test(e2e): retire docker-unreachable gateway script

.github/workflows/regression-e2e.yaml

@@ -21,7 +21,7 @@ on:
       jobs:
         description: >-
           Comma-separated regression job names to run (empty = all).
-          Valid: dashboard-remote-bind-e2e,gateway-health-honest-e2e,docker-unreachable-gateway-start-e2e,gateway-drift-preflight-e2e,openshell-version-pin-e2e,onboard-inference-smoke-e2e,strict-tool-call-probe-e2e,model-router-provider-routed-inference-e2e,openclaw-plugin-runtime-exdev-e2e,whatsapp-qr-compact-e2e
+          Valid: dashboard-remote-bind-e2e,gateway-health-honest-e2e,gateway-drift-preflight-e2e,openshell-version-pin-e2e,onboard-inference-smoke-e2e,strict-tool-call-probe-e2e,model-router-provider-routed-inference-e2e,openclaw-plugin-runtime-exdev-e2e,whatsapp-qr-compact-e2e
         required: false
         type: string
         default: ""
@@ -46,7 +46,6 @@ jobs:
     outputs:
       dashboard: ${{ steps.select.outputs.dashboard }}
       gateway: ${{ steps.select.outputs.gateway }}
-      docker_unreachable_gateway_start: ${{ steps.select.outputs.docker_unreachable_gateway_start }}
       gateway_drift_preflight: ${{ steps.select.outputs.gateway_drift_preflight }}
       openshell_version_pin: ${{ steps.select.outputs.openshell_version_pin }}
       onboard_inference_smoke: ${{ steps.select.outputs.onboard_inference_smoke }}
@@ -81,12 +80,6 @@ jobs:
             echo "gateway=false" >> "$GITHUB_OUTPUT"
           fi
 
-          if [ -z "$normalized" ] || includes_job "docker-unreachable-gateway-start-e2e"; then
-            echo "docker_unreachable_gateway_start=true" >> "$GITHUB_OUTPUT"
-          else
-            echo "docker_unreachable_gateway_start=false" >> "$GITHUB_OUTPUT"
-          fi
-
           if [ -z "$normalized" ] || includes_job "gateway-drift-preflight-e2e"; then
             echo "gateway_drift_preflight=true" >> "$GITHUB_OUTPUT"
           else
@@ -194,44 +187,6 @@ jobs:
             /tmp/nemoclaw-e2e-gateway-health-honest-process.log
           if-no-files-found: ignore
 
-  # ── Docker-unreachable gateway-start E2E ───────────────────
-  # Coverage guard for #2347's fallback path. Preflight catches Docker
-  # daemon outages before step [2/8] on the normal path; this focused
-  # regression simulates Docker dying during `openshell gateway start` and
-  # verifies onboard exits before health polling with Docker recovery advice.
-  docker-unreachable-gateway-start-e2e:
-    needs: select_regression_jobs
-    if: >-
-      github.repository == 'NVIDIA/NemoClaw' &&
-      needs.select_regression_jobs.outputs.docker_unreachable_gateway_start == 'true'
-    runs-on: ubuntu-latest
-    timeout-minutes: 15
-    steps:
-      - name: Checkout
-        uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3
-
-      - name: Setup Node
-        uses: actions/setup-node@v6
-        with:
-          node-version: "22"
-
-      - name: Run Docker-unreachable gateway-start E2E test
-        env:
-          NEMOCLAW_NON_INTERACTIVE: "1"
-          NEMOCLAW_ACCEPT_THIRD_PARTY_SOFTWARE: "1"
-        run: bash test/e2e/test-docker-unreachable-gateway-start.sh
-
-      - name: Upload Docker-unreachable gateway-start logs on failure
-        if: failure()
-        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
-        with:
-          name: docker-unreachable-gateway-start-logs
-          path: |
-            /tmp/nemoclaw-e2e-docker-unreachable-gateway-start.log
-            /tmp/nemoclaw-e2e-docker-unreachable-gateway-start-node.log
-            /tmp/nemoclaw-e2e-docker-unreachable-gateway-start-openshell.log
-          if-no-files-found: ignore
-
   # ── OpenShell version-pin E2E ──────────────────────────────
   # Coverage guard for #3474. If a host has sticky OpenShell 0.0.45 on PATH
   # but this NemoClaw release supports only <=0.0.44, install-openshell.sh

ci/test-file-size-budget.json

@@ -11,7 +11,7 @@
     "test/nemoclaw-start.test.ts": 5289,
     "test/onboard-messaging.test.ts": 2097,
     "test/onboard-selection.test.ts": 6922,
-    "test/onboard.test.ts": 4874,
+    "test/onboard.test.ts": 4783,
     "test/policies.test.ts": 2763
   }
 }

