snapshot"
-	ec2volumeapi "github.com/mulgadc/spinifex/spinifex/domains/ec2/awsapi/volume"
 	ec2vpcapi "github.com/mulgadc/spinifex/spinifex/domains/ec2/awsapi/vpc"
 	"github.com/mulgadc/spinifex/spinifex/foundation/aws/errors"
 	"github.com/nats-io/nats.go"
@@ -30,13 +25,10 @@ import (
 // accounting correct without teardown having to reimplement any of it.
 func EC2Reapers(nc *nats.Conn, expectedNodes int, nodeIDs []string) []Reaper {
 	return []Reaper{
-		&instanceReaper{nc: nc, nodeIDs: nodeIDs},
-
 		// Snapshots before volumes: a volume with a live snapshot refuses to
 		// delete, because snapshot-backed clones read from its chunk prefix.
 		&snapshotReaper{nc: nc},
 		&imageReaper{nc: nc},
-		&volumeReaper{nc: nc, nodeIDs: nodeIDs},
 
 		// Reverse of creation order. The VPC cannot go until everything
 		// inside it has, and the gateway enforces that.
@@ -55,107 +47,6 @@ func EC2Reapers(nc *nats.Conn, expectedNodes int, nodeIDs []string) []Reaper {
 	}
 }
 
-// terminalInstanceStates are the states an instance no longer holds anything
-// in. Anything else is still occupying a volume, an address and a host.
-var terminalInstanceStates = map[string]bool{"terminated": true, "shutting-down": true}
-
-type instanceReaper struct {
-	nc *nats.Conn
-	// nodeIDs is the configured node set; all must answer before an empty
-	// listing counts as drained.
-	nodeIDs []string
-
-	// describeOpts tunes the inventory fan-out; production leaves it empty.
-	describeOpts []ec2instanceapi.DescribeOption
-}
-
-func (r *instanceReaper) Kind() string { return "instance" }
-func (r *instanceReaper) Stage() Stage { return StageCompute }
-
-func (r *instanceReaper) List(ctx context.Context, accountID string) ([]Resource, error) {
-	// Complete or nothing: a partial answer would read as "no instances left"
-	// and let storage deletion start while a guest still holds a volume.
-	out, err := ec2instanceapi.DescribeAccountInstancesComplete(ctx, r.nc, r.nodeIDs, accountID, r.describeOpts...)
-	if err != nil {
-		return nil, err
-	}
-
-	var found []Resource
-	for _, reservation := range out.Reservations {
-		for _, instance := range reservation.Instances {
-			if instance == nil || instance.InstanceId == nil {
-				continue
-			}
-			state := ""
-			if instance.State != nil && instance.State.Name != nil {
-				state = *instance.State.Name
-			}
-			if terminalInstanceStates[state] {
-				continue
-			}
-			found = append(found, Resource{Kind: r.Kind(), ID: *instance.InstanceId, Detail: state})
-		}
-	}
-	return found, nil
-}
-
-func (r *instanceReaper) Delete(ctx context.Context, accountID string, resource Resource, _ bool) error {
-	_, err := ec2instanceapi.TerminateInstances(ctx, &ec2.TerminateInstancesInput{
-		InstanceIds: []*string{aws.String(resource.ID)},
-	}, r.nc, accountID)
-	return ignoreAlreadyGone(err)
-}
-
-type volumeReaper struct {
-	nc *nats.Conn
-	// nodeIDs is the configured node set; every one must answer, by identity,
-	// before a volume's holder can be proven gone or the delete guard passes.
-	nodeIDs []string
-}
-
-func (r *volumeReaper) Kind() string { return "volume" }
-func (r *volumeReaper) Stage() Stage { return StageStorage }
-
-func (r *volumeReaper) List(ctx context.Context, accountID string) ([]Resource, error) {
-	out, err := ec2volumeapi.DescribeVolumes(ctx, &ec2.DescribeVolumesInput{}, r.nc, accountID)
-	if err != nil {
-		return nil, err
-	}
-
-	var found []Resource
-	for _, volume := range out.Volumes {
-		if volume == nil || volume.VolumeId == nil {
-			continue
-		}
-		resource := Resource{Kind: r.Kind(), ID: *volume.VolumeId}
-		// The attachment is the reason a volume will not go, so carry it: it
-		// is the difference between "stuck" and "stuck on i-abc".
-		for _, attachment := range volume.Attachments {
-			if attachment != nil && attachment.InstanceId != nil {
-				resource.Detail = "attached to " + *attachment.InstanceId
-				break
-			}
-		}
-		found = append(found, resource)
-	}
-	return found, nil
-}
-
-func (r *volumeReaper) Delete(ctx context.Context, accountID string, resource Resource, force bool) error {
-	_, err := ec2volumeapi.DeleteVolume(ctx, &ec2.DeleteVolumeInput{
-		VolumeId: aws.String(resource.ID),
-	}, r.nc, r.nodeIDs, accountID)
-	if isAlreadyGone(err) {
-		return nil
-	}
-	if !force {
-		return err
-	}
-	// Force path: the ordinary delete refuses an attached volume, and a holder
-	// that is gone but still on the record would otherwise strand it forever.
-	return forceDeleteVolume(ctx, r.nc, r.nodeIDs, accountID, resource.ID, err)
-}
-
 type snapshotReaper struct{ nc *nats.Conn }
 
 func (r *snapshotReaper) Kind() string { return "snapshot" }
@@ -599,131 +490,3 @@ func ignoreAlreadyGone(err error) error {
 	}
 	return err
 }
-
-// forceDetachSubject clears a volume's attachment in the control plane only.
-// It is answered by any node, unlike the ordinary detach, which routes to the
-// host running the instance and so cannot help when that host is the problem.
-const forceDetachSubject = "ec2.ForceDetachVolume"
-
-// forceDetachTimeout bounds the metadata-only detach. It touches one document
-// and needs nothing from the guest, so a slow answer means the cluster is
-// unwell rather than that the work is large.
-const forceDetachTimeout = 30 * time.Second
-
-// forceDeleteVolume clears an attachment the ordinary delete refuses to
-// override, then deletes, but only once every instance the volume names is
-// proven absent. Anything short of proof leaves the volume and its record alone.
-func forceDeleteVolume(ctx context.Context, nc *nats.Conn, nodeIDs []string, accountID, volumeID string, refusal error) error {
-	holders, err := volumeHolders(ctx, nc, accountID, volumeID)
-	if err != nil {
-		if isAlreadyGone(err) {
-			return nil
-		}
-		return fmt.Errorf("volume %s retained: cannot read its attachment: %w", volumeID, err)
-	}
-	if len(holders) == 0 {
-		// No holder on record, so the refusal is not an attachment and
-		// clearing the record would change nothing but the evidence.
-		return refusal
-	}
-	for _, instanceID := range holders {
-		if err := proveInstanceAbsent(ctx, nc, nodeIDs, accountID, instanceID); err != nil {
-			return fmt.Errorf("volume %s retained: %w", volumeID, err)
-		}
-	}
-
-	_, err = natsmsg.NATSRequest[ec2.VolumeAttachment](ctx, nc, forceDetachSubject,
-		&ec2.DetachVolumeInput{VolumeId: aws.String(volumeID), Force: aws.Bool(true)},
-		forceDetachTimeout, accountID)
-	if err != nil && !isAlreadyGone(err) {
-		return fmt.Errorf("force detach %s: %w", volumeID, err)
-	}
-
-	// The ordinary delete re-checks that the record is unattached and that no
-	// live instance maps the volume, so a re-attach in between still refuses.
-	_, err = ec2volumeapi.DeleteVolume(ctx, &ec2.DeleteVolumeInput{
-		VolumeId: aws.String(volumeID),
-	}, nc, nodeIDs, accountID)
-	return ignoreAlreadyGone(err)
-}
-
-// volumeHolders returns the instances the volume's own record says hold it,
-// read fresh rather than from the listing that queued the delete.
-func volumeHolders(ctx context.Context, nc *nats.Conn, accountID, volumeID string) ([]string, error) {
-	out, err := ec2volumeapi.DescribeVolumes(ctx, &ec2.DescribeVolumesInput{
-		VolumeIds: []*string{aws.String(volumeID)},
-	}, nc, accountID)
-	if err != nil {
-		return nil, err
-	}
-	var holders []string
-	for _, volume := range out.Volumes {
-		if volume == nil || aws.StringValue(volume.VolumeId) != volumeID {
-			continue
-		}
-		for _, attachment := range volume.Attachments {
-			if attachment != nil && aws.StringValue(attachment.InstanceId) != "" {
-				holders = append(holders, aws.StringValue(attachment.InstanceId))
-			}
-		}
-	}
-	return holders, nil
-}
-
-// proveInstanceAbsent accepts only a definitive InvalidInstanceID.NotFound or
-// a terminated copy from a describe naming the instance, then confirms both on
-// a listing every configured node answered by identity. Anything else is not proof.
-func proveInstanceAbsent(ctx context.Context, nc *nats.Conn, nodeIDs []string, accountID, instanceID string) error {
-	if len(nodeIDs) == 0 {
-		return fmt.Errorf("cannot prove holding instance %s is gone: the configured node set is unknown", instanceID)
-	}
-	named, err := ec2instanceapi.DescribeInstancesChecked(ctx, &ec2.DescribeInstancesInput{
-		InstanceIds: []*string{aws.String(instanceID)},
-	}, nc, 0, nodeIDs, accountID)
-	if err != nil {
-		if code, ok := awserrors.ResolveErrorCode(err); !ok || code != awserrors.ErrorInvalidInstanceIDNotFound {
-			return fmt.Errorf("cannot prove holding instance %s is gone: %w", instanceID, err)
-		}
-	} else {
-		found, err := requireTerminated(named.Reservations, instanceID)
-		if err != nil {
-			return err
-		}
-		if !found {
-			return fmt.Errorf("cannot prove holding instance %s is gone: the describe naming it returned neither the instance nor NotFound", instanceID)
-		}
-	}
-
-	// The named describe stops collecting once the id turns up, and a lone
-	// node's 4xx can surface as NotFound, so neither answer is complete alone.
-	listing, err := ec2instanceapi.DescribeAccountInstancesComplete(ctx, nc, nodeIDs, accountID)
-	if err != nil {
-		return fmt.Errorf("cannot prove holding instance %s is gone: %w", instanceID, err)
-	}
-	_, err = requireTerminated(listing.Reservations, instanceID)
-	return err
-}
-
-// requireTerminated reports whether instanceID appears in reservations, and
-// fails if any copy of it is in a state other than terminated.
-func requireTerminated(reservations []*ec2.Reservation, instanceID string) (found bool, err error) {
-	for _, reservation := range reservations {
-		if reservation == nil {
-			continue
-		}
-		for _, instance := range reservation.Instances {
-			if instance == nil || aws.StringValue(instance.InstanceId) != instanceID {
-				continue
-			}
-			found = true
-			state := "in an unknown state"
-			if instance.State != nil && aws.StringValue(instance.State.Name) != "" {
-				state = aws.StringValue(instance.State.Name)
-			}
-			if state != ec2.InstanceStateNameTerminated {
-				return true, fmt.Errorf("holding instance %s is still %s", instanceID, state)
-			}
-		}
-	}
-	return found, nil
-}

spinifex/operator/accountlifecycle/reapers_ec2_extra.go

@@ -7,66 +7,21 @@ import (
 	"github.com/aws/aws-sdk-go/service/ec2"
 	ec2eigw "github.com/mulgadc/spinifex/spinifex/domains/ec2/eigw"
 	ec2natgw "github.com/mulgadc/spinifex/spinifex/domains/ec2/natgw"
-	ec2spotinstance "github.com/mulgadc/spinifex/spinifex/domains/ec2/spotinstance"
 	"github.com/nats-io/nats.go"
 )
 
 // EC2ExtraReapers returns the EC2 reapers that hang off the VPC rather than off
 // an instance.
 //
-// Spot requests come first and in the compute stage: a live request can launch
-// a replacement after the instance reaper has drained, so leaving one running
-// makes teardown race its own fulfilment rather than merely leak.
+// Spot requests are not here: the EC2 compute cleanup cancels them before it
+// terminates anything, so none can launch a replacement.
 func EC2ExtraReapers(nc *nats.Conn) []Reaper {
 	return []Reaper{
-		&spotRequestReaper{svc: ec2spotinstance.NewNATSSpotInstanceService(nc)},
 		&natGatewayReaper{svc: ec2natgw.NewNATSNatGatewayService(nc)},
 		&egressOnlyIGWReaper{svc: ec2eigw.NewNATSEgressOnlyIGWService(nc)},
 	}
 }
 
-// terminalSpotStates are the states a request can no longer launch from.
-// Anything else may still produce an instance.
-var terminalSpotStates = map[string]bool{"cancelled": true, "closed": true, "failed": true}
-
-type spotRequestReaper struct {
-	svc ec2spotinstance.SpotInstanceService
-}
-
-func (r *spotRequestReaper) Kind() string { return "spot-request" }
-func (r *spotRequestReaper) Stage() Stage { return StageCompute }
-
-func (r *spotRequestReaper) List(ctx context.Context, accountID string) ([]Resource, error) {
-	out, err := r.svc.DescribeSpotInstanceRequests(ctx, &ec2.DescribeSpotInstanceRequestsInput{}, accountID)
-	if err != nil {
-		return nil, err
-	}
-
-	var found []Resource
-	for _, request := range out.SpotInstanceRequests {
-		if request == nil || request.SpotInstanceRequestId == nil {
-			continue
-		}
-		state := aws.StringValue(request.State)
-		if terminalSpotStates[state] {
-			continue
-		}
-		found = append(found, Resource{
-			Kind:   r.Kind(),
-			ID:     aws.StringValue(request.SpotInstanceRequestId),
-			Detail: state,
-		})
-	}
-	return found, nil
-}
-
-func (r *spotRequestReaper) Delete(ctx context.Context, accountID string, resource Resource, _ bool) error {
-	_, err := r.svc.CancelSpotInstanceRequests(ctx, &ec2.CancelSpotInstanceRequestsInput{
-		SpotInstanceRequestIds: []*string{aws.String(resource.ID)},
-	}, accountID)
-	return ignoreAlreadyGone(err)
-}
-
 // deletedNatGatewayStates are the states a NAT gateway has already released its
 // address and its subnet placement in.
 var deletedNatGatewayStates = map[string]bool{"deleted": true}

spinifex/operator/accountlifecycle/reapers_ec2_extra_test.go

@@ -12,36 +12,11 @@ import (
 	"github.com/aws/aws-sdk-go/service/ec2"
 	ec2eigw "github.com/mulgadc/spinifex/spinifex/domains/ec2/eigw"
 	ec2natgw "github.com/mulgadc/spinifex/spinifex/domains/ec2/natgw"
-	ec2spotinstance "github.com/mulgadc/spinifex/spinifex/domains/ec2/spotinstance"
 	awserrors "github.com/mulgadc/spinifex/spinifex/foundation/aws/errors"
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
 )
 
-// The embedded interfaces are deliberately nil: a reaper that calls anything
-// beyond the two methods it is supposed to use panics rather than passing.
-type fakeSpot struct {
-	ec2spotinstance.SpotInstanceService
-
-	requests  []*ec2.SpotInstanceRequest
-	cancelled []string
-	err       error
-}
-
-func (f *fakeSpot) DescribeSpotInstanceRequests(_ context.Context, _ *ec2.DescribeSpotInstanceRequestsInput, _ string) (*ec2.DescribeSpotInstanceRequestsOutput, error) {
-	return &ec2.DescribeSpotInstanceRequestsOutput{SpotInstanceRequests: f.requests}, nil
-}
-
-func (f *fakeSpot) CancelSpotInstanceRequests(_ context.Context, in *ec2.CancelSpotInstanceRequestsInput, _ string) (*ec2.CancelSpotInstanceRequestsOutput, error) {
-	if f.err != nil {
-		return nil, f.err
-	}
-	for _, id := range in.SpotInstanceRequestIds {
-		f.cancelled = append(f.cancelled, aws.StringValue(id))
-	}
-	return &ec2.CancelSpotInstanceRequestsOutput{}, nil
-}
-
 type fakeNatGateways struct {
 	ec2natgw.NatGatewayService
 
@@ -82,42 +57,6 @@ func (f *fakeEgressGateways) DeleteEgressOnlyInternetGateway(_ context.Context,
 	return &ec2.DeleteEgressOnlyInternetGatewayOutput{}, nil
 }
 
-// A cancelled request cannot launch anything, so listing it would leave the
-// stage waiting on a row that never clears — spot requests are never removed
-// from the account, only moved to a terminal state.
-func TestSpotRequestReaperSkipsTerminalRequests(t *testing.T) {
-	svc := &fakeSpot{requests: []*ec2.SpotInstanceRequest{
-		{SpotInstanceRequestId: aws.String("sir-open"), State: aws.String("open")},
-		{SpotInstanceRequestId: aws.String("sir-active"), State: aws.String("active")},
-		{SpotInstanceRequestId: aws.String("sir-cancelled"), State: aws.String("cancelled")},
-		{SpotInstanceRequestId: aws.String("sir-closed"), State: aws.String("closed")},
-		{SpotInstanceRequestId: aws.String("sir-failed"), State: aws.String("failed")},
-	}}
-	reaper := &spotRequestReaper{svc: svc}
-
-	assert.Equal(t, StageCompute, reaper.Stage())
-
-	found, err := reaper.List(t.Context(), "000000000002")
-	require.NoError(t, err)
-	require.Len(t, found, 2)
-	assert.Equal(t, "sir-open", found[0].ID)
-	assert.Equal(t, "open", found[0].Detail)
-	assert.Equal(t, "sir-active", found[1].ID)
-
-	require.NoError(t, reaper.Delete(t.Context(), "000000000002", found[0], false))
-	assert.Equal(t, []string{"sir-open"}, svc.cancelled)
-}
-
-// A request cancelled by someone else between the listing and the delete is
-// the ordinary second-pass case, not a failure.
-func TestSpotRequestReaperTreatsAMissingRequestAsCancelled(t *testing.T) {
-	svc := &fakeSpot{err: errors.New("InvalidSpotInstanceRequestID.NotFound: no such request")}
-	reaper := &spotRequestReaper{svc: svc}
-
-	err := reaper.Delete(t.Context(), "000000000002", Resource{ID: "sir-gone"}, false)
-	assert.NoError(t, err)
-}
-
 // A deleting gateway still holds its address and its subnet, so it has to keep
 // listing until it is actually gone or the address reaper runs too early.
 func TestNatGatewayReaperKeepsDeletingGateways(t *testing.T) {

