test: improve Docker aggregate scheduling

This commit is contained in:
Peter Steinberger
2026-04-24 23:44:00 +01:00
parent 99cfa50451
commit a7696b496a
5 changed files with 452 additions and 120 deletions

View File

@@ -1043,6 +1043,7 @@ run_update_scenario() {
if ! timeout "$DOCKER_RUN_TIMEOUT" docker run --rm \
-e COREPACK_ENABLE_DOWNLOAD_PROMPT=0 \
-e OPENCLAW_BUNDLED_CHANNEL_UPDATE_BASELINE_VERSION="$UPDATE_BASELINE_VERSION" \
-e "OPENCLAW_BUNDLED_CHANNEL_UPDATE_TARGETS=${OPENCLAW_BUNDLED_CHANNEL_UPDATE_TARGETS:-telegram,discord,slack,feishu,memory-lancedb,acpx}" \
"${PACKAGE_DOCKER_ARGS[@]}" \
-i "$IMAGE_NAME" bash -s >"$run_log" 2>&1 <<'EOF'
set -euo pipefail
@@ -1056,6 +1057,7 @@ export OPENCLAW_UPDATE_PACKAGE_SPEC=""
TOKEN="bundled-channel-update-token"
PORT="18790"
UPDATE_TARGETS="${OPENCLAW_BUNDLED_CHANNEL_UPDATE_TARGETS:-telegram,discord,slack,feishu,memory-lancedb,acpx}"
package_root() {
printf "%s/openclaw" "$(npm root -g)"
@@ -1291,7 +1293,16 @@ run_update_and_capture() {
fi
}
should_run_update_target() {
local target="$1"
case ",$UPDATE_TARGETS," in
*",all,"* | *",$target,"*) return 0 ;;
*) return 1 ;;
esac
}
echo "Installing current candidate as update baseline..."
echo "Update targets: $UPDATE_TARGETS"
npm install -g "$package_tgz" --no-fund --no-audit >/tmp/openclaw-update-baseline-install.log 2>&1
command -v openclaw >/dev/null
baseline_root="$(package_root)"
@@ -1299,76 +1310,88 @@ test -d "$baseline_root/dist/extensions/telegram"
test -d "$baseline_root/dist/extensions/feishu"
test -d "$baseline_root/dist/extensions/acpx"
echo "Replicating configured Telegram missing-runtime state..."
write_config telegram
assert_no_dep_available telegram grammy
set +e
openclaw doctor --non-interactive >/tmp/openclaw-baseline-doctor.log 2>&1
baseline_doctor_status=$?
set -e
echo "baseline doctor exited with $baseline_doctor_status"
remove_runtime_dep telegram grammy
assert_no_dep_available telegram grammy
if should_run_update_target telegram; then
echo "Replicating configured Telegram missing-runtime state..."
write_config telegram
assert_no_dep_available telegram grammy
set +e
openclaw doctor --non-interactive >/tmp/openclaw-baseline-doctor.log 2>&1
baseline_doctor_status=$?
set -e
echo "baseline doctor exited with $baseline_doctor_status"
remove_runtime_dep telegram grammy
assert_no_dep_available telegram grammy
echo "Updating from baseline to current candidate; candidate doctor must repair Telegram deps..."
run_update_and_capture telegram /tmp/openclaw-update-telegram.json
cat /tmp/openclaw-update-telegram.json
assert_update_ok /tmp/openclaw-update-telegram.json "$candidate_version"
assert_dep_available telegram grammy
echo "Updating from baseline to current candidate; candidate doctor must repair Telegram deps..."
run_update_and_capture telegram /tmp/openclaw-update-telegram.json
cat /tmp/openclaw-update-telegram.json
assert_update_ok /tmp/openclaw-update-telegram.json "$candidate_version"
assert_dep_available telegram grammy
echo "Mutating installed package: remove Telegram deps, then update-mode doctor repairs them..."
remove_runtime_dep telegram grammy
assert_no_dep_available telegram grammy
if ! OPENCLAW_UPDATE_IN_PROGRESS=1 openclaw doctor --non-interactive >/tmp/openclaw-update-mode-doctor.log 2>&1; then
echo "update-mode doctor failed while repairing Telegram deps" >&2
cat /tmp/openclaw-update-mode-doctor.log >&2
exit 1
echo "Mutating installed package: remove Telegram deps, then update-mode doctor repairs them..."
remove_runtime_dep telegram grammy
assert_no_dep_available telegram grammy
if ! OPENCLAW_UPDATE_IN_PROGRESS=1 openclaw doctor --non-interactive >/tmp/openclaw-update-mode-doctor.log 2>&1; then
echo "update-mode doctor failed while repairing Telegram deps" >&2
cat /tmp/openclaw-update-mode-doctor.log >&2
exit 1
fi
assert_dep_available telegram grammy
fi
assert_dep_available telegram grammy
echo "Mutating config to Discord and rerunning same-version update path..."
