feat: split IAM e2e (#372)

.github/workflows/e2e-suites.yml

@@ -3,7 +3,7 @@
 # Restructures the looped e2e.yml into env-grouped job blocks so each suite
 # surfaces as its own node in the Actions graph (mulga-siv-283):
 #
-#   single_provision ─┬─ single_suite (matrix: single, cert, lb, eks)
+#   single_provision ─┬─ single_suite (matrix: single, cert, lb, eks, iam)
 #                     └─ single_teardown (if: always)
 #   multi_provision  ─┬─ multi_suite  (matrix: multinode)
 #                     └─ multi_teardown (if: always)
@@ -27,7 +27,7 @@ on:
       single_suites:
         description: "Single-node suites to run as matrix legs"
         required: false
-        default: "single cert lb eks"
+        default: "single cert lb eks iam"
       run_multinode:
         description: "Also run the multinode env block"
         type: boolean
@@ -114,7 +114,7 @@ jobs:
       - name: Compute suite matrix
         id: suites
         run: |
-          LIST="${{ inputs.single_suites || 'single cert lb eks' }}"
+          LIST="${{ inputs.single_suites || 'single cert lb eks iam' }}"
           echo "list=${LIST}" >> "$GITHUB_OUTPUT"
           JSON=$(printf '%s\n' $LIST | jq -R . | jq -cs .)
           echo "matrix=${JSON}" >> "$GITHUB_OUTPUT"

.github/workflows/e2e.yml

@@ -15,9 +15,9 @@ on:
   workflow_dispatch:
     inputs:
       suites:
-        description: "Space-separated Make targets to run (e.g. 'e2e-single e2e-multinode e2e-cert e2e-lb e2e-eks')"
+        description: "Space-separated Make targets to run (e.g. 'e2e-single e2e-multinode e2e-cert e2e-lb e2e-eks e2e-iam')"
         required: false
-        default: "e2e-single e2e-multinode e2e-cert e2e-lb"
+        default: "e2e-single e2e-multinode e2e-cert e2e-lb e2e-iam"
       test_run:
         description: "Optional regex for -test.run (e.g. '^TestMultinodeIPSec$'). Empty = run all tests in suite."
         required: false
@@ -29,7 +29,7 @@ permissions:
 
 env:
   GIT_BRANCH: ${{ github.ref_name }}
-  SUITES: ${{ inputs.suites || 'e2e-single e2e-multinode e2e-cert e2e-lb' }}
+  SUITES: ${{ inputs.suites || 'e2e-single e2e-multinode e2e-cert e2e-lb e2e-iam' }}
   TEST_RUN: ${{ inputs.test_run || '' }}
 
 jobs:
@@ -75,7 +75,7 @@ jobs:
     # suites='e2e-multinode' only). Push triggers and the default SUITES list
     # both include single/cert/lb so the inputs.suites null-check keeps the
     # job running on plain PR pushes.
-    if: github.event_name != 'workflow_dispatch' || contains(github.event.inputs.suites, 'e2e-single') || contains(github.event.inputs.suites, 'e2e-cert') || contains(github.event.inputs.suites, 'e2e-lb') || contains(github.event.inputs.suites, 'e2e-eks')
+    if: github.event_name != 'workflow_dispatch' || contains(github.event.inputs.suites, 'e2e-single') || contains(github.event.inputs.suites, 'e2e-cert') || contains(github.event.inputs.suites, 'e2e-lb') || contains(github.event.inputs.suites, 'e2e-eks') || contains(github.event.inputs.suites, 'e2e-iam')
     runs-on: [self-hosted, ci-single]
     timeout-minutes: 45
     env:
@@ -590,12 +590,15 @@ jobs:
         run: |
           set -euo pipefail
           source "${GITHUB_WORKSPACE}/spinifex/.github/scripts/ci-fmt.sh"
+          # single + iam are single-node-only suites — their fixtures skip
+          # unless SPINIFEX_MODE=single, so building/running them on multinode
+          # is wasted work.
           # eks is excluded on multinode: EKS KV is a clustered (Raft) JetStream
           # stream and create-cluster needs an elected leader, but multinode
           # JetStream clustering isn't solid yet (mulga-omq/z2z/5yg). EKS is
           # validated single-node only for now; re-enable here once clustered
           # JetStream + multinode EKS are a deliberate target.
-          SUITES_BUILD=$(echo "${SUITES}" | sed 's/e2e-//g' | tr ' ' '\n' | grep -vE '^(single|eks)$' | tr '\n' ' ')
+          SUITES_BUILD=$(echo "${SUITES}" | sed 's/e2e-//g' | tr ' ' '\n' | grep -vE '^(single|eks|iam)$' | tr '\n' ' ')
           if [ -z "$(echo "${SUITES_BUILD}" | tr -d ' ')" ]; then
             echo "no multinode-eligible suites requested (SUITES=${SUITES}); nothing to build"
             exit 0
@@ -678,7 +681,7 @@ jobs:
         id: e2e
         run: |
           set +e
-          SUITES_RUN=$(echo "${SUITES}" | sed 's/e2e-//g' | tr ' ' '\n' | grep -vE '^(single|eks)$' | tr '\n' ' ')
+          SUITES_RUN=$(echo "${SUITES}" | sed 's/e2e-//g' | tr ' ' '\n' | grep -vE '^(single|eks|iam)$' | tr '\n' ' ')
           if [ -z "$(echo "${SUITES_RUN}" | tr -d ' ')" ]; then
             echo "no multinode-eligible suites requested (SUITES=${SUITES}); skipping"
             echo "rc=0" >> "$GITHUB_OUTPUT"

docs/service-interfaces.yaml

@@ -284,6 +284,14 @@ suites:
     # into per-test fixture-driven coverage so the selector can pick
     # subtests, not just whole suites.
 
+  e2e-iam:
+    path: tests/e2e/iam
+    covers: [awsgw, spinifex_daemon, vpcd, viperblockd, predastore]
+    # "Everything IAM": the IAM/STS control plane plus the instance-identity
+    # tests (instance-profile association + IMDS), which boot guest VMs — OVN
+    # via vpcd, AMI via predastore, root volume via viperblockd, lifecycle via
+    # spinifex_daemon. Single-mode, VM-booting leg.
+
   e2e-ddil:
     path: tests/e2e/ddil
     covers: [awsgw, spinifex_daemon, vpcd, viperblockd, predastore]

tests/e2e/Makefile

@@ -1,4 +1,4 @@
-PKG         ?= ./cert/... ./lb/... ./eks/... ./multinode/... ./baremetal/... ./reboot/... ./tofu/...
+PKG         ?= ./cert/... ./lb/... ./eks/... ./iam/... ./multinode/... ./baremetal/... ./reboot/... ./tofu/...
 ARTIFACT_DIR ?= /tmp/spinifex-e2e-artifacts
 TIMEOUT     ?= 30m
 MODE        ?= single
@@ -11,9 +11,9 @@ SUITE_TAG   ?= $(notdir $(patsubst %/...,%,$(firstword $(PKG))))
 # named suite into BIN_DIR/<suite>.test so CI can ship binaries (not source +
 # go toolchain) to the remote VM.
 BIN_DIR      ?= $(CURDIR)/_bin
-SUITES_BUILD ?= cert lb eks single gpu
+SUITES_BUILD ?= cert lb eks single iam gpu
 
-.PHONY: e2e e2e-cert e2e-lb e2e-eks e2e-single e2e-multinode e2e-bm e2e-reboot e2e-tofu e2e-gpu e2e-all clean e2e-build
+.PHONY: e2e e2e-cert e2e-lb e2e-eks e2e-single e2e-iam e2e-multinode e2e-bm e2e-reboot e2e-tofu e2e-gpu e2e-all clean e2e-build
 
 ifeq ($(GOTESTSUM),1)
 E2E_CMD = gotestsum \
@@ -36,6 +36,7 @@ e2e-cert:       ; $(MAKE) e2e PKG=./cert/...
 e2e-lb:         ; $(MAKE) e2e PKG=./lb/...
 e2e-eks:        ; $(MAKE) e2e PKG=./eks/...
 e2e-single:     ; $(MAKE) e2e PKG=./single/...
+e2e-iam:        ; $(MAKE) e2e PKG=./iam/...
 e2e-multinode:  ; $(MAKE) e2e PKG=./multinode/... MODE=multinode
 e2e-bm:         ; $(MAKE) e2e PKG=./baremetal/... MODE=baremetal
 e2e-reboot:     ; $(MAKE) e2e PKG=./reboot/...

tests/e2e/harness/iam.go

@@ -0,0 +1,70 @@
+//go:build e2e
+
+package harness
+
+import (
+	"testing"
+
+	"github.com/aws/aws-sdk-go/aws"
+	"github.com/aws/aws-sdk-go/service/iam"
+	"github.com/aws/aws-sdk-go/service/sts"
+	"github.com/stretchr/testify/require"
+)
+
+// IAMRoleARN constructs arn:aws:iam::<account>:role/<name>.
+func IAMRoleARN(account, name string) string {
+	return "arn:aws:iam::" + account + ":role/" + name
+}
+
+// IAMPolicyARN builds the canonical policy ARN for a given account + policy
+// "key" (a bare name, or "path/name" without the leading slash).
+func IAMPolicyARN(account, key string) string {
+	return "arn:aws:iam::" + account + ":policy/" + key
+}
+
+// IAMAccountID returns the active account ID via STS GetCallerIdentity. Unlike
+// deriving it from a created IAM user's ARN, this is side-effect-free, so suites
+// sharing one account can call it concurrently without minting colliding users.
+func IAMAccountID(t *testing.T, c *AWSClient) string {
+	t.Helper()
+	out, err := c.STS.GetCallerIdentity(&sts.GetCallerIdentityInput{})
+	require.NoError(t, err, "sts:GetCallerIdentity")
+	require.NotEmpty(t, aws.StringValue(out.Account), "empty account in GetCallerIdentity response")
+	return aws.StringValue(out.Account)
+}
+
+// IAMDeleteRoleAndProfilesBestEffort tears down every fragment of a role +
+// instance-profile graph a suite might have left behind. Each step swallows
+// errors so a missing fragment never cascades; usable both as a pre-test sweep
+// and as a fixture-teardown cleanup. Order matters: unbind role-from-profile
+// before deleting the profile, detach policies before deleting the role.
+func IAMDeleteRoleAndProfilesBestEffort(c *AWSClient, roleName string, profileNames []string, policyARNs ...string) {
+	for _, p := range profileNames {
+		_, _ = c.IAM.RemoveRoleFromInstanceProfile(&iam.RemoveRoleFromInstanceProfileInput{
+			InstanceProfileName: aws.String(p),
+			RoleName:            aws.String(roleName),
+		})
+		_, _ = c.IAM.DeleteInstanceProfile(&iam.DeleteInstanceProfileInput{
+			InstanceProfileName: aws.String(p),
+		})
+	}
+	for _, arn := range policyARNs {
+		_, _ = c.IAM.DetachRolePolicy(&iam.DetachRolePolicyInput{
+			RoleName:  aws.String(roleName),
+			PolicyArn: aws.String(arn),
+		})
+	}
+	// Defensive: pull any other attached policies so the final DeleteRole isn't
+	// blocked by a stray attach from a partial run.
+	if attached, err := c.IAM.ListAttachedRolePolicies(&iam.ListAttachedRolePoliciesInput{
+		RoleName: aws.String(roleName),
+	}); err == nil {
+		for _, p := range attached.AttachedPolicies {
+			_, _ = c.IAM.DetachRolePolicy(&iam.DetachRolePolicyInput{
+				RoleName:  aws.String(roleName),
+				PolicyArn: p.PolicyArn,
+			})
+		}
+	}
+	_, _ = c.IAM.DeleteRole(&iam.DeleteRoleInput{RoleName: aws.String(roleName)})
+}