spinifex/operator/accountlifecycle/reapers_ec2_inventory_test.go

@@ -1,198 +0,0 @@
-package accountlifecycle
-
-//test:in-package — the instance reaper is unexported, and whether its listing
-// can be mistaken for an empty account is the safety property under test.
-
-import (
-	"context"
-	"encoding/json"
-	"testing"
-	"time"
-
-	"github.com/aws/aws-sdk-go/aws"
-	"github.com/aws/aws-sdk-go/service/ec2"
-	"github.com/mulgadc/spinifex/internal/testkit"
-	ec2instanceapi "github.com/mulgadc/spinifex/spinifex/domains/ec2/awsapi/instance"
-	"github.com/mulgadc/spinifex/spinifex/foundation/aws/errors"
-	"github.com/mulgadc/spinifex/spinifex/foundation/messaging/nats"
-	"github.com/nats-io/nats.go"
-	"github.com/stretchr/testify/assert"
-	"github.com/stretchr/testify/require"
-)
-
-var inventoryNodes = []string{"node-1", "node-2"}
-
-// testInventoryReaper builds the instance reaper for a two-node cluster, with
-// a short fan-out deadline so a silent node costs the test little.
-func testInventoryReaper(nc *nats.Conn) *instanceReaper {
-	return &instanceReaper{nc: nc, nodeIDs: inventoryNodes, describeOpts: []ec2instanceapi.DescribeOption{
-		ec2instanceapi.WithFanoutTimeout(300 * time.Millisecond),
-	}}
-}
-
-// answerAsNode replies on subject the way a daemon does, carrying the node ID
-// header the fan-out attributes frames by.
-func answerAsNode(t *testing.T, nc *nats.Conn, subject, nodeID string, payload []byte) {
-	t.Helper()
-	sub, err := nc.Subscribe(subject, func(msg *nats.Msg) {
-		reply := nats.NewMsg(msg.Reply)
-		reply.Data = payload
-		if nodeID != "" {
-			reply.Header.Set(natsmsg.NodeIDHeader, nodeID)
-		}
-		_ = msg.RespondMsg(reply)
-	})
-	require.NoError(t, err)
-	t.Cleanup(func() { _ = sub.Unsubscribe() })
-}
-
-func describeOutput(t *testing.T, instanceIDs ...string) []byte {
-	t.Helper()
-	out := ec2.DescribeInstancesOutput{Reservations: []*ec2.Reservation{}}
-	if len(instanceIDs) > 0 {
-		reservation := &ec2.Reservation{ReservationId: aws.String("r-" + instanceIDs[0])}
-		for _, id := range instanceIDs {
-			reservation.Instances = append(reservation.Instances, &ec2.Instance{
-				InstanceId: aws.String(id),
-				State:      &ec2.InstanceState{Name: aws.String("running")},
-			})
-		}
-		out.Reservations = append(out.Reservations, reservation)
-	}
-	payload, err := json.Marshal(out)
-	require.NoError(t, err)
-	return payload
-}
-
-// emptyInstanceBuckets answers the stopped and terminated bucket queries with
-// nothing, so only the running-instance fan-out decides completeness.
-func emptyInstanceBuckets(t *testing.T, nc *nats.Conn) {
-	t.Helper()
-	answerAsNode(t, nc, "ec2.DescribeStoppedInstances", "", describeOutput(t))
-	answerAsNode(t, nc, "ec2.DescribeTerminatedInstances", "", describeOutput(t))
-}
-
-// A node that never answers is the timeout case: its running guests are absent
-// from the aggregate, so an empty answer from the rest is not evidence the
-// account has no instances.
-func TestInstanceReaperFailsClosedWhenANodeDoesNotAnswer(t *testing.T) {
-	t.Parallel()
-	_, nc := testutil.StartTestNATS(t)
-	emptyInstanceBuckets(t, nc)
-	answerAsNode(t, nc, "ec2.DescribeInstances", "node-1", describeOutput(t))
-	require.NoError(t, nc.Flush())
-
-	found, err := testInventoryReaper(nc).List(testCtx(t), testAccountID)
-
-	require.Error(t, err, "a listing missing a node must not read as an empty account, got %v", found)
-	code, _, ok := awserrors.ResolveErrorDetail(err)
-	require.True(t, ok)
-	assert.Equal(t, awserrors.ErrorServiceUnavailable, code)
-}
-
-// The partial list is the dangerous shape: the reaper would terminate what it
-// can see and then report the stage drained on the next, equally partial pass.
-func TestInstanceReaperFailsClosedOnAPartialList(t *testing.T) {
-	t.Parallel()
-	_, nc := testutil.StartTestNATS(t)
-	emptyInstanceBuckets(t, nc)
-	answerAsNode(t, nc, "ec2.DescribeInstances", "node-1", describeOutput(t, "i-on-node-1"))
-	require.NoError(t, nc.Flush())
-
-	found, err := testInventoryReaper(nc).List(testCtx(t), testAccountID)
-
-	require.Error(t, err, "a partial listing must not be returned as the inventory, got %v", found)
-}
-
-// An error frame from a node is no better than silence: it says nothing about
-// what that node is running.
-func TestInstanceReaperFailsClosedWhenANodeAnswersWithAnError(t *testing.T) {
-	_, nc := testutil.StartTestNATS(t)
-	emptyInstanceBuckets(t, nc)
-	answerAsNode(t, nc, "ec2.DescribeInstances", "node-1", describeOutput(t))
-	answerAsNode(t, nc, "ec2.DescribeInstances", "node-2", awserrors.GenerateErrorPayload(awserrors.ErrorServerInternal))
-	require.NoError(t, nc.Flush())
-
-	_, err := testInventoryReaper(nc).List(testCtx(t), testAccountID)
-
-	require.Error(t, err)
-}
-
-// Completeness is judged against the configured node set, not a count of
-// replies: two frames from one node while another is silent is still missing
-// a node.
-func TestInstanceReaperFailsClosedWhenOneNodeAnswersTwice(t *testing.T) {
-	t.Parallel()
-	_, nc := testutil.StartTestNATS(t)
-	emptyInstanceBuckets(t, nc)
-	answerAsNode(t, nc, "ec2.DescribeInstances", "node-1", describeOutput(t))
-	answerAsNode(t, nc, "ec2.DescribeInstances", "node-1", describeOutput(t))
-	require.NoError(t, nc.Flush())
-
-	_, err := testInventoryReaper(nc).List(testCtx(t), testAccountID)
-
-	require.Error(t, err)
-}
-
-// Stopped instances live in a bucket, not on a node. A failed bucket query
-// hides them as surely as a silent node hides running ones.
-func TestInstanceReaperFailsClosedWhenTheStoppedBucketFails(t *testing.T) {
-	_, nc := testutil.StartTestNATS(t)
-	answerAsNode(t, nc, "ec2.DescribeStoppedInstances", "", awserrors.GenerateErrorPayload(awserrors.ErrorServerInternal))
-	answerAsNode(t, nc, "ec2.DescribeTerminatedInstances", "", describeOutput(t))
-	answerAsNode(t, nc, "ec2.DescribeInstances", "node-1", describeOutput(t))
-	answerAsNode(t, nc, "ec2.DescribeInstances", "node-2", describeOutput(t))
-	require.NoError(t, nc.Flush())
-
-	_, err := testInventoryReaper(nc).List(testCtx(t), testAccountID)
-
-	require.Error(t, err)
-}
-
-// The healthy cluster still lists: every configured node answered, so the
-// union is the account's inventory.
-func TestInstanceReaperListsEveryNodeWhenAllAnswer(t *testing.T) {
-	_, nc := testutil.StartTestNATS(t)
-	emptyInstanceBuckets(t, nc)
-	answerAsNode(t, nc, "ec2.DescribeInstances", "node-1", describeOutput(t, "i-on-node-1"))
-	answerAsNode(t, nc, "ec2.DescribeInstances", "node-2", describeOutput(t, "i-on-node-2"))
-	require.NoError(t, nc.Flush())
-
-	started := time.Now()
-	found, err := testInventoryReaper(nc).List(testCtx(t), testAccountID)
-	require.NoError(t, err)
-
-	var ids []string
-	for _, resource := range found {
-		ids = append(ids, resource.ID)
-	}
-	assert.ElementsMatch(t, []string{"i-on-node-1", "i-on-node-2"}, ids)
-	assert.Less(t, time.Since(started), 2*time.Second, "a fully answered fan-out must not wait out its deadline")
-}
-
-// End to end through the engine: with a node silent, the compute stage must
-// not read as drained and the account record must survive.
-func TestTeardownDoesNotReportComputeDrainedWhileANodeIsSilent(t *testing.T) {
-	t.Parallel()
-	_, nc := testutil.StartTestNATS(t)
-	emptyInstanceBuckets(t, nc)
-	answerAsNode(t, nc, "ec2.DescribeInstances", "node-1", describeOutput(t))
-	require.NoError(t, nc.Flush())
-
-	accounts := newFakeAccounts()
-	volumes := &fakeReaper{kind: "volume", stage: StageStorage,
-		resources: []Resource{{Kind: "volume", ID: "vol-held-by-node-2"}}}
-	engine := quietEngine(accounts, testInventoryReaper(nc), volumes)
-
-	ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)
-	defer cancel()
-	result, err := engine.Teardown(ctx, request())
-
-	require.Error(t, err, "teardown must not succeed on an incomplete instance inventory")
-	assert.ErrorContains(t, err, "list instance")
-	assert.False(t, accounts.deleted, "the account record must survive an incomplete inventory")
-	for _, stage := range result.Stages {
-		assert.NotEqual(t, StageCompute, stage.Stage, "the compute stage must not be reported finished")
-	}
-	assert.Equal(t, 0, volumes.deletesAttempted, "storage must not be touched before compute is proven drained")
-}

spinifex/operator/accountlifecycle/reapers_ec2_test.go

@@ -106,29 +106,6 @@ func (c *fakeCluster) fail(subject, code string) {
 	c.errorCode[subject] = code
 }
 
-// handle answers subject from state the test owns, for doubles whose answer
-// must change as the reaper acts — an attachment a force detach clears.
-func (c *fakeCluster) handle(subject string, handler func(data []byte) (any, string)) {
-	c.mu.Lock()
-	defer c.mu.Unlock()
-	c.handlers[subject] = handler
-}
-
-// echoAsNode adds a second reply to every request on subject, attributed to
-// node, so one node can answer twice while another stays silent.
-func (c *fakeCluster) echoAsNode(t *testing.T, subject, node string) {
-	t.Helper()
-	sub, err := c.nc.Subscribe(subject, func(msg *nats.Msg) {
-		out := nats.NewMsg(msg.Reply)
-		out.Data = []byte(`{}`)
-		out.Header.Set(natsmsg.NodeIDHeader, node)
-		_ = msg.RespondMsg(out)
-	})
-	require.NoError(t, err)
-	t.Cleanup(func() { _ = sub.Unsubscribe() })
-	require.NoError(t, c.nc.Flush())
-}
-
 func (c *fakeCluster) called() []string {
 	c.mu.Lock()
 	defer c.mu.Unlock()
@@ -150,13 +127,6 @@ func testCtx(t *testing.T) context.Context {
 	return ctx
 }
 
-// silentClusterCtx outlasts the full fan-out deadline a silent node costs.
-func silentClusterCtx(t *testing.T) context.Context {
-	ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)
-	t.Cleanup(cancel)
-	return ctx
-}
-
 // The stage graph is what keeps teardown converging, so it is asserted rather
 // than left to the order the constructor happens to list.
 func TestEC2ReapersAreOrderedByDependency(t *testing.T) {
@@ -168,16 +138,10 @@ func TestEC2ReapersAreOrderedByDependency(t *testing.T) {
 	}
 
 	assert.Equal(t, []string{
-		"instance",
-		"snapshot", "image", "volume",
+		"snapshot", "image",
 		"address", "network-interface", "internet-gateway", "route-table", "subnet", "security-group", "vpc",
 	}, kinds)
 
-	// A volume with a live snapshot refuses to delete, so the reverse order
-	// would strand every snapshot-backed volume in the account.
-	assert.Less(t, indexOfKind(reapers, "snapshot"), indexOfKind(reapers, "volume"))
-	assert.Equal(t, StageCompute, reapers[indexOfKind(reapers, "instance")].Stage())
-	assert.Equal(t, StageStorage, reapers[indexOfKind(reapers, "volume")].Stage())
 	assert.Equal(t, StageNetwork, reapers[indexOfKind(reapers, "vpc")].Stage())
 }
 
@@ -190,333 +154,6 @@ func indexOfKind(reapers []Reaper, kind string) int {
 	return -1
 }
 
-// A terminated instance holds nothing. Listing it would keep the compute stage
-// polling until its timeout and then report it stuck, for no reason.
-func TestInstanceReaperSkipsTerminalStates(t *testing.T) {
-	cluster := newFakeCluster(t)
-	cluster.reply("ec2.DescribeInstances", ec2.DescribeInstancesOutput{
-		Reservations: []*ec2.Reservation{{Instances: []*ec2.Instance{
-			{InstanceId: aws.String("i-running"), State: &ec2.InstanceState{Name: aws.String("running")}},
-			{InstanceId: aws.String("i-gone"), State: &ec2.InstanceState{Name: aws.String("terminated")}},
-			{InstanceId: aws.String("i-going"), State: &ec2.InstanceState{Name: aws.String("shutting-down")}},
-			{State: &ec2.InstanceState{Name: aws.String("running")}},
-		}}},
-	})
-
-	reaper := &instanceReaper{nc: cluster.nc, nodeIDs: []string{fakeClusterNode}}
-	found, err := reaper.List(testCtx(t), "000000000042")
-	require.NoError(t, err)
-
-	require.Len(t, found, 1)
-	assert.Equal(t, "i-running", found[0].ID)
-	assert.Equal(t, "running", found[0].Detail)
-}
-
-// Termination is addressed to the node hosting the instance, which is what
-// makes a stuck host the case --force exists for.
-func TestInstanceReaperTerminates(t *testing.T) {
-	cluster := newFakeCluster(t)
-	reaper := &instanceReaper{nc: cluster.nc, nodeIDs: []string{fakeClusterNode}}
-
-	require.NoError(t, reaper.Delete(testCtx(t), "000000000042",
-		Resource{Kind: "instance", ID: "i-0abc0123456789def"}, false))
-
-	assert.Contains(t, cluster.called(), "ec2.cmd.i-0abc0123456789def")
-}
-
-// The attachment is the reason a volume will not go, so it has to reach the
-// stuck report — "stuck" and "stuck on i-abc" are different bug reports.
-func TestVolumeReaperCarriesItsAttachment(t *testing.T) {
-	cluster := newFakeCluster(t)
-	cluster.reply("ec2.DescribeVolumes", ec2.DescribeVolumesOutput{
-		Volumes: []*ec2.Volume{
-			{VolumeId: aws.String("vol-free")},
-			{VolumeId: aws.String("vol-held"), Attachments: []*ec2.VolumeAttachment{
-				{InstanceId: aws.String("i-abc")},
-			}},
-		},
-	})
-
-	reaper := newVolumeReaper(cluster.nc, fakeClusterNode)
-	found, err := reaper.List(testCtx(t), "000000000042")
-	require.NoError(t, err)
-
-	require.Len(t, found, 2)
-	assert.Empty(t, found[0].Detail)
-	assert.Equal(t, "attached to i-abc", found[1].Detail)
-}
-
-// Without --force an attached volume must surface the refusal, not swallow it:
-// a stuck resource is a bug we want reported rather than papered over.
-func TestVolumeReaperReportsRefusalWithoutForce(t *testing.T) {
-	cluster := newFakeCluster(t)
-	cluster.fail("ec2.DeleteVolume", "VolumeInUse")
-
-	reaper := newVolumeReaper(cluster.nc, fakeClusterNode)
-	err := reaper.Delete(testCtx(t), "000000000042", Resource{Kind: "volume", ID: "vol-held"}, false)
-
-	require.Error(t, err)
-	assert.NotContains(t, cluster.called(), forceDetachSubject)
-}
-
-// heldVolume is one volume record as its owner keeps it: the ordinary delete
-// refuses while the record names an instance, and a force detach clears it.
-type heldVolume struct {
-	mu       sync.Mutex
-	id       string
-	holder   string
-	deleted  bool
-	detaches int
-}
-
-func newHeldVolume(cluster *fakeCluster, id, holder string) *heldVolume {
-	volume := &heldVolume{id: id, holder: holder}
-	cluster.handle("ec2.DescribeVolumes", func(data []byte) (any, string) {
-		var input ec2.DescribeVolumesInput
-		_ = json.Unmarshal(data, &input)
-		volume.mu.Lock()
-		defer volume.mu.Unlock()
-		if volume.deleted {
-			if len(input.VolumeIds) > 0 {
-				return nil, awserrors.ErrorInvalidVolumeNotFound
-			}
-			return ec2.DescribeVolumesOutput{}, ""
-		}
-		described := &ec2.Volume{VolumeId: aws.String(volume.id), State: aws.String("available")}
-		if volume.holder != "" {
-			described.State = aws.String("in-use")
-			described.Attachments = []*ec2.VolumeAttachment{
-				{VolumeId: aws.String(volume.id), InstanceId: aws.String(volume.holder), State: aws.String("attached")},
-			}
-		}
-		return ec2.DescribeVolumesOutput{Volumes: []*ec2.Volume{described}}, ""
-	})
-	cluster.handle(forceDetachSubject, func([]byte) (any, string) {
-		volume.mu.Lock()
-		defer volume.mu.Unlock()
-		volume.detaches++
-		previous := volume.holder
-		volume.holder = ""
-		return ec2.VolumeAttachment{VolumeId: aws.String(volume.id), InstanceId: aws.String(previous)}, ""
-	})
-	cluster.handle("ec2.DeleteVolume", func([]byte) (any, string) {
-		volume.mu.Lock()
-		defer volume.mu.Unlock()
-		switch {
-		case volume.deleted:
-			return nil, awserrors.ErrorInvalidVolumeNotFound
-		case volume.holder != "":
-			return nil, awserrors.ErrorVolumeInUse
-		}
-		volume.deleted = true
-		return ec2.DeleteVolumeOutput{}, ""
-	})
-	return volume
-}
-
-func (v *heldVolume) state() (deleted bool, holder string, detaches int) {
-	v.mu.Lock()
-	defer v.mu.Unlock()
-	return v.deleted, v.holder, v.detaches
-}
-
-func instancesIn(state string, instanceIDs ...string) ec2.DescribeInstancesOutput {
-	reservation := &ec2.Reservation{ReservationId: aws.String("r-" + state)}
-	for _, id := range instanceIDs {
-		reservation.Instances = append(reservation.Instances, &ec2.Instance{
-			InstanceId: aws.String(id), State: &ec2.InstanceState{Name: aws.String(state)},
-		})
-	}
-	return ec2.DescribeInstancesOutput{Reservations: []*ec2.Reservation{reservation}}
-}
-
-// Force must not clear an attachment a live guest still holds. The instance
-// carries no block device mapping for the volume, so the attachment record is
-// the only thing standing between the guest and a deleted disk.
-func TestVolumeReaperForceRetainsAVolumeALiveInstanceHolds(t *testing.T) {
-	t.Parallel()
-	for _, state := range []string{"running", "stopped", "shutting-down", "pending"} {
-		t.Run(state, func(t *testing.T) {
-			t.Parallel()
-			cluster := newFakeCluster(t)
-			volume := newHeldVolume(cluster, "vol-held", "i-live")
-			subject := "ec2.DescribeInstances"
-			if state == "stopped" {
-				subject = "ec2.DescribeStoppedInstances"
-			}
-			cluster.reply(subject, instancesIn(state, "i-live"))
-
-			reaper := newVolumeReaper(cluster.nc, fakeClusterNode)
-			err := reaper.Delete(testCtx(t), "000000000042", Resource{Kind: "volume", ID: "vol-held"}, true)
-
-			require.Error(t, err)
-			assert.Contains(t, err.Error(), "vol-held")
-			assert.Contains(t, err.Error(), "i-live")
-			assert.Contains(t, err.Error(), state)
-			deleted, holder, detaches := volume.state()
-			assert.False(t, deleted, "a volume a live guest holds must survive teardown")
-			assert.Equal(t, "i-live", holder, "the attachment must not be cleared under a live guest")
-			assert.Zero(t, detaches)
-		})
-	}
-}
-
-// "Cannot prove gone" is not "gone": every way the proof can fall short keeps
-// the volume and its attachment exactly as they were.
-func TestVolumeReaperForceRetainsAVolumeWhenTheHolderCannotBeProvenGone(t *testing.T) {
-	t.Parallel()
-	cases := []struct {
-		name  string
-		setup func(cluster *fakeCluster)
-		nodes []string
-	}{
-		{"configured node set unknown", func(*fakeCluster) {}, []string{}},
-		{"node describe fails", func(cluster *fakeCluster) {
-			cluster.fail("ec2.DescribeInstances", awserrors.ErrorServerInternal)
-		}, nil},
-		{"terminated bucket unreachable", func(cluster *fakeCluster) {
-			cluster.fail("ec2.DescribeTerminatedInstances", awserrors.ErrorServerInternal)
-		}, nil},
-		{"stopped bucket unreachable", func(cluster *fakeCluster) {
-			cluster.fail("ec2.DescribeStoppedInstances", awserrors.ErrorServerInternal)
-		}, nil},
-		{"sources disagree", func(cluster *fakeCluster) {
-			cluster.reply("ec2.DescribeTerminatedInstances", instancesIn("terminated", "i-held"))
-			cluster.reply("ec2.DescribeInstances", instancesIn("running", "i-held"))
-		}, nil},
-		{"volume describe fails", func(cluster *fakeCluster) {
-			cluster.fail("ec2.DescribeVolumes", awserrors.ErrorServerInternal)
-		}, nil},
-	}
-	for _, tc := range cases {
-		t.Run(tc.name, func(t *testing.T) {
-			t.Parallel()
-			nodes := tc.nodes
-			if nodes == nil {
-				nodes = []string{fakeClusterNode}
-			}
-			assertForceRetainsTheHeldVolume(t, tc.setup, nodes)
-		})
-	}
-}
-
-// assertForceRetainsTheHeldVolume force-deletes a volume i-held holds and
-// asserts the volume and its attachment survive untouched.
-func assertForceRetainsTheHeldVolume(t *testing.T, setup func(cluster *fakeCluster), nodes []string) {
-	t.Helper()
-	cluster := newFakeCluster(t)
-	volume := newHeldVolume(cluster, "vol-held", "i-held")
-	setup(cluster)
-
-	reaper := newVolumeReaper(cluster.nc, nodes...)
-	err := reaper.Delete(silentClusterCtx(t), "000000000042", Resource{Kind: "volume", ID: "vol-held"}, true)
-
-	require.Error(t, err)
-	assert.Contains(t, err.Error(), "vol-held")
-	deleted, holder, detaches := volume.state()
-	assert.False(t, deleted)
-	assert.Equal(t, "i-held", holder)
-	assert.Zero(t, detaches)
-}
-
-// The deadlock force exists for: the instance is gone but its volume record
-// still names it, so the ordinary delete refuses forever. Once the holder is
-// proven absent the attachment is cleared and the volume deleted.
-func TestVolumeReaperForceDeletesOnceTheHolderIsProvenGone(t *testing.T) {
-	t.Parallel()
-	cases := map[string]func(cluster *fakeCluster){
-		"terminated": func(cluster *fakeCluster) {
-			cluster.reply("ec2.DescribeTerminatedInstances", instancesIn("terminated", "i-gone"))
-		},
-		"not found": func(*fakeCluster) {},
-	}
-	for name, setup := range cases {
-		t.Run(name, func(t *testing.T) {
-			t.Parallel()
-			cluster := newFakeCluster(t)
-			volume := newHeldVolume(cluster, "vol-held", "i-gone")
-			setup(cluster)
-
-			reaper := newVolumeReaper(cluster.nc, fakeClusterNode)
-			require.NoError(t, reaper.Delete(testCtx(t), "000000000042",
-				Resource{Kind: "volume", ID: "vol-held"}, true))
-
-			deleted, holder, detaches := volume.state()
-			assert.True(t, deleted)
-			assert.Empty(t, holder)
-			assert.Equal(t, 1, detaches)
-
-			var input ec2.DetachVolumeInput
-			cluster.request(t, forceDetachSubject, &input)
-			assert.Equal(t, "vol-held", aws.StringValue(input.VolumeId))
-		})
-	}
-}
-
-// With no holder on record, the refusal is about something else — a snapshot,
-// a mapping the delete guard found — and clearing the record cannot help.
-func TestVolumeReaperForceLeavesAnUnattachedRefusalAlone(t *testing.T) {
-	cluster := newFakeCluster(t)
-	cluster.reply("ec2.DescribeVolumes", ec2.DescribeVolumesOutput{
-		Volumes: []*ec2.Volume{{VolumeId: aws.String("vol-snapped"), State: aws.String("available")}},
-	})
-	cluster.fail("ec2.DeleteVolume", awserrors.ErrorVolumeInUse)
-
-	reaper := newVolumeReaper(cluster.nc, fakeClusterNode)
-	err := reaper.Delete(testCtx(t), "000000000042", Resource{Kind: "volume", ID: "vol-snapped"}, true)
-
-	require.Error(t, err)
-	assert.Contains(t, err.Error(), awserrors.ErrorVolumeInUse)
-	assert.NotContains(t, cluster.called(), forceDetachSubject)
-}
-
-// End to end through the engine: a forced teardown whose guest is still up
-// leaves the volume, names why, and keeps the account TERMINATING.
-func TestForcedTeardownKeepsTheAccountWhileALiveGuestHoldsAVolume(t *testing.T) {
-	t.Parallel()
-	cluster := newFakeCluster(t)
-	volume := newHeldVolume(cluster, "vol-held", "i-live")
-	cluster.reply("ec2.DescribeInstances", instancesIn("running", "i-live"))
-
-	accounts := newFakeAccounts()
-	engine := quietEngine(accounts, newVolumeReaper(cluster.nc, fakeClusterNode))
-	req := request()
-	req.Force = true
-
-	result, err := engine.Teardown(testCtx(t), req)
-
-	require.ErrorIs(t, err, ErrResourcesStuck)
-	require.Equal(t, 1, result.StuckCount())
-	var stuck Stuck
-	for _, stage := range result.Stages {
-		if len(stage.Stuck) > 0 {
-			stuck = stage.Stuck[0]
-		}
-	}
-	assert.Equal(t, "vol-held", stuck.Resource.ID)
-	assert.Contains(t, stuck.Reason, "vol-held")
-	assert.Contains(t, stuck.Reason, "i-live")
-	assert.Contains(t, stuck.Reason, "running")
-	assert.False(t, result.AccountDeleted)
-	assert.False(t, accounts.deleted)
-	assert.Equal(t, AccountStatusTerminating, accounts.account.Status)
-	deleted, holder, _ := volume.state()
-	assert.False(t, deleted)
-	assert.Equal(t, "i-live", holder)
-}
-
-// A volume that is already gone is a success. Teardown re-runs after a crash
-// and would otherwise never finish what the first pass started.
-func TestVolumeReaperTreatsAMissingVolumeAsDeleted(t *testing.T) {
-	cluster := newFakeCluster(t)
-	cluster.fail("ec2.DeleteVolume", "InvalidVolume.NotFound")
-
-	reaper := newVolumeReaper(cluster.nc, fakeClusterNode)
-
-	assert.NoError(t, reaper.Delete(testCtx(t), "000000000042",
-		Resource{Kind: "volume", ID: "vol-gone"}, false))
-}
-
 // Owner-scoped: an image the account can merely see is not its property and
 // must survive its deletion.
 func TestImageReaperListsOnlyOwnedImages(t *testing.T) {
@@ -781,9 +418,9 @@ func TestSubnetAndVPCReapersReportTheirVPC(t *testing.T) {
 // empty account is how a stage reports itself drained while resources remain.
 func TestAListingFailureIsReported(t *testing.T) {
 	cluster := newFakeCluster(t)
-	cluster.fail("ec2.DescribeVolumes", "InternalError")
+	cluster.fail("ec2.DescribeSnapshots", "InternalError")
 