write_config discord
remove_runtime_dep discord discord-api-types
assert_no_dep_available discord discord-api-types
run_update_and_capture discord /tmp/openclaw-update-discord.json
cat /tmp/openclaw-update-discord.json
assert_update_ok /tmp/openclaw-update-discord.json "$candidate_version"
assert_dep_available discord discord-api-types
if should_run_update_target discord; then
echo "Mutating config to Discord and rerunning same-version update path..."
write_config discord
remove_runtime_dep discord discord-api-types
assert_no_dep_available discord discord-api-types
run_update_and_capture discord /tmp/openclaw-update-discord.json
cat /tmp/openclaw-update-discord.json
assert_update_ok /tmp/openclaw-update-discord.json "$candidate_version"
assert_dep_available discord discord-api-types
fi
echo "Mutating config to Slack and rerunning same-version update path..."
write_config slack
remove_runtime_dep slack @slack/web-api
assert_no_dep_available slack @slack/web-api
run_update_and_capture slack /tmp/openclaw-update-slack.json
cat /tmp/openclaw-update-slack.json
assert_update_ok /tmp/openclaw-update-slack.json "$candidate_version"
assert_dep_available slack @slack/web-api
if should_run_update_target slack; then
echo "Mutating config to Slack and rerunning same-version update path..."
write_config slack
remove_runtime_dep slack @slack/web-api
assert_no_dep_available slack @slack/web-api
run_update_and_capture slack /tmp/openclaw-update-slack.json
cat /tmp/openclaw-update-slack.json
assert_update_ok /tmp/openclaw-update-slack.json "$candidate_version"
assert_dep_available slack @slack/web-api
fi
echo "Mutating config to Feishu and rerunning same-version update path..."
write_config feishu
remove_runtime_dep feishu @larksuiteoapi/node-sdk
assert_no_dep_available feishu @larksuiteoapi/node-sdk
run_update_and_capture feishu /tmp/openclaw-update-feishu.json
cat /tmp/openclaw-update-feishu.json
assert_update_ok /tmp/openclaw-update-feishu.json "$candidate_version"
assert_dep_available feishu @larksuiteoapi/node-sdk
if should_run_update_target feishu; then
echo "Mutating config to Feishu and rerunning same-version update path..."
write_config feishu
remove_runtime_dep feishu @larksuiteoapi/node-sdk
assert_no_dep_available feishu @larksuiteoapi/node-sdk
run_update_and_capture feishu /tmp/openclaw-update-feishu.json
cat /tmp/openclaw-update-feishu.json
assert_update_ok /tmp/openclaw-update-feishu.json "$candidate_version"
assert_dep_available feishu @larksuiteoapi/node-sdk
fi
echo "Mutating config to memory-lancedb and rerunning same-version update path..."
write_config memory-lancedb
remove_runtime_dep memory-lancedb @lancedb/lancedb
assert_no_dep_available memory-lancedb @lancedb/lancedb
run_update_and_capture memory-lancedb /tmp/openclaw-update-memory-lancedb.json
cat /tmp/openclaw-update-memory-lancedb.json
assert_update_ok /tmp/openclaw-update-memory-lancedb.json "$candidate_version"
assert_dep_available memory-lancedb @lancedb/lancedb
if should_run_update_target memory-lancedb; then
echo "Mutating config to memory-lancedb and rerunning same-version update path..."
write_config memory-lancedb
remove_runtime_dep memory-lancedb @lancedb/lancedb
assert_no_dep_available memory-lancedb @lancedb/lancedb
run_update_and_capture memory-lancedb /tmp/openclaw-update-memory-lancedb.json
cat /tmp/openclaw-update-memory-lancedb.json
assert_update_ok /tmp/openclaw-update-memory-lancedb.json "$candidate_version"
assert_dep_available memory-lancedb @lancedb/lancedb
fi
echo "Removing ACPX runtime package and rerunning same-version update path..."
remove_runtime_dep acpx acpx
assert_no_dep_available acpx acpx
run_update_and_capture acpx /tmp/openclaw-update-acpx.json
cat /tmp/openclaw-update-acpx.json
assert_update_ok /tmp/openclaw-update-acpx.json "$candidate_version"
assert_dep_available acpx acpx
if should_run_update_target acpx; then
echo "Removing ACPX runtime package and rerunning same-version update path..."