src/lib/onboard/gateway-start-failure-integration.test.ts

@@ -5,31 +5,24 @@
 // startGatewayWithOptions() takes when `openshell gateway start` reports
 // the Docker daemon is not reachable. See src/lib/onboard.ts:2233.
 //
-// This file replaces test/e2e/test-docker-unreachable-gateway-start.sh,
-// which was structurally a Node-process unit test of startGateway() with a
-// PATH-shimmed openshell binary, not a sandbox-lifecycle e2e.
+// This helper-level suite preserves coverage from the former
+// test/e2e/test-docker-unreachable-gateway-start.sh, which was structurally
+// a Node-process unit test of startGateway() with a PATH-shimmed openshell
+// binary, not a sandbox-lifecycle e2e.
 //
 // Original regression: NemoClaw #2347.
 // Owning migration issue: NemoClaw #4355.
 //
-// Coverage strategy: prove the contract through two layers:
+// Coverage strategy: prove the helper-level contract through two layers:
 //
 //   1. Unit tests of the already-exported helpers (printDockerDaemonRecovery,
 //      handleFinalGatewayStartFailure with dockerUnreachable=true).
 //   2. A composition test that runs the same helper sequence the call site
 //      uses (classify → handleFinal → exitProcess(1)).
 //
-// What this file deliberately DOES NOT cover (gap documented as it.todo
-// at the bottom of the suite, with a follow-up issue tracking the
-// refactor): direct executable proof that startGatewayWithOptions, on a
-// docker-unreachable streamGatewayStart() result, (a) throws
-// pRetry.AbortError instead of retrying, (b) never logs "Waiting for
-// gateway health...", and (c) never calls openshell `status` or
-// `gateway info` probes. Closing that gap requires either mocking the
-// ~10 module-internal closures `startGatewayWithOptions` touches before
-// reaching streamGatewayStart, or extracting the inner pRetry async body
-// into an exported helper that takes streamGatewayStart as a DI
-// parameter. Both changes are out of scope for the retirement PR.
+// The caller-level process regression that drives startGateway() through a
+// PATH-shimmed openshell binary lives in
+// test/onboard-gateway-docker-unreachable.test.ts.
 
 import { describe, expect, it, vi } from "vitest";
 // `handleFinalGatewayStartFailure` is exposed via `module.exports = {...}` at
@@ -310,45 +303,6 @@ describe("startGatewayWithOptions docker-unreachable abort (#2347)", () => {
     });
   });
 
-  // ── Documented coverage gap: caller-level contracts that need a
-  //    behavior-level seam (out of scope for this retirement PR) ─────────
-  //
-  // The legacy bash script directly executed `startGateway()` with a
-  // PATH-shimmed openshell binary and asserted runtime behavior of the
-  // call site. The unit + composition tests above cover the same ground
-  // for everything reachable through already-exported helpers, but three
-  // call-site contracts cannot be proven without driving the actual
-  // `startGatewayWithOptions` orchestrator past its ~200 lines of
-  // gateway-reuse / ssh-keygen / known_hosts / docker-driver-detect
-  // preamble. Doing that in a focused way requires either:
-  //
-  //   (a) ~10 vi.mock() calls on module-internal closures inside
-  //       onboard.ts (brittle), or
-  //   (b) a small refactor extracting the inner pRetry async body of
-  //       startGatewayWithOptions into an exported helper that takes
-  //       streamGatewayStart as a DI parameter.
-  //
-  // Both are out of scope for this retirement PR. Follow-up issue
-  // [#5113](https://github.com/NVIDIA/NemoClaw/issues/5113) tracks
-  // landing option (b) and converting these `it.todo` placeholders into
-  // real assertions. Until then, the
-  // primary safety net for these contracts is code review on
-  // src/lib/onboard.ts:startGatewayWithOptions plus the existing
-  // classifyGatewayStartFailure / printDockerDaemonRecovery /
-  // handleFinalGatewayStartFailure unit tests above.
-
-  describe("call-site contracts (caller-level coverage gap)", () => {
-    it.todo(
-      "startGateway aborts via pRetry.AbortError without entering health-poll loop on docker-unreachable streamGatewayStart output",
-    );
-    it.todo(
-      "startGateway never invokes openshell `status` or `gateway info` after docker-unreachable streamGatewayStart output",
-    );
-    it.todo(
-      "startGateway forwards dockerUnreachable=true to handleFinalGatewayStartFailure (no doctor logs collection, no destroyGateway cleanup) when streamGatewayStart returns docker-unreachable signature",
-    );
-  });
-
   // ── Sanity: classifyGatewayStartFailure recognises both signatures ─────
   // (Already covered in gateway-start-failure.test.ts; this is a pinning
   // assertion for the two strings the legacy script generated, kept here so

test/e2e-scenario/migration/legacy-inventory.json

@@ -266,17 +266,6 @@
       "deletionReady": false,
       "notes": "Initial completeness row; classify detailed coverage and deletion evidence in the owning migration issue before deleting."
     },
-    {
-      "legacyScript": "test/e2e/test-docker-unreachable-gateway-start.sh",
-      "domain": "sandbox-lifecycle",
-      "ownerIssue": "#4355",
-      "status": "not-migrated",
-      "targetVitestScenarios": [],
-      "bridgeProbes": [],
-      "retiredReason": "",
-      "deletionReady": false,
-      "notes": "PR #5109 added Vitest helper coverage at src/lib/onboard/gateway-start-failure-integration.test.ts (two-layer strategy: unit tests of printDockerDaemonRecovery + handleFinalGatewayStartFailure, plus a composition test running the same classify -> handleFinal -> exitProcess(1) sequence the call site uses). Full retirement (delete this script + remove the docker-unreachable-gateway-start-e2e job in regression-e2e.yaml) was deferred per PR Review Advisor feedback because three caller-level contracts (no health-poll loop entry, no openshell status/gateway-info probes after gateway-start failure, dockerUnreachable forwarded to handleFinalGatewayStartFailure) are not directly proven without a behavior-level seam in startGatewayWithOptions. Follow-up issue #5113 tracks extracting attemptGatewayStart and converting the three it.todo placeholders into real assertions; deletion + retirement happen in a follow-up PR after that lands. Refs: #2347, #4355, #5113."
-    },
     {
       "legacyScript": "test/e2e/test-docs-validation.sh",
       "domain": "support-diagnostics",