-	_, err := (newVolumeReaper(cluster.nc, fakeClusterNode)).List(testCtx(t), "000000000042")
+	_, err := (&snapshotReaper{nc: cluster.nc}).List(testCtx(t), "000000000042")
 
 	assert.Error(t, err)
 }
@@ -931,7 +568,3 @@ func TestEC2ReapersRemoveInterfacesBeforeSubnets(t *testing.T) {
 	require.NotEqual(t, -1, eni, "there is no network-interface reaper")
 	require.Less(t, eni, subnet, "interfaces must be reaped before the subnets holding them")
 }
-
-func newVolumeReaper(nc *nats.Conn, nodeIDs ...string) *volumeReaper {
-	return &volumeReaper{nc: nc, nodeIDs: nodeIDs}
-}

spinifex/operator/accountlifecycle/reapers_ecs_test.go

@@ -5,6 +5,7 @@ package accountlifecycle
 // registered in.
 
 import (
+	"context"
 	"encoding/json"
 	"github.com/mulgadc/spinifex/spinifex/foundation/aws/errors"
 	"sync"
@@ -13,6 +14,7 @@ import (
 	"github.com/aws/aws-sdk-go/aws"
 	"github.com/aws/aws-sdk-go/service/ecs"
 	"github.com/mulgadc/spinifex/internal/testkit"
+	"github.com/mulgadc/spinifex/spinifex/domains/ec2/accountcleanup"
 	ecsdomain "github.com/mulgadc/spinifex/spinifex/domains/ecs"
 	"github.com/nats-io/nats.go"
 	"github.com/stretchr/testify/assert"
@@ -173,32 +175,37 @@ func TestECSClusterNameFromARN(t *testing.T) {
 	assert.Empty(t, ecsClusterName(""))
 }
 
+// orderedCompute records when the engine reaches the instance cleanup.
+type orderedCompute struct {
+	ComputeStage
+
+	order *[]string
+}
+
+func (o *orderedCompute) CleanupCompute(ctx context.Context, req *accountcleanup.ComputeRequest) (*accountcleanup.ComputeResult, error) {
+	*o.order = append(*o.order, "instances")
+	return o.ComputeStage.CleanupCompute(ctx, req)
+}
+
 // Container instances are ordinary EC2 instances in the tenant's account, so
-// the instance reaper takes them — but the tasks riding on them have to be
-// stopped first, which is what puts ECS ahead of it in the compute stage.
-func TestECSReapersRunBeforeTheInstanceReaper(t *testing.T) {
-	reapers := ECSReapers(nil)
-	reapers = append(reapers, EC2Reapers(nil, 1, []string{"node-1"})...)
+// the compute cleanup terminates them — but the tasks riding on them have to be
+// stopped first, so the compute-stage reapers must run before that cleanup.
+func TestECSClustersAreDeletedBeforeTheInstancesAreTerminated(t *testing.T) {
+	var order []string
+	cluster := &fakeReaper{
+		kind: "ecs-cluster", stage: StageCompute, order: &order,
+		resources: []Resource{{Kind: "ecs-cluster", ID: "web"}},
+	}
+	accounts := newFakeAccounts()
+	engine := quietEngine(accounts, cluster)
+	compute, _ := cleanStages()
+	engine.Compute = &orderedCompute{ComputeStage: compute, order: &order}
 
-	ecsIndex := indexOfKind(reapers, "ecs-cluster")
-	instanceIndex := indexOfKind(reapers, "instance")
+	_, err := engine.Teardown(context.Background(), request())
 
-	require.NotEqual(t, -1, ecsIndex)
-	require.NotEqual(t, -1, instanceIndex)
-	assert.Less(t, ecsIndex, instanceIndex)
-	assert.Equal(t, StageCompute, reapers[ecsIndex].Stage())
-
-	// Both are in the same stage, so the engine's within-stage ordering is
-	// what enforces this rather than the stage graph.
-	assert.Equal(t, reapers[instanceIndex].Stage(), reapers[ecsIndex].Stage())
-}
-
-// SortReapers is what the engine applies, and a stable sort is the only reason
-// the registration order above survives it.
-func TestSortReapersKeepsECSAheadOfInstances(t *testing.T) {
-	reapers := ECSReapers(nil)
-	reapers = append(reapers, EC2Reapers(nil, 1, []string{"node-1"})...)
-	SortReapers(reapers)
-
-	assert.Less(t, indexOfKind(reapers, "ecs-cluster"), indexOfKind(reapers, "instance"))
+	require.NoError(t, err)
+	require.Equal(t, []string{"ecs-cluster/web", "instances"}, order)
+	for _, reaper := range ECSReapers(nil) {
+		assert.Equal(t, StageCompute, reaper.Stage())
+	}
 }

spinifex/operator/admin/account_delete.go

@@ -124,12 +124,12 @@ func (d Deps) deleteAccount(ctx context.Context, body []byte) (any, error) {
 		return nil, errors.New(awserrors.ErrorInternalError)
 	}
 
-	if !req.DryRun && d.PlatformCleaner == nil {
+	if !req.DryRun && (d.PlatformCleaner == nil || d.ComputeCleaner == nil || d.StorageCleaner == nil) {
 		slog.Error("DeleteAccount: platform cleanup not available")
 		return nil, errors.New(awserrors.ErrorServiceUnavailable)
 	}
 
-	engine, err := accountteardown.NewClusterEngine(ctx, d.NATS, accountteardown.ClusterNodes{Expected: d.ExpectedNodes, DaemonIDs: d.NodeIDs, GatewayIDs: d.GatewayNodeIDs}, d.Termination, d.BucketStore, d.PlatformCleaner)
+	engine, err := accountteardown.NewClusterEngine(ctx, d.NATS, accountteardown.ClusterNodes{Expected: d.ExpectedNodes, DaemonIDs: d.NodeIDs, GatewayIDs: d.GatewayNodeIDs}, d.Termination, d.BucketStore, d.PlatformCleaner, accountteardown.WithComputeCleanup(d.ComputeCleaner, d.StorageCleaner))
 	if err != nil {
 		slog.Error("DeleteAccount: failed to build teardown engine", "err", err)
 		return nil, errors.New(awserrors.ErrorServiceUnavailable)

spinifex/operator/admin/account_delete_dryrun_test.go

@@ -26,7 +26,7 @@ func TestAdminDeleteAccountDryRunAnswersTheInventoryAndCleansNothing(t *testing.
 	gotJSON, err := json.Marshal(got)
 	require.NoError(t, err)
 
-	engine, err := accountteardown.NewClusterEngine(t.Context(), gw.NATS, accountteardown.ClusterNodes{Expected: gw.ExpectedNodes, DaemonIDs: gw.NodeIDs, GatewayIDs: gw.GatewayNodeIDs}, gw.Termination, gw.BucketStore, gw.PlatformCleaner)
+	engine, err := accountteardown.NewClusterEngine(t.Context(), gw.NATS, accountteardown.ClusterNodes{Expected: gw.ExpectedNodes, DaemonIDs: gw.NodeIDs, GatewayIDs: gw.GatewayNodeIDs}, gw.Termination, gw.BucketStore, gw.PlatformCleaner, accountteardown.WithComputeCleanup(gw.ComputeCleaner, gw.StorageCleaner))
 	require.NoError(t, err)
 	inventory, err := engine.Inventory(t.Context(), account.AccountID)
 	require.NoError(t, err)

spinifex/operator/admin/account_delete_job_test.go

@@ -8,7 +8,9 @@ import (
 	"context"
 	"encoding/json"
 	operatorv1 "github.com/mulgadc/spinifex/contracts/operator/v1"
+	"github.com/mulgadc/spinifex/spinifex/domains/ec2/accountcleanup"
 	"strings"
+	"sync/atomic"
 	"testing"
 	"time"
 
@@ -148,6 +150,8 @@ func newDeleteAccountGateway(t *testing.T) Deps {
 		GatewayNodeIDs:  []string{"node-1"},
 		BucketStore:     emptyBucketStore{},
 		PlatformCleaner: &panicCleaner{},
+		ComputeCleaner:  emptyCompute{},
+		StorageCleaner:  emptyStorage{},
 	}
 }
 
@@ -607,3 +611,85 @@ func TestDeletionTrackerHeartbeatStops(t *testing.T) {
 	tracker.persist(t.Context())
 	assert.NotEmpty(t, describeDeletion(t, gw, "000000000042").UpdatedAt)
 }
+
+// emptyCompute and emptyStorage stand in for the EC2 cleanup of an account that
+// holds no instance, spot request or volume.
+type (
+	emptyCompute struct{}
+	emptyStorage struct{}
+)
+
+func cleanOutcome(accountID string) *accountcleanup.Outcome {
+	return &accountcleanup.Outcome{AccountID: accountID, Status: accountcleanup.StatusClean}
+}
+
+func (emptyCompute) CleanupCompute(_ context.Context, req *accountcleanup.ComputeRequest) (*accountcleanup.ComputeResult, error) {
+	return cleanOutcome(req.AccountID), nil
+}
+
+func (emptyCompute) ComputeRemaining(_ context.Context, accountID string) (*accountcleanup.Outcome, error) {
+	return cleanOutcome(accountID), nil
+}
+
+func (emptyStorage) CleanupStorage(_ context.Context, req *accountcleanup.StorageRequest) (*accountcleanup.StorageResult, error) {
+	return cleanOutcome(req.AccountID), nil
+}
+
+func (emptyStorage) StorageRemaining(_ context.Context, accountID string) (*accountcleanup.Outcome, error) {
+	return cleanOutcome(accountID), nil
+}
+
+// blockedCompute holds an instance the cleanup cannot release until unblocked,
+// the way a managed service still owning a worker does.
+type blockedCompute struct {
+	blocked atomic.Bool
+}
+
+func (b *blockedCompute) outcome(accountID string) *accountcleanup.Outcome {
+	out := cleanOutcome(accountID)
+	if b.blocked.Load() {
+		out.Items = []accountcleanup.Item{{
+			Kind: accountcleanup.KindInstance, ResourceID: "i-held", State: "running",
+			Disposition: accountcleanup.DispositionBlocked, Reason: accountcleanup.ReasonManagedServiceOwner,
+		}}
+	}
+	return out
+}
+
+func (b *blockedCompute) CleanupCompute(_ context.Context, req *accountcleanup.ComputeRequest) (*accountcleanup.ComputeResult, error) {
+	return b.outcome(req.AccountID), nil
+}
+
+func (b *blockedCompute) ComputeRemaining(_ context.Context, accountID string) (*accountcleanup.Outcome, error) {
+	return b.outcome(accountID), nil
+}
+
+// A job that ends FAILED naming a blocker the loop cannot clear keeps the
+// account, and an operator rerun after the blocker clears converges it.
+func TestAdminDeleteAccountFailedJobConvergesOnOperatorRerun(t *testing.T) {
+	t.Parallel()
+	gw := newDeleteAccountGateway(t)
+	compute := &blockedCompute{}
+	compute.blocked.Store(true)
+	gw.ComputeCleaner = compute
+	account := newTenantAccount(t, gw, "tenant@example.com")
+
+	_, err := gw.deleteAccount(t.Context(),
+		deleteAccountBody(t, account.AccountID, "tenant@example.com", strings.Repeat("a", 32)))
+	require.NoError(t, err)
+	require.Eventually(t, func() bool {
+		return describeDeletion(t, gw, account.AccountID).State == operatorv1.DeletionStateFailed
+	}, 60*time.Second, 50*time.Millisecond)
+	failed := describeDeletion(t, gw, account.AccountID)
+	assert.NotEmpty(t, failed.Error)
+	_, stillThere := gw.IAM.GetAccount(account.AccountID)
+	require.NoError(t, stillThere, "a FAILED job must keep the account")
+
+	compute.blocked.Store(false)
+	_, err = gw.deleteAccount(t.Context(),
+		deleteAccountBody(t, account.AccountID, "tenant@example.com", strings.Repeat("b", 32)))
+	require.NoError(t, err)
+	require.Eventually(t, func() bool {
+		return describeDeletion(t, gw, account.AccountID).State == operatorv1.DeletionStateCompleted
+	}, 60*time.Second, 50*time.Millisecond)
+}

spinifex/operator/admin/account_delete_resumer.go

@@ -99,7 +99,7 @@ func (d Deps) resumeDeletion(
 	revision uint64,
 	now time.Time,
 ) bool {
-	engine, err := accountteardown.NewClusterEngine(ctx, d.NATS, accountteardown.ClusterNodes{Expected: d.ExpectedNodes, DaemonIDs: d.NodeIDs, GatewayIDs: d.GatewayNodeIDs}, d.Termination, d.BucketStore, d.PlatformCleaner)
+	engine, err := accountteardown.NewClusterEngine(ctx, d.NATS, accountteardown.ClusterNodes{Expected: d.ExpectedNodes, DaemonIDs: d.NodeIDs, GatewayIDs: d.GatewayNodeIDs}, d.Termination, d.BucketStore, d.PlatformCleaner, accountteardown.WithComputeCleanup(d.ComputeCleaner, d.StorageCleaner))
 	if err != nil {
 		// Nothing was written, so a later scan retries.
 		slog.Error("DeletionResumer: failed to build teardown engine", "account_id", existing.AccountID, "operation_id", existing.OperationID, "err", err)

spinifex/operator/admin/registration.go

@@ -57,7 +57,11 @@ type Deps struct {
 	Termination accountteardown.IAMBackend
 	// PlatformCleaner is the in-process EC2 platform cleanup every DeleteAccount
 	// runs. Nil refuses a request that is not a dry run.
-	PlatformCleaner   accountteardown.PlatformCleaner
+	PlatformCleaner accountteardown.PlatformCleaner
+	// ComputeCleaner and StorageCleaner are the EC2 instance, spot request and
+	// volume cleanup. Nil refuses a request that is not a dry run.
+	ComputeCleaner    accountteardown.ComputeStage
+	StorageCleaner    accountteardown.StorageStage
 	Quota             *quota.Service
 	ExpectedNodes     int
 	NodeIDs           []string

spinifex/runtime/compute/lambda/testlink/control.go

@@ -151,6 +151,9 @@ type ControlServer struct {
 	// IOTimeout bounds reading the request and writing the reply.
 	IOTimeout time.Duration
 	Handle    func(ctx context.Context, req []byte, cred Credential) Reply
+	// AcceptStopped, if set, runs once Serve stops accepting, before it waits
+	// for the connections it started.
+	AcceptStopped func()
 }
 
 // Serve accepts until ctx ends or the listener fails, then waits for every
@@ -164,6 +167,9 @@ func (s *ControlServer) Serve(ctx context.Context, ln *net.UnixListener) error {
 	defer func() { _ = ln.Close() }()
 	var wg sync.WaitGroup
 	defer wg.Wait()
+	if s.AcceptStopped != nil {
+		defer s.AcceptStopped()
+	}
 	slots := make(chan struct{}, MaxAdmitted)
 	for {
 		select {

spinifex/runtime/compute/lambda/testlink/control_test.go

@@ -436,18 +436,23 @@ func TestControl_WriteErrorAfterDeliveryIsUnknown(t *testing.T) {
 	require.Equal(t, testCred, s.cred.Load())
 }
 
-// legacyControlName is the protocol name before the request carried the image
-// digest; an old daemon still serves it and ignores the field.
-const legacyControlName = "spinifex-lambda-testlink-control"
+// Earlier control protocol names: unversioned, before the request carried the
+// image digest, and /2, before the 6 MiB payload and frame bounds.
+const (
+	legacyControlName = "spinifex-lambda-testlink-control"
+	v2ControlName     = "spinifex-lambda-testlink-control/2"
+)
 
 // A gateway and daemon on different control protocol names never connect, in
 // either direction. The scratch prefix isolates the binds; the names after it
-// are the production constant and its predecessor.
+// are the production constant and its predecessors.
 func TestControl_MixedProtocolNamesNeverConnect(t *testing.T) {
 	prefix := controlName(t) + "/"
 	for name, tc := range map[string]struct{ daemon, gateway string }{
-		"old daemon, new gateway": {prefix + legacyControlName, prefix + ControlName},
-		"new daemon, old gateway": {prefix + ControlName, prefix + legacyControlName},
+		"unversioned daemon, new gateway": {prefix + legacyControlName, prefix + ControlName},
+		"new daemon, unversioned gateway": {prefix + ControlName, prefix + legacyControlName},
+		"/2 daemon, new gateway":          {prefix + v2ControlName, prefix + ControlName},
+		"new daemon, /2 gateway":          {prefix + ControlName, prefix + v2ControlName},
 	} {
 		t.Run(name, func(t *testing.T) {
 			s, _ := serveControl(t, tc.daemon, os.Getuid(), nil)
@@ -457,3 +462,45 @@ func TestControl_MixedProtocolNamesNeverConnect(t *testing.T) {
 		})
 	}
 }
+
+func TestControl_AcceptStoppedRunsBeforeTheDrain(t *testing.T) {
+	name := controlName(t)
+	ln, err := ListenControl(name)
+	require.NoError(t, err)
+	inHandle, release := make(chan struct{}), make(chan struct{})
+	stopped := make(chan struct{})
+	srv := &ControlServer{GatewayUID: os.Getuid(), RequestLimit: testRequestLimit, ReplyLimit: 4096, IOTimeout: 5 * time.Second,
+		Handle: func(context.Context, []byte, Credential) Reply {
+			close(inHandle)
+			<-release
+			return Reply{Outcome: OutcomeDone}
+		},
+		AcceptStopped: func() { close(stopped) }}
+	done := make(chan error, 1)
+	go func() { done <- srv.Serve(context.Background(), ln) }()
+	go func() { _, _ = client(name, os.Getuid()).Exchange(context.Background(), []byte("{}"), testCred, soon()) }()
+	select {
+	case <-inHandle:
+	case <-time.After(5 * time.Second):
+		t.Fatal("the request never reached Handle")
+	}
+
+	require.NoError(t, ln.Close(), "the listener fails under an outstanding invocation")
+	select {
+	case <-stopped:
+	case <-time.After(5 * time.Second):
+		t.Fatal("AcceptStopped did not run while the invocation was outstanding")
+	}
+	select {
+	case err := <-done:
+		t.Fatalf("Serve returned before its connection finished: %v", err)
+	default:
+	}
+	close(release)
+	select {
+	case err := <-done:
+		require.Error(t, err, "a listener failure is reported")
+	case <-time.After(5 * time.Second):
+		t.Fatal("Serve did not return after its connection finished")
+	}
+}

spinifex/runtime/compute/lambda/testlink/daemon.go

@@ -90,10 +90,17 @@ type Daemon struct {
 	// InitializerSHA256 the digest it must have.
 	Initializer       string
 	InitializerSHA256 string
+	// Firecracker, when set, is the Firecracker binary each attempt boots
+	// under instead of QEMU.
+	Firecracker string
 
-	busy atomic.Bool
+	busy    atomic.Bool
+	serving atomic.Bool
 }
 
+// Serving reports whether the control socket is accepting invocations.
+func (d *Daemon) Serving() bool { return d.serving.Load() }
+
 // unresolved names an attempt whose marker is still on disk. A marker goes
 // only once the attempt's VM stop is proven and everything it held is
 // released, so it outlives a daemon restart; nothing resolves it on its own.
@@ -108,6 +115,12 @@ func (d *Daemon) unresolved() (string, error) {
 	return strings.TrimSuffix(filepath.Base(markers[0]), markerSuffix), nil
 }
 
+// validHandler is the handler a guest run accepts: 1..128 bytes, no
+// whitespace or NUL.
+func validHandler(h string) bool {
+	return h != "" && len(h) <= 128 && strings.IndexFunc(h, unicode.IsSpace) < 0 && !strings.ContainsRune(h, 0)
+}
+
 // controlIOTimeout bounds reading a request from, and writing a reply to, the
 // gateway on the control socket.
 const controlIOTimeout = 30 * time.Second
@@ -119,14 +132,24 @@ func (d *Daemon) ServeControl(ctx context.Context, name string, gatewayUID int)
 	if err != nil {
 		return err
 	}
-	srv := &ControlServer{GatewayUID: gatewayUID, RequestLimit: MaxFrameBytes, ReplyLimit: MaxFrameBytes,
-		IOTimeout: controlIOTimeout, Handle: d.Handle}
+	d.serve(ctx, ln, d.controlServer(gatewayUID))
+	return nil
+}
+
+// controlServer answers the gateway with d.Handle, and stops d reporting
+// serving as soon as it stops accepting, before its connections drain.
+func (d *Daemon) controlServer(gatewayUID int) *ControlServer {
+	return &ControlServer{GatewayUID: gatewayUID, RequestLimit: MaxFrameBytes, ReplyLimit: MaxFrameBytes,
+		IOTimeout: controlIOTimeout, Handle: d.Handle, AcceptStopped: func() { d.serving.Store(false) }}
+}
+
+func (d *Daemon) serve(ctx context.Context, ln *net.UnixListener, srv *ControlServer) {
+	d.serving.Store(true)
 	go func() {
 		if err := srv.Serve(ctx, ln); err != nil {
 			slog.Error("Lambda test link: control socket stopped; not serving new invocations", "err", err)
 		}
 	}()
-	return nil
 }
 
 // Handle runs one request under its grant, giving the guest cred in its first
@@ -293,8 +316,7 @@ func CheckRequest(req Request) error {
 	if req.RequestID == "" || req.RevisionID == "" {
 		return errors.New("request and revision IDs are required")
 	}
-	if req.Handler == "" || len(req.Handler) > 128 || strings.IndexFunc(req.Handler, unicode.IsSpace) >= 0 ||
-		strings.ContainsRune(req.Handler, 0) {
+	if !validHandler(req.Handler) {
 		return errors.New("handler must be 1..128 characters with no whitespace")
 	}
 	if req.TimeoutMs <= 0 || req.TimeoutMs > maxTimeoutMs {
@@ -404,24 +426,50 @@ func (d *Daemon) run(ctx context.Context, req Request, code []byte, lib *os.File
 	}
 	defer func() { _ = ln.Close() }()
 
-	cmd, err := cfg.Execute()
-	if err != nil {
-		return refused("VM command build failed"), d.release(a), marker
-	}
-	// A launch delayed past the fixed deadline is refused before QEMU starts.
+	// A launch delayed past the fixed deadline is refused before the VM starts.
 	if expired(hardDeadline) {
 		return refused(ErrDeadlineExpired.Error()), d.release(a), marker
 	}
-	hk, err := armLaunch(cmd, d.Image, lib, hardDeadline)
-	if err != nil {
-		return refused("hard-kill setup failed"), d.release(a), marker
+	var cmd *exec.Cmd
+	var hk *hardKill
+	var fcAck *net.UnixConn
+	fc := &vmHandle{fd: -1}
+	if d.Firecracker == "" {
+		if cmd, err = cfg.Execute(); err != nil {
+			return refused("VM command build failed"), d.release(a), marker
+		}
+		if hk, err = armLaunch(cmd, d.Image, lib, hardDeadline); err != nil {
+			return refused("hard-kill setup failed"), d.release(a), marker
+		}
+		defer hk.close()
+	} else {
+		var child *os.File
+		if fcAck, child, err = ackPair(); err != nil {
+			return refused("hard-kill setup failed"), d.release(a), marker
+		}
+		defer func() { _ = fcAck.Close() }()
+		cmd, err = d.supervisorCommand(filepath.Join(a.dir, fcConfigFile), hardDeadline, child)
+		defer func() { _ = child.Close() }()
+		if err != nil {
+			return refused("VM command build failed"), d.release(a), marker
+		}
 	}
-	defer hk.close()
 	var result Reply
 	unrecorded := false
 	out, runErr := attempt.Run(ctx, cmd, timeoutFor(req), func(sctx context.Context, pidfd int) error {
+		vmPidfd, vmPID := pidfd, cmd.Process.Pid
+		if fcAck != nil {
+			defer fc.callbackDone()
+			// Firecracker is the VM; its supervisor acks only once both kernel kills are armed.
+			fd, pid, err := awaitFirecracker(sctx, fcAck, cmd.Process.Pid, armWait)
+			if err != nil {
+				return err
+			}
+			fc.set(fd)
+			vmPidfd, vmPID = fd, pid
+		}
 		// The owner holds the launch identity before the guest is sent anything.
-		l, err := launchIdentity(pidfd, cmd.Process.Pid, a.tap)
+		l, err := launchIdentity(vmPidfd, vmPID, a.tap)
 		if err != nil {
 			return err
 		}
@@ -430,13 +478,20 @@ func (d *Daemon) run(ctx context.Context, req Request, code []byte, lib *os.File
 			return fmt.Errorf("%w: %w", ErrLaunchNotRecorded, err)
 		}
 		// No frame reaches the guest until QEMU proves its own kill timer is armed.
-		if err := hk.await(sctx, cmd.Process.Pid, armWait); err != nil {
-			return err
+		if hk != nil {
+			if err := hk.await(sctx, cmd.Process.Pid, armWait); err != nil {
+				return err
+			}
 		}
 		r, err := d.session(sctx, req, gen, ln, a.dir, hardDeadline, d.Grants.Accept)
 		result = r
 		return err
 	})
+	if fcAck != nil {
+		// Always settled, so a received pidfd is closed or handed to the callback.
+		fcGone := settleFirecracker(out.SessionDone, fc, cmd.Process != nil)
+		out.Stopped = out.Stopped && fcGone
+	}
 	if !out.Stopped {
 		slog.Error("Lambda test link: VM stop not proven; marker, tap, address and files held",
 			"attempt", req.AttemptID, "tap", tapName, "err", runErr)
@@ -527,6 +582,22 @@ func (d *Daemon) prepare(req Request, code []byte, dir, tap, mac, ip string) (*v
 	entries["test-package-sha256"] = []byte(req.CodeSHA256)
 	entries["test-grant"] = grantEntry(req.Grant)
 	entries["test-handler"] = []byte(req.Handler)
+	netcfgBytes := fmt.Appendf(nil, "NIC0_MAC=%s\nNIC0_CIDR=%s/24\nNIC0_DEFAULT=0\n", mac, ip)
+	if d.Firecracker != "" {
+		archive := make(map[string][]byte, len(entries)+1)
+		for name, value := range entries {
+			archive[lambdav1.FwCfgPrefix+name] = value
+		}
+		archive[fwCfgNetcfg] = netcfgBytes
+		cfgDisk := filepath.Join(dir, configArchive)
+		if err := writeConfigArchive(cfgDisk, archive); err != nil {
+			return fail(err)
+		}
+		if err := writeFirecrackerConfig(filepath.Join(dir, fcConfigFile), diskPath, cfgDisk, tap, mac, int(max(req.MemoryMiB, 128))); err != nil {
+			return fail(err)
+		}
+		return nil, gen, ln, nil
+	}
 	var fwCfg []vm.FwCfgEntry
 	for name, value := range entries {
 		path := filepath.Join(dir, "fwcfg-"+name)
@@ -536,7 +607,7 @@ func (d *Daemon) prepare(req Request, code []byte, dir, tap, mac, ip string) (*v
 		fwCfg = append(fwCfg, vm.FwCfgEntry{Name: lambdav1.FwCfgPrefix + name, File: path})
 	}
 	netcfg := filepath.Join(dir, "netcfg")
-	if err := os.WriteFile(netcfg, fmt.Appendf(nil, "NIC0_MAC=%s\nNIC0_CIDR=%s/24\nNIC0_DEFAULT=0\n", mac, ip), 0o600); err != nil {
+	if err := os.WriteFile(netcfg, netcfgBytes, 0o600); err != nil {
 		return fail(err)
 	}
 	fwCfg = append(fwCfg, vm.FwCfgEntry{Name: fwCfgNetcfg, File: netcfg})
@@ -588,6 +659,7 @@ func (d *Daemon) session(ctx context.Context, req Request, gen *guestconn.Genera
 	}
 	defer func() { _ = conn.Close() }()
 	_ = os.Remove(filepath.Join(dir, "fwcfg-"+lambdav1.EntryGuestKey))
+	_ = os.Remove(filepath.Join(dir, configArchive))
 
 	// The function timeout runs on the host from dispatch; the boot allowance
 	// covers startup only, so a guest that stops answering is cut off here.

spinifex/runtime/compute/lambda/testlink/firecracker.go

@@ -0,0 +1,396 @@
+//go:build lambdatestlink
+
+package testlink
+
+import (
+	"archive/tar"
+	"bytes"
+	"context"
+	"encoding/json"
+	"errors"
+	"fmt"
+	"net"
+	"os"
+	"os/exec"
+	"path/filepath"
+	"runtime"
+	"sort"
+	"strconv"
+	"strings"
+	"sync"
+	"syscall"
+	"time"
+	"unsafe"
+
+	"golang.org/x/sys/unix"
+)
+
+// firecrackerArg0 is the argv[0] that turns the spx binary into the
+// Firecracker supervisor, so the supervisor is the already-pinned binary.
+const firecrackerArg0 = "spx-lambda-firecracker"
+
+// The supervisor inherits the sealed kernel and initramfs as fds 3 and 4,
+// which it passes on to Firecracker unchanged, and its ack socket as fd 5.
+const (
+	fcKernelFD = 3
+	fcInitrdFD = 4
+	fcAckFD    = 5
+	fcAck      = "armed"
+)
+
+// firecrackerCmdline is the guest kernel command line; reboot=k lets a guest
+// reboot end Firecracker, and Firecracker adds the virtio-mmio devices itself.
+const firecrackerCmdline = "console=ttyS0 quiet loglevel=3 reboot=k panic=1 pci=off"
+
+const (
+	configArchive = "config.tar"
+	fcConfigFile  = "firecracker.json"
+	// fcStopWait bounds proving Firecracker gone after its supervisor stopped.
+	fcStopWait = 5 * time.Second
+)
+
+func init() {
+	if filepath.Base(os.Args[0]) == firecrackerArg0 {
+		os.Exit(supervise(os.Args[1:]))
+	}
+}
+
+// supervise arms a CLOCK_BOOTTIME SIGKILL on itself at the absolute deadline,
+// then starts Firecracker with SIGKILL as its parent-death signal, so the
+// kernel ends both at the deadline even if this process and the daemon stop.
+func supervise(args []string) int {
+	if len(args) != 3 {
+		return 2
+	}
+	deadline, err := strconv.ParseInt(args[2], 10, 64)
+	if err != nil {
+		return 2
+	}
+	unix.CloseOnExec(fcAckFD)
+	// The parent-death signal follows the forking thread, which never exits.
+	runtime.LockOSThread()
+	if err := armBoottimeKill(time.Duration(deadline)); err != nil {
+		fmt.Fprintln(os.Stderr, "firecracker supervisor:", err)
+		return 3
+	}
+	pidfd := -1
+	cmd := exec.Command(args[0], "--no-api", "--config-file", args[1], "--level", "Warning")
+	cmd.Stdout, cmd.Stderr = os.Stdout, os.Stderr
+	cmd.ExtraFiles = []*os.File{os.NewFile(fcKernelFD, "kernel"), os.NewFile(fcInitrdFD, "initrd")}
+	cmd.SysProcAttr = &syscall.SysProcAttr{Pdeathsig: syscall.SIGKILL, PidFD: &pidfd}
+	if err := cmd.Start(); err != nil {
+		fmt.Fprintln(os.Stderr, "firecracker supervisor: start:", err)
+		return 4
+	}
+	if err := unix.Sendmsg(fcAckFD, []byte(fcAck), unix.UnixRights(pidfd), nil, 0); err != nil {
+		_ = cmd.Process.Kill()
+		_ = cmd.Wait()
+		return 5
+	}
+	if err := cmd.Wait(); err != nil {
+		var exit *exec.ExitError
+		if errors.As(err, &exit) {
+			return exit.ExitCode()
+		}
+		return 6
+	}
+	return 0
+}
+
+// sigevent is the kernel's struct sigevent on 64-bit Linux.
+type sigevent struct {
+	value  uintptr
+	signo  int32
+	notify int32
+	_      [48]byte
+}
+
+const (
+	sigevSignal  = 0
+	timerAbstime = 1
+)
+
+// armBoottimeKill arms a POSIX timer that delivers SIGKILL to this process at
+// the absolute CLOCK_BOOTTIME deadline; a passed deadline fires at once.
+func armBoottimeKill(deadline time.Duration) error {
+	ev := sigevent{signo: int32(unix.SIGKILL), notify: sigevSignal}
+	var id int32
+	if _, _, e := unix.Syscall(unix.SYS_TIMER_CREATE, unix.CLOCK_BOOTTIME,
+		uintptr(unsafe.Pointer(&ev)), uintptr(unsafe.Pointer(&id))); e != 0 {
+		return fmt.Errorf("timer_create: %w", e)
+	}
+	spec := unix.ItimerSpec{Value: unix.NsecToTimespec(int64(deadline))}
+	if _, _, e := unix.Syscall6(unix.SYS_TIMER_SETTIME, uintptr(id), timerAbstime,
+		uintptr(unsafe.Pointer(&spec)), 0, 0, 0); e != 0 {
+		return fmt.Errorf("timer_settime: %w", e)
+	}
+	return nil
+}
+
+// supervisorCommand builds the supervisor for one attempt: the spx binary
+// under firecrackerArg0, the sealed kernel and initramfs, and the ack socket.
+func (d *Daemon) supervisorCommand(configPath string, deadline time.Duration, ack *os.File) (*exec.Cmd, error) {
+	if d.Image == nil || d.Image.Kernel == nil || d.Image.Initrd == nil {
+		return nil, errors.New("runtime image not prepared")
+	}
+	self, err := os.Executable()
+	if err != nil {
+		return nil, err
+	}
+	cmd := exec.Command(self, d.Firecracker, configPath, strconv.FormatInt(int64(deadline), 10))
+	cmd.Args[0] = firecrackerArg0
+	cmd.Stdout, cmd.Stderr = os.Stdout, os.Stderr
+	cmd.ExtraFiles = []*os.File{d.Image.Kernel, d.Image.Initrd, ack}
+	return cmd, nil
+}
+
+// ackPair is the socket the supervisor acks on; the daemon keeps one end.
+func ackPair() (daemonEnd *net.UnixConn, child *os.File, err error) {
+	fds, err := unix.Socketpair(unix.AF_UNIX, unix.SOCK_SEQPACKET|unix.SOCK_CLOEXEC, 0)
+	if err != nil {
+		return nil, nil, err
+	}
+	child = os.NewFile(uintptr(fds[1]), "fc-ack")
+	f := os.NewFile(uintptr(fds[0]), "fc-ack-daemon")
+	c, err := net.FileConn(f)
+	_ = f.Close()
+	if err != nil {
+		_ = child.Close()
+		return nil, nil, err
+	}
+	return c.(*net.UnixConn), child, nil
+}
+
+// awaitFirecracker reads the supervisor's ack and the pidfd it carries, and
+// checks that pidfd names a live child of the supervisor.
+func awaitFirecracker(ctx context.Context, ack *net.UnixConn, supervisor int, wait time.Duration) (pidfd, pid int, err error) {
+	deadline := time.Now().Add(wait)
+	if d, ok := ctx.Deadline(); ok && d.Before(deadline) {
+		deadline = d
+	}
+	if err := ack.SetReadDeadline(deadline); err != nil {
+		return -1, 0, err
+	}
+	buf := make([]byte, 16)
+	oob := make([]byte, unix.CmsgSpace(4))
+	n, oobn, _, _, err := ack.ReadMsgUnix(buf, oob)
+	if err != nil {
+		return -1, 0, fmt.Errorf("%w: %w", ErrNotArmed, err)
+	}
+	msgs, err := unix.ParseSocketControlMessage(oob[:oobn])
+	if err != nil || len(msgs) != 1 {
+		return -1, 0, fmt.Errorf("%w: no pidfd in ack", ErrNotArmed)
+	}
+	fds, err := unix.ParseUnixRights(&msgs[0])
+	if err != nil || len(fds) != 1 {
+		return -1, 0, fmt.Errorf("%w: no pidfd in ack", ErrNotArmed)
+	}
+	pidfd = fds[0]
+	unix.CloseOnExec(pidfd)
+	if string(buf[:n]) != fcAck {
+		_ = unix.Close(pidfd)
+		return -1, 0, fmt.Errorf("%w: unexpected ack", ErrNotArmed)
+	}
+	if pid, err = pidfdPID(pidfd); err == nil {
+		err = checkParent(pid, supervisor)
+	}
+	if err != nil {
+		_ = unix.Close(pidfd)
+		return -1, 0, fmt.Errorf("%w: %w", ErrNotArmed, err)
+	}
+	return pidfd, pid, nil
+}
+
+// pidfdPID reads the PID a pidfd refers to from its fdinfo.
+func pidfdPID(pidfd int) (int, error) {
+	b, err := os.ReadFile(fmt.Sprintf("/proc/self/fdinfo/%d", pidfd))
+	if err != nil {
+		return 0, err
+	}
+	for line := range strings.SplitSeq(string(b), "\n") {
+		if v, ok := strings.CutPrefix(line, "Pid:"); ok {
+			pid, err := strconv.Atoi(strings.TrimSpace(v))
+			if err != nil || pid <= 0 {
+				return 0, fmt.Errorf("pidfd names no live process (%q)", strings.TrimSpace(v))
+			}
+			return pid, nil
+		}
+	}
+	return 0, errors.New("pidfd fdinfo has no Pid")
+}
+
+func checkParent(pid, parent int) error {
+	b, err := os.ReadFile(fmt.Sprintf("/proc/%d/status", pid))
+	if err != nil {
+		return err
+	}
+	for line := range strings.SplitSeq(string(b), "\n") {
+		if v, ok := strings.CutPrefix(line, "PPid:"); ok {
+			if strings.TrimSpace(v) != strconv.Itoa(parent) {
+				return fmt.Errorf("process %d is not the supervisor's child", pid)
+			}
+			return nil
+		}
+	}
+	return errors.New("process status has no PPid")
+}
+
+// stopFirecracker kills Firecracker through its pidfd and reports whether its
+// exit was proven; the supervisor's own exit proves nothing about it.
+func stopFirecracker(pidfd int) bool {
+	if pidfd < 0 {
+		return false
+	}
+	_ = unix.PidfdSendSignal(pidfd, unix.SIGKILL, nil, 0)
+	exited, err := pidfdExited(pidfd, fcStopWait)
+	return err == nil && exited
+}
+
+// settleFirecracker reports whether Firecracker is proven gone once Run has
+// returned. A supervisor that never started launched nothing; otherwise only
+// Firecracker's own pidfd proves its exit, since the supervisor's says nothing.
+func settleFirecracker(sessionDone <-chan struct{}, fc *vmHandle, supervisorStarted bool) bool {
+	select {
+	case <-sessionDone:
+	default:
+		// The callback may still receive the pidfd; it closes it, and the attempt is held.
+		fc.abandon()
+		return false
+	}
+	fd := fc.take()
+	if fd < 0 {
+		return !supervisorStarted
+	}
+	defer func() { _ = unix.Close(fd) }()
+	return stopFirecracker(fd)
+}
+
+// writeConfigArchive writes entries as a tar archive in the fw_cfg sysfs
+// layout, by_name/<entry>/raw, which the guest's init extracts.
+func writeConfigArchive(path string, entries map[string][]byte) error {
+	names := make([]string, 0, len(entries))
+	for name := range entries {
+		names = append(names, name)
+	}
+	sort.Strings(names)
+	var buf bytes.Buffer
+	tw := tar.NewWriter(&buf)
+	dirs := map[string]bool{}
+	for _, name := range names {
+		file := "by_name/" + name + "/raw"
+		var parents []string
+		for dir := filepath.Dir(file); dir != "." && !dirs[dir]; dir = filepath.Dir(dir) {
+			parents = append(parents, dir)
+			dirs[dir] = true
+		}
+		for i := len(parents) - 1; i >= 0; i-- {
+			if err := tw.WriteHeader(&tar.Header{Typeflag: tar.TypeDir, Name: parents[i] + "/", Mode: 0o700}); err != nil {
+				return err
+			}
+		}
+		b := entries[name]
+		if err := tw.WriteHeader(&tar.Header{Typeflag: tar.TypeReg, Name: file, Mode: 0o600, Size: int64(len(b))}); err != nil {
+			return err
+		}
+		if _, err := tw.Write(b); err != nil {
+			return err
+		}
+	}
+	if err := tw.Close(); err != nil {
+		return err
+	}
+	return os.WriteFile(path, buf.Bytes(), 0o600)
+}
+
+type fcConfig struct {
+	BootSource struct {
+		Kernel   string `json:"kernel_image_path"`
+		Initrd   string `json:"initrd_path"`
+		BootArgs string `json:"boot_args"`
+	} `json:"boot-source"`
+	Drives  []fcDrive `json:"drives"`
+	Machine struct {
+		VCPUs  int  `json:"vcpu_count"`
+		MemMiB int  `json:"mem_size_mib"`
+		SMT    bool `json:"smt"`
+	} `json:"machine-config"`
+	Net []fcNet `json:"network-interfaces"`
+}
+
+type fcDrive struct {
+	ID       string `json:"drive_id"`
+	Path     string `json:"path_on_host"`
+	Root     bool   `json:"is_root_device"`
+	ReadOnly bool   `json:"is_read_only"`
+}
+
+type fcNet struct {
+	ID  string `json:"iface_id"`
+	MAC string `json:"guest_mac"`
+	Tap string `json:"host_dev_name"`
+}
+
+// writeFirecrackerConfig writes the attempt's Firecracker config: the sealed
+// kernel and initramfs, the package disk then the configuration disk, and the tap.
+func writeFirecrackerConfig(path, pkg, cfgDisk, tap, mac string, memMiB int) error {
+	var c fcConfig
+	c.BootSource.Kernel = fmt.Sprintf("/proc/self/fd/%d", fcKernelFD)
+	c.BootSource.Initrd = fmt.Sprintf("/proc/self/fd/%d", fcInitrdFD)
+	c.BootSource.BootArgs = firecrackerCmdline
+	c.Drives = []fcDrive{{ID: "pkg", Path: pkg, ReadOnly: true}, {ID: "cfg", Path: cfgDisk}}
+	c.Machine.VCPUs, c.Machine.MemMiB = 1, memMiB
+	c.Net = []fcNet{{ID: "net0", MAC: mac, Tap: tap}}
+	b, err := json.Marshal(c)
+	if err != nil {
+		return err
+	}
+	return os.WriteFile(path, b, 0o600)
+}
+
+// vmHandle owns Firecracker's pidfd across the session callback, which can
+// still be running when attempt.Run returns; whichever side finishes last
+// closes an abandoned fd.
+type vmHandle struct {
+	mu        sync.Mutex
+	fd        int
+	finished  bool
+	abandoned bool
+}
+
+func (h *vmHandle) set(fd int) {
+	h.mu.Lock()
+	defer h.mu.Unlock()
+	h.fd = fd
+}
+
+// callbackDone runs as the callback returns.
+func (h *vmHandle) callbackDone() {
+	h.mu.Lock()
+	defer h.mu.Unlock()
+	h.finished = true
+	h.closeIfAbandoned()
+}
+
+// abandon gives up the fd while the callback may still run.
+func (h *vmHandle) abandon() {
+	h.mu.Lock()
+	defer h.mu.Unlock()
+	h.abandoned = true
+	h.closeIfAbandoned()
+}
+
+func (h *vmHandle) closeIfAbandoned() {
+	if h.abandoned && h.finished && h.fd >= 0 {
+		_ = unix.Close(h.fd)
+		h.fd = -1
+	}
+}
+
+// take hands the fd to teardown once the callback has returned.
+func (h *vmHandle) take() int {
+	h.mu.Lock()
+	defer h.mu.Unlock()
+	fd := h.fd
+	h.fd = -1
+	return fd
+}