remove_runtime_dep acpx acpx
assert_no_dep_available acpx acpx
run_update_and_capture acpx /tmp/openclaw-update-acpx.json
cat /tmp/openclaw-update-acpx.json
assert_update_ok /tmp/openclaw-update-acpx.json "$candidate_version"
assert_dep_available acpx acpx
fi
echo "bundled channel runtime deps Docker update E2E passed"
EOF

View File

@@ -11,6 +11,14 @@ const DEFAULT_TAIL_PARALLELISM = 10;
const DEFAULT_FAILURE_TAIL_LINES = 80;
const DEFAULT_LANE_TIMEOUT_MS = 120 * 60 * 1000;
const DEFAULT_LANE_START_STAGGER_MS = 2_000;
const DEFAULT_STATUS_INTERVAL_MS = 30_000;
const DEFAULT_PREFLIGHT_RUN_TIMEOUT_MS = 60_000;
const DEFAULT_TIMINGS_FILE = path.join(ROOT_DIR, ".artifacts/docker-tests/lane-timings.json");
const LIVE_PROFILE_TIMEOUT_MS = 20 * 60 * 1000;
const LIVE_CLI_TIMEOUT_MS = 12 * 60 * 1000;
const LIVE_ACP_TIMEOUT_MS = 20 * 60 * 1000;
const OPENWEBUI_TIMEOUT_MS = 20 * 60 * 1000;
const BUNDLED_UPDATE_TIMEOUT_MS = 20 * 60 * 1000;
const DEFAULT_RESOURCE_LIMITS = {
docker: DEFAULT_PARALLELISM,
live: 4,
@@ -24,14 +32,17 @@ const bundledChannelLaneCommand =
function lane(name, command, options = {}) {
return {
command,
estimateSeconds: options.estimateSeconds,
name,
resources: options.resources ?? [],
timeoutMs: options.timeoutMs,
weight: options.weight ?? 1,
};
}
function liveLane(name, command, options = {}) {
return lane(name, command, {
...options,
resources: ["live", ...(options.resources ?? [])],
weight: options.weight ?? 3,
});
@@ -39,6 +50,7 @@ function liveLane(name, command, options = {}) {
function npmLane(name, command, options = {}) {
return lane(name, command, {
...options,
resources: ["npm", ...(options.resources ?? [])],
weight: options.weight ?? 2,
});
@@ -46,6 +58,7 @@ function npmLane(name, command, options = {}) {
function serviceLane(name, command, options = {}) {
return lane(name, command, {
...options,
resources: ["service", ...(options.resources ?? [])],
weight: options.weight ?? 2,
});
@@ -70,8 +83,34 @@ const bundledScenarioLanes = [
`OPENCLAW_BUNDLED_CHANNELS=memory-lancedb ${bundledChannelLaneCommand}`,
),
npmLane(
"bundled-channel-update",
"OPENCLAW_SKIP_DOCKER_BUILD=1 OPENCLAW_BUNDLED_CHANNEL_SCENARIOS=0 OPENCLAW_BUNDLED_CHANNEL_UPDATE_SCENARIO=1 OPENCLAW_BUNDLED_CHANNEL_ROOT_OWNED_SCENARIO=0 OPENCLAW_BUNDLED_CHANNEL_SETUP_ENTRY_SCENARIO=0 OPENCLAW_BUNDLED_CHANNEL_LOAD_FAILURE_SCENARIO=0 OPENCLAW_BUNDLED_CHANNEL_DISABLED_CONFIG_SCENARIO=0 pnpm test:docker:bundled-channel-deps",
"bundled-channel-update-telegram",
"OPENCLAW_SKIP_DOCKER_BUILD=1 OPENCLAW_BUNDLED_CHANNEL_SCENARIOS=0 OPENCLAW_BUNDLED_CHANNEL_UPDATE_SCENARIO=1 OPENCLAW_BUNDLED_CHANNEL_UPDATE_TARGETS=telegram OPENCLAW_BUNDLED_CHANNEL_ROOT_OWNED_SCENARIO=0 OPENCLAW_BUNDLED_CHANNEL_SETUP_ENTRY_SCENARIO=0 OPENCLAW_BUNDLED_CHANNEL_LOAD_FAILURE_SCENARIO=0 OPENCLAW_BUNDLED_CHANNEL_DISABLED_CONFIG_SCENARIO=0 pnpm test:docker:bundled-channel-deps",
{ timeoutMs: BUNDLED_UPDATE_TIMEOUT_MS },
),
npmLane(
"bundled-channel-update-discord",
"OPENCLAW_SKIP_DOCKER_BUILD=1 OPENCLAW_BUNDLED_CHANNEL_SCENARIOS=0 OPENCLAW_BUNDLED_CHANNEL_UPDATE_SCENARIO=1 OPENCLAW_BUNDLED_CHANNEL_UPDATE_TARGETS=discord OPENCLAW_BUNDLED_CHANNEL_ROOT_OWNED_SCENARIO=0 OPENCLAW_BUNDLED_CHANNEL_SETUP_ENTRY_SCENARIO=0 OPENCLAW_BUNDLED_CHANNEL_LOAD_FAILURE_SCENARIO=0 OPENCLAW_BUNDLED_CHANNEL_DISABLED_CONFIG_SCENARIO=0 pnpm test:docker:bundled-channel-deps",
{ timeoutMs: BUNDLED_UPDATE_TIMEOUT_MS },
),
npmLane(
"bundled-channel-update-slack",
"OPENCLAW_SKIP_DOCKER_BUILD=1 OPENCLAW_BUNDLED_CHANNEL_SCENARIOS=0 OPENCLAW_BUNDLED_CHANNEL_UPDATE_SCENARIO=1 OPENCLAW_BUNDLED_CHANNEL_UPDATE_TARGETS=slack OPENCLAW_BUNDLED_CHANNEL_ROOT_OWNED_SCENARIO=0 OPENCLAW_BUNDLED_CHANNEL_SETUP_ENTRY_SCENARIO=0 OPENCLAW_BUNDLED_CHANNEL_LOAD_FAILURE_SCENARIO=0 OPENCLAW_BUNDLED_CHANNEL_DISABLED_CONFIG_SCENARIO=0 pnpm test:docker:bundled-channel-deps",
{ timeoutMs: BUNDLED_UPDATE_TIMEOUT_MS },
),
npmLane(
"bundled-channel-update-feishu",
"OPENCLAW_SKIP_DOCKER_BUILD=1 OPENCLAW_BUNDLED_CHANNEL_SCENARIOS=0 OPENCLAW_BUNDLED_CHANNEL_UPDATE_SCENARIO=1 OPENCLAW_BUNDLED_CHANNEL_UPDATE_TARGETS=feishu OPENCLAW_BUNDLED_CHANNEL_ROOT_OWNED_SCENARIO=0 OPENCLAW_BUNDLED_CHANNEL_SETUP_ENTRY_SCENARIO=0 OPENCLAW_BUNDLED_CHANNEL_LOAD_FAILURE_SCENARIO=0 OPENCLAW_BUNDLED_CHANNEL_DISABLED_CONFIG_SCENARIO=0 pnpm test:docker:bundled-channel-deps",
{ timeoutMs: BUNDLED_UPDATE_TIMEOUT_MS },
),
npmLane(
"bundled-channel-update-memory-lancedb",
"OPENCLAW_SKIP_DOCKER_BUILD=1 OPENCLAW_BUNDLED_CHANNEL_SCENARIOS=0 OPENCLAW_BUNDLED_CHANNEL_UPDATE_SCENARIO=1 OPENCLAW_BUNDLED_CHANNEL_UPDATE_TARGETS=memory-lancedb OPENCLAW_BUNDLED_CHANNEL_ROOT_OWNED_SCENARIO=0 OPENCLAW_BUNDLED_CHANNEL_SETUP_ENTRY_SCENARIO=0 OPENCLAW_BUNDLED_CHANNEL_LOAD_FAILURE_SCENARIO=0 OPENCLAW_BUNDLED_CHANNEL_DISABLED_CONFIG_SCENARIO=0 pnpm test:docker:bundled-channel-deps",
{ timeoutMs: BUNDLED_UPDATE_TIMEOUT_MS },
),
npmLane(
"bundled-channel-update-acpx",
"OPENCLAW_SKIP_DOCKER_BUILD=1 OPENCLAW_BUNDLED_CHANNEL_SCENARIOS=0 OPENCLAW_BUNDLED_CHANNEL_UPDATE_SCENARIO=1 OPENCLAW_BUNDLED_CHANNEL_UPDATE_TARGETS=acpx OPENCLAW_BUNDLED_CHANNEL_ROOT_OWNED_SCENARIO=0 OPENCLAW_BUNDLED_CHANNEL_SETUP_ENTRY_SCENARIO=0 OPENCLAW_BUNDLED_CHANNEL_LOAD_FAILURE_SCENARIO=0 OPENCLAW_BUNDLED_CHANNEL_DISABLED_CONFIG_SCENARIO=0 pnpm test:docker:bundled-channel-deps",
{ timeoutMs: BUNDLED_UPDATE_TIMEOUT_MS },
),
npmLane(
"bundled-channel-root-owned",
@@ -93,22 +132,25 @@ const bundledScenarioLanes = [
const lanes = [
liveLane("live-models", "OPENCLAW_SKIP_DOCKER_BUILD=1 pnpm test:docker:live-models", {
timeoutMs: LIVE_PROFILE_TIMEOUT_MS,
weight: 4,
}),
liveLane("live-gateway", "OPENCLAW_SKIP_DOCKER_BUILD=1 pnpm test:docker:live-gateway", {
timeoutMs: LIVE_PROFILE_TIMEOUT_MS,
weight: 4,
}),
liveLane(
"live-cli-backend-claude",
"OPENCLAW_SKIP_DOCKER_BUILD=1 pnpm test:docker:live-cli-backend:claude",
{ resources: ["npm"], weight: 3 },
{ resources: ["npm"], timeoutMs: LIVE_CLI_TIMEOUT_MS, weight: 3 },
),
liveLane(
"live-cli-backend-gemini",
"OPENCLAW_SKIP_DOCKER_BUILD=1 pnpm test:docker:live-cli-backend:gemini",
{ resources: ["npm"], weight: 3 },
{ resources: ["npm"], timeoutMs: LIVE_CLI_TIMEOUT_MS, weight: 3 },
),
serviceLane("openwebui", "OPENCLAW_SKIP_DOCKER_BUILD=1 pnpm test:docker:openwebui", {
timeoutMs: OPENWEBUI_TIMEOUT_MS,
weight: 5,
}),
serviceLane("onboard", "OPENCLAW_SKIP_DOCKER_BUILD=1 pnpm test:docker:onboard", {
@@ -145,35 +187,37 @@ const exclusiveLanes = [
serviceLane(
"openai-web-search-minimal",
"OPENCLAW_SKIP_DOCKER_BUILD=1 pnpm test:docker:openai-web-search-minimal",
{ timeoutMs: 8 * 60 * 1000 },
),
liveLane(
"live-codex-harness",
"OPENCLAW_SKIP_DOCKER_BUILD=1 pnpm test:docker:live-codex-harness",
{ resources: ["npm"], weight: 3 },
{ resources: ["npm"], timeoutMs: LIVE_ACP_TIMEOUT_MS, weight: 3 },
),
liveLane("live-codex-bind", "OPENCLAW_SKIP_DOCKER_BUILD=1 pnpm test:docker:live-codex-bind", {
resources: ["npm"],
timeoutMs: LIVE_ACP_TIMEOUT_MS,
weight: 3,
}),
liveLane(
"live-cli-backend-codex",
"OPENCLAW_SKIP_DOCKER_BUILD=1 pnpm test:docker:live-cli-backend:codex",
{ resources: ["npm"], weight: 3 },
{ resources: ["npm"], timeoutMs: LIVE_CLI_TIMEOUT_MS, weight: 3 },
),
liveLane(
"live-acp-bind-claude",
"OPENCLAW_SKIP_DOCKER_BUILD=1 pnpm test:docker:live-acp-bind:claude",
{ resources: ["npm"], weight: 3 },
{ resources: ["npm"], timeoutMs: LIVE_ACP_TIMEOUT_MS, weight: 3 },
),
liveLane(
"live-acp-bind-codex",
"OPENCLAW_SKIP_DOCKER_BUILD=1 pnpm test:docker:live-acp-bind:codex",
{ resources: ["npm"], weight: 3 },
{ resources: ["npm"], timeoutMs: LIVE_ACP_TIMEOUT_MS, weight: 3 },
),
liveLane(
"live-acp-bind-gemini",
"OPENCLAW_SKIP_DOCKER_BUILD=1 pnpm test:docker:live-acp-bind:gemini",
{ resources: ["npm"], weight: 3 },
{ resources: ["npm"], timeoutMs: LIVE_ACP_TIMEOUT_MS, weight: 3 },
),
];
@@ -240,7 +284,8 @@ function laneResources(poolLane) {
function laneSummary(poolLane) {
const resources = laneResources(poolLane).join(",");
return `${poolLane.name}(w=${laneWeight(poolLane)} r=${resources})`;
const timeout = poolLane.timeoutMs ? ` timeout=${Math.round(poolLane.timeoutMs / 1000)}s` : "";
return `${poolLane.name}(w=${laneWeight(poolLane)} r=${resources}${timeout})`;
}
function sleep(ms) {
@@ -277,6 +322,91 @@ function shellQuote(value) {
return `'${String(value).replaceAll("'", "'\\''")}'`;
}
function timingSeconds(timingStore, poolLane) {
const fromStore = timingStore?.lanes?.[poolLane.name]?.durationSeconds;
if (typeof fromStore === "number" && Number.isFinite(fromStore) && fromStore > 0) {
return fromStore;
}
return poolLane.estimateSeconds ?? 0;
}
function orderLanes(poolLanes, timingStore) {
return poolLanes
.map((poolLane, index) => ({ index, poolLane, seconds: timingSeconds(timingStore, poolLane) }))
.toSorted((a, b) => b.seconds - a.seconds || a.index - b.index)
.map(({ poolLane }) => poolLane);
}
async function loadTimingStore(file, enabled) {
if (!enabled) {
return { enabled: false, file, lanes: {}, version: 1 };
}
const raw = await readFile(file, "utf8").catch(() => "");
if (!raw.trim()) {
return { enabled: true, file, lanes: {}, version: 1 };
}
try {
const parsed = JSON.parse(raw);
return {
enabled: true,
file,
lanes: parsed && typeof parsed.lanes === "object" && parsed.lanes ? parsed.lanes : {},
version: 1,
};
} catch (error) {
console.warn(
`WARN: ignoring unreadable Docker lane timings ${file}: ${
error instanceof Error ? error.message : String(error)
}`,
);
return { enabled: true, file, lanes: {}, version: 1 };
}
}
async function writeTimingStore(timingStore, results) {
if (!timingStore.enabled || results.length === 0) {
return;
}
const next = {
lanes: { ...timingStore.lanes },
updatedAt: new Date().toISOString(),
version: 1,
};
for (const result of results) {
if (!result || typeof result.elapsedSeconds !== "number") {
continue;
}
next.lanes[result.name] = {
durationSeconds: result.elapsedSeconds,
status: result.status,
timedOut: result.timedOut,
updatedAt: new Date().toISOString(),
};
}
await mkdir(path.dirname(timingStore.file), { recursive: true });
await fs.promises.writeFile(timingStore.file, `${JSON.stringify(next, null, 2)}\n`);
timingStore.lanes = next.lanes;
console.log(`==> Docker lane timings: ${timingStore.file}`);
}
function printLaneManifest(label, poolLanes, timingStore) {
console.log(`==> ${label} lanes (${poolLanes.length})`);
for (const [index, poolLane] of poolLanes.entries()) {
const seconds = timingSeconds(timingStore, poolLane);
const estimate = seconds > 0 ? ` last=${Math.round(seconds)}s` : "";
console.log(` ${index + 1}. ${laneSummary(poolLane)}${estimate}`);
}
}
function dockerPreflightContainerNames(raw) {
return raw
.split(/\r?\n/)
.map((line) => line.trim().split(/\s+/, 1)[0])
.filter((name) =>
/^(?:openclaw-(?:gateway-e2e|openwebui|openwebui-gateway|config-reload-e2e)-)/.test(name),
);
}
function runShellCommand({ command, env, label, logFile, timeoutMs }) {
return new Promise((resolve) => {
const child = spawn("bash", ["-lc", command], {
@@ -329,6 +459,44 @@ function runShellCommand({ command, env, label, logFile, timeoutMs }) {
});
}
function runShellCaptureCommand({ command, env, label, timeoutMs }) {
return new Promise((resolve) => {
const child = spawn("bash", ["-lc", command], {
cwd: ROOT_DIR,
detached: process.platform !== "win32",
env,
stdio: ["ignore", "pipe", "pipe"],
});
activeChildren.add(child);
let stdout = "";
let stderr = "";
let timedOut = false;
const timeoutTimer =
timeoutMs > 0
? setTimeout(() => {
timedOut = true;
terminateChild(child, "SIGTERM");
setTimeout(() => terminateChild(child, "SIGKILL"), 10_000).unref?.();
}, timeoutMs)
: undefined;
timeoutTimer?.unref?.();
child.stdout.on("data", (chunk) => {
stdout += String(chunk);
});
child.stderr.on("data", (chunk) => {
stderr += String(chunk);
});
child.on("close", (status, signal) => {
if (timeoutTimer) {
clearTimeout(timeoutTimer);
}
activeChildren.delete(child);
const exitCode = typeof status === "number" ? status : signal ? 128 : 1;
resolve({ label, signal, status: exitCode, stderr, stdout, timedOut });
});
});
}
async function runForeground(label, command, env) {
console.log(`==> ${label}`);
const result = await runShellCommand({ command, env, label });
@@ -357,6 +525,66 @@ async function runForegroundGroup(entries, env) {
}
}
async function runDockerPreflight(baseEnv, options) {
if (!options.enabled) {
console.log("==> Docker preflight: skipped");
return;
}
console.log("==> Docker preflight");
const version = await runShellCaptureCommand({
command: "docker version --format '{{.Server.Version}}'",
env: baseEnv,
label: "docker-version",
timeoutMs: 20_000,
});
if (version.status !== 0) {
throw new Error(
`Docker preflight failed: docker version status=${version.status}\n${version.stderr}${version.stdout}`,
);
}
console.log(`==> Docker server: ${version.stdout.trim()}`);
if (options.cleanup) {
const stale = await runShellCaptureCommand({
command:
"docker ps -a --filter status=created --filter status=exited --filter status=dead --format '{{.Names}} {{.Status}}'",
env: baseEnv,
label: "docker-stale-list",
timeoutMs: 20_000,
});
if (stale.status === 0) {
const names = dockerPreflightContainerNames(stale.stdout);
if (names.length > 0) {
console.log(`==> Docker preflight cleanup: ${names.join(", ")}`);
const cleanup = await runShellCommand({
command: `docker rm -f ${names.map(shellQuote).join(" ")}`,
env: baseEnv,
label: "docker-stale-cleanup",
timeoutMs: 90_000,
});
if (cleanup.status !== 0) {
throw new Error(`Docker preflight cleanup failed with status ${cleanup.status}`);
}
}
}
}
const startedAt = Date.now();
const run = await runShellCommand({
command: "docker run --rm alpine:3.20 true",
env: baseEnv,
label: "docker-run-smoke",
timeoutMs: options.runTimeoutMs,
});
const elapsedSeconds = Math.round((Date.now() - startedAt) / 1000);
if (run.status !== 0) {
throw new Error(
`Docker preflight failed: docker run alpine:3.20 true status=${run.status} elapsed=${elapsedSeconds}s`,
);
}
console.log(`==> Docker preflight run: ${elapsedSeconds}s`);
}
async function prepareBundledChannelPackage(baseEnv, logDir) {
if (baseEnv.OPENCLAW_BUNDLED_CHANNEL_PACKAGE_TGZ) {
console.log(`==> Bundled channel package: ${baseEnv.OPENCLAW_BUNDLED_CHANNEL_PACKAGE_TGZ}`);
@@ -411,8 +639,9 @@ function laneEnv(name, baseEnv, logDir) {
return env;
}
async function runLane(lane, baseEnv, logDir, timeoutMs) {
async function runLane(lane, baseEnv, logDir, fallbackTimeoutMs) {
const { command, name } = lane;
const timeoutMs = lane.timeoutMs ?? fallbackTimeoutMs;
const logFile = path.join(logDir, `${name}.log`);
const env = laneEnv(name, baseEnv, logDir);
await mkdir(env.OPENCLAW_DOCKER_CLI_TOOLS_DIR, { recursive: true });
@@ -422,6 +651,7 @@ async function runLane(lane, baseEnv, logDir, timeoutMs) {
[
`==> [${name}] cli tools dir: ${env.OPENCLAW_DOCKER_CLI_TOOLS_DIR}`,
`==> [${name}] cache dir: ${env.OPENCLAW_DOCKER_CACHE_HOME_DIR}`,
`==> [${name}] timeout: ${timeoutMs}ms`,
"",
].join("\n"),
);
@@ -441,6 +671,7 @@ async function runLane(lane, baseEnv, logDir, timeoutMs) {
command,
logFile,
name,
elapsedSeconds,
status: result.status,
timedOut: result.timedOut,
};
@@ -448,6 +679,7 @@ async function runLane(lane, baseEnv, logDir, timeoutMs) {
async function runLanePool(poolLanes, baseEnv, logDir, parallelism, options) {
const failures = [];
const results = [];
const pending = [...poolLanes];
const running = new Set();
const active = {
@@ -455,8 +687,26 @@ async function runLanePool(poolLanes, baseEnv, logDir, parallelism, options) {
resources: new Map(),
weight: 0,
};
const activeLanes = new Map();
let lastLaneStartAt = 0;
let laneStartQueue = Promise.resolve();
const statusTimer =
options.statusIntervalMs > 0
? setInterval(() => {
const runningSummary = [...activeLanes.values()]
.map((entry) => `${entry.name}:${Math.round((Date.now() - entry.startedAt) / 1000)}s`)
.join(", ");
const resources = [...active.resources.entries()]
.map(([resource, value]) => `${resource}=${value}`)
.join(" ");
console.log(
`==> [${options.poolLabel}] active=${active.count} pending=${pending.length} ${resources}${
runningSummary ? ` lanes=${runningSummary}` : ""
}`,
);
}, options.statusIntervalMs)
: undefined;
statusTimer?.unref?.();
async function waitForLaneStartSlot() {
if (options.startStaggerMs <= 0) {
@@ -517,56 +767,66 @@ async function runLanePool(poolLanes, baseEnv, logDir, parallelism, options) {
async function startLane(poolLane) {
await waitForLaneStartSlot();
reserve(poolLane);
activeLanes.set(poolLane.name, { name: poolLane.name, startedAt: Date.now() });
let promise;
promise = runLane(poolLane, baseEnv, logDir, options.timeoutMs)
.then((result) => ({ lane: poolLane, promise, result }))
.finally(() => {
activeLanes.delete(poolLane.name);
release(poolLane);
});
running.add(promise);
}
while (pending.length > 0 || running.size > 0) {
let started = false;
if (!options.failFast || failures.length === 0) {
for (let index = 0; index < pending.length; ) {
const candidate = pending[index];
if (!canStartLane(candidate)) {
index += 1;
continue;
}
pending.splice(index, 1);
await startLane(candidate);
started = true;
}
}
if (started) {
continue;
}
if (running.size === 0) {
const blocked = pending.map(laneSummary).join(", ");
throw new Error(`No Docker lanes fit scheduler limits: ${blocked}`);
}
const { promise, result } = await Promise.race(running);
running.delete(promise);
if (result.status !== 0) {
failures.push(result);
}
if (options.failFast && failures.length > 0) {
const remainingResults = await Promise.all(running);
running.clear();
for (const remaining of remainingResults) {
if (remaining.result.status !== 0) {
failures.push(remaining.result);
try {
while (pending.length > 0 || running.size > 0) {
let started = false;
if (!options.failFast || failures.length === 0) {
for (let index = 0; index < pending.length; ) {
const candidate = pending[index];
if (!canStartLane(candidate)) {
index += 1;
continue;
}
pending.splice(index, 1);
await startLane(candidate);
started = true;
}
}
break;
if (started) {
continue;
}
if (running.size === 0) {
const blocked = pending.map(laneSummary).join(", ");
throw new Error(`No Docker lanes fit scheduler limits: ${blocked}`);
}
const { promise, result } = await Promise.race(running);
running.delete(promise);
results.push(result);
if (result.status !== 0) {
failures.push(result);
}
if (options.failFast && failures.length > 0) {
const remainingResults = await Promise.all(running);
running.clear();
for (const remaining of remainingResults) {
results.push(remaining.result);
if (remaining.result.status !== 0) {
failures.push(remaining.result);
}
}
break;
}
}
} finally {
if (statusTimer) {
clearInterval(statusTimer);
}
}
return failures;
return { failures, results };
}
async function tailFile(file, lines) {
@@ -640,7 +900,24 @@ async function main() {
DEFAULT_LANE_START_STAGGER_MS,
"OPENCLAW_DOCKER_ALL_START_STAGGER_MS",
);
const statusIntervalMs = parseNonNegativeInt(
process.env.OPENCLAW_DOCKER_ALL_STATUS_INTERVAL_MS,
DEFAULT_STATUS_INTERVAL_MS,
"OPENCLAW_DOCKER_ALL_STATUS_INTERVAL_MS",
);
const preflightRunTimeoutMs = parsePositiveInt(
process.env.OPENCLAW_DOCKER_ALL_PREFLIGHT_RUN_TIMEOUT_MS,
DEFAULT_PREFLIGHT_RUN_TIMEOUT_MS,
"OPENCLAW_DOCKER_ALL_PREFLIGHT_RUN_TIMEOUT_MS",
);
const failFast = parseBool(process.env.OPENCLAW_DOCKER_ALL_FAIL_FAST, true);
const dryRun = parseBool(process.env.OPENCLAW_DOCKER_ALL_DRY_RUN, false);
const preflightEnabled = parseBool(process.env.OPENCLAW_DOCKER_ALL_PREFLIGHT, true);
const preflightCleanup = parseBool(process.env.OPENCLAW_DOCKER_ALL_PREFLIGHT_CLEANUP, true);
const timingsEnabled = parseBool(process.env.OPENCLAW_DOCKER_ALL_TIMINGS, true);
const timingsFile = path.resolve(
process.env.OPENCLAW_DOCKER_ALL_TIMINGS_FILE || DEFAULT_TIMINGS_FILE,
);
const runId = process.env.OPENCLAW_DOCKER_ALL_RUN_ID || utcStampForPath();
const logDir = path.resolve(
process.env.OPENCLAW_DOCKER_ALL_LOG_DIR ||
@@ -655,12 +932,24 @@ async function main() {
appendExtension(baseEnv, "acpx");
appendExtension(baseEnv, "codex");
const timingStore = await loadTimingStore(timingsFile, timingsEnabled);
const orderedLanes = orderLanes(lanes, timingStore);
const orderedTailLanes = orderLanes(tailLanes, timingStore);
console.log(`==> Docker test logs: ${logDir}`);
console.log(`==> Parallelism: ${parallelism}`);
console.log(`==> Tail parallelism: ${tailParallelism}`);
console.log(`==> Lane timeout: ${laneTimeoutMs}ms`);
console.log(`==> Lane start stagger: ${laneStartStaggerMs}ms`);
console.log(`==> Status interval: ${statusIntervalMs}ms`);
console.log(`==> Fail fast: ${failFast ? "yes" : "no"}`);
console.log(`==> Dry run: ${dryRun ? "yes" : "no"}`);
console.log(
`==> Docker preflight: ${preflightEnabled ? "yes" : "no"}${
preflightCleanup ? " cleanup=yes" : " cleanup=no"
}`,
);
console.log(`==> Docker lane timings: ${timingStore.enabled ? timingsFile : "disabled"}`);
console.log(`==> Live-test bundled plugin deps: ${baseEnv.OPENCLAW_DOCKER_BUILD_EXTENSIONS}`);
const schedulerOptions = parseSchedulerOptions(process.env, parallelism);
const tailSchedulerOptions = parseSchedulerOptions(process.env, tailParallelism);
@@ -670,6 +959,18 @@ async function main() {
console.log(
`==> Tail scheduler: weight=${tailSchedulerOptions.weightLimit} docker=${tailSchedulerOptions.resourceLimits.docker} live=${tailSchedulerOptions.resourceLimits.live} npm=${tailSchedulerOptions.resourceLimits.npm} service=${tailSchedulerOptions.resourceLimits.service}`,
);
printLaneManifest("Main", orderedLanes, timingStore);
printLaneManifest("Tail", orderedTailLanes, timingStore);
if (dryRun) