test: stagger docker aggregate starts

steipete · steipete · commit 38caa6832d4e · 2026-04-24T08:38:41.000+01:00
diff --git a/docs/ci.md b/docs/ci.md
@@ -91,7 +91,7 @@ Jobs are ordered so cheap checks fail before expensive ones run:
 Scope logic lives in `scripts/ci-changed-scope.mjs` and is covered by unit tests in `src/scripts/ci-changed-scope.test.ts`.
 CI workflow edits validate the Node CI graph plus workflow linting, but do not force Windows, Android, or macOS native builds by themselves; those platform lanes stay scoped to platform source changes.
 Windows Node checks are scoped to Windows-specific process/path wrappers, npm/pnpm/UI runner helpers, package manager config, and the CI workflow surfaces that execute that lane; unrelated source, plugin, install-smoke, and test-only changes stay on the Linux Node lanes so they do not reserve a 16-vCPU Windows worker for coverage that is already exercised by the normal test shards.
-The separate `install-smoke` workflow reuses the same scope script through its own `preflight` job. It splits smoke coverage into `run_fast_install_smoke` and `run_full_install_smoke`. Pull requests run the fast path for Docker/package surfaces, bundled plugin package/manifest changes, and core plugin/channel/gateway/Plugin SDK surfaces that the Docker smoke jobs exercise. Source-only bundled plugin changes, test-only edits, and docs-only edits do not reserve Docker workers. The fast path builds the root Dockerfile image once, checks the CLI, runs the container gateway-network e2e, verifies a bundled extension build arg, and runs the bounded bundled-plugin Docker profile under a 120-second command timeout. The full path keeps QR package install and installer Docker/update coverage for nightly scheduled runs, manual dispatches, workflow-call release checks, and pull requests that truly touch installer/package/Docker surfaces. `main` pushes, including merge commits, do not force the full path; when changed-scope logic would request full coverage on a push, the workflow keeps the fast Docker smoke and leaves the full install smoke to nightly or release validation. The slow Bun global install image-provider smoke is separately gated by `run_bun_global_install_smoke`; it runs on the nightly schedule and from the release checks workflow, and manual `install-smoke` dispatches can opt into it, but pull requests and `main` pushes do not run it. QR and installer Docker tests keep their own install-focused Dockerfiles. Local `test:docker:all` prebuilds one shared live-test image and one shared `scripts/e2e/Dockerfile` built-app image, then runs the live/E2E smoke lanes in parallel with `OPENCLAW_SKIP_DOCKER_BUILD=1`; tune the default main-pool concurrency of 8 with `OPENCLAW_DOCKER_ALL_PARALLELISM` and the provider-sensitive tail-pool concurrency of 8 with `OPENCLAW_DOCKER_ALL_TAIL_PARALLELISM`. The local aggregate stops scheduling new pooled lanes after the first failure by default, and each lane has a 120-minute timeout overrideable with `OPENCLAW_DOCKER_ALL_LANE_TIMEOUT_MS`. The reusable live/E2E workflow mirrors the shared-image pattern by building and pushing one SHA-tagged GHCR Docker E2E image before the Docker matrix, then running the matrix with `OPENCLAW_SKIP_DOCKER_BUILD=1`. The scheduled live/E2E workflow runs the full release-path Docker suite daily. The full bundled update/channel matrix remains manual/full-suite because it performs repeated real npm update and doctor repair passes.
+The separate `install-smoke` workflow reuses the same scope script through its own `preflight` job. It splits smoke coverage into `run_fast_install_smoke` and `run_full_install_smoke`. Pull requests run the fast path for Docker/package surfaces, bundled plugin package/manifest changes, and core plugin/channel/gateway/Plugin SDK surfaces that the Docker smoke jobs exercise. Source-only bundled plugin changes, test-only edits, and docs-only edits do not reserve Docker workers. The fast path builds the root Dockerfile image once, checks the CLI, runs the container gateway-network e2e, verifies a bundled extension build arg, and runs the bounded bundled-plugin Docker profile under a 120-second command timeout. The full path keeps QR package install and installer Docker/update coverage for nightly scheduled runs, manual dispatches, workflow-call release checks, and pull requests that truly touch installer/package/Docker surfaces. `main` pushes, including merge commits, do not force the full path; when changed-scope logic would request full coverage on a push, the workflow keeps the fast Docker smoke and leaves the full install smoke to nightly or release validation. The slow Bun global install image-provider smoke is separately gated by `run_bun_global_install_smoke`; it runs on the nightly schedule and from the release checks workflow, and manual `install-smoke` dispatches can opt into it, but pull requests and `main` pushes do not run it. QR and installer Docker tests keep their own install-focused Dockerfiles. Local `test:docker:all` prebuilds one shared live-test image and one shared `scripts/e2e/Dockerfile` built-app image, then runs the live/E2E smoke lanes in parallel with `OPENCLAW_SKIP_DOCKER_BUILD=1`; tune the default main-pool concurrency of 8 with `OPENCLAW_DOCKER_ALL_PARALLELISM` and the provider-sensitive tail-pool concurrency of 8 with `OPENCLAW_DOCKER_ALL_TAIL_PARALLELISM`. Lane starts are staggered by 2 seconds by default to avoid local Docker daemon create storms; override with `OPENCLAW_DOCKER_ALL_START_STAGGER_MS=0` or another millisecond value. The local aggregate stops scheduling new pooled lanes after the first failure by default, and each lane has a 120-minute timeout overrideable with `OPENCLAW_DOCKER_ALL_LANE_TIMEOUT_MS`. The reusable live/E2E workflow mirrors the shared-image pattern by building and pushing one SHA-tagged GHCR Docker E2E image before the Docker matrix, then running the matrix with `OPENCLAW_SKIP_DOCKER_BUILD=1`. The scheduled live/E2E workflow runs the full release-path Docker suite daily. The full bundled update/channel matrix remains manual/full-suite because it performs repeated real npm update and doctor repair passes.
 