tests/e2e/iam/assumed_role_enforcement_test.go

@@ -1,6 +1,6 @@
 //go:build e2e
 
-package single
+package iam
 
 import (
 	"testing"
@@ -28,13 +28,13 @@ const (
 // live session is allowed once a policy is attached (policies resolved per request).
 func runAssumedRoleControlPlaneEnforcement(t *testing.T, fix *Fixture) {
 	harness.Phase(t, "Single — Assumed-role control-plane enforcement")
-	adminAccount := iamEnsureAdminAccountID(t, fix)
-	roleARN := iamRoleARN(adminAccount, areRoleName)
-	policyARN := iamPolicyARN(adminAccount, arePolicyName)
+	adminAccount := harness.IAMAccountID(t, fix.AWS)
+	roleARN := harness.IAMRoleARN(adminAccount, areRoleName)
+	policyARN := harness.IAMPolicyARN(adminAccount, arePolicyName)
 
 	// Pre-sweep in case a previous run left fragments.
 	sweep := func() {
-		iamDeleteRoleAndProfilesBestEffort(fix, areRoleName, nil, policyARN)
+		harness.IAMDeleteRoleAndProfilesBestEffort(fix.AWS, areRoleName, nil, policyARN)
 		_, _ = fix.AWS.IAM.DeletePolicy(&iam.DeletePolicyInput{PolicyArn: aws.String(policyARN)})
 	}
 	sweep()

tests/e2e/iam/datapath_helpers_test.go

@@ -0,0 +1,230 @@
+//go:build e2e
+
+package iam
+
+import (
+	"bufio"
+	"bytes"
+	"fmt"
+	"net"
+	"os"
+	"os/exec"
+	"path/filepath"
+	"strconv"
+	"strings"
+	"sync/atomic"
+	"testing"
+	"time"
+
+	"github.com/aws/aws-sdk-go/aws"
+	"github.com/aws/aws-sdk-go/service/ec2"
+	"github.com/mulgadc/spinifex/tests/e2e/harness"
+)
+
+// SSH / datapath helpers ported from the single suite so the IMDS and
+// instance-profile tests run self-contained inside iam. The single suite keeps
+// its own copy — other single tests still use these. They are separate test
+// binaries (separate processes), so the sticky-skip state below is per-suite.
+
+// sshDatapathBroken is set the first time an SSH probe in this suite times out.
+// Downstream SSH-dependent assertions call requireSSHHealthy(t) and skip rather
+// than re-running the same multi-minute timeout against the same broken datapath.
+var sshDatapathBroken atomic.Bool
+
+// sshReadyBudget bounds the SSH-handshake probes. Widen via
+// SPINIFEX_SSH_READY_TIMEOUT (Go duration); default 3m bounds shared runners.
+var sshReadyBudget = resolveSSHReadyBudget(3 * time.Minute)
+
+func resolveSSHReadyBudget(def time.Duration) time.Duration {
+	if v := os.Getenv("SPINIFEX_SSH_READY_TIMEOUT"); v != "" {
+		if d, err := time.ParseDuration(v); err == nil && d > 0 {
+			return d
+		}
+	}
+	return def
+}
+
+// requireSSHHealthy skips the calling test if a prior SSH probe in this suite
+// has already failed, cutting the cascade cost of a broken datapath.
+func requireSSHHealthy(t *testing.T) {
+	t.Helper()
+	if sshDatapathBroken.Load() {
+		t.Skipf("skipping: earlier SSH probe failed; " +
+			"not retrying to keep suite time bounded")
+	}
+}
+
+// waitForSSHReady probes a full SSH handshake (BatchMode + ConnectTimeout)
+// against host:port, retrying until the daemon completes banner exchange.
+// TCP-port reachability alone is insufficient — sshd accepts the connect
+// while pam/cloud-init are still finishing, and the first real runSSH then
+// hits "Connection timed out during banner exchange".
+func waitForSSHReady(t *testing.T, host string, port int, keyPath string) {
+	t.Helper()
+	requireSSHHealthy(t)
+	addr := net.JoinHostPort(host, strconv.Itoa(port))
+	harness.Step(t, "waiting for SSH handshake %s", addr)
+	if !trySSHReady(host, port, keyPath, sshReadyBudget) {
+		sshDatapathBroken.Store(true)
+		t.Fatalf("Eventually: condition not met within %s: "+
+			"[SSH handshake %s never completed] "+
+			"(sticky-skip enabled for downstream)", sshReadyBudget, addr)
+	}
+}
+
+// waitForSSHHandshake is an alias kept for the imds_test.go call site.
+func waitForSSHHandshake(t *testing.T, host string, port int, keyPath string) {
+	waitForSSHReady(t, host, port, keyPath)
+}
+
+// trySSHReady mirrors waitForSSHReady but returns reachability as a bool
+// instead of t.Fatal-ing on timeout.
+func trySSHReady(host string, port int, keyPath string, timeout time.Duration) bool {
+	if sshDatapathBroken.Load() {
+		return false
+	}
+	deadline := time.Now().Add(timeout)
+	for {
+		cmd := exec.Command("ssh",
+			"-o", "StrictHostKeyChecking=no",
+			"-o", "UserKnownHostsFile=/dev/null",
+			"-o", "LogLevel=ERROR",
+			"-o", "ConnectTimeout=3",
+			"-o", "BatchMode=yes",
+			"-p", strconv.Itoa(port),
+			"-i", keyPath,
+			"ec2-user@"+host,
+			"true")
+		if cmd.Run() == nil {
+			return true
+		}
+		if time.Now().After(deadline) {
+			return false
+		}
+		time.Sleep(2 * time.Second)
+	}
+}
+
+// runSSH is a thin wrapper around `ssh` matching the option set used by
+// harness.LsblkRootGiB. Returns stdout. t.Fatal on non-zero exit so callers
+// can chain assertions on the output without nil-checking err.
+func runSSH(t *testing.T, tgt harness.SSHTarget, command string) string {
+	t.Helper()
+	args := []string{
+		"-o", "StrictHostKeyChecking=no",
+		"-o", "UserKnownHostsFile=/dev/null",
+		"-o", "LogLevel=ERROR",
+		"-o", "ConnectTimeout=5",
+		"-o", "BatchMode=yes",
+		"-p", strconv.Itoa(tgt.Port),
+		"-i", tgt.KeyPath,
+		tgt.User + "@" + tgt.Host,
+		command,
+	}
+	var stdout, stderr bytes.Buffer
+	cmd := exec.Command("ssh", args...)
+	cmd.Stdout = &stdout
+	cmd.Stderr = &stderr
+	if err := cmd.Run(); err != nil {
+		t.Fatalf("ssh %s@%s:%d %q failed: %v\nstderr: %s",
+			tgt.User, tgt.Host, tgt.Port, command, err, stderr.String())
+	}
+	return stdout.String()
+}
+
+// runSSHCombined runs `command` over SSH against tgt and returns combined
+// stdout+stderr regardless of exit status. Unlike runSSH it does not t.Fatal
+// on non-zero exit — needed for probes (e.g. curl status codes) where a
+// non-zero exit is an expected outcome.
+func runSSHCombined(tgt harness.SSHTarget, command string) (string, error) {
+	args := []string{
+		"-o", "StrictHostKeyChecking=no",
+		"-o", "UserKnownHostsFile=/dev/null",
+		"-o", "LogLevel=ERROR",
+		"-o", "ConnectTimeout=5",
+		"-o", "BatchMode=yes",
+		"-p", strconv.Itoa(tgt.Port),
+		"-i", tgt.KeyPath,
+		tgt.User + "@" + tgt.Host,
+		command,
+	}
+	var combined bytes.Buffer
+	cmd := exec.Command("ssh", args...)
+	cmd.Stdout = &combined
+	cmd.Stderr = &combined
+	err := cmd.Run()
+	return combined.String(), err
+}
+
+// primaryENI returns the NetworkInterfaceId of an instance's first ENI.
+// t.Fatal if the instance has no ENI — every running EC2 instance must.
+func primaryENI(t *testing.T, inst *ec2.Instance) string {
+	t.Helper()
+	if len(inst.NetworkInterfaces) == 0 {
+		t.Fatalf("instance %s has no NetworkInterfaces", aws.StringValue(inst.InstanceId))
+	}
+	eni := aws.StringValue(inst.NetworkInterfaces[0].NetworkInterfaceId)
+	if eni == "" {
+		t.Fatalf("instance %s primary ENI has empty NetworkInterfaceId", aws.StringValue(inst.InstanceId))
+	}
+	return eni
+}
+
+// waitForInstanceStateSoft is the cleanup-time analogue of
+// harness.WaitForInstanceState — no t.Fatal, just polls and returns.
+func waitForInstanceStateSoft(c *harness.AWSClient, id, target string, timeout time.Duration) error {
+	deadline := time.Now().Add(timeout)
+	for {
+		out, err := c.EC2.DescribeInstances(&ec2.DescribeInstancesInput{
+			InstanceIds: []*string{aws.String(id)},
+		})
+		if err == nil && len(out.Reservations) > 0 && len(out.Reservations[0].Instances) > 0 {
+			if aws.StringValue(out.Reservations[0].Instances[0].State.Name) == target {
+				return nil
+			}
+		}
+		if time.Now().After(deadline) {
+			return fmt.Errorf("instance %s did not reach %s within %s", id, target, timeout)
+		}
+		time.Sleep(2 * time.Second)
+	}
+}
+
+// detectPoolMode reads external_mode from spinifex.toml. Defaults to false
+// (dev_networking) which is the single-node CI fixture. Any non-empty
+// external_mode value ("pool" / "nat") means external IPAM is in play, which
+// gates the IMDS probe-VPC public-IP path.
+func detectPoolMode(env *harness.Env) bool {
+	cfg := os.ExpandEnv("$HOME/spinifex/config/spinifex.toml")
+	if env.ConfigDir != "" {
+		cfg = filepath.Join(env.ConfigDir, "spinifex.toml")
+	}
+	f, err := os.Open(cfg)
+	if err != nil {
+		return false
+	}
+	defer f.Close()
+
+	scanner := bufio.NewScanner(f)
+	inNetwork := false
+	for scanner.Scan() {
+		line := strings.TrimSpace(scanner.Text())
+		if strings.HasPrefix(line, "[") {
+			inNetwork = line == "[network]"
+			continue
+		}
+		if !inNetwork {
+			continue
+		}
+		if !strings.HasPrefix(line, "external_mode") {
+			continue
+		}
+		// external_mode = "pool" — quoted value, anything non-empty == pool mode.
+		if i := bytes.IndexByte([]byte(line), '='); i >= 0 {
+			val := strings.TrimSpace(line[i+1:])
+			val = strings.Trim(val, "\"'")
+			return val != ""
+		}
+	}
+	return false
+}