spinifex/runtime/compute/lambda/testlink/firecracker_test.go

@@ -0,0 +1,190 @@
+//go:build lambdatestlink
+
+//test:in-package drives the unexported supervisor, ack and pidfd helpers against real processes
+
+package testlink
+
+import (
+	"archive/tar"
+	"bytes"
+	"context"
+	"errors"
+	"io"
+	"os"
+	"os/exec"
+	"path/filepath"
+	"strconv"
+	"strings"
+	"syscall"
+	"testing"
+	"time"
+
+	lambdav1 "github.com/mulgadc/spinifex/contracts/lambda/v1"
+	"github.com/mulgadc/spinifex/spinifex/agents/lambda/hostlink"
+	"github.com/mulgadc/spinifex/spinifex/runtime/compute/lambda/attempt"
+	"github.com/mulgadc/spinifex/spinifex/runtime/compute/lambda/guestconn"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"golang.org/x/sys/unix"
+)
+
+// supervised starts this test binary as the Firecracker supervisor over a
+// stand-in VMM that runs until killed, and returns the supervisor and the
+// VMM's pidfd from the ack.
+func supervised(t *testing.T, deadline time.Duration) (*exec.Cmd, int) {
+	t.Helper()
+	dir := t.TempDir()
+	vmm := filepath.Join(dir, "firecracker")
+	require.NoError(t, os.WriteFile(vmm, []byte("#!/bin/sh\nexec tail -f /dev/null\n"), 0o755))
+	kernel, err := os.Open(os.DevNull)
+	require.NoError(t, err)
+	initrd, err := os.Open(os.DevNull)
+	require.NoError(t, err)
+	t.Cleanup(func() { _ = kernel.Close(); _ = initrd.Close() })
+	d := &Daemon{Firecracker: vmm, Image: &Image{Kernel: kernel, Initrd: initrd}}
+	ack, child, err := ackPair()
+	require.NoError(t, err)
+	t.Cleanup(func() { _ = ack.Close() })
+	cmd, err := d.supervisorCommand(filepath.Join(dir, fcConfigFile), deadline, child)
+	require.NoError(t, err)
+	cmd.Path = os.Args[0]
+	cmd.Stdout, cmd.Stderr = io.Discard, io.Discard
+	require.NoError(t, cmd.Start())
+	_ = child.Close()
+	t.Cleanup(func() { _ = cmd.Process.Kill(); _ = cmd.Wait() })
+
+	pidfd, pid, err := awaitFirecracker(context.Background(), ack, cmd.Process.Pid, armWait)
+	require.NoError(t, err)
+	t.Cleanup(func() { _ = unix.PidfdSendSignal(pidfd, unix.SIGKILL, nil, 0); _ = unix.Close(pidfd) })
+	require.NotEqual(t, cmd.Process.Pid, pid)
+	return cmd, pidfd
+}
+
+func boottime(t *testing.T) time.Duration {
+	t.Helper()
+	now, err := bootNow()
+	require.NoError(t, err)
+	return now
+}
+
+func TestFirecrackerSupervisorKillsAtBoottimeDeadlineWhileStopped(t *testing.T) {
+	cmd, pidfd := supervised(t, boottime(t)+1500*time.Millisecond)
+
+	// The kernel holds a SIGKILL timer on CLOCK_BOOTTIME (7), so the deadline counts suspend.
+	timers, err := os.ReadFile("/proc/" + strconv.Itoa(cmd.Process.Pid) + "/timers")
+	require.NoError(t, err)
+	assert.Contains(t, string(timers), "signal: 9/")
+	assert.Contains(t, string(timers), "ClockID: 7")
+
+	// A stopped supervisor still loses the VMM at the deadline: both kills come from the kernel.
+	require.NoError(t, cmd.Process.Signal(syscall.SIGSTOP))
+	exited, err := pidfdExited(pidfd, 5*time.Second)
+	require.NoError(t, err)
+	assert.True(t, exited, "VMM outlived its deadline")
+	state, err := cmd.Process.Wait()
+	require.NoError(t, err)
+	assert.Equal(t, syscall.SIGKILL, state.Sys().(syscall.WaitStatus).Signal())
+}
+
+func TestFirecrackerDiesWithItsSupervisor(t *testing.T) {
+	cmd, pidfd := supervised(t, boottime(t)+time.Hour)
+	exited, err := pidfdExited(pidfd, 0)
+	require.NoError(t, err)
+	require.False(t, exited)
+
+	require.NoError(t, cmd.Process.Kill())
+	_, _ = cmd.Process.Wait()
+	// Only the kernel's parent-death signal can end the VMM here; nothing else signals it.
+	exited, err = pidfdExited(pidfd, 5*time.Second)
+	require.NoError(t, err)
+	assert.True(t, exited, "VMM outlived its supervisor")
+}
+
+// A supervisor that never started launched nothing, so the attempt is
+// released; one that started without handing over a pidfd may have started
+// Firecracker, so the attempt is held.
+func TestSettleFirecracker_OnlyANeverStartedSupervisorIsProvenStopped(t *testing.T) {
+	cancelled, cancel := context.WithCancel(context.Background())
+	cancel()
+	cases := map[string]struct {
+		ctx  context.Context
+		cmd  func() *exec.Cmd
+		want bool
+	}{
+		"cancelled before Start": {cancelled, func() *exec.Cmd { return exec.Command("/bin/true") }, true},
+		"failed Start":           {context.Background(), func() *exec.Cmd { return exec.Command(filepath.Join(t.TempDir(), "missing")) }, true},
+		"started, no pidfd":      {context.Background(), func() *exec.Cmd { return exec.Command("/bin/true") }, false},
+	}
+	for name, tc := range cases {
+		t.Run(name, func(t *testing.T) {
+			cmd := tc.cmd()
+			fc := &vmHandle{fd: -1}
+			out, _ := attempt.Run(tc.ctx, cmd, 10*time.Second, func(context.Context, int) error {
+				fc.callbackDone()
+				return errors.New("supervisor exited before its ack")
+			})
+			require.True(t, out.Stopped, "the supervisor itself is proven stopped")
+			<-out.SessionDone
+			assert.Equal(t, tc.want, settleFirecracker(out.SessionDone, fc, cmd.Process != nil))
+		})
+	}
+}
+
+func TestFirecrackerAckRequiresTheSupervisorsChild(t *testing.T) {
+	ack, child, err := ackPair()
+	require.NoError(t, err)
+	defer func() { _ = ack.Close() }()
+	self, err := unix.PidfdOpen(os.Getpid(), 0)
+	require.NoError(t, err)
+	require.NoError(t, unix.Sendmsg(int(child.Fd()), []byte(fcAck), unix.UnixRights(self), nil, 0))
+	_ = unix.Close(self)
+	_ = child.Close()
+
+	_, _, err = awaitFirecracker(context.Background(), ack, 1, time.Second)
+	require.ErrorIs(t, err, ErrNotArmed)
+	assert.Contains(t, err.Error(), "not the supervisor's child")
+}
+
+// The archive the daemon writes is read by the guest's own identity reader
+// once extracted, as the runtime image's init extracts it.
+func TestConfigArchiveIsReadByTheGuestReaders(t *testing.T) {
+	gen, err := guestconn.NewGeneration("ap-southeast-2/000000000001/fn/env/1", time.Hour)
+	require.NoError(t, err)
+	entries := map[string][]byte{fwCfgNetcfg: []byte("NIC0_MAC=02:00:00:00:00:01\n")}
+	for name, b := range gen.GuestEntries("192.0.2.1:4000") {
+		entries[lambdav1.FwCfgPrefix+name] = b
+	}
+	path := filepath.Join(t.TempDir(), configArchive)
+	require.NoError(t, writeConfigArchive(path, entries))
+
+	root := filepath.Join(t.TempDir(), "cfg")
+	extract(t, path, root)
+	_, err = hostlink.LoadIdentity(filepath.Join(root, "by_name", lambdav1.FwCfgPrefix))
+	require.NoError(t, err)
+	netcfg, err := os.ReadFile(filepath.Join(root, "by_name", fwCfgNetcfg, "raw"))
+	require.NoError(t, err)
+	assert.Equal(t, entries[fwCfgNetcfg], netcfg)
+}
+
+func extract(t *testing.T, archive, root string) {
+	t.Helper()
+	b, err := os.ReadFile(archive)
+	require.NoError(t, err)
+	tr := tar.NewReader(bytes.NewReader(b))
+	for {
+		h, err := tr.Next()
+		if err == io.EOF {
+			return
+		}
+		require.NoError(t, err)
+		require.False(t, strings.Contains(h.Name, ".."), h.Name)
+		p := filepath.Join(root, h.Name)
+		if h.Typeflag == tar.TypeDir {
+			require.NoError(t, os.MkdirAll(p, 0o700))
+			continue
+		}
+		data, err := io.ReadAll(tr)
+		require.NoError(t, err)
+		require.NoError(t, os.WriteFile(p, data, 0o600))
+	}
+}

spinifex/runtime/compute/lambda/testlink/readiness.go

@@ -0,0 +1,38 @@
+//go:build lambdatestlink
+
+package testlink
+
+import (
+	"context"
+
+	"github.com/mulgadc/spinifex/spinifex/domains/lambda/function"
+)
+
+// Readiness admits a function the test link can run, and only while d serves
+// the pinned image on the cluster's only node, so no gateway or daemon is
+// handed an Active function it cannot run.
+func (d *Daemon) Readiness(imagePin string, nodes int) function.Readiness {
+	return func(_ context.Context, rec function.Record) error {
+		if rec.EnvCipher != "" {
+			return notReady("The function cannot run here: test-link Lambda does not support environment variables.")
+		}
+		if !validHandler(rec.Handler) {
+			return notReady("The function cannot run here: test-link Lambda runs only a handler of 1 to 128 characters with no whitespace.")
+		}
+		if rec.Code.Size <= 0 || rec.Code.Size > MaxCodeBytes {
+			return notReady("The function cannot run here: its code is larger than test-link Lambda accepts.")
+		}
+		if nodes != 1 {
+			return notReady("The function cannot run here: test-link Lambda activates only on a single-node cluster.")
+		}
+		if !d.Serving() || d.Image == nil || d.Image.Digest != imagePin {
+			return notReady(function.ReasonCreating)
+		}
+		return nil
+	}
+}
+
+// notReady is an unmet activation condition, worded as the StateReason.
+type notReady string
+
+func (n notReady) Error() string { return string(n) }

spinifex/runtime/compute/lambda/testlink/readiness_test.go

@@ -0,0 +1,173 @@
+//go:build lambdatestlink
+
+//test:in-package — reuses the package fixtures writeImage, testImageDigest and controlName
+package testlink
+
+import (
+	"context"
+	"os"
+	"strings"
+	"testing"
+	"time"
+
+	"github.com/mulgadc/spinifex/internal/testkit"
+	"github.com/mulgadc/spinifex/spinifex/domains/lambda/function"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+const (
+	readyAccount = "000000000001"
+	readyRegion  = "ap-southeast-2"
+)
+
+// storedFunction creates a function through the real owner and records its
+// code as stored, leaving it Pending the way CreateFunction does.
+func storedFunction(t *testing.T, svc *function.MetaService, name string) {
+	t.Helper()
+	storedFunctionWith(t, svc, name, "", 9)
+}
+
+func storedFunctionWith(t *testing.T, svc *function.MetaService, name, envCipher string, size int64) {
+	t.Helper()
+	storedFunctionHandler(t, svc, name, "bootstrap", envCipher, size)
+}
+
+func storedFunctionHandler(t *testing.T, svc *function.MetaService, name, handler, envCipher string, size int64) {
+	t.Helper()
+	created, err := svc.Create(t.Context(), &function.CreateRequest{Region: readyRegion, Record: function.Record{
+		FunctionID: "id-" + name, FunctionName: name, Role: "arn:aws:iam::000000000001:role/r",
+		Runtime: "provided.al2023", Handler: handler, Timeout: 3, MemorySize: 128, EnvCipher: envCipher,
+		Code: function.Code{RevisionID: "rev-" + name},
+	}}, readyAccount)
+	require.NoError(t, err)
+	_, err = svc.CodeStored(t.Context(), &function.CodeStoredRequest{Region: readyRegion, FunctionName: name,
+		Revision: created.Revision, SHA256: testImageDigest, Size: size}, readyAccount)
+	require.NoError(t, err)
+}
+
+func functionState(t *testing.T, svc *function.MetaService, name string) function.Record {
+	t.Helper()
+	got, err := svc.Get(t.Context(), &function.GetRequest{Region: readyRegion, FunctionName: name}, readyAccount)
+	require.NoError(t, err)
+	require.True(t, got.Found)
+	return got.Record
+}
+
+func readyOwner(t *testing.T) *function.MetaService {
+	t.Helper()
+	_, nc, _ := testutil.StartTestJetStream(t)
+	return function.NewMetaService(testutil.NewJetStream(t, nc))
+}
+
+func TestReadiness_ActivatesOnlyOnceTheDaemonServesThePinnedImage(t *testing.T) {
+	svc := readyOwner(t)
+	im, err := PrepareImage(writeImage(t), testImageDigest)
+	require.NoError(t, err)
+	t.Cleanup(im.Close)
+	d := &Daemon{Node: "node1", Image: im}
+	svc.SetReadiness(d.Readiness(testImageDigest, 1))
+	storedFunction(t, svc, "hello")
+
+	require.NoError(t, svc.ActivatePending(t.Context()))
+	waiting := functionState(t, svc, "hello")
+	assert.Equal(t, function.StatePending, waiting.State, "a prepared image that is not served activates nothing")
+	assert.Equal(t, function.ReasonCreating, waiting.StateReason)
+
+	ctx, cancel := context.WithCancel(t.Context())
+	require.NoError(t, d.ServeControl(ctx, controlName(t), os.Getuid()))
+	require.NoError(t, svc.ActivatePending(t.Context()))
+	active := functionState(t, svc, "hello")
+	assert.Equal(t, function.StateActive, active.State)
+	assert.Empty(t, active.StateReason)
+
+	cancel()
+	require.Eventually(t, func() bool { return !d.Serving() }, 5*time.Second, 10*time.Millisecond,
+		"a stopped control socket is reported as not serving")
+	storedFunction(t, svc, "later")
+	require.NoError(t, svc.ActivatePending(t.Context()))
+	assert.Equal(t, function.StatePending, functionState(t, svc, "later").State)
+}
+
+func TestReadiness_RefusesAnotherPinOrAnotherNode(t *testing.T) {
+	im, err := PrepareImage(writeImage(t), testImageDigest)
+	require.NoError(t, err)
+	t.Cleanup(im.Close)
+	d := &Daemon{Node: "node1", Image: im}
+	require.NoError(t, d.ServeControl(t.Context(), controlName(t), os.Getuid()))
+
+	for _, tc := range []struct {
+		name   string
+		pin    string
+		nodes  int
+		reason string
+	}{
+		{"other pin", "0000000000000000000000000000000000000000000000000000000000000000", 1, function.ReasonCreating},
+		{"two nodes", testImageDigest, 2, "The function cannot run here: test-link Lambda activates only on a single-node cluster."},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			svc := readyOwner(t)
+			svc.SetReadiness(d.Readiness(tc.pin, tc.nodes))
+			storedFunction(t, svc, "hello")
+			require.NoError(t, svc.ActivatePending(t.Context()))
+			got := functionState(t, svc, "hello")
+			assert.Equal(t, function.StatePending, got.State)
+			assert.Equal(t, tc.reason, got.StateReason)
+		})
+	}
+}
+
+func TestReadiness_RefusesFunctionsTheTestLinkCannotRun(t *testing.T) {
+	im, err := PrepareImage(writeImage(t), testImageDigest)
+	require.NoError(t, err)
+	t.Cleanup(im.Close)
+	d := &Daemon{Node: "node1", Image: im}
+	require.NoError(t, d.ServeControl(t.Context(), controlName(t), os.Getuid()))
+	svc := readyOwner(t)
+	svc.SetReadiness(d.Readiness(testImageDigest, 1))
+	storedFunctionWith(t, svc, "env", "sealed", 9)
+	storedFunctionWith(t, svc, "huge", "", MaxCodeBytes+1)
+	storedFunctionWith(t, svc, "edge", "", MaxCodeBytes)
+	storedFunctionHandler(t, svc, "spaced", "boot\u00a0strap", "", 9)
+	storedFunctionHandler(t, svc, "longhandler", strings.Repeat("h", 129), "", 9)
+	storedFunctionHandler(t, svc, "edgehandler", strings.Repeat("h", 128), "", 9)
+
+	require.NoError(t, svc.ActivatePending(t.Context()))
+	assert.Equal(t, "The function cannot run here: test-link Lambda does not support environment variables.", functionState(t, svc, "env").StateReason)
+	assert.Equal(t, function.StatePending, functionState(t, svc, "env").State)
+	assert.Equal(t, "The function cannot run here: its code is larger than test-link Lambda accepts.", functionState(t, svc, "huge").StateReason)
+	assert.Equal(t, function.StatePending, functionState(t, svc, "huge").State)
+	assert.Equal(t, function.StateActive, functionState(t, svc, "edge").State, "the largest accepted code copy activates")
+	for _, name := range []string{"spaced", "longhandler"} {
+		assert.Equal(t, function.StatePending, functionState(t, svc, name).State, name)
+		assert.Equal(t, "The function cannot run here: test-link Lambda runs only a handler of 1 to 128 characters with no whitespace.", functionState(t, svc, name).StateReason, name)
+	}
+	assert.Equal(t, function.StateActive, functionState(t, svc, "edgehandler").State, "the longest accepted handler activates")
+}
+
+func TestReadiness_ServingEndsWhenAcceptingStops(t *testing.T) {
+	d := &Daemon{Node: "node1"}
+	name := controlName(t)
+	ln, err := ListenControl(name)
+	require.NoError(t, err)
+	inHandle, release := make(chan struct{}), make(chan struct{})
+	srv := d.controlServer(os.Getuid())
+	srv.Handle = func(context.Context, []byte, Credential) Reply {
+		close(inHandle)
+		<-release
+		return Reply{Outcome: OutcomeDone}
+	}
+	d.serve(t.Context(), ln, srv)
+	require.True(t, d.Serving())
+	go func() { _, _ = (&ControlClient{Name: name, DaemonUID: os.Getuid(), RequestLimit: MaxFrameBytes, ReplyLimit: MaxFrameBytes, IOTimeout: 5 * time.Second}).Exchange(context.Background(), []byte("{}"), testCred, time.Now().Add(5*time.Second)) }()
+	select {
+	case <-inHandle:
+	case <-time.After(5 * time.Second):
+		t.Fatal("the request never reached Handle")
+	}
+	defer close(release)
+
+	require.NoError(t, ln.Close(), "the listener fails under an outstanding invocation")
+	assert.Eventually(t, func() bool { return !d.Serving() }, 2*time.Second, 10*time.Millisecond,
+		"serving ends while the invocation is still held, not after the drain")
+}

spinifex/runtime/daemon/daemon.go

@@ -46,6 +46,7 @@ import (
 	acmdomain "github.com/mulgadc/spinifex/spinifex/domains/acm"
 	"github.com/mulgadc/spinifex/spinifex/domains/dns"
 	ec2account "github.com/mulgadc/spinifex/spinifex/domains/ec2/account"
+	ec2instanceapi "github.com/mulgadc/spinifex/spinifex/domains/ec2/awsapi/instance"
 	ec2eigw "github.com/mulgadc/spinifex/spinifex/domains/ec2/eigw"
 	ec2eip "github.com/mulgadc/spinifex/spinifex/domains/ec2/eip"
 	ec2igw "github.com/mulgadc/spinifex/spinifex/domains/ec2/igw"
@@ -1052,6 +1053,7 @@ func (d *Daemon) subscribeAll() error {
 		{"ec2.terminate", handleNATSRequest(d.node, d.instanceService.TerminateStoppedInstance), "spinifex-workers"},
 		{"ec2.DescribeStoppedInstances", handleNATSRequest(d.node, d.instanceService.DescribeStoppedInstances), "spinifex-workers"},
 		{"ec2.DescribeAccountInstanceRecords", handleNATSRequest(d.node, d.instanceService.DescribeAccountInstanceRecords), "spinifex-workers"},
+		{ec2instanceapi.AccountTeardownOutstandingSubject, handleNATSRequest(d.node, d.instanceService.AccountTeardownOutstandingRPC), "spinifex-workers"},
 		{"ec2.DescribeTerminatedInstances", handleNATSRequest(d.node, d.instanceService.DescribeTerminatedInstances), "spinifex-workers"},
 		// these fan out to all nodes and gateway aggregates the results. The
 		// handler only sees per-daemon local state (vmMgr/stoppedStore), so
@@ -2331,6 +2333,7 @@ func (d *Daemon) startCluster() error {
 	}
 
 	d.elbv2Service.StartLifecycleReaper(context.Background())
+	d.elbv2Service.StartDataPathReconciler(d.ctx, d.node)
 
 	d.eksService, err = initServiceWithRetry("EKS service", func() (*ekscluster.EKSServiceImpl, error) {
 		return ekscluster.NewEKSServiceImpl(d.buildEKSServiceDeps())

spinifex/runtime/daemon/daemon_handlers.go

@@ -81,11 +81,18 @@ func respondWithError(nodeID string, msg *nats.Msg, errCode string) {
 // an opaque ServerInternal, leaving the reason visible only in the daemon log.
 // Mirrors natsmsg.ServeNATSRequestCtx, which has always preserved the message.
 func respondWithServiceError(nodeID string, msg *nats.Msg, err error) {
+	respondWithServiceErrorHeader(nodeID, msg, err, nil)
+}
+
+// respondWithServiceErrorHeader is respondWithServiceError with headers the
+// caller wants on the reply in addition to the ones the error implies.
+func respondWithServiceErrorHeader(nodeID string, msg *nats.Msg, err error, base nats.Header) {
 	_, message, _ := awserrors.ResolveErrorDetail(err)
 	payload := awserrors.GenerateErrorPayloadWithMessage(awserrors.ValidErrorCodeFromError(err), message)
 	// The AWS body stays as it is; the header is how ec2.start tells its
 	// forwarder this refusal must not be answered by starting elsewhere.
 	extra := nats.Header{}
+	maps.Copy(extra, base)
 	if _, ok := errors.AsType[*ec2instance.UnsealedWritesError](err); ok {
 		extra.Set(natsmsg.UnsealedWritesHeader, "true")
 	}
@@ -246,6 +253,10 @@ func (d *Daemon) dispatchEC2Command(msg *nats.Msg) (string, string) {
 		return name, d.handleSetInstanceTags(ctx, msg, command, instance)
 	case command.Attributes.SetInstanceMonitoring:
 		return name, d.handleSetInstanceMonitoring(ctx, msg, command, instance)
+	case command.Attributes.ModifyInstanceMetadataOptions:
+		return name, d.handleModifyInstanceMetadataOptions(ctx, msg, command, instance)
+	case command.Attributes.SetDisableApiTermination:
+		return name, d.handleSetDisableApiTermination(ctx, msg, command, instance)
 	case command.Attributes.StartInstance:
 		opCtx, opSpan := startOpSpan(ctx, "ec2.StartInstance", instance.ID)
 		err := d.instanceService.StartInstance(opCtx, instance, command)
@@ -314,6 +325,10 @@ func ec2CommandName(command ec2v1.EC2InstanceCommand) string {
 		return "RemoveInstanceTags"
 	case command.Attributes.SetInstanceMonitoring:
 		return "SetInstanceMonitoring"
+	case command.Attributes.ModifyInstanceMetadataOptions:
+		return "ModifyInstanceMetadataOptions"
+	case command.Attributes.SetDisableApiTermination:
+		return "SetDisableApiTermination"
 	case command.Attributes.StartInstance:
 		return "StartInstance"
 	case command.Attributes.RebootInstance:

spinifex/runtime/daemon/daemon_handlers_instance.go

@@ -360,3 +360,76 @@ func relayStartReply(msg, resp *nats.Msg) error {
 	maps.Copy(reply.Header, resp.Header)
 	return msg.RespondMsg(reply)
 }
+
+// handledHeader marks a reply as sent by the instance's owner, so the
+// forwarder can tell an owner's answer from a stale or missing responder.
+func handledHeader() nats.Header {
+	h := nats.Header{}
+	h.Set(ec2v1.HandledHeader, "1")
+	return h
+}
+
+// respondHandledError answers with err and the handled marker.
+func respondHandledError(nodeID string, msg *nats.Msg, err error) string {
+	respondWithServiceErrorHeader(nodeID, msg, err, handledHeader())
+	return outcomeForError(err)
+}
+
+// respondHandledJSON answers with a JSON body and the handled marker.
+func respondHandledJSON(nodeID string, msg *nats.Msg, out any) string {
+	body, err := json.Marshal(out)
+	if err != nil {
+		return respondHandledError(nodeID, msg, errors.New(awserrors.ErrorServerInternal))
+	}
+	respondNATSMsgWithHeader(nodeID, msg, body, handledHeader())
+	return outcomeSuccess
+}
+
+// handleModifyInstanceMetadataOptions applies a metadata-options change on the
+// node that runs the instance, through the same service method as the
+// gateway-local path.
+func (d *Daemon) handleModifyInstanceMetadataOptions(ctx context.Context, msg *nats.Msg, command ec2v1.EC2InstanceCommand, instance *vm.VM) string {
+	data := command.InstanceMetadataOptionsData
+	if data == nil {
+		return respondHandledError(d.node, msg, errors.New(awserrors.ErrorMissingParameter))
+	}
+	input := &ec2.ModifyInstanceMetadataOptionsInput{
+		InstanceId:              aws.String(instance.ID),
+		HttpPutResponseHopLimit: data.HttpPutResponseHopLimit,
+	}
+	if data.HttpTokens != "" {
+		input.HttpTokens = aws.String(data.HttpTokens)
+	}
+	if data.HttpEndpoint != "" {
+		input.HttpEndpoint = aws.String(data.HttpEndpoint)
+	}
+	if data.HttpProtocolIpv6 != "" {
+		input.HttpProtocolIpv6 = aws.String(data.HttpProtocolIpv6)
+	}
+	if data.InstanceMetadataTags != "" {
+		input.InstanceMetadataTags = aws.String(data.InstanceMetadataTags)
+	}
+	out, err := d.instanceService.ModifyInstanceMetadataOptions(ctx, input, natsmsg.AccountIDFromMsg(msg))
+	if err != nil {
+		return respondHandledError(d.node, msg, err)
+	}
+	return respondHandledJSON(d.node, msg, out)
+}
+
+// handleSetDisableApiTermination sets termination protection on the node that
+// runs the instance. Absent data is a validation error, never "false".
+func (d *Daemon) handleSetDisableApiTermination(ctx context.Context, msg *nats.Msg, command ec2v1.EC2InstanceCommand, instance *vm.VM) string {
+	data := command.DisableApiTerminationData
+	if data == nil {
+		return respondHandledError(d.node, msg, errors.New(awserrors.ErrorMissingParameter))
+	}
+	input := &ec2.ModifyInstanceAttributeInput{
+		InstanceId:            aws.String(instance.ID),
+		DisableApiTermination: &ec2.AttributeBooleanValue{Value: aws.Bool(data.Value)},
+	}
+	out, err := d.instanceService.ModifyInstanceAttribute(ctx, input, natsmsg.AccountIDFromMsg(msg))
+	if err != nil {
+		return respondHandledError(d.node, msg, err)
+	}
+	return respondHandledJSON(d.node, msg, out)
+}

spinifex/runtime/daemon/daemon_handlers_owner_routing_test.go

@@ -0,0 +1,323 @@
+package daemon
+
+import (
+	"bytes"
+	"encoding/json"
+	"log/slog"
+	"reflect"
+	"strings"
+	"testing"
+
+	"github.com/aws/aws-sdk-go/aws"
+	"github.com/aws/aws-sdk-go/service/ec2"
+	"github.com/mulgadc/spinifex/contracts/ec2/v1"
+	"github.com/mulgadc/spinifex/internal/testkit"
+	ec2instanceapi "github.com/mulgadc/spinifex/spinifex/domains/ec2/awsapi/instance"
+	ec2instance "github.com/mulgadc/spinifex/spinifex/domains/ec2/instance"
+	"github.com/mulgadc/spinifex/spinifex/foundation/aws/errors"
+	"github.com/mulgadc/spinifex/spinifex/providers/objectstore"
+	"github.com/mulgadc/spinifex/spinifex/runtime/compute/vm"
+	vmmock "github.com/mulgadc/spinifex/spinifex/runtime/compute/vm/mock"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func metadataCommand(id string, data *ec2v1.InstanceMetadataOptionsData) ec2v1.EC2InstanceCommand {
+	return ec2v1.EC2InstanceCommand{
+		ID:                          id,
+		Attributes:                  ec2v1.EC2CommandAttributes{ModifyInstanceMetadataOptions: true},
+		InstanceMetadataOptionsData: data,
+	}
+}
+
+func terminationCommand(id string, data *ec2v1.DisableApiTerminationData) ec2v1.EC2InstanceCommand {
+	return ec2v1.EC2InstanceCommand{
+		ID:                        id,
+		Attributes:                ec2v1.EC2CommandAttributes{SetDisableApiTermination: true},
+		DisableApiTerminationData: data,
+	}
+}
+
+func TestOwnerDispatch_ModifyInstanceMetadataOptions(t *testing.T) {
+	t.Run("success applies the change and sets the handled header", func(t *testing.T) {
+		const id = "i-or-meta-ok"
+		d, _ := monitoringTestDaemon(t, id, false)
+		cmd := metadataCommand(id, &ec2v1.InstanceMetadataOptionsData{
+			HttpTokens:              ec2.HttpTokensStateOptional,
+			HttpPutResponseHopLimit: aws.Int64(3),
+		})
+
+		reply := requestHandler(t, d.natsConn, "ec2.cmd."+id, d.handleEC2Events, testAccountID, mustMarshal(t, cmd))
+		assert.Equal(t, "1", reply.Header.Get(ec2v1.HandledHeader))
+
+		got, ok := d.vmMgr.Get(id)
+		require.True(t, ok)
+		require.NotNil(t, got.Instance.MetadataOptions)
+		assert.Equal(t, ec2.HttpTokensStateOptional, aws.StringValue(got.Instance.MetadataOptions.HttpTokens))
+		assert.Equal(t, int64(3), aws.Int64Value(got.Instance.MetadataOptions.HttpPutResponseHopLimit))
+	})
+
+	t.Run("service error still sets the handled header", func(t *testing.T) {
+		const id = "i-or-meta-bad"
+		d, _ := monitoringTestDaemon(t, id, false)
+		cmd := metadataCommand(id, &ec2v1.InstanceMetadataOptionsData{HttpTokens: "bogus"})
+
+		reply := requestHandler(t, d.natsConn, "ec2.cmd."+id, d.handleEC2Events, testAccountID, mustMarshal(t, cmd))
+		assert.Equal(t, "1", reply.Header.Get(ec2v1.HandledHeader))
+		assert.Equal(t, awserrors.ErrorUnsupportedOperation, decodeError(t, reply.Data)["Code"])
+	})
+
+	t.Run("missing data is a validation error with the handled header", func(t *testing.T) {
+		const id = "i-or-meta-nodata"
+		d, _ := monitoringTestDaemon(t, id, false)
+
+		reply := requestHandler(t, d.natsConn, "ec2.cmd."+id, d.handleEC2Events, testAccountID, mustMarshal(t, metadataCommand(id, nil)))
+		assert.Equal(t, "1", reply.Header.Get(ec2v1.HandledHeader))
+		assert.Equal(t, awserrors.ErrorMissingParameter, decodeError(t, reply.Data)["Code"])
+	})
+}
+
+func TestOwnerDispatch_SetDisableApiTermination(t *testing.T) {
+	protection := func(t *testing.T, d *Daemon, id string) *bool {
+		t.Helper()
+		got, ok := d.vmMgr.Get(id)
+		require.True(t, ok)
+		return got.RunInstancesInput.DisableApiTermination
+	}
+
+	t.Run("true then explicit false are both applied", func(t *testing.T) {
+		const id = "i-or-term-ok"
+		d, _ := monitoringTestDaemon(t, id, false)
+
+		reply := requestHandler(t, d.natsConn, "ec2.cmd."+id, d.handleEC2Events, testAccountID,
+			mustMarshal(t, terminationCommand(id, &ec2v1.DisableApiTerminationData{Value: true})))
+		assert.Equal(t, "1", reply.Header.Get(ec2v1.HandledHeader))
+		require.NotNil(t, protection(t, d, id))
+		assert.True(t, *protection(t, d, id))
+
+		reply = requestHandler(t, d.natsConn, "ec2.cmd."+id, d.handleEC2Events, testAccountID,
+			mustMarshal(t, terminationCommand(id, &ec2v1.DisableApiTerminationData{Value: false})))
+		assert.Equal(t, "1", reply.Header.Get(ec2v1.HandledHeader))
+		require.NotNil(t, protection(t, d, id))
+		assert.False(t, *protection(t, d, id))
+	})
+
+	t.Run("missing data is an error and never clears protection", func(t *testing.T) {
+		const id = "i-or-term-nodata"
+		d, _ := monitoringTestDaemon(t, id, false)
+		d.vmMgr.UpdateState(id, func(v *vm.VM) { v.RunInstancesInput.DisableApiTermination = aws.Bool(true) })
+
+		reply := requestHandler(t, d.natsConn, "ec2.cmd."+id, d.handleEC2Events, testAccountID, mustMarshal(t, terminationCommand(id, nil)))
+		assert.Equal(t, "1", reply.Header.Get(ec2v1.HandledHeader))
+		assert.Equal(t, awserrors.ErrorMissingParameter, decodeError(t, reply.Data)["Code"])
+		require.NotNil(t, protection(t, d, id))
+		assert.True(t, *protection(t, d, id))
+	})
+}
+
+// ownerRoutingDaemon is a daemon on url with its own vmMgr and the shared
+// stopped store.
+func ownerRoutingDaemon(t *testing.T, url string, stopped *vmmock.StateStore) *Daemon {
+	t.Helper()
+	d := createTestDaemon(t, url)
+	d.instanceService = ec2instance.NewInstanceServiceImpl(
+		d.config, d.resourceMgr.instanceTypes, d.natsConn,
+		objectstore.NewMemoryObjectStore(), d.vmMgr, d.resourceMgr, stopped)
+	return d
+}
+
+func TestModifyInstanceMetadataOptions_NonOwnerReceivesQueueOwnerAnswers(t *testing.T) {
+	const id = "i-or-multinode"
+	ns, gateway := testutil.StartTestNATS(t)
+	stopped := vmmock.New()
+
+	// B runs the instance; A does not and is the only queue-group member.
+	b := ownerRoutingDaemon(t, ns.ClientURL(), stopped)
+	b.vmMgr.Insert(&vm.VM{
+		ID:        id,
+		Status:    vm.StateRunning,
+		AccountID: testAccountID,
+		Instance:  &ec2.Instance{InstanceId: aws.String(id)},
+	})
+	subscribeOwner(t, b, id)
+
+	a := ownerRoutingDaemon(t, ns.ClientURL(), stopped)
+	sub, err := a.natsConn.QueueSubscribe("ec2.ModifyInstanceMetadataOptions", "spinifex-workers",
+		asMsgHandler(handleNATSRequest(a.node, a.instanceService.ModifyInstanceMetadataOptions)))
+	require.NoError(t, err)
+	t.Cleanup(func() { _ = sub.Unsubscribe() })
+	require.NoError(t, a.natsConn.Flush())
+
+	for range 9 {
+		out, err := ec2instanceapi.ModifyInstanceMetadataOptions(t.Context(), &ec2.ModifyInstanceMetadataOptionsInput{
+			InstanceId:              aws.String(id),
+			HttpPutResponseHopLimit: aws.Int64(3),
+		}, gateway, testAccountID)
+		require.NoError(t, err)
+		assert.Equal(t, id, aws.StringValue(out.InstanceId))
+		assert.Equal(t, int64(3), aws.Int64Value(out.InstanceMetadataOptions.HttpPutResponseHopLimit))
+	}
+
+	got, ok := b.vmMgr.Get(id)
+	require.True(t, ok)
+	assert.Equal(t, int64(3), aws.Int64Value(got.Instance.MetadataOptions.HttpPutResponseHopLimit))
+	_, onA := a.vmMgr.Get(id)
+	assert.False(t, onA, "the queue-group member never held the instance")
+}
+
+// n1CommandType is EC2InstanceCommand as the previous release defined it: the
+// current type minus the two attributes and two data blocks this change adds.
+func n1CommandType() reflect.Type {
+	newFields := map[string]bool{
+		"ModifyInstanceMetadataOptions": true, "SetDisableApiTermination": true,
+		"InstanceMetadataOptionsData": true, "DisableApiTerminationData": true,
+	}
+	strip := func(t reflect.Type, swap map[string]reflect.Type) reflect.Type {
+		var fields []reflect.StructField
+		for f := range t.Fields() {
+			if newFields[f.Name] {
+				continue
+			}
+			if st, ok := swap[f.Name]; ok {
+				f.Type = st
+			}
+			fields = append(fields, f)
+		}
+		return reflect.StructOf(fields)
+	}
+	attrs := strip(reflect.TypeFor[ec2v1.EC2CommandAttributes](), nil)
+	return strip(reflect.TypeFor[ec2v1.EC2InstanceCommand](), map[string]reflect.Type{"Attributes": attrs})
+}
+
+// An N-1 owner decodes a new-variant payload with the unknown attribute and
+// data dropped, so dispatch sees no attribute and takes the default branch.
+func TestS9_N1DispatchRejectsNewVariantsWithoutWriting(t *testing.T) {
+	var logs bytes.Buffer
+	prev := slog.Default()
+	slog.SetDefault(slog.New(slog.NewTextHandler(&logs, &slog.HandlerOptions{Level: slog.LevelWarn})))
+	t.Cleanup(func() { slog.SetDefault(prev) })
+
+	variants := map[string]func(id string) ec2v1.EC2InstanceCommand{
+		"metadata options": func(id string) ec2v1.EC2InstanceCommand {
+			return metadataCommand(id, &ec2v1.InstanceMetadataOptionsData{HttpPutResponseHopLimit: aws.Int64(3)})
+		},
+		"termination protection": func(id string) ec2v1.EC2InstanceCommand {
+			return terminationCommand(id, &ec2v1.DisableApiTerminationData{Value: true})
+		},
+	}
+	for name, build := range variants {
+		t.Run(name, func(t *testing.T) {
+			id := "i-or-n1-" + strings.ReplaceAll(name, " ", "-")
+			d, _ := monitoringTestDaemon(t, id, false)
+			before, _ := d.vmMgr.Get(id)
+			beforeInstance := *before.Instance
+			logs.Reset()
+
+			// Current encoder -> N-1 decoder -> what the N-1 owner re-dispatches.
+			n1 := reflect.New(n1CommandType())
+			require.NoError(t, json.Unmarshal(mustMarshal(t, build(id)), n1.Interface()))
+			require.Equal(t, id, n1.Elem().FieldByName("ID").String())
+			payload := mustMarshal(t, n1.Interface())
+			assert.NotContains(t, string(payload), "modify_instance_metadata_options")
+			assert.NotContains(t, string(payload), "set_disable_api_termination")
+
+			reply := requestHandler(t, d.natsConn, "ec2.cmd."+id, d.handleEC2Events, testAccountID, payload)
+			assert.Equal(t, awserrors.ErrorServerInternal, decodeError(t, reply.Data)["Code"])
+			assert.Empty(t, reply.Header.Get(ec2v1.HandledHeader), "an N-1 rejection carries no handled marker")
+			assert.Contains(t, logs.String(), "Unhandled EC2 instance command")
+
+			after, _ := d.vmMgr.Get(id)
+			assert.Equal(t, beforeInstance, *after.Instance)
+			assert.Nil(t, after.RunInstancesInput.DisableApiTermination)
+		})
+	}
+}
+
+func TestModifyInstanceMetadataOptions_NoOwnerAppliesToRealStoppedRecord(t *testing.T) {
+	const id = "i-or-stopped-real"
+	ns, gateway := testutil.StartTestNATS(t)
+	stopped := vmmock.New()
+	stopped.Stopped[id] = stoppedVMFixture(id, testAccountID)
+
+	// No daemon runs the instance; the only queue-group member is the real service.
+	a := ownerRoutingDaemon(t, ns.ClientURL(), stopped)
+	sub, err := a.natsConn.QueueSubscribe("ec2.ModifyInstanceMetadataOptions", "spinifex-workers",
+		asMsgHandler(handleNATSRequest(a.node, a.instanceService.ModifyInstanceMetadataOptions)))
+	require.NoError(t, err)
+	t.Cleanup(func() { _ = sub.Unsubscribe() })
+	require.NoError(t, a.natsConn.Flush())
+
+	out, err := ec2instanceapi.ModifyInstanceMetadataOptions(t.Context(), &ec2.ModifyInstanceMetadataOptionsInput{
+		InstanceId:              aws.String(id),
+		HttpPutResponseHopLimit: aws.Int64(2),
+	}, gateway, testAccountID)
+	require.NoError(t, err)
+	assert.Equal(t, int64(2), aws.Int64Value(out.InstanceMetadataOptions.HttpPutResponseHopLimit))
+
+	reloaded, err := stopped.LoadStoppedInstance(id)
+	require.NoError(t, err)
+	require.NotNil(t, reloaded.Instance.MetadataOptions)
+	assert.Equal(t, int64(2), aws.Int64Value(reloaded.Instance.MetadataOptions.HttpPutResponseHopLimit))
+	require.NotNil(t, stopped.WroteStopped[id], "the service persisted the stopped record")
+}
+
+func TestOwnerDispatch_RejectsOtherAccountWithoutWriting(t *testing.T) {
+	const id = "i-or-otheracct"
+	d, _ := monitoringTestDaemon(t, id, false)
+	before, _ := d.vmMgr.Get(id)
+	beforeInstance := *before.Instance
+
+	cmds := map[string]ec2v1.EC2InstanceCommand{
+		"metadata options": metadataCommand(id, &ec2v1.InstanceMetadataOptionsData{HttpPutResponseHopLimit: aws.Int64(3)}),
+		"termination":      terminationCommand(id, &ec2v1.DisableApiTerminationData{Value: true}),
+	}
+	for name, cmd := range cmds {
+		reply := requestHandler(t, d.natsConn, "ec2.cmd."+id, d.handleEC2Events, "999999999999", mustMarshal(t, cmd))
+		assert.Equal(t, awserrors.ErrorInvalidInstanceIDNotFound, decodeError(t, reply.Data)["Code"], name)
+	}
+
+	after, _ := d.vmMgr.Get(id)
+	assert.Equal(t, beforeInstance, *after.Instance)
+	assert.Nil(t, after.RunInstancesInput.DisableApiTermination)
+}
+
+func TestModifyInstanceAttribute_DisableApiTerminationRoutedToOwner(t *testing.T) {
+	const id = "i-or-termroute"
+	ns, gateway := testutil.StartTestNATS(t)
+	stopped := vmmock.New()
+
+	b := ownerRoutingDaemon(t, ns.ClientURL(), stopped)
+	b.vmMgr.Insert(&vm.VM{
+		ID:                id,
+		Status:            vm.StateRunning,
+		AccountID:         testAccountID,
+		Instance:          &ec2.Instance{InstanceId: aws.String(id)},
+		RunInstancesInput: &ec2.RunInstancesInput{},
+	})
+	subscribeOwner(t, b, id)
+
+	a := ownerRoutingDaemon(t, ns.ClientURL(), stopped)
+	sub, err := a.natsConn.QueueSubscribe("ec2.ModifyInstanceAttribute", "spinifex-workers",
+		asMsgHandler(handleNATSRequest(a.node, a.instanceService.ModifyInstanceAttribute)))
+	require.NoError(t, err)
+	t.Cleanup(func() { _ = sub.Unsubscribe() })
+	require.NoError(t, a.natsConn.Flush())
+
+	for range 9 {
+		_, err := ec2instanceapi.ModifyInstanceAttribute(t.Context(), &ec2.ModifyInstanceAttributeInput{
+			InstanceId:            aws.String(id),
+			DisableApiTermination: &ec2.AttributeBooleanValue{Value: aws.Bool(true)},
+		}, gateway, testAccountID)
+		require.NoError(t, err)
+	}
+
+	got, ok := b.vmMgr.Get(id)
+	require.True(t, ok)
+	require.NotNil(t, got.RunInstancesInput.DisableApiTermination)
+	assert.True(t, *got.RunInstancesInput.DisableApiTermination)
+
+	// The node that enforces the guard now refuses the terminate.
+	reply := requestHandler(t, b.natsConn, "ec2.cmd."+id+".enforce", b.handleEC2Events, testAccountID,
+		mustMarshal(t, ec2v1.EC2InstanceCommand{ID: id, Attributes: ec2v1.EC2CommandAttributes{TerminateInstance: true}}))
+	assert.Equal(t, awserrors.ErrorOperationNotPermitted, decodeError(t, reply.Data)["Code"])
+}

spinifex/runtime/daemon/daemon_lambda_testlink.go

@@ -3,14 +3,18 @@
 package daemon
 
 import (
+	"crypto/sha256"
 	"crypto/tls"
 	"crypto/x509"
+	"encoding/hex"
 	"errors"
 	"fmt"
+	"io"
 	"log/slog"
 	"net/http"
 	"os"
 	"path/filepath"
+	"time"
 
 	"github.com/mulgadc/spinifex/spinifex/domains/lambda/execution"
 	"github.com/mulgadc/spinifex/spinifex/domains/network/host"
@@ -27,6 +31,28 @@ var lambdaHardKillSHA256 string
 // image bytes and does not authenticate where they came from.
 var lambdaImageDigest string
 
+// lambdaFirecrackerSHA256 pins the Firecracker v1.17.0 x86_64 release binary.
+// A node whose test-link image directory holds a firecracker binary boots
+// every attempt under it, and does not serve unless the binary matches.
+const lambdaFirecrackerSHA256 = "99ad0f5cd0514a88aad0e9ae8cfdb3cc3b4ab9d190e1194602406c786b5de7a5"
+
+// checkFirecracker refuses a Firecracker binary other than the pinned release.
+func checkFirecracker(path string) error {
+	f, err := os.Open(path)
+	if err != nil {
+		return fmt.Errorf("firecracker binary: %w", err)
+	}
+	defer func() { _ = f.Close() }()
+	h := sha256.New()
+	if _, err := io.Copy(h, f); err != nil {
+		return fmt.Errorf("firecracker binary: %w", err)
+	}
+	if got := hex.EncodeToString(h.Sum(nil)); got != lambdaFirecrackerSHA256 {
+		return errors.New("firecracker binary does not match the pinned release")
+	}
+	return nil
+}
+
 // startLambdaTestLink serves controlled-test Lambda invocations for this node.
 // Recovery needs only the owner store, taps and addresses, so it starts even
 // when the bridge address, initializer, pin or download trust would not serve.
@@ -70,9 +96,20 @@ func (d *Daemon) startLambdaTestLink() {
 		InitializerSHA256: lambdaHardKillSHA256,
 		HTTP:              httpClient,
 	}
+	fc := filepath.Join(imageDir, "firecracker")
+	if _, err := os.Lstat(fc); err == nil {
+		tl.Firecracker = fc
+	}
 	err = runLambdaTestLink(
 		func() { go tl.RecoverLoop(d.ctx, testlink.RecoverFirst, testlink.RecoverMax) },
-		func() error { return lambdaTestLinkReady(tl, imageDir, lambdaImageDigest, trustErr) },
+		func() error {
+			if tl.Firecracker != "" {
+				if err := checkFirecracker(tl.Firecracker); err != nil {
+					return err
+				}
+			}
+			return lambdaTestLinkReady(tl, imageDir, lambdaImageDigest, trustErr)
+		},
 		func() error {
 			gatewayUID, err := testlink.LookupUID(testlink.GatewayUser)
 			if err != nil {
@@ -88,8 +125,17 @@ func (d *Daemon) startLambdaTestLink() {
 	}
 	slog.Warn("Lambda test link serving: test-only, not for production; the guest does not verify owner grant authority",
 		"node", d.node, "image_digest", tl.Image.Digest)
+	if d.lambdaMetaService != nil && d.clusterConfig != nil {
+		d.lambdaMetaService.SetReadiness(tl.Readiness(lambdaImageDigest, len(d.clusterConfig.Nodes)))
+		go d.lambdaMetaService.RunActivation(d.ctx, lambdaActivationInterval)
+	}
 }
 
+// lambdaActivationInterval bounds how long a function whose activation was
+// missed, by a crash or an earlier refusal, waits for the next pass.
+const lambdaActivationInterval = 30 * time.Second
+
+
 // lambdaTestLinkReady runs the serving checks and, only once they pass,
 // prepares the pinned image into tl; a refusal leaves tl.Image nil.
 func lambdaTestLinkReady(tl *testlink.Daemon, imageDir, imagePin string, trustErr error) error {

spinifex/runtime/daemon/daemon_mgmt_ip.go

@@ -154,8 +154,17 @@ func (a *MgmtIPAllocator) BindKV(jsManager *JetStreamManager, node string) {
 // an address is free cluster-wide on its own — and is refused otherwise.
 func (a *MgmtIPAllocator) Allocate(instanceID string) (string, error) {
 	a.mu.Lock()
+	// Addresses this node's own instances hold stay off the table even when the
+	// cluster record has lost their entries.
+	held := make([]MgmtIPEntry, 0, len(a.allocated))
+	for id, ip := range a.allocated {
+		held = append(held, MgmtIPEntry{IP: ip, InstanceID: id})
+	}
 	if ip, ok := a.allocated[instanceID]; ok {
 		a.mu.Unlock()
+		if other := heldByOther(held, ip, instanceID); other != "" {
+			return "", fmt.Errorf("mgmt IP %s for %s is also held locally by %s", ip, instanceID, other)
+		}
 		return ip, nil
 	}
 	jsManager, node, subnet := a.jsManager, a.node, a.subnet
@@ -171,29 +180,7 @@ func (a *MgmtIPAllocator) Allocate(instanceID string) (string, error) {
 	// CAS write that follows is a harmless no-op against the same revision.
 	var allocatedIP string
 	var mutateErr error
-	_, err := updateMgmtIPAMWithRetry(jsManager, subnet, func(r *MgmtIPRecord) {
-		if r.Subnet == "" {
-			r.Subnet = subnet
-		}
-
-		// Idempotent: an instance that already holds an address (allocated
-		// by this node earlier, or by another node before a restart lost
-		// the local cache) keeps it rather than taking a second one.
-		for _, e := range r.Allocated {
-			if e.InstanceID == instanceID {
-				allocatedIP = e.IP
-				return
-			}
-		}
-
-		ip, findErr := nextFreeIP(a.baseIP, a.rangeMin, a.rangeMax, r.Allocated)
-		if findErr != nil {
-			mutateErr = findErr
-			return
-		}
-		r.Allocated = append(r.Allocated, MgmtIPEntry{IP: ip, InstanceID: instanceID, Node: node})
-		allocatedIP = ip
-	}, true)
+	_, err := updateMgmtIPAMWithRetry(jsManager, subnet, a.allocMutate(instanceID, node, subnet, held, &allocatedIP, &mutateErr), true)
 	if err != nil {
 		return "", fmt.Errorf("mgmt IP CAS update for subnet %s: %w", subnet, err)
 	}
@@ -209,6 +196,16 @@ func (a *MgmtIPAllocator) Allocate(instanceID string) (string, error) {
 	return allocatedIP, nil
 }
 
+// heldByOther names a cached instance other than id holding ip, or "".
+func heldByOther(held []MgmtIPEntry, ip, id string) string {
+	for _, e := range held {
+		if e.IP == ip && e.InstanceID != id {
+			return e.InstanceID
+		}
+	}
+	return ""
+}
+
 // nextFreeIP scans rangeMin-rangeMax on the /24 rooted at baseIP, skipping
 // every IP already present in allocated regardless of which entry's Node it
 // belongs to — the range is shared
@@ -511,3 +508,37 @@ func (d *Daemon) primeMgmtNeighEntry(instanceID, mgmtIP, mgmtMAC string) {
 			"instanceId", instanceID, "mgmtIP", mgmtIP, "mgmtMAC", mgmtMAC, "bridge", bridge, "err", err)
 	}
 }
+
+// allocMutate returns the CAS callback for Allocate. Results go out through
+// ipOut and errOut because casUpdate's mutate has no return value.
+func (a *MgmtIPAllocator) allocMutate(instanceID, node, subnet string, held []MgmtIPEntry, ipOut *string, errOut *error) func(*MgmtIPRecord) {
+	return func(r *MgmtIPRecord) {
+		// Each CAS attempt starts clean, so a retry never returns an earlier attempt's result.
+		*ipOut, *errOut = "", nil
+		if r.Subnet == "" {
+			r.Subnet = subnet
+		}
+
+		// Idempotent: an instance that already holds an address (allocated
+		// by this node earlier, or by another node before a restart lost
+		// the local cache) keeps it rather than taking a second one.
+		for _, e := range r.Allocated {
+			if e.InstanceID == instanceID {
+				if other := heldByOther(held, e.IP, instanceID); other != "" {
+					*errOut = fmt.Errorf("mgmt IP %s recorded for %s is held locally by %s", e.IP, instanceID, other)
+					return
+				}
+				*ipOut = e.IP
+				return
+			}
+		}
+
+		ip, findErr := nextFreeIP(a.baseIP, a.rangeMin, a.rangeMax, append(held, r.Allocated...))
+		if findErr != nil {
+			*errOut = findErr
+			return
+		}
+		r.Allocated = append(r.Allocated, MgmtIPEntry{IP: ip, InstanceID: instanceID, Node: node})
+		*ipOut = ip
+	}
+}

spinifex/runtime/daemon/daemon_mgmt_ip_test.go

@@ -472,6 +472,76 @@ func TestMgmtIPAllocator_Allocate_IdempotentAcrossNodes(t *testing.T) {
 	}
 }
 
+// An instance whose cluster-record entry has gone (the record lost it while the
+// instance kept running) still holds its address on this node, so a new
+// allocation must not be handed that address.
+func TestMgmtIPAllocator_Allocate_SkipsLocallyHeldWhenRecordLost(t *testing.T) {
+	t.Parallel()
+	jsm := newTestMgmtJSM(t)
+
+	a, err := NewMgmtIPAllocator("10.41.8.1")
+	if err != nil {
+		t.Fatal(err)
+	}
+	a.BindKV(jsm, "node-1")
+	cleanupMgmtIPAM(t, jsm, a)
+
+	held, err := a.Allocate("i-system")
+	if err != nil {
+		t.Fatal(err)
+	}
+	if err := jsm.clusterKV.Delete(context.Background(), mgmtIPAMKeyPrefix+a.subnet); err != nil {
+		t.Fatalf("drop record: %v", err)
+	}
+
+	got, err := a.Allocate("lambda-attempt")
+	if err != nil {
+		t.Fatal(err)
+	}
+	if got == held {
+		t.Fatalf("new allocation reused %s, still held by i-system", got)
+	}
+	if again, err := a.Allocate("i-system"); err != nil || again != held {
+		t.Errorf("holder re-allocate = %q, %v; want %q", again, err, held)
+	}
+}
+
+// Allocate refuses a recorded entry whose address another local instance holds.
+func TestMgmtIPAllocator_Allocate_RefusesRecordedConflictWithLocalHolder(t *testing.T) {
+	t.Parallel()
+	jsm := newTestMgmtJSM(t)
+
+	a, err := NewMgmtIPAllocator("10.42.8.1")
+	if err != nil {
+		t.Fatal(err)
+	}
+	a.BindKV(jsm, "node-1")
+	cleanupMgmtIPAM(t, jsm, a)
+
+	held, err := a.Allocate("i-system")
+	if err != nil {
+		t.Fatal(err)
+	}
+	other, err := NewMgmtIPAllocator("10.42.8.2")
+	if err != nil {
+		t.Fatal(err)
+	}
+	other.BindKV(jsm, "node-2")
+	if err := jsm.clusterKV.Delete(context.Background(), mgmtIPAMKeyPrefix+a.subnet); err != nil {
+		t.Fatalf("drop record: %v", err)
+	}
+	// The record now names the held address for a different instance.
+	other.Claim("lambda-attempt", held)
+
+	if got, err := a.Allocate("lambda-attempt"); err == nil {
+		t.Fatalf("Allocate returned %s, which i-system holds locally", got)
+	}
+	a.Claim("i-dup", held)
+	if got, err := a.Allocate("i-dup"); err == nil {
+		t.Fatalf("cached duplicate returned %s, which i-system holds locally", got)
+	}
+}
+
 // TestMgmtIPAllocator_Release_FreesAddressClusterWide proves Release's
 // effect is visible cluster-wide: a second allocator, not the one that
 // released the address, can immediately take it.
@@ -868,8 +938,9 @@ func TestMgmtIPAllocator_ClaimLetsTheNewHolderReleaseItLater(t *testing.T) {
 		t.Errorf("the holder could not release its own address: entry still %+v", entry)
 	}
 
-	// And the address is genuinely back, not merely unrecorded.
-	reused, err := node1.Allocate("i-next")
+	// And the address is genuinely back, not merely unrecorded. node-1's cache
+	// still names its dropped copy, so the holder that released it allocates.
+	reused, err := node2.Allocate("i-next")
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -1110,3 +1181,34 @@ func TestMgmtIPAllocator_ReleaseDropsLocalOnForeignHolderAndKVFailure(t *testing
 		t.Errorf("local entries after release with KV down = %d, want 0", node3.AllocatedCount())
 	}
 }
+
+// A CAS retry that sees the conflicting entry gone must report its own
+// result, not the refusal left by the earlier attempt.
+func TestMgmtIPAllocator_AllocMutate_RetryClearsEarlierConflict(t *testing.T) {
+	a, err := NewMgmtIPAllocator("10.30.8.1")
+	if err != nil {
+		t.Fatal(err)
+	}
+	held := []MgmtIPEntry{{IP: "10.30.8.10", InstanceID: "i-sqs"}}
+	var ip string
+	var mutateErr error
+	mutate := a.allocMutate("i-fn", "node1", "10.30.8.0/24", held, &ip, &mutateErr)
+
+	mutate(&MgmtIPRecord{Allocated: []MgmtIPEntry{{IP: "10.30.8.10", InstanceID: "i-fn", Node: "node1"}}})
+	if mutateErr == nil || ip != "" {
+		t.Fatalf("conflicted attempt: ip=%q err=%v, want refusal", ip, mutateErr)
+	}
+
+	clean := &MgmtIPRecord{}
+	mutate(clean)
+	if mutateErr != nil {
+		t.Fatalf("retry kept the earlier refusal: %v", mutateErr)
+	}
+	if ip == "" || ip == "10.30.8.10" {
+		t.Fatalf("retry allocated %q, want a free address other than the held one", ip)
+	}
+	want := MgmtIPEntry{IP: ip, InstanceID: "i-fn", Node: "node1"}
+	if len(clean.Allocated) != 1 || clean.Allocated[0] != want {
+		t.Fatalf("record = %+v, want [%+v]", clean.Allocated, want)
+	}
+}

spinifex/runtime/daemon/daemon_system_dispatch.go

@@ -97,6 +97,14 @@ func (d *Daemon) serveSystemLaunch(msg *nats.Msg, launch systemLaunchFunc) strin
 			"instanceId", output.InstanceID, "err", subErr)
 	}
 
+	d.mu.Lock()
+	rebootErr := d.subscribeSystemRebootLocked(output.InstanceID)
+	d.mu.Unlock()
+	if rebootErr != nil {
+		slog.Error("system.LaunchInstance: failed to subscribe reboot subject",
+			"instanceId", output.InstanceID, "err", rebootErr)
+	}
+
 	respondWithSystemLaunchOutput(msg, output)
 	return outcomeSuccess
 }
@@ -175,21 +183,121 @@ func (d *Daemon) terminateSystemInstanceRemote(instanceID string) error {
 		return fmt.Errorf("decode routed terminate reply %s: %w", instanceID, err)
 	}
 	if env.Error != "" {
-		if strings.Contains(env.Error, systeminstance.ErrSystemInstanceNotFound.Error()) {
-			return fmt.Errorf("%w: %s", systeminstance.ErrSystemInstanceNotFound, instanceID)
-		}
-		if strings.Contains(env.Error, systeminstance.ErrSystemInstanceNotSystem.Error()) {
-			return fmt.Errorf("%w: %s", systeminstance.ErrSystemInstanceNotSystem, instanceID)
-		}
-		if strings.Contains(env.Error, systeminstance.ErrSystemInstanceOwnershipUnknown.Error()) {
-			return fmt.Errorf("%w: %s", systeminstance.ErrSystemInstanceOwnershipUnknown, instanceID)
-		}
-		return fmt.Errorf("routed terminate %s: %s", instanceID, env.Error)
+		return routedSystemError("terminate", instanceID, env.Error)
 	}
 	slog.Info("terminateSystemInstanceRemote: owner terminated VM", "instanceId", instanceID)
 	return nil
 }
 
+// routedSystemError maps an owner's error text back to the sentinel it carried,
+// so callers keep telling gone, customer-owned and unreadable apart.
+func routedSystemError(op, instanceID, text string) error {
+	for _, sentinel := range []error{
+		systeminstance.ErrSystemInstanceNotFound,
+		systeminstance.ErrSystemInstanceNotSystem,
+		systeminstance.ErrSystemInstanceOwnershipUnknown,
+	} {
+		if strings.Contains(text, sentinel.Error()) {
+			return fmt.Errorf("%w: %s", sentinel, instanceID)
+		}
+	}
+	return fmt.Errorf("routed %s %s: %s", op, instanceID, text)
+}
+
+// rebootSystemInstanceRemote routes a reboot to the owning daemon over
+// system.RebootInstance.{id}. No responder means the owner is unreachable and
+// is reported as ErrSystemInstanceNotFound; the caller retries and must not
+// count it as a repair.
+func (d *Daemon) rebootSystemInstanceRemote(instanceID string) error {
+	if d.natsConn == nil {
+		return fmt.Errorf("%w: %s", systeminstance.ErrSystemInstanceNotFound, instanceID)
+	}
+	subject := systemRebootSubject(instanceID)
+	reply, err := d.natsConn.Request(subject, nil, systemRebootRemoteTimeout)
+	if err != nil {
+		if errors.Is(err, nats.ErrNoResponders) {
+			return fmt.Errorf("%w: %s", systeminstance.ErrSystemInstanceNotFound, instanceID)
+		}
+		return fmt.Errorf("route reboot %s: %w", instanceID, err)
+	}
+	var env systemInstanceTerminateEnvelope
+	if err := json.Unmarshal(reply.Data, &env); err != nil {
+		return fmt.Errorf("decode routed reboot reply %s: %w", instanceID, err)
+	}
+	if env.Error != "" {
+		return routedSystemError("reboot", instanceID, env.Error)
+	}
+	return nil
+}
+
+// systemRebootRemoteTimeout caps a routed reboot round trip. The owner admits
+// the reboot and returns; the guest restart runs in the background.
+const systemRebootRemoteTimeout = 15 * time.Second
+
+func systemRebootSubject(instanceID string) string {
+	return "system.RebootInstance." + instanceID
+}
+
+// subscribeSystemRebootLocked registers the owning-node subscription for
+// system.RebootInstance.{instanceID}. Idempotent; the caller holds d.mu.
+func (d *Daemon) subscribeSystemRebootLocked(instanceID string) error {
+	if d.natsConn == nil {
+		return nil
+	}
+	subject := systemRebootSubject(instanceID)
+	if _, exists := d.natsSubscriptions[subject]; exists {
+		return nil
+	}
+	sub, err := d.natsConn.Subscribe(subject, d.handleSystemRebootInstance)
+	if err != nil {
+		return fmt.Errorf("subscribe %s: %w", subject, err)
+	}
+	d.natsSubscriptions[subject] = sub
+	return nil
+}
+
+// subscribeSystemReboot is subscribeSystemRebootLocked for callers without d.mu.
+func (d *Daemon) subscribeSystemReboot(instanceID string) error {
+	d.mu.Lock()
+	defer d.mu.Unlock()
+	return d.subscribeSystemRebootLocked(instanceID)
+}
+
+// handleSystemRebootInstance is the NATS subscriber for
+// system.RebootInstance.{instanceID}. Only the owning daemon subscribes.
+func (d *Daemon) handleSystemRebootInstance(msg *nats.Msg) {
+	d.systemDispatchWg.Go(func() {
+		defer func() {
+			if r := recover(); r != nil {
+				slog.Error("system.RebootInstance: handler panic", "subject", msg.Subject, "panic", r)
+				respondWithSystemRebootReply(msg, awserrors.ErrorServerInternal)
+			}
+		}()
+		id, ok := strings.CutPrefix(msg.Subject, "system.RebootInstance.")
+		if !ok || id == "" {
+			respondWithSystemRebootReply(msg, awserrors.ErrorInvalidInstanceIDNotFound)
+			return
+		}
+		// The local body, not the routing wrapper: only the owner subscribes.
+		if err := d.rebootSystemInstanceLocal(id); err != nil {
+			respondWithSystemRebootReply(msg, err.Error())
+			return
+		}
+		respondWithSystemRebootReply(msg, "")
+	})
+}
+
+func respondWithSystemRebootReply(msg *nats.Msg, errMsg string) {
+	payload, err := json.Marshal(systemInstanceTerminateEnvelope{Error: errMsg})
+	if err != nil {
+		slog.Error("system.RebootInstance: marshal reply failed", "err", err)
+		return
+	}
+	if err := msg.Respond(payload); err != nil {
+		slog.Error("system.RebootInstance: respond failed", "err", err)
+	}
+}
+
 // subscribeSystemTerminate registers an owning-node subscription for
 // system.TerminateInstance.{instanceID}. Idempotent — calling for an
 // already-bound subscription is a no-op.
@@ -266,6 +374,10 @@ func (d *Daemon) serveSystemTerminateInstance(msg *nats.Msg) {
 		}
 		delete(d.natsSubscriptions, subject)
 	}
+	if sub, exists := d.natsSubscriptions[systemRebootSubject(parts)]; exists {
+		_ = sub.Unsubscribe()
+		delete(d.natsSubscriptions, systemRebootSubject(parts))
+	}
 	d.mu.Unlock()
 
 	respondWithSystemTerminateOK(msg)

spinifex/runtime/daemon/daemon_system_instance.go

@@ -682,13 +682,19 @@ func (d *Daemon) reclaimSystemInstanceEIP(instanceID string, cls systeminstance.
 // customer ID is refused as not a system instance, an unknown ID answered as
 // gone and an unreadable record as ownership unknown; nothing is touched.
 func (d *Daemon) requireSystemInstance(instanceID string) (systeminstance.Classification, error) {
+	return d.requireSystemInstanceFor("TerminateSystemInstance", instanceID)
+}
+
+// requireSystemInstanceFor is requireSystemInstance with the calling operation
+// named in the refusal log.
+func (d *Daemon) requireSystemInstanceFor(op, instanceID string) (systeminstance.Classification, error) {
 	cls, err := d.classifySystemInstance(instanceID)
 	if err != nil {
-		slog.Warn("TerminateSystemInstance: refused, ownership could not be read", "instanceId", instanceID, "err", err)
+		slog.Warn(op+": refused, ownership could not be read", "instanceId", instanceID, "err", err)
 		return cls, fmt.Errorf("%w: %s: %w", systeminstance.ErrSystemInstanceOwnershipUnknown, instanceID, err)
 	}
 	if cls.Ownership != systeminstance.OwnershipSystem {
-		slog.Warn("TerminateSystemInstance: refused, not a system instance", "instanceId", instanceID, "ownership", cls.Ownership)
+		slog.Warn(op+": refused, not a system instance", "instanceId", instanceID, "ownership", cls.Ownership)
 		if cls.Ownership == systeminstance.OwnershipCustomer {
 			return cls, fmt.Errorf("%w: %s", systeminstance.ErrSystemInstanceNotSystem, instanceID)
 		}
@@ -738,6 +744,43 @@ func (d *Daemon) terminateSystemInstanceLocal(instanceID string) error {
 	return nil
 }
 
+// RebootSystemInstance restarts a load balancer VM into a fresh QEMU, keeping
+// its instance ID, ENIs and addresses. When this node does not run the VM the
+// request is routed to the owning daemon; no responder means the owner is
+// unreachable, reported as ErrSystemInstanceNotFound for the caller to retry.
+func (d *Daemon) RebootSystemInstance(instanceID string) error {
+	if _, err := d.requireSystemInstanceFor("RebootSystemInstance", instanceID); err != nil {
+		return err
+	}
+	if _, exists := d.vmMgr.Get(instanceID); exists {
+		return d.rebootSystemInstanceLocal(instanceID)
+	}
+	return d.rebootSystemInstanceRemote(instanceID)
+}
+
+// rebootSystemInstanceLocal restarts a VM owned by this node. It refuses unless
+// the VM is a load balancer in the system account, and never terminates.
+func (d *Daemon) rebootSystemInstanceLocal(instanceID string) error {
+	if _, err := d.requireSystemInstanceFor("RebootSystemInstance", instanceID); err != nil {
+		return err
+	}
+	instance, exists := d.vmMgr.Get(instanceID)
+	if !exists {
+		return fmt.Errorf("%w: %s", systeminstance.ErrSystemInstanceNotFound, instanceID)
+	}
+	if instance.ManagedBy != tags.ManagedByELBv2 || instance.AccountID != awsidentifiers.GlobalAccountID {
+		slog.Warn("RebootSystemInstance: refused, not a load balancer system VM",
+			"instanceId", instanceID, "managedBy", instance.ManagedBy)
+		return fmt.Errorf("%w: %s", systeminstance.ErrSystemInstanceNotSystem, instanceID)
+	}
+	if err := d.vmMgr.Reboot(context.Background(), instanceID); err != nil {
+		slog.Error("RebootSystemInstance: vmMgr.Reboot failed", "instanceId", instanceID, "err", err)
+		return fmt.Errorf("reboot: %w", err)
+	}
+	slog.Info("RebootSystemInstance admitted", "instanceId", instanceID)
+	return nil
+}
+
 // prepareInstanceRelaunch makes this node's world match an instance record it is
 // about to launch from.
 //
@@ -1093,3 +1136,5 @@ func buildNetcfgBlob(nics []elbv2domain.NICConfig) (string, error) {
 	}
 	return sb.String(), nil
 }
+
+var _ systeminstance.SystemInstanceLauncher = (*Daemon)(nil)

spinifex/runtime/daemon/daemon_system_reboot_test.go

@@ -0,0 +1,209 @@
+package daemon
+
+import (
+	"strings"
+	"sync"
+	"testing"
+	"time"
+
+	"github.com/mulgadc/spinifex/spinifex/domains/ec2/systeminstance"
+	elbv2domain "github.com/mulgadc/spinifex/spinifex/domains/elbv2"
+	awsidentifiers "github.com/mulgadc/spinifex/spinifex/foundation/aws/identifiers"
+	"github.com/mulgadc/spinifex/spinifex/foundation/aws/tags"
+	"github.com/mulgadc/spinifex/spinifex/runtime/compute/qmp"
+	"github.com/mulgadc/spinifex/spinifex/runtime/compute/vm"
+	"github.com/nats-io/nats.go"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+// withMonitor gives a test VM a monitor client. A reboot ends the old process
+// and marks that client dead, which makes "Reboot ran" observable.
+func withMonitor(v *vm.VM) *qmp.QMPClient {
+	v.QMPClient = &qmp.QMPClient{}
+	return v.QMPClient
+}
+
+func monitorDead(q *qmp.QMPClient) bool {
+	q.Mu.Lock()
+	defer q.Mu.Unlock()
+	return q.Dead
+}
+
+// Verifies: ADR-0003:S7.
+func TestS7_LBRepairRefusesACustomerInstance(t *testing.T) {
+	cases := []struct {
+		name string
+		inst *vm.VM
+		loc  recordLocation
+	}{
+		{"running customer", &vm.VM{ID: "i-r-running", Status: vm.StateRunning, AccountID: testAccountID}, atLive},
+		{"stopped customer", &vm.VM{ID: "i-r-stopped", Status: vm.StateStopped, AccountID: testAccountID}, atStopped},
+		{"load balancer manager in a customer account", &vm.VM{ID: "i-r-lbcust", Status: vm.StateRunning, AccountID: testAccountID, ManagedBy: tags.ManagedByELBv2}, atLive},
+		{"customer account holding an EKS tag", &vm.VM{ID: "i-r-eks-cust", Status: vm.StateRunning, AccountID: testAccountID, ManagedBy: tags.ManagedByEKS}, atLive},
+		{"system VM that is not a load balancer", &vm.VM{ID: "i-r-eks", Status: vm.StateRunning, AccountID: awsidentifiers.GlobalAccountID, ManagedBy: tags.ManagedByEKS}, atLive},
+	}
+	for _, route := range []string{"local", "wrapper", "routed", "nats launcher"} {
+		t.Run(route, func(t *testing.T) {
+			d := guardDaemon(t)
+			terminated := countTerminations(d)
+			launcher := elbv2domain.NewNATSSystemInstanceLauncher(d.natsConn, 5*time.Second)
+
+			for _, c := range cases {
+				mon := withMonitor(c.inst)
+				stage(t, d, c.inst, c.loc)
+				var err error
+				switch route {
+				case "local":
+					err = d.rebootSystemInstanceLocal(c.inst.ID)
+				case "wrapper":
+					err = d.RebootSystemInstance(c.inst.ID)
+				case "routed":
+					require.NoError(t, d.subscribeSystemReboot(c.inst.ID))
+					err = d.rebootSystemInstanceRemote(c.inst.ID)
+				case "nats launcher":
+					require.NoError(t, d.subscribeSystemReboot(c.inst.ID))
+					err = launcher.RebootSystemInstance(c.inst.ID)
+				}
+				require.Error(t, err, c.name)
+				assert.Contains(t, err.Error(), systeminstance.ErrSystemInstanceNotSystem.Error(), c.name)
+				assert.NotContains(t, err.Error(), systeminstance.ErrSystemInstanceNotFound.Error(), c.name)
+				d.vmMgr.WaitForBackgroundWork()
+				assert.False(t, monitorDead(mon), "%s: Reboot must never be reached", c.name)
+			}
+			assert.Zero(t, terminated.Load())
+		})
+	}
+}
+
+// Verifies: ADR-0003:S7.
+func TestS7_LBRepairRebootsAndNeverTerminates(t *testing.T) {
+	for _, route := range []string{"wrapper", "routed", "nats launcher"} {
+		t.Run(route, func(t *testing.T) {
+			d := guardDaemon(t)
+			var mu sync.Mutex
+			var moves []vm.InstanceState
+			d.vmMgr.SetDeps(vm.Deps{NodeID: d.node, TransitionState: func(_ *vm.VM, to vm.InstanceState) error {
+				mu.Lock()
+				defer mu.Unlock()
+				moves = append(moves, to)
+				return nil
+			}})
+			eip := &sysInstReclaimingEIP{}
+			d.eipService = eip
+			lb := &vm.VM{ID: "i-lb-" + strings.ReplaceAll(route, " ", "-"), Status: vm.StateRunning, AccountID: awsidentifiers.GlobalAccountID, ManagedBy: tags.ManagedByELBv2}
+			mon := withMonitor(lb)
+			stage(t, d, lb, atLive)
+
+			var err error
+			switch route {
+			case "wrapper":
+				err = d.RebootSystemInstance(lb.ID)
+			case "routed":
+				require.NoError(t, d.subscribeSystemReboot(lb.ID))
+				err = d.rebootSystemInstanceRemote(lb.ID)
+			case "nats launcher":
+				require.NoError(t, d.subscribeSystemReboot(lb.ID))
+				err = elbv2domain.NewNATSSystemInstanceLauncher(d.natsConn, 5*time.Second).RebootSystemInstance(lb.ID)
+			}
+			require.NoError(t, err)
+
+			require.Eventually(t, func() bool { return monitorDead(mon) }, 5*time.Second, 10*time.Millisecond, "the owner restarts the guest")
+			d.vmMgr.WaitForBackgroundWork()
+			assert.NotEqual(t, vm.StateTerminated, d.vmMgr.Status(lb))
+			mu.Lock()
+			defer mu.Unlock()
+			assert.NotContains(t, moves, vm.StateShuttingDown, "a repair never terminates")
+			assert.NotContains(t, moves, vm.StateTerminated, "a repair never terminates")
+			assert.Empty(t, eip.released, "a repair never releases an address")
+			assert.Empty(t, eip.reclaimed)
+		})
+	}
+}
+
+func TestRebootSystemInstance_NoOwnerAndUnreadableRecord(t *testing.T) {
+	t.Run("no owner subscribed", func(t *testing.T) {
+		d := guardDaemon(t)
+		err := d.rebootSystemInstanceRemote("i-nobody-runs-this")
+		require.ErrorIs(t, err, systeminstance.ErrSystemInstanceNotFound)
+
+		err = elbv2domain.NewNATSSystemInstanceLauncher(d.natsConn, 2*time.Second).RebootSystemInstance("i-nobody-runs-this")
+		require.Error(t, err)
+		assert.ErrorIs(t, err, nats.ErrNoResponders, "an unreachable owner stays distinguishable")
+	})
+
+	t.Run("a record this node cannot read", func(t *testing.T) {
+		d := guardDaemon(t)
+		lb := &vm.VM{ID: "i-lb-unreadable", Status: vm.StateRunning, AccountID: awsidentifiers.GlobalAccountID, ManagedBy: tags.ManagedByELBv2}
+		mon := withMonitor(lb)
+		d.vmMgr.Insert(lb)
+		d.ownership = fakeOwnership{err: errInjected}
+
+		err := d.RebootSystemInstance(lb.ID)
+		require.ErrorIs(t, err, systeminstance.ErrSystemInstanceOwnershipUnknown)
+		d.vmMgr.WaitForBackgroundWork()
+		assert.False(t, monitorDead(mon))
+	})
+
+	t.Run("a VM that is not running", func(t *testing.T) {
+		d := guardDaemon(t)
+		lb := &vm.VM{ID: "i-lb-stopped", Status: vm.StateStopped, AccountID: awsidentifiers.GlobalAccountID, ManagedBy: tags.ManagedByELBv2}
+		stage(t, d, lb, atLive)
+		require.Error(t, d.RebootSystemInstance(lb.ID))
+	})
+}
+
+func TestRebootSystemInstanceRemote_ReplyAndConnectionFailures(t *testing.T) {
+	t.Run("no connection reads as an unreachable owner", func(t *testing.T) {
+		d := guardDaemon(t)
+		d.natsConn = nil
+		require.ErrorIs(t, d.rebootSystemInstanceRemote("i-lb"), systeminstance.ErrSystemInstanceNotFound)
+	})
+
+	t.Run("an undecodable reply is an error, not a reboot", func(t *testing.T) {
+		d := guardDaemon(t)
+		sub, err := d.natsConn.Subscribe(systemRebootSubject("i-garbled"), func(msg *nats.Msg) {
+			_ = msg.Respond([]byte("{not json"))
+		})
+		require.NoError(t, err)
+		defer func() { _ = sub.Unsubscribe() }()
+
+		err = d.rebootSystemInstanceRemote("i-garbled")
+		require.Error(t, err)
+		assert.NotErrorIs(t, err, systeminstance.ErrSystemInstanceNotFound)
+	})
+
+	t.Run("an owner that refuses is reported to the caller", func(t *testing.T) {
+		d := guardDaemon(t)
+		sub, err := d.natsConn.Subscribe(systemRebootSubject("i-refusing"), func(msg *nats.Msg) {
+			respondWithSystemRebootReply(msg, "not a system instance")
+		})
+		require.NoError(t, err)
+		defer func() { _ = sub.Unsubscribe() }()
+
+		require.Error(t, d.rebootSystemInstanceRemote("i-refusing"))
+	})
+}
+
+func TestHandleSystemRebootInstance_RejectsAMalformedSubject(t *testing.T) {
+	d := guardDaemon(t)
+	sub, err := d.natsConn.Subscribe("system.Elsewhere.i-lb", d.handleSystemRebootInstance)
+	require.NoError(t, err)
+	defer func() { _ = sub.Unsubscribe() }()
+
+	reply, err := d.natsConn.Request("system.Elsewhere.i-lb", nil, 2*time.Second)
+	require.NoError(t, err)
+	assert.Contains(t, string(reply.Data), "InvalidInstanceID.NotFound")
+	d.systemDispatchWg.Wait()
+}
+
+func TestSubscribeSystemReboot_IdempotentAndSkippedWithoutConnection(t *testing.T) {
+	d := guardDaemon(t)
+	require.NoError(t, d.subscribeSystemReboot("i-lb"))
+	require.NoError(t, d.subscribeSystemReboot("i-lb"))
+	assert.Contains(t, d.natsSubscriptions, systemRebootSubject("i-lb"))
+
+	d.natsConn = nil
+	require.NoError(t, d.subscribeSystemReboot("i-other"))
+	assert.NotContains(t, d.natsSubscriptions, systemRebootSubject("i-other"))
+}

spinifex/runtime/daemon/nats_classification_test.go

@@ -137,6 +137,8 @@ func TestEC2CommandNameCoversEveryCommand(t *testing.T) {
 		{"SetInstanceTags", func(a *ec2v1.EC2CommandAttributes) { a.SetInstanceTags = true }},
 		{"RemoveInstanceTags", func(a *ec2v1.EC2CommandAttributes) { a.RemoveInstanceTags = true }},
 		{"SetInstanceMonitoring", func(a *ec2v1.EC2CommandAttributes) { a.SetInstanceMonitoring = true }},
+		{"ModifyInstanceMetadataOptions", func(a *ec2v1.EC2CommandAttributes) { a.ModifyInstanceMetadataOptions = true }},
+		{"SetDisableApiTermination", func(a *ec2v1.EC2CommandAttributes) { a.SetDisableApiTermination = true }},
 		{"StartInstance", func(a *ec2v1.EC2CommandAttributes) { a.StartInstance = true }},
 		{"RebootInstance", func(a *ec2v1.EC2CommandAttributes) { a.RebootInstance = true }},
 		{"StopInstance", func(a *ec2v1.EC2CommandAttributes) { a.StopInstance = true }},