 Local changed-lane logic lives in `scripts/changed-lanes.mjs` and is executed by `scripts/check-changed.mjs`. That local gate is stricter about architecture boundaries than the broad CI platform scope: core production changes run core prod typecheck plus core tests, core test-only changes run only core test typecheck/tests, extension production changes run extension prod typecheck plus extension tests, and extension test-only changes run only extension test typecheck/tests. Public Plugin SDK or plugin-contract changes expand to extension validation because extensions depend on those core contracts. Release metadata-only version bumps run targeted version/config/root-dependency checks. Unknown root/config changes fail safe to all lanes.
 
diff --git a/docs/reference/test.md b/docs/reference/test.md
@@ -32,7 +32,7 @@ title: "Tests"
 - Gateway integration: opt-in via `OPENCLAW_TEST_INCLUDE_GATEWAY=1 pnpm test` or `pnpm test:gateway`.
 - `pnpm test:e2e`: Runs gateway end-to-end smoke tests (multi-instance WS/HTTP/node pairing). Defaults to `threads` + `isolate: false` with adaptive workers in `vitest.e2e.config.ts`; tune with `OPENCLAW_E2E_WORKERS=<n>` and set `OPENCLAW_E2E_VERBOSE=1` for verbose logs.
 - `pnpm test:live`: Runs provider live tests (minimax/zai). Requires API keys and `LIVE=1` (or provider-specific `*_LIVE_TEST=1`) to unskip.
-- `pnpm test:docker:all`: Builds the shared live-test image and Docker E2E image once, then runs the Docker smoke lanes with `OPENCLAW_SKIP_DOCKER_BUILD=1` at concurrency 8 by default. Tune the main pool with `OPENCLAW_DOCKER_ALL_PARALLELISM=<n>` and the provider-sensitive tail pool with `OPENCLAW_DOCKER_ALL_TAIL_PARALLELISM=<n>`; both default to 8. The runner stops scheduling new pooled lanes after the first failure unless `OPENCLAW_DOCKER_ALL_FAIL_FAST=0` is set, and each lane has a 120-minute timeout overrideable with `OPENCLAW_DOCKER_ALL_LANE_TIMEOUT_MS`. Per-lane logs are written under `.artifacts/docker-tests/<run-id>/`.
+- `pnpm test:docker:all`: Builds the shared live-test image and Docker E2E image once, then runs the Docker smoke lanes with `OPENCLAW_SKIP_DOCKER_BUILD=1` at concurrency 8 by default. Tune the main pool with `OPENCLAW_DOCKER_ALL_PARALLELISM=<n>` and the provider-sensitive tail pool with `OPENCLAW_DOCKER_ALL_TAIL_PARALLELISM=<n>`; both default to 8. Lane starts are staggered by 2 seconds by default to avoid local Docker daemon create storms; override with `OPENCLAW_DOCKER_ALL_START_STAGGER_MS=<ms>`. The runner stops scheduling new pooled lanes after the first failure unless `OPENCLAW_DOCKER_ALL_FAIL_FAST=0` is set, and each lane has a 120-minute timeout overrideable with `OPENCLAW_DOCKER_ALL_LANE_TIMEOUT_MS`. Per-lane logs are written under `.artifacts/docker-tests/<run-id>/`.
 - `pnpm test:docker:openwebui`: Starts Dockerized OpenClaw + Open WebUI, signs in through Open WebUI, checks `/api/models`, then runs a real proxied chat through `/api/chat/completions`. Requires a usable live model key (for example OpenAI in `~/.profile`), pulls an external Open WebUI image, and is not expected to be CI-stable like the normal unit/e2e suites.
 - `pnpm test:docker:mcp-channels`: Starts a seeded Gateway container and a second client container that spawns `openclaw mcp serve`, then verifies routed conversation discovery, transcript reads, attachment metadata, live event queue behavior, outbound send routing, and Claude-style channel + permission notifications over the real stdio bridge. The Claude notification assertion reads the raw stdio MCP frames directly so the smoke reflects what the bridge actually emits.
 