tests/e2e/iam/helpers_test.go

@@ -0,0 +1,82 @@
+//go:build e2e
+
+package iam
+
+import (
+	"testing"
+
+	"github.com/aws/aws-sdk-go/aws"
+	"github.com/aws/aws-sdk-go/service/ec2"
+	"github.com/mulgadc/spinifex/tests/e2e/harness"
+	"github.com/stretchr/testify/require"
+)
+
+// Instance-bootstrap helpers wrapping harness.Discover*/Ensure* so the IMDS and
+// instance-profile tests can self-bootstrap their VM prerequisites. Results are
+// memoized by harness.Fixture. The single suite keeps its own copy of these —
+// other single tests still use them.
+
+// needInstanceTypeArch returns the nano instance type and its architecture.
+func needInstanceTypeArch(t *testing.T, fix *Fixture) (instanceType, arch string) {
+	t.Helper()
+	return harness.DiscoverNanoInstanceType(t, fix.Harness)
+}
+
+// needAMI returns the discovered architecture-appropriate Ubuntu AMI ID.
+func needAMI(t *testing.T, fix *Fixture) string {
+	t.Helper()
+	_, arch := needInstanceTypeArch(t, fix)
+	return harness.DiscoverUbuntuAMI(t, fix.Harness, arch)
+}
+
+// needKeyPair returns the memoized primary EC2 key pair name + PEM path.
+func needKeyPair(t *testing.T, fix *Fixture) (name, pemPath string) {
+	t.Helper()
+	return harness.EnsureKeyPair(t, fix.Harness, fix.TmpDir)
+}
+
+// needInstance bootstraps the suite's primary running instance and returns
+// the *ec2.Instance plus its root volume ID. Idempotent via harness.EnsureInstance.
+func needInstance(t *testing.T, fix *Fixture) (inst *ec2.Instance, rootVolumeID string) {
+	t.Helper()
+	instType, _ := needInstanceTypeArch(t, fix)
+	keyName, _ := needKeyPair(t, fix)
+	ami := needAMI(t, fix)
+	vpc := harness.EnsureDefaultVPC(t, fix.Harness)
+	require.NotEmpty(t, vpc.SGID, "default SG ID required")
+	harness.AuthorizeSSHIngress(t, fix.AWS, vpc.SGID)
+
+	instanceID := harness.EnsureInstance(t, fix.Harness, harness.InstanceSpec{
+		AMIID:        ami,
+		InstanceType: instType,
+		KeyName:      keyName,
+		SubnetID:     vpc.SubnetID,
+		SGID:         vpc.SGID,
+	})
+	require.NotEmpty(t, instanceID, "EnsureInstance returned empty InstanceId")
+
+	descOut, err := fix.AWS.EC2.DescribeInstances(&ec2.DescribeInstancesInput{
+		InstanceIds: []*string{aws.String(instanceID)},
+	})
+	require.NoError(t, err, "describe-instances %s", instanceID)
+	require.NotEmpty(t, descOut.Reservations, "no reservations for %s", instanceID)
+	require.NotEmpty(t, descOut.Reservations[0].Instances, "no instances for %s", instanceID)
+	inst = descOut.Reservations[0].Instances[0]
+
+	// Match BDM by RootDeviceName; fall back to first BDM if empty.
+	rootDev := aws.StringValue(inst.RootDeviceName)
+	for _, bdm := range inst.BlockDeviceMappings {
+		if rootDev != "" && aws.StringValue(bdm.DeviceName) != rootDev {
+			continue
+		}
+		if bdm.Ebs != nil {
+			rootVolumeID = aws.StringValue(bdm.Ebs.VolumeId)
+			break
+		}
+	}
+	if rootVolumeID == "" && len(inst.BlockDeviceMappings) > 0 && inst.BlockDeviceMappings[0].Ebs != nil {
+		rootVolumeID = aws.StringValue(inst.BlockDeviceMappings[0].Ebs.VolumeId)
+	}
+	require.NotEmpty(t, rootVolumeID, "could not resolve root volume from BlockDeviceMappings")
+	return inst, rootVolumeID
+}

tests/e2e/iam/iam_helpers_test.go

@@ -1,6 +1,6 @@
 //go:build e2e
 
-package single
+package iam
 
 import (
 	"sync"
@@ -36,10 +36,6 @@ var (
 	iamCharlieKeyID     string
 	iamCharlieKeySecret string
 	iamCharlieKeyErr    error
-
-	iamAdminAccountOnce sync.Once
-	iamAdminAccountID   string
-	iamAdminAccountErr  error
 )
 
 // iamEnsureAlice ensures the canonical "alice" IAM user exists. The user is
@@ -129,26 +125,6 @@ func iamEnsureCharlieKey(t *testing.T, fix *Fixture) (keyID, secret string) {
 	return iamCharlieKeyID, iamCharlieKeySecret
 }
 
-// iamEnsureAdminAccountID returns the active account ID, derived from the
-// ARN of a probe IAM resource. Uses GetUser on the canonical alice user so
-// no policy churn is needed; alice is created on first call.
-func iamEnsureAdminAccountID(t *testing.T, fix *Fixture) string {
-	t.Helper()
-	iamEnsureAlice(t, fix)
-	iamAdminAccountOnce.Do(func() {
-		out, err := fix.AWS.IAM.GetUser(&iam.GetUserInput{UserName: aws.String(iamUserAlice)})
-		if err != nil {
-			iamAdminAccountErr = err
-			return
-		}
-		iamAdminAccountID = iamAccountFromARN(t, aws.StringValue(out.User.Arn))
-	})
-	if iamAdminAccountErr != nil {
-		t.Fatalf("iamEnsureAdminAccountID: %v", iamAdminAccountErr)
-	}
-	return iamAdminAccountID
-}
-
 // iamCreateUserIdempotent creates a user, treating EntityAlreadyExists as success.
 func iamCreateUserIdempotent(fix *Fixture, name, path string) error {
 	input := &iam.CreateUserInput{UserName: aws.String(name)}

tests/e2e/iam/iam_roles_test.go

@@ -1,6 +1,6 @@
 //go:build e2e
 
-package single
+package iam
 
 import (
 	"testing"
@@ -32,17 +32,17 @@ const (
 // which recreates the same names.
 func runIAMRolesAndProfiles(t *testing.T, fix *Fixture) {
 	harness.Phase(t, "Single — IAM Roles & Instance Profiles")
-	adminAccount := iamEnsureAdminAccountID(t, fix)
-	roleARN := iamRoleARN(adminAccount, iamRoleAppName)
-	adminPolicyARN := iamPolicyARN(adminAccount, iamPolicyAdministrator)
+	adminAccount := harness.IAMAccountID(t, fix.AWS)
+	roleARN := harness.IAMRoleARN(adminAccount, iamRoleAppName)
+	adminPolicyARN := harness.IAMPolicyARN(adminAccount, iamPolicyAdministrator)
 
 	// Defensive sweep — previous failed run may have left fragments. Order
 	// matters: detach role from profile before delete-instance-profile,
 	// detach policy from role before delete-role.
-	iamDeleteRoleAndProfilesBestEffort(fix, iamRoleAppName,
+	harness.IAMDeleteRoleAndProfilesBestEffort(fix.AWS, iamRoleAppName,
 		[]string{iamProfileAppName, iamProfileOtherName, iamProfileEngName}, adminPolicyARN)
 	fix.Harness.RegisterCleanup(func() {
-		iamDeleteRoleAndProfilesBestEffort(fix, iamRoleAppName,
+		harness.IAMDeleteRoleAndProfilesBestEffort(fix.AWS, iamRoleAppName,
 			[]string{iamProfileAppName, iamProfileOtherName, iamProfileEngName}, adminPolicyARN)
 	})
 