diff --git a/scripts/test-docker-all.mjs b/scripts/test-docker-all.mjs
@@ -10,6 +10,7 @@ const DEFAULT_PARALLELISM = 8;
 const DEFAULT_TAIL_PARALLELISM = 8;
 const DEFAULT_FAILURE_TAIL_LINES = 80;
 const DEFAULT_LANE_TIMEOUT_MS = 120 * 60 * 1000;
+const DEFAULT_LANE_START_STAGGER_MS = 2_000;
 
 const lanes = [
   ["live-models", "OPENCLAW_SKIP_DOCKER_BUILD=1 pnpm test:docker:live-models"],
@@ -68,13 +69,30 @@ function parsePositiveInt(raw, fallback, label) {
   return parsed;
 }
 
+function parseNonNegativeInt(raw, fallback, label) {
+  if (!raw) {
+    return fallback;
+  }
+  const parsed = Number(raw);
+  if (!Number.isInteger(parsed) || parsed < 0) {
+    throw new Error(`${label} must be a non-negative integer. Got: ${JSON.stringify(raw)}`);
+  }
+  return parsed;
+}
+
 function parseBool(raw, fallback) {
   if (raw === undefined || raw === "") {
     return fallback;
   }
   return !/^(?:0|false|no)$/i.test(raw);
 }
 
+function sleep(ms) {
+  return new Promise((resolve) => {
+    setTimeout(resolve, ms);
+  });
+}
+
 function utcStampForPath() {
   return new Date().toISOString().replaceAll("-", "").replaceAll(":", "").replace(/\..*$/, "Z");
 }
@@ -209,13 +227,34 @@ async function runLane(lane, baseEnv, logDir, timeoutMs) {
 async function runLanePool(poolLanes, baseEnv, logDir, parallelism, options) {
   const failures = [];
   let nextIndex = 0;
+  let lastLaneStartAt = 0;
+  let laneStartQueue = Promise.resolve();
+
+  async function waitForLaneStartSlot() {
+    if (options.startStaggerMs <= 0) {
+      return;
+    }
+    const previous = laneStartQueue;
+    let releaseQueue;
+    laneStartQueue = new Promise((resolve) => {
+      releaseQueue = resolve;
+    });
+    await previous;
+    const waitMs = Math.max(0, lastLaneStartAt + options.startStaggerMs - Date.now());
+    if (waitMs > 0) {
+      await sleep(waitMs);
+    }
+    lastLaneStartAt = Date.now();
+    releaseQueue();
+  }
 
   async function worker() {
     while (nextIndex < poolLanes.length) {
       if (options.failFast && failures.length > 0) {
         return;
       }
       const lane = poolLanes[nextIndex++];
+      await waitForLaneStartSlot();
       const result = await runLane(lane, baseEnv, logDir, options.timeoutMs);
       if (result.status !== 0) {
         failures.push(result);
@@ -285,6 +324,11 @@ async function main() {
     DEFAULT_LANE_TIMEOUT_MS,
     "OPENCLAW_DOCKER_ALL_LANE_TIMEOUT_MS",
   );
+  const laneStartStaggerMs = parseNonNegativeInt(
+    process.env.OPENCLAW_DOCKER_ALL_START_STAGGER_MS,
+    DEFAULT_LANE_START_STAGGER_MS,
+    "OPENCLAW_DOCKER_ALL_START_STAGGER_MS",
+  );
   const failFast = parseBool(process.env.OPENCLAW_DOCKER_ALL_FAIL_FAST, true);
   const runId = process.env.OPENCLAW_DOCKER_ALL_RUN_ID || utcStampForPath();
   const logDir = path.resolve(
@@ -304,6 +348,7 @@ async function main() {
   console.log(`==> Parallelism: ${parallelism}`);
   console.log(`==> Tail parallelism: ${tailParallelism}`);
   console.log(`==> Lane timeout: ${laneTimeoutMs}ms`);
+  console.log(`==> Lane start stagger: ${laneStartStaggerMs}ms`);
   console.log(`==> Fail fast: ${failFast ? "yes" : "no"}`);
   console.log(`==> Live-test bundled plugin deps: ${baseEnv.OPENCLAW_DOCKER_BUILD_EXTENSIONS}`);
 
@@ -314,7 +359,7 @@ async function main() {
     baseEnv,
   );
 
-  const options = { failFast, timeoutMs: laneTimeoutMs };
+  const options = { failFast, startStaggerMs: laneStartStaggerMs, timeoutMs: laneTimeoutMs };
   const failures = await runLanePool(lanes, baseEnv, logDir, parallelism, options);
   if (failFast && failures.length > 0) {
     await printFailureSummary(failures, tailLines);