@@ -159,7 +159,7 @@ func runIAMRolesAndProfiles(t *testing.T, fix *Fixture) {
 	harness.ExpectError(t, "NoSuchEntity", func() error {
 		_, e := fix.AWS.IAM.AttachRolePolicy(&iam.AttachRolePolicyInput{
 			RoleName:  aws.String(iamRoleAppName),
-			PolicyArn: aws.String(iamPolicyARN(adminAccount, "Ghost")),
+			PolicyArn: aws.String(harness.IAMPolicyARN(adminAccount, "Ghost")),
 		})
 		return e
 	})
@@ -329,44 +329,3 @@ func runIAMRolesAndProfiles(t *testing.T, fix *Fixture) {
 		return e
 	})
 }
-
-// iamRoleARN constructs arn:aws:iam::<acct>:role/<name>.
-func iamRoleARN(account, name string) string {
-	return "arn:aws:iam::" + account + ":role/" + name
-}
-
-// iamDeleteRoleAndProfilesBestEffort tears down every fragment of a role +
-// profile graph the suite might have left behind. Each step swallows errors
-// so a missing fragment doesn't cascade. Used both as a pre-test sweep and
-// as a fixture-teardown cleanup.
-func iamDeleteRoleAndProfilesBestEffort(fix *Fixture, roleName string, profileNames []string, policyARNs ...string) {
-	for _, p := range profileNames {
-		// Drop role-from-profile binding (idempotent).
-		_, _ = fix.AWS.IAM.RemoveRoleFromInstanceProfile(&iam.RemoveRoleFromInstanceProfileInput{
-			InstanceProfileName: aws.String(p),
-			RoleName:            aws.String(roleName),
-		})
-		_, _ = fix.AWS.IAM.DeleteInstanceProfile(&iam.DeleteInstanceProfileInput{
-			InstanceProfileName: aws.String(p),
-		})
-	}
-	for _, arn := range policyARNs {
-		_, _ = fix.AWS.IAM.DetachRolePolicy(&iam.DetachRolePolicyInput{
-			RoleName:  aws.String(roleName),
-			PolicyArn: aws.String(arn),
-		})
-	}
-	// Defensive: pull any other attached policies we don't know about so the
-	// final DeleteRole isn't blocked by a stray attach from a partial run.
-	if attached, err := fix.AWS.IAM.ListAttachedRolePolicies(&iam.ListAttachedRolePoliciesInput{
-		RoleName: aws.String(roleName),
-	}); err == nil {
-		for _, p := range attached.AttachedPolicies {
-			_, _ = fix.AWS.IAM.DetachRolePolicy(&iam.DetachRolePolicyInput{
-				RoleName:  aws.String(roleName),
-				PolicyArn: p.PolicyArn,
-			})
-		}
-	}
-	_, _ = fix.AWS.IAM.DeleteRole(&iam.DeleteRoleInput{RoleName: aws.String(roleName)})
-}

tests/e2e/iam/iam_test.go

@@ -1,6 +1,6 @@
 //go:build e2e
 
-package single
+package iam
 
 import (
 	"strings"
@@ -81,24 +81,13 @@ const (
             "Resource": "*"
         }]
     }`
-
-	iamDocEC2ReadOnlyV2 = `{
-        "Version": "2012-10-17",
-        "Statement": [{
-            "Effect": "Allow",
-            "Action": ["ec2:DescribeInstances", "ec2:DescribeVolumes", "ec2:DescribeVpcs", "ec2:DescribeImages"],
-            "Resource": "*"
-        }]
-    }`
 )
 
 // runIAMUserCRUD ports run-e2e.sh IAM Phase 1 (~1405–1459):
-// CreateUser, GetUser, ListUsers, UpdateUser (rename+path), DeleteUser
-// plus the EntityAlreadyExists / NoSuchEntity negative paths. Each
-// created user is scheduled for cleanup so a mid-phase failure can't
-// poison later runs. UpdateUser exercises an out-of-band rename +
-// rename-back so Phase 2+ continue to find "alice" by the well-known
-// name.
+// CreateUser, GetUser, ListUsers, DeleteUser plus the
+// EntityAlreadyExists / NoSuchEntity negative paths. Each created user
+// is scheduled for cleanup so a mid-phase failure can't poison later
+// runs.
 func runIAMUserCRUD(t *testing.T, fix *Fixture) {
 	harness.Phase(t, "Single — IAM User CRUD")
 
@@ -169,34 +158,6 @@ func runIAMUserCRUD(t *testing.T, fix *Fixture) {
 	require.Len(t, eng.Users, 1,
 		"expected 1 user under %s, got %d (names=%v)", iamUserBobPath, len(eng.Users), iamUserNames(eng.Users))
 
-	// UpdateUser rename+path round-trip: daemon does not implement the
-	// UpdateUser action yet (returns InvalidAction). Skip-gate the block
-	// so the rest of Phase 1 still runs; flip to a real assertion once
-	// the daemon handler lands.
-	t.Run("UpdateUser", func(t *testing.T) {
-		t.Skip("daemon UpdateUser handler not implemented — InvalidAction")
-		const renamed = "alice-renamed"
-		const renamedPath = "/staff/"
-		harness.Step(t, "update-user %q -> %q path=%q", iamUserAlice, renamed, renamedPath)
-		_, err := fix.AWS.IAM.UpdateUser(&iam.UpdateUserInput{
-			UserName:    aws.String(iamUserAlice),
-			NewUserName: aws.String(renamed),
-			NewPath:     aws.String(renamedPath),
-		})
-		require.NoError(t, err, "update-user rename+path")
-		check, err := fix.AWS.IAM.GetUser(&iam.GetUserInput{UserName: aws.String(renamed)})
-		require.NoError(t, err, "get-user after rename")
-		assert.Equal(t, renamedPath, aws.StringValue(check.User.Path), "path not updated")
-
-		harness.Step(t, "update-user %q -> %q (restore)", renamed, iamUserAlice)
-		_, err = fix.AWS.IAM.UpdateUser(&iam.UpdateUserInput{
-			UserName:    aws.String(renamed),
-			NewUserName: aws.String(iamUserAlice),
-			NewPath:     aws.String("/"),
-		})
-		require.NoError(t, err, "update-user restore")
-	})
-
 	// DeleteUser idempotency: create a throwaway user, delete it, then
 	// delete it again and assert the second delete surfaces NoSuchEntity
 	// rather than silently succeeding (matches AWS semantics).
@@ -315,25 +276,28 @@ func runIAMAccessKeyLifecycle(t *testing.T, fix *Fixture) {
 	require.Len(t, after.AccessKeyMetadata, 1, "alice should have 1 key after delete")
 }
 
-// runIAMUserAuthentication ports IAM Phase 3 (~1538–1579):
-// build a scoped AWS client with alice's key and confirm signing
-// works (active key) / fails (deactivated key, bad secret, bogus ID).
-// Also creates bob's key so Phase 5 enforcement / Phase 7 cleanup
-// can use it.
+// runIAMUserAuthentication ports IAM Phase 3 (~1538–1579): build a scoped
+// AWS client with alice's key and confirm signing is honoured (active key)
+// / rejected (deactivated key, bad secret, bogus ID). Alice has no policy in
+// this phase, so the authenticated call is authz-denied — AccessDenied (not
+// an authn error) proves the active key signed and was accepted. Also creates
+// bob's key so Phase 5 enforcement / Phase 7 cleanup can use it.
 func runIAMUserAuthentication(t *testing.T, fix *Fixture) {
 	harness.Phase(t, "Single — IAM User Authentication")
-	// Daemon does not yet honour active-key signatures created via the IAM
-	// API — every scoped DescribeInstances returns 403.
-	t.Skip("daemon IAM signature/principal lookup not implemented")
 	aliceKeyID, aliceSecret := iamEnsureAliceKey(t, fix)
 
-	harness.Step(t, "scoped client (alice) describe-instances — active key OK")
+	// Active, correctly-signed key with no policy: the request authenticates
+	// then default-denies. AccessDenied (vs the InvalidClientTokenId /
+	// SignatureDoesNotMatch below) is the signal that authn succeeded.
+	harness.Step(t, "scoped client (alice) describe-instances — active key authenticates (AccessDenied, no policy)")
 	aliceCli := harness.NewAWSClientWithCreds(t, fix.Env, aliceKeyID, aliceSecret)
-	_, err := aliceCli.EC2.DescribeInstances(&ec2.DescribeInstancesInput{})
-	require.NoError(t, err, "alice describe-instances with active key")
+	harness.ExpectError(t, "AccessDenied", func() error {
+		_, e := aliceCli.EC2.DescribeInstances(&ec2.DescribeInstancesInput{})
+		return e
+	})
 
 	harness.Step(t, "deactivate alice key, expect InvalidClientTokenId")
-	_, err = fix.AWS.IAM.UpdateAccessKey(&iam.UpdateAccessKeyInput{
+	_, err := fix.AWS.IAM.UpdateAccessKey(&iam.UpdateAccessKeyInput{
 		UserName:    aws.String(iamUserAlice),
 		AccessKeyId: aws.String(aliceKeyID),
 		Status:      aws.String(iam.StatusTypeInactive),
@@ -375,11 +339,9 @@ func runIAMUserAuthentication(t *testing.T, fix *Fixture) {
 }
 
 // runIAMPolicyCRUD ports IAM Phase 4 (~1583–1698):
-// CreatePolicy (5 variants), CreatePolicyVersion / ListPolicyVersions /
-// SetDefaultPolicyVersion / DeletePolicyVersion (extends bash, per
-// task spec), GetPolicy, GetPolicyVersion, ListPolicies plus
-// EntityAlreadyExists / MalformedPolicyDocument / NoSuchEntity errors.
-// Captures the admin account ID into the Fixture for Phase 5–7.
+// CreatePolicy (5 variants), GetPolicy, GetPolicyVersion, ListPolicies
+// plus EntityAlreadyExists / MalformedPolicyDocument / NoSuchEntity
+// errors. Captures the admin account ID into the Fixture for Phase 5–7.
 func runIAMPolicyCRUD(t *testing.T, fix *Fixture) {
 	harness.Phase(t, "Single — IAM Policy CRUD")
 
@@ -416,7 +378,7 @@ func runIAMPolicyCRUD(t *testing.T, fix *Fixture) {
 			if p.path != "" {
 				key = p.path[1:] + p.name
 			}
-			iamDeletePolicyBestEffort(fix, iamPolicyARN(adminAccount, key))
+			iamDeletePolicyBestEffort(fix, harness.IAMPolicyARN(adminAccount, key))
 		}
 	})
 	harness.Detail(t, "policy", iamPolicyEC2ReadOnly, "arn", ec2roArn, "account", adminAccount)
@@ -490,7 +452,7 @@ func runIAMPolicyCRUD(t *testing.T, fix *Fixture) {
 	require.Equal(t, iamPolicyEC2ReadOnly, aws.StringValue(got.Policy.PolicyName))
 
 	harness.Step(t, "get-policy nonexistent (expect NoSuchEntity)")
-	ghostArn := iamPolicyARN(adminAccount, "Ghost")
+	ghostArn := harness.IAMPolicyARN(adminAccount, "Ghost")
 	harness.ExpectError(t, "NoSuchEntity", func() error {
 		_, e := fix.AWS.IAM.GetPolicy(&iam.GetPolicyInput{PolicyArn: aws.String(ghostArn)})
 		return e
@@ -504,47 +466,6 @@ func runIAMPolicyCRUD(t *testing.T, fix *Fixture) {
 	require.NoError(t, err, "get-policy-version v1")
 	require.Equal(t, "v1", aws.StringValue(pv.PolicyVersion.VersionId))
 
-	// Policy version lifecycle — extends the bash driver (task spec).
-	// Round-trip: create v2 (as default), list, set v1 default, delete v2.
-	// Daemon does not implement CreatePolicyVersion / ListPolicyVersions /
-	// SetDefaultPolicyVersion / DeletePolicyVersion yet (InvalidAction).
-	// Skip-gate until the handlers land.
-	t.Run("PolicyVersionLifecycle", func(t *testing.T) {
-		t.Skip("daemon CreatePolicyVersion handler not implemented — InvalidAction")
-		harness.Step(t, "create-policy-version %s v2 (default)", iamPolicyEC2ReadOnly)
-		cv, err := fix.AWS.IAM.CreatePolicyVersion(&iam.CreatePolicyVersionInput{
-			PolicyArn:      aws.String(ec2roArn),
-			PolicyDocument: aws.String(iamDocEC2ReadOnlyV2),
-			SetAsDefault:   aws.Bool(true),
-		})
-		require.NoError(t, err, "create-policy-version v2")
-		v2 := aws.StringValue(cv.PolicyVersion.VersionId)
-		require.NotEmpty(t, v2, "empty new VersionId")
-		harness.Detail(t, "new_version", v2)
-
-		harness.Step(t, "list-policy-versions %s", iamPolicyEC2ReadOnly)
-		versions, err := fix.AWS.IAM.ListPolicyVersions(&iam.ListPolicyVersionsInput{
-			PolicyArn: aws.String(ec2roArn),
-		})
-		require.NoError(t, err, "list-policy-versions")
-		require.GreaterOrEqual(t, len(versions.Versions), 2,
-			"expected >=2 versions, got %d", len(versions.Versions))
-
-		harness.Step(t, "set-default-policy-version %s v1", iamPolicyEC2ReadOnly)
-		_, err = fix.AWS.IAM.SetDefaultPolicyVersion(&iam.SetDefaultPolicyVersionInput{
-			PolicyArn: aws.String(ec2roArn),
-			VersionId: aws.String("v1"),
-		})
-		require.NoError(t, err, "set-default-policy-version v1")
-
-		harness.Step(t, "delete-policy-version %s %s", iamPolicyEC2ReadOnly, v2)
-		_, err = fix.AWS.IAM.DeletePolicyVersion(&iam.DeletePolicyVersionInput{
-			PolicyArn: aws.String(ec2roArn),
-			VersionId: aws.String(v2),
-		})
-		require.NoError(t, err, "delete-policy-version %s", v2)
-	})
-
 	harness.Step(t, "list-policies (>=5)")
 	all, err := fix.AWS.IAM.ListPolicies(&iam.ListPoliciesInput{})
 	require.NoError(t, err, "list-policies")
@@ -557,20 +478,17 @@ func runIAMPolicyCRUD(t *testing.T, fix *Fixture) {
 // (default deny / explicit allow / wildcard / explicit deny / root
 // bypass / prefix wildcard / FullAdmin), DetachUserPolicy.
 func runIAMPolicyAttachmentEnforcement(t *testing.T, fix *Fixture) {
-	harness.Phase(t, "Single — IAM Policy Attachment IAM Phase 5 — Policy Attachment & Enforcement Enforcement")
-	// Enforcement depends on scoped-credential signing which the daemon
-	// doesn't honour yet.
-	t.Skip("daemon IAM scoped-credential enforcement not implemented")
-	adminAccount := iamEnsureAdminAccountID(t, fix)
+	harness.Phase(t, "Single — IAM Policy Attachment & Enforcement")
+	adminAccount := harness.IAMAccountID(t, fix.AWS)
 	aliceKeyID, aliceSecret := iamEnsureAliceKey(t, fix)
 	bobKeyID, bobSecret := iamEnsureBobKey(t, fix)
 	charlieKeyID, charlieSecret := iamEnsureCharlieKey(t, fix)
 
-	ec2roArn := iamPolicyARN(adminAccount, iamPolicyEC2ReadOnly)
-	iamroArn := iamPolicyARN(adminAccount, iamPolicyIAMReadOnly)
-	denyArn := iamPolicyARN(adminAccount, iamPolicyDenyTerminate)
-	descAllArn := iamPolicyARN(adminAccount, iamPolicyEC2DescribeAll)
-	fullAdminArn := iamPolicyARN(adminAccount, iamPolicyFullAdminPath[1:]+iamPolicyFullAdmin)
+	ec2roArn := harness.IAMPolicyARN(adminAccount, iamPolicyEC2ReadOnly)
+	iamroArn := harness.IAMPolicyARN(adminAccount, iamPolicyIAMReadOnly)
+	denyArn := harness.IAMPolicyARN(adminAccount, iamPolicyDenyTerminate)
+	descAllArn := harness.IAMPolicyARN(adminAccount, iamPolicyEC2DescribeAll)
+	fullAdminArn := harness.IAMPolicyARN(adminAccount, iamPolicyFullAdminPath[1:]+iamPolicyFullAdmin)
 
 	harness.Detail(t, "charlie_key", charlieKeyID)
 	aliceCli := harness.NewAWSClientWithCreds(t, fix.Env, aliceKeyID, aliceSecret)
@@ -617,7 +535,7 @@ func runIAMPolicyAttachmentEnforcement(t *testing.T, fix *Fixture) {
 	harness.ExpectError(t, "NoSuchEntity", func() error {
 		_, e := fix.AWS.IAM.AttachUserPolicy(&iam.AttachUserPolicyInput{
 			UserName:  aws.String(iamUserAlice),
-			PolicyArn: aws.String(iamPolicyARN(adminAccount, "Ghost")),
+			PolicyArn: aws.String(harness.IAMPolicyARN(adminAccount, "Ghost")),
 		})
 		return e
 	})
@@ -730,15 +648,12 @@ func runIAMPolicyAttachmentEnforcement(t *testing.T, fix *Fixture) {
 // still-attached policy must surface DeleteConflict; after detach +
 // delete the policy is gone (NoSuchEntity).
 func runIAMPolicyLifecycle(t *testing.T, fix *Fixture) {
-	harness.Phase(t, "Single — IAM Policy Lifecycle (Detach IAM Phase 6 — Policy Lifecycle (Detach & Delete) Delete)")
-	// DetachUserPolicy returns 404 NoSuchEntity even for confirmed attachments —
-	// the daemon's attachment ledger isn't persisted across the API boundary.
-	t.Skip("daemon DetachUserPolicy not implemented")
-	adminAccount := iamEnsureAdminAccountID(t, fix)
+	harness.Phase(t, "Single — IAM Policy Lifecycle (Detach & Delete)")
+	adminAccount := harness.IAMAccountID(t, fix.AWS)
 	aliceKeyID, aliceSecret := iamEnsureAliceKey(t, fix)
 
-	descAllArn := iamPolicyARN(adminAccount, iamPolicyEC2DescribeAll)
-	denyArn := iamPolicyARN(adminAccount, iamPolicyDenyTerminate)
+	descAllArn := harness.IAMPolicyARN(adminAccount, iamPolicyEC2DescribeAll)
+	denyArn := harness.IAMPolicyARN(adminAccount, iamPolicyDenyTerminate)
 
 	aliceCli := harness.NewAWSClientWithCreds(t, fix.Env, aliceKeyID, aliceSecret)
 
@@ -785,7 +700,7 @@ func runIAMCleanup(t *testing.T, fix *Fixture) {
 
 	// Resolve account ID before deleting users — iamEnsureAdminAccountID
 	// probes via GetUser(alice), and the user is gone after the loop below.
-	adminAccount := iamEnsureAdminAccountID(t, fix)
+	adminAccount := harness.IAMAccountID(t, fix.AWS)
 
 	// Users: delete their keys first, then the user. iamDeleteUserBestEffort
 	// handles already-deleted resources gracefully.
@@ -819,7 +734,7 @@ func runIAMCleanup(t *testing.T, fix *Fixture) {
 			// stripping the leading '/' matches iamPolicyARN's expectation.
 			key = p.path[1:] + p.name
 		}
-		arn := iamPolicyARN(adminAccount, key)
+		arn := harness.IAMPolicyARN(adminAccount, key)
 		harness.Step(t, "cleanup policy %s", arn)
 		iamDeletePolicyBestEffort(fix, arn)
 	}
@@ -945,9 +860,3 @@ func iamAccountFromARN(t *testing.T, arn string) string {
 	require.NotEmpty(t, parts[4], "empty account-id in ARN %q", arn)
 	return parts[4]
 }
-
-// iamPolicyARN builds the canonical policy ARN for a given account +
-// policy "key" (name or "path/name" without the leading slash).
-func iamPolicyARN(account, key string) string {
-	return "arn:aws:iam::" + account + ":policy/" + key
-}

tests/e2e/iam/imds_test.go

@@ -1,6 +1,6 @@
 //go:build e2e
 
-package single
+package iam
 
 import (
 	"encoding/base64"
@@ -27,6 +27,12 @@ const (
 	imdsRoleName    = "imds-e2e-role"
 	imdsProfileName = "imds-e2e-profile"
 
+	// imdsAdminPolicyName is the bootstrap managed policy attached to the IMDS
+	// role so the minted instance-role creds round-trip a real API call.
+	imdsAdminPolicyName = "AdministratorAccess"
+	// imdsTrustPolicyEC2 lets ec2.amazonaws.com assume the IMDS instance role.
+	imdsTrustPolicyEC2 = `{"Version":"2012-10-17","Statement":[{"Effect":"Allow","Principal":{"Service":"ec2.amazonaws.com"},"Action":"sts:AssumeRole"}]}`
+
 	// metaURL is the AWS-compatible link-local IMDS endpoint, answered L2 on each
 	// subnet's own switch from the host via SO_BINDTODEVICE on the
 	// imds-h-<shortSubnetID> veth (no router, no proxy-ARP, no /32 route).
@@ -87,7 +93,7 @@ func runIMDS(t *testing.T, fix *Fixture) {
 	harness.SkipIfNoOVN(t)
 	requireSSHHealthy(t)
 
-	adminAccount := iamEnsureAdminAccountID(t, fix)
+	adminAccount := harness.IAMAccountID(t, fix.AWS)
 	keyName, keyPath := needKeyPair(t, fix)
 	imdsEnsureRoleProfile(t, fix, adminAccount)
 
@@ -525,16 +531,16 @@ func imdsAttachInternet(t *testing.T, c *harness.AWSClient, vpcID, subnetID stri
 // sweeps any residue from a prior run first.
 func imdsEnsureRoleProfile(t *testing.T, fix *Fixture, adminAccount string) {
 	t.Helper()
-	adminPolicyARN := iamPolicyARN(adminAccount, iamPolicyAdministrator)
-	iamDeleteRoleAndProfilesBestEffort(fix, imdsRoleName, []string{imdsProfileName}, adminPolicyARN)
+	adminPolicyARN := harness.IAMPolicyARN(adminAccount, imdsAdminPolicyName)
+	harness.IAMDeleteRoleAndProfilesBestEffort(fix.AWS, imdsRoleName, []string{imdsProfileName}, adminPolicyARN)
 	fix.Harness.RegisterCleanup(func() {
-		iamDeleteRoleAndProfilesBestEffort(fix, imdsRoleName, []string{imdsProfileName}, adminPolicyARN)
+		harness.IAMDeleteRoleAndProfilesBestEffort(fix.AWS, imdsRoleName, []string{imdsProfileName}, adminPolicyARN)
 	})
 
 	harness.Step(t, "create-role %q (trust=ec2.amazonaws.com) + profile %q", imdsRoleName, imdsProfileName)
 	_, err := fix.AWS.IAM.CreateRole(&iam.CreateRoleInput{
 		RoleName:                 aws.String(imdsRoleName),
-		AssumeRolePolicyDocument: aws.String(iamTrustPolicyEC2Standard),
+		AssumeRolePolicyDocument: aws.String(imdsTrustPolicyEC2),
 		Description:              aws.String("E2E IMDS instance-role credentials"),
 	})
 	require.NoError(t, err, "create-role")

tests/e2e/iam/instance_iam_profile_test.go

@@ -1,6 +1,6 @@
 //go:build e2e
 
-package single
+package iam
 
 import (
 	"strings"
@@ -14,24 +14,36 @@ import (
 	"github.com/stretchr/testify/require"
 )
 
+// IAM identifiers owned by this test. They use a single-iamprofile- namespace
+// distinct from the control-plane iam suite's app-role/app-profile so the two
+// suites can run concurrently against the same account without colliding on
+// account-flat IAM names.
+const (
+	singleIAMProfileRole        = "single-iamprofile-role"
+	singleIAMProfileApp         = "single-iamprofile-profile"
+	singleIAMProfileOther       = "single-iamprofile-other-profile"
+	singleIAMProfileAdminPolicy = "AdministratorAccess"
+	singleIAMProfileTrustPolicy = `{"Version":"2012-10-17","Statement":[{"Effect":"Allow","Principal":{"Service":"ec2.amazonaws.com"},"Action":"sts:AssumeRole"}]}`
+)
+
 // runIAMInstanceProfileAssociation ports run-e2e.sh IAM Phase 9: EC2 IAM
 // instance-profile association lifecycle. Reuses the singleton VM for the
 // post-launch Associate/Replace/Disassociate cycle, then launches a single
 // short-lived VM whose sole purpose is to exercise (a) RunInstances with
 // --iam-instance-profile and (b) auto-disassociate on terminate.
 //
-// Pre-state: TestIAMRolesAndProfiles already tore everything down — this
-// test recreates the role + two profiles, asserts the EC2 surface, then
-// cleans up so subsequent tests see no residue.
+// Self-bootstrapping: it creates its own role + two profiles under the
+// single-iamprofile- namespace, asserts the EC2 surface, then cleans up so
+// subsequent tests see no residue.
 func runIAMInstanceProfileAssociation(t *testing.T, fix *Fixture) {
 	harness.Phase(t, "Single — EC2 IAM Instance Profile Association")
 
-	adminAccount := iamEnsureAdminAccountID(t, fix)
-	adminPolicyARN := iamPolicyARN(adminAccount, iamPolicyAdministrator)
+	adminAccount := harness.IAMAccountID(t, fix.AWS)
+	adminPolicyARN := harness.IAMPolicyARN(adminAccount, singleIAMProfileAdminPolicy)
 
 	// Defensive cleanup of stale state from a prior run.
-	iamDeleteRoleAndProfilesBestEffort(fix, iamRoleAppName,
-		[]string{iamProfileAppName, iamProfileOtherName}, adminPolicyARN)
+	harness.IAMDeleteRoleAndProfilesBestEffort(fix.AWS, singleIAMProfileRole,
+		[]string{singleIAMProfileApp, singleIAMProfileOther}, adminPolicyARN)
 
 	// Singleton instance — reused for the post-launch Associate/Replace
 	// cycle. Must be running before we proceed.
@@ -43,27 +55,27 @@ func runIAMInstanceProfileAssociation(t *testing.T, fix *Fixture) {
 	// Build role + two profiles. The second profile is the Replace target.
 	harness.Step(t, "setup: create-role + 2 instance profiles")
 	_, err := fix.AWS.IAM.CreateRole(&iam.CreateRoleInput{
-		RoleName:                 aws.String(iamRoleAppName),
-		AssumeRolePolicyDocument: aws.String(iamTrustPolicyEC2Standard),
+		RoleName:                 aws.String(singleIAMProfileRole),
+		AssumeRolePolicyDocument: aws.String(singleIAMProfileTrustPolicy),
 	})
 	require.NoError(t, err, "create-role")
 
 	// Even though root bypasses iam:PassRole, attach AdministratorAccess
 	// to match production usage and exercise the PassRole code path.
 	_, err = fix.AWS.IAM.AttachRolePolicy(&iam.AttachRolePolicyInput{
-		RoleName:  aws.String(iamRoleAppName),
+		RoleName:  aws.String(singleIAMProfileRole),
 		PolicyArn: aws.String(adminPolicyARN),
 	})
 	require.NoError(t, err, "attach AdministratorAccess to role")
 
-	for _, p := range []string{iamProfileAppName, iamProfileOtherName} {
+	for _, p := range []string{singleIAMProfileApp, singleIAMProfileOther} {
 		_, err = fix.AWS.IAM.CreateInstanceProfile(&iam.CreateInstanceProfileInput{
 			InstanceProfileName: aws.String(p),
 		})
 		require.NoError(t, err, "create-instance-profile %s", p)
 		_, err = fix.AWS.IAM.AddRoleToInstanceProfile(&iam.AddRoleToInstanceProfileInput{
 			InstanceProfileName: aws.String(p),
-			RoleName:            aws.String(iamRoleAppName),
+			RoleName:            aws.String(singleIAMProfileRole),
 		})
 		require.NoError(t, err, "add-role-to-instance-profile %s", p)
 	}
@@ -86,15 +98,15 @@ func runIAMInstanceProfileAssociation(t *testing.T, fix *Fixture) {
 					&ec2.DisassociateIamInstanceProfileInput{AssociationId: a.AssociationId})
 			}
 		}
-		iamDeleteRoleAndProfilesBestEffort(fix, iamRoleAppName,
-			[]string{iamProfileAppName, iamProfileOtherName}, adminPolicyARN)
+		harness.IAMDeleteRoleAndProfilesBestEffort(fix.AWS, singleIAMProfileRole,
+			[]string{singleIAMProfileApp, singleIAMProfileOther}, adminPolicyARN)
 	})
 
 	// ---- Associate (post-launch) on singleton ----
-	harness.Step(t, "associate-iam-instance-profile %s -> %s", iamProfileAppName, singletonID)
+	harness.Step(t, "associate-iam-instance-profile %s -> %s", singleIAMProfileApp, singletonID)
 	assocOut, err := fix.AWS.EC2.AssociateIamInstanceProfile(&ec2.AssociateIamInstanceProfileInput{
 		InstanceId:         aws.String(singletonID),
-		IamInstanceProfile: &ec2.IamInstanceProfileSpecification{Name: aws.String(iamProfileAppName)},
+		IamInstanceProfile: &ec2.IamInstanceProfileSpecification{Name: aws.String(singleIAMProfileApp)},
 	})
 	require.NoError(t, err, "associate-iam-instance-profile")
 	require.NotNil(t, assocOut.IamInstanceProfileAssociation, "missing association in response")
@@ -108,7 +120,7 @@ func runIAMInstanceProfileAssociation(t *testing.T, fix *Fixture) {
 	harness.ExpectError(t, "IamInstanceProfileAlreadyAssociated", func() error {
 		_, e := fix.AWS.EC2.AssociateIamInstanceProfile(&ec2.AssociateIamInstanceProfileInput{
 			InstanceId:         aws.String(singletonID),
-			IamInstanceProfile: &ec2.IamInstanceProfileSpecification{Name: aws.String(iamProfileAppName)},
+			IamInstanceProfile: &ec2.IamInstanceProfileSpecification{Name: aws.String(singleIAMProfileApp)},
 		})
 		return e
 	})
@@ -124,7 +136,7 @@ func runIAMInstanceProfileAssociation(t *testing.T, fix *Fixture) {
 	ip := desc.Reservations[0].Instances[0].IamInstanceProfile
 	require.NotNil(t, ip, "DescribeInstances must populate IamInstanceProfile when bound")
 	require.True(t,
-		strings.HasSuffix(aws.StringValue(ip.Arn), ":instance-profile/"+iamProfileAppName),
+		strings.HasSuffix(aws.StringValue(ip.Arn), ":instance-profile/"+singleIAMProfileApp),
 		"unexpected IamInstanceProfile.Arn %q", aws.StringValue(ip.Arn))
 
 	// DescribeIamInstanceProfileAssociations by id.
@@ -143,17 +155,17 @@ func runIAMInstanceProfileAssociation(t *testing.T, fix *Fixture) {
 	harness.Step(t, "delete-instance-profile while bound (expect DeleteConflict)")
 	harness.ExpectError(t, "DeleteConflict", func() error {
 		_, e := fix.AWS.IAM.DeleteInstanceProfile(&iam.DeleteInstanceProfileInput{
-			InstanceProfileName: aws.String(iamProfileAppName),
+			InstanceProfileName: aws.String(singleIAMProfileApp),
 		})
 		return e
 	})
 
 	// Replace → other-profile, must mint a new assoc id.
-	harness.Step(t, "replace-iam-instance-profile-association -> %s", iamProfileOtherName)
+	harness.Step(t, "replace-iam-instance-profile-association -> %s", singleIAMProfileOther)
 	replaceOut, err := fix.AWS.EC2.ReplaceIamInstanceProfileAssociation(
 		&ec2.ReplaceIamInstanceProfileAssociationInput{
 			AssociationId:      aws.String(assocID),
-			IamInstanceProfile: &ec2.IamInstanceProfileSpecification{Name: aws.String(iamProfileOtherName)},
+			IamInstanceProfile: &ec2.IamInstanceProfileSpecification{Name: aws.String(singleIAMProfileOther)},
 		})
 	require.NoError(t, err, "replace-iam-instance-profile-association")
 	newAssocID := aws.StringValue(replaceOut.IamInstanceProfileAssociation.AssociationId)
@@ -166,7 +178,7 @@ func runIAMInstanceProfileAssociation(t *testing.T, fix *Fixture) {
 		_, e := fix.AWS.EC2.ReplaceIamInstanceProfileAssociation(
 			&ec2.ReplaceIamInstanceProfileAssociationInput{
 				AssociationId:      aws.String(assocID),
-				IamInstanceProfile: &ec2.IamInstanceProfileSpecification{Name: aws.String(iamProfileAppName)},
+				IamInstanceProfile: &ec2.IamInstanceProfileSpecification{Name: aws.String(singleIAMProfileApp)},
 			})
 		return e
 	})
@@ -200,7 +212,7 @@ func runIAMInstanceProfileAssociation(t *testing.T, fix *Fixture) {
 	// Reuse the singleton's launch attributes so we don't need to rediscover
 	// AMI / type / key / subnet / SG — guarantees the ephemeral VM lands on a
 	// known-good combination on the local single-node fixture.
-	ephID := runInstanceWithProfile(t, fix, inst, iamProfileAppName)
+	ephID := runInstanceWithProfile(t, fix, inst, singleIAMProfileApp)
 	defer func() {
 		// Belt-and-braces: even if the auto-disassociate assertion fails, we
 		// must not leave a VM running. Terminate is idempotent.

tests/e2e/iam/main_test.go

@@ -0,0 +1,108 @@
+//go:build e2e
+
+// Package iam is the standalone "everything IAM" E2E suite, split out of the
+// single-node suite so it builds and runs as its own iam.test binary. Every
+// entry point is a top-level Test* in tests_test.go. The control-plane phases
+// (IAM/STS CRUD) are zero-VM; the instance-identity tail (instance-profile
+// association + IMDS) boots guest VMs, so the suite is a single-mode,
+// VM-booting matrix leg covering the awsgw IAM/STS surface plus the
+// instance-profile + IMDS datapath.
+package iam
+
+import (
+	"os"
+	"sync"
+	"testing"
+
+	"github.com/mulgadc/spinifex/tests/e2e/harness"
+)
+
+// Package-scoped singleton fixture. Initialised lazily by the first call to
+// requireIAMFixture. TestMain drains its cleanup chain after m.Run() so every
+// resource ensured during the run is torn down at process exit, regardless of
+// which Test* created it.
+var (
+	pkgFixOnce sync.Once
+	pkgFix     *Fixture
+	pkgFixErr  error
+)
+
+// TestMain owns process-level lifecycle for the singleton fixture. The
+// singleton itself is built lazily by requireIAMFixture so a test run with
+// SPINIFEX_E2E unset stays cheap (no AWS dial, no temp dir).
+func TestMain(m *testing.M) {
+	code := m.Run()
+	if pkgFix != nil {
+		if pkgFix.Harness != nil {
+			pkgFix.Harness.Close()
+		}
+		if pkgFix.TmpDir != "" {
+			_ = os.RemoveAll(pkgFix.TmpDir)
+		}
+	}
+	os.Exit(code)
+}
+
+// requireIAMFixture returns the package-scoped Fixture singleton, building it on
+// first call. Skips the calling test if SPINIFEX_E2E is unset or the cluster
+// mode is not single-node. Fails the test if init itself errors (e.g. AWS
+// client construction). PoolMode is detected so the instance-identity tests can
+// gate their probe-VPC public-IP path.
+func requireIAMFixture(t *testing.T) *Fixture {
+	t.Helper()
+	pkgFixOnce.Do(func() {
+		if os.Getenv("SPINIFEX_E2E") == "" {
+			return
+		}
+		env := harness.LoadEnv(t)
+		if env.Mode != harness.ModeSingle {
+			return
+		}
+		awsCli := harness.NewAWSClient(t, env)
+		h, herr := harness.NewProcessFixture(awsCli)
+		if herr != nil {
+			pkgFixErr = herr
+			return
+		}
+		tmpDir, terr := os.MkdirTemp("", "iam-pkgfix-*")
+		if terr != nil {
+			pkgFixErr = terr
+			return
+		}
+		pkgFix = &Fixture{
+			Env:      env,
+			AWS:      awsCli,
+			Harness:  h,
+			TmpDir:   tmpDir,
+			PoolMode: detectPoolMode(env),
+		}
+	})
+	if pkgFixErr != nil {
+		t.Fatalf("iam fixture init failed: %v", pkgFixErr)
+	}
+	if pkgFix == nil {
+		t.Skip("iam fixture unavailable (SPINIFEX_E2E unset or mode != single)")
+	}
+	return pkgFix
+}
+
+// Fixture carries the per-process state shared across every Test* in this
+// package. Only environment-level slots — every per-phase resource ID is
+// memoized on Harness (harness.Fixture) or the package-local iamEnsure* helpers.
+type Fixture struct {
+	Env      *harness.Env
+	AWS      *harness.AWSClient
+	Harness  *harness.Fixture // memoized Ensure* fixture; spans the whole process.
+	TmpDir   string           // package-scoped scratch dir; survives every Test* in the package.
+	PoolMode bool             // external IPAM in play; gates the IMDS probe-VPC public-IP path.
+}
+
+// ArtifactDir returns the artifact directory for the *currently running* test.
+// It must be derived per-call from the live t (not memoized on the singleton
+// Fixture): the process fixture is built lazily during whichever test runs
+// first, so a stored dir would freeze to that test's name and every later test
+// would write into a stale — and, once that test passes, pruned — directory.
+func (f *Fixture) ArtifactDir(t *testing.T) string {
+	t.Helper()
+	return harness.ArtifactDir(t, f.Env)
+}

tests/e2e/iam/sts_test.go

@@ -1,6 +1,6 @@
 //go:build e2e
 
-package single
+package iam
 
 import (
 	"fmt"
@@ -54,7 +54,7 @@ const (
 // blocks (Condition / NotPrincipal / NotAction) via the SDK.
 func runSTS(t *testing.T, fix *Fixture) {
 	harness.Phase(t, "Single — STS AssumeRole + GetCallerIdentity")
-	adminAccount := iamEnsureAdminAccountID(t, fix)
+	adminAccount := harness.IAMAccountID(t, fix.AWS)
 
 	// Bootstrap-creds GetCallerIdentity — sanity check the live endpoint and
 	// capture the active ARN for the detail log so failures elsewhere are
@@ -151,11 +151,11 @@ func runSTS(t *testing.T, fix *Fixture) {
 		stsRoleNameForbidNotAction,
 	}
 	for _, name := range stsRoles {
-		iamDeleteRoleAndProfilesBestEffort(fix, name, nil)
+		harness.IAMDeleteRoleAndProfilesBestEffort(fix.AWS, name, nil)
 	}
 	fix.Harness.RegisterCleanup(func() {
 		for _, name := range stsRoles {
-			iamDeleteRoleAndProfilesBestEffort(fix, name, nil)
+			harness.IAMDeleteRoleAndProfilesBestEffort(fix.AWS, name, nil)
 		}
 	})
 
@@ -167,7 +167,7 @@ func runSTS(t *testing.T, fix *Fixture) {
 	})
 	require.NoError(t, err, "create-role")
 	roleARN := aws.StringValue(createOut.Role.Arn)
-	require.Equal(t, iamRoleARN(adminAccount, stsRoleName), roleARN,
+	require.Equal(t, harness.IAMRoleARN(adminAccount, stsRoleName), roleARN,
 		"role ARN must follow arn:aws:iam::<acct>:role/<name>")
 
 	// Happy path. Verify the wire-format invariants: ASIA prefix, non-empty
@@ -375,7 +375,7 @@ func runSTS(t *testing.T, fix *Fixture) {
 	// → trust-policy match → new ASIA mint) is not.
 	chainedTrustPolicy := fmt.Sprintf(
 		`{"Version":"2012-10-17","Statement":[{"Effect":"Allow","Principal":{"AWS":%q},"Action":"sts:AssumeRole"}]}`,
-		iamRoleARN(adminAccount, stsRoleName),
+		harness.IAMRoleARN(adminAccount, stsRoleName),
 	)
 	harness.Step(t, "create-role %q (trust=role/%s for chained assume)",
 		stsRoleNameChain, stsRoleName)
@@ -445,7 +445,7 @@ func runSTS(t *testing.T, fix *Fixture) {
 	harness.Step(t, "assume-role missing role (expect NoSuchEntity)")
 	harness.ExpectError(t, "NoSuchEntity", func() error {
 		_, e := fix.AWS.STS.AssumeRole(&sts.AssumeRoleInput{
-			RoleArn: aws.String(iamRoleARN(adminAccount,
+			RoleArn: aws.String(harness.IAMRoleARN(adminAccount,
 				"sts-e2e-no-such-role")),
 			RoleSessionName: aws.String("ghost"),
 		})

tests/e2e/iam/tests_test.go

@@ -0,0 +1,82 @@
+//go:build e2e
+
+package iam
+
+import "testing"
+
+// Top-level Test* entry points for the IAM/STS + instance-identity suite. Each
+// delegates to a runX function in the matching <name>_test.go file. Kept as
+// separate top-level Test* (not one inline Test) so `go test -run TestX` and the
+// suite selector can address each individually.
+//
+// Execution is sequential in source order: these tests share the package-scoped
+// IAM state (users/policies/roles are account-flat), and TestIAMCleanup tears
+// down the user/policy graph the earlier phases build. The parallelism win of
+// the split is the suite-as-matrix-leg, not intra-suite concurrency — sub-tests
+// inside each runIAM* still fan out with t.Parallel where it is safe.
+//
+// The final two entries (TestIAMInstanceProfileAssociation, TestIMDS) boot guest
+// VMs. They self-bootstrap their own roles/profiles/VPCs under collision-safe
+// namespaces and run last, after the control-plane phases.
+
+func TestIAMUserCRUD(t *testing.T) {
+	runIAMUserCRUD(t, requireIAMFixture(t))
+}
+
+func TestIAMAccessKeyLifecycle(t *testing.T) {
+	runIAMAccessKeyLifecycle(t, requireIAMFixture(t))
+}
+
+func TestIAMUserAuthentication(t *testing.T) {
+	runIAMUserAuthentication(t, requireIAMFixture(t))
+}
+
+func TestIAMPolicyCRUD(t *testing.T) {
+	runIAMPolicyCRUD(t, requireIAMFixture(t))
+}
+
+func TestIAMPolicyAttachmentEnforcement(t *testing.T) {
+	runIAMPolicyAttachmentEnforcement(t, requireIAMFixture(t))
+}
+
+func TestIAMPolicyLifecycle(t *testing.T) {
+	runIAMPolicyLifecycle(t, requireIAMFixture(t))
+}
+
+func TestIAMCleanup(t *testing.T) {
+	runIAMCleanup(t, requireIAMFixture(t))
+}
+
+func TestIAMRolesAndProfiles(t *testing.T) {
+	runIAMRolesAndProfiles(t, requireIAMFixture(t))
+}
+
+// TestSTSAssumeRoleAndGetCallerIdentity is sequential to avoid racing trust-policy
+// mutations against a parallel AssumeRole.
+func TestSTSAssumeRoleAndGetCallerIdentity(t *testing.T) {
+	runSTS(t, requireIAMFixture(t))
+}
+
+// TestAssumedRoleControlPlaneEnforcement verifies a zero-policy assumed-role is
+// denied and permitted once a policy is attached. Sequential to avoid racing
+// the mid-test grant.
+func TestAssumedRoleControlPlaneEnforcement(t *testing.T) {
+	runAssumedRoleControlPlaneEnforcement(t, requireIAMFixture(t))
+}
+
+// TestIAMInstanceProfileAssociation exercises the EC2 IAM instance-profile
+// association lifecycle (associate/replace/disassociate, RunInstances
+// --iam-instance-profile, auto-disassociate on terminate). It boots its own
+// instance and self-bootstraps a role + two profiles under the
+// single-iamprofile- namespace, distinct from this suite's app-role/app-profile.
+// Sequential: it boots VMs and mutates instance state.
+func TestIAMInstanceProfileAssociation(t *testing.T) {
+	runIAMInstanceProfileAssociation(t, requireIAMFixture(t))
+}
+
+// TestIMDS exercises IMDSv2 end-to-end: token issuance, metadata surface,
+// instance-role credentials, OVN datapath, and cross-VPC isolation. Sequential
+// because it launches profile-bound VMs and creates fresh VPCs (imds-e2e-*).
+func TestIMDS(t *testing.T) {
+	runIMDS(t, requireIAMFixture(t))
+}

tests/e2e/manifest-lint/baseline.txt

@@ -7,6 +7,13 @@ create	tests/e2e/eks/eks_test.go	CreateVpc	#1
 create	tests/e2e/gpu/gpu_test.go	RunInstances	#1
 create	tests/e2e/gpu/gpu_test.go	RunInstances	#2
 create	tests/e2e/gpu/helpers_test.go	RunInstances	#1
+create	tests/e2e/iam/iam_test.go	CreateKeyPair	#1
+create	tests/e2e/iam/imds_test.go	CreateInternetGateway	#1
+create	tests/e2e/iam/imds_test.go	CreateRouteTable	#1
+create	tests/e2e/iam/imds_test.go	CreateSubnet	#1
+create	tests/e2e/iam/imds_test.go	CreateVpc	#1
+create	tests/e2e/iam/imds_test.go	RunInstances	#1
+create	tests/e2e/iam/instance_iam_profile_test.go	RunInstances	#1
 create	tests/e2e/lb/lb_test.go	CreateInternetGateway	#1
 create	tests/e2e/lb/lb_test.go	CreateKeyPair	#1
 create	tests/e2e/lb/lb_test.go	CreateListener	#1
@@ -72,13 +79,6 @@ create	tests/e2e/single/baseline_test.go	CreateVpc	#1
 create	tests/e2e/single/baseline_test.go	RunInstances	#1
 create	tests/e2e/single/datapath_test.go	CreateSecurityGroup	#1
 create	tests/e2e/single/datapath_test.go	RunInstances	#1
-create	tests/e2e/single/iam_test.go	CreateKeyPair	#1
-create	tests/e2e/single/imds_test.go	CreateInternetGateway	#1
-create	tests/e2e/single/imds_test.go	CreateRouteTable	#1
-create	tests/e2e/single/imds_test.go	CreateSubnet	#1
-create	tests/e2e/single/imds_test.go	CreateVpc	#1
-create	tests/e2e/single/imds_test.go	RunInstances	#1
-create	tests/e2e/single/instance_iam_profile_test.go	RunInstances	#1
 create	tests/e2e/single/keypair_test.go	ImportKeyPair	#1
 create	tests/e2e/single/lifecycle_test.go	CreateVolume	#1
 create	tests/e2e/single/lifecycle_test.go	RunInstances	#1

tests/e2e/single/discovery_test.go

@@ -6,6 +6,7 @@ import (
 	"errors"
 	"fmt"
 	"os"
+	"strings"
 	"testing"
 	"time"
 
@@ -90,7 +91,9 @@ func runEnvironment(t *testing.T, fix *Fixture) {
 
 // runClusterStatsCLI exercises the spx CLI cluster surface (`get nodes`,
 // `top nodes`, `get vms`). Single-node only — multinode mode is tested by a
-// different scenario. Maps to run-e2e.sh ~95–127.
+// different scenario. The `get vms` baseline is concurrency-tolerant: it asserts
+// the empty state only when no instance rows are present, so suites booting VMs
+// in parallel (iam/lb) don't break it. Maps to run-e2e.sh ~95–127.
 func runClusterStatsCLI(t *testing.T, fix *Fixture) {
 	harness.Phase(t, "Single — Cluster Stats CLI")
 	if fix.Env.Mode != harness.ModeSingle {
@@ -108,10 +111,17 @@ func runClusterStatsCLI(t *testing.T, fix *Fixture) {
 	// actually rendered.
 	assert.Contains(t, top, "0/", "spx top nodes should report resource stats\n%s", top)
 
-	harness.Step(t, "spx get vms (empty before any launches)")
+	// spx get vms is node-wide: a hard "No VMs found" baseline breaks once the
+	// iam/lb suites boot VMs concurrently. Assert the empty state only when no
+	// instance rows are present; otherwise log (the CLI is still exercised).
+	harness.Step(t, "spx get vms (baseline; tolerant of concurrent suites)")
 	vms := harness.SpxGetVMs(t)
-	assert.Contains(t, vms, "No VMs found",
-		"spx get vms should be empty before Phase 5\n%s", vms)
+	if strings.Contains(vms, "i-") {
+		t.Logf("spx get vms lists instance-like rows at baseline (may be from concurrent suites):\n%s", vms)
+	} else {
+		assert.Contains(t, vms, "No VMs found",
+			"spx get vms with no instances present should report the empty state\n%s", vms)
+	}
 }
 
 // runDiscovery re-asserts describe-regions / describe-availability-zones

tests/e2e/single/tests_test.go

@@ -69,7 +69,8 @@ func TestReplaceRouteConvergence(t *testing.T) {
 
 // --- Sequential: singleton VM lifecycle ---
 
-// TestClusterStatsCLI asserts 0-VM baseline state before any instance launches.
+// TestClusterStatsCLI exercises the spx cluster CLI; its get-vms baseline is
+// concurrency-tolerant (other suites may have VMs on the node).
 func TestClusterStatsCLI(t *testing.T) {
 	runClusterStatsCLI(t, requireSingleNodeFixture(t))
 }
@@ -185,68 +186,6 @@ func TestSGToSGDatapath(t *testing.T) {
 	runSGToSGDatapath(t, requireSingleNodeFixture(t))
 }
 
-// IAM Tests share the package-scoped IAM fixture; sub-tests inside each
-// runIAM* already use t.Parallel for fan-out.
-func TestIAMUserCRUD(t *testing.T) {
-	runIAMUserCRUD(t, requireSingleNodeFixture(t))
-}
-
-func TestIAMAccessKeyLifecycle(t *testing.T) {
-	runIAMAccessKeyLifecycle(t, requireSingleNodeFixture(t))
-}
-
-func TestIAMUserAuthentication(t *testing.T) {
-	runIAMUserAuthentication(t, requireSingleNodeFixture(t))
-}
-
-func TestIAMPolicyCRUD(t *testing.T) {
-	runIAMPolicyCRUD(t, requireSingleNodeFixture(t))
-}
-
-func TestIAMPolicyAttachmentEnforcement(t *testing.T) {
-	runIAMPolicyAttachmentEnforcement(t, requireSingleNodeFixture(t))
-}
-
-func TestIAMPolicyLifecycle(t *testing.T) {
-	runIAMPolicyLifecycle(t, requireSingleNodeFixture(t))
-}
-
-func TestIAMCleanup(t *testing.T) {
-	runIAMCleanup(t, requireSingleNodeFixture(t))
-}
-
-// TestIAMRolesAndProfiles + TestIAMInstanceProfileAssociation are sequential:
-// the second recreates role/profile names the first tore down.
-func TestIAMRolesAndProfiles(t *testing.T) {
-	runIAMRolesAndProfiles(t, requireSingleNodeFixture(t))
-}
-
-// TestIAMInstanceProfileAssociation mutates the singleton VM, so it cannot
-// share the parallel bucket.
-func TestIAMInstanceProfileAssociation(t *testing.T) {
-	runIAMInstanceProfileAssociation(t, requireSingleNodeFixture(t))
-}
-
-// TestSTSAssumeRoleAndGetCallerIdentity is sequential to avoid racing trust-policy
-// mutations against a parallel AssumeRole.
-func TestSTSAssumeRoleAndGetCallerIdentity(t *testing.T) {
-	runSTS(t, requireSingleNodeFixture(t))
-}
-
-// TestAssumedRoleControlPlaneEnforcement verifies a zero-policy assumed-role
-// is denied and permitted once a policy is attached. Sequential to avoid
-// racing the mid-test grant.
-func TestAssumedRoleControlPlaneEnforcement(t *testing.T) {
-	runAssumedRoleControlPlaneEnforcement(t, requireSingleNodeFixture(t))
-}
-
-// TestIMDS exercises IMDSv2 end-to-end: token issuance, metadata surface,
-// instance-role credentials, OVN datapath, and cross-VPC isolation. Sequential
-// because it launches profile-bound VMs and creates a fresh VPC.
-func TestIMDS(t *testing.T) {
-	runIMDS(t, requireSingleNodeFixture(t))
-}
-
 // TestFinalClusterStats runs as the last sequential test.
 func TestFinalClusterStats(t *testing.T) {
 	runFinalClusterStats(t, requireSingleNodeFixture(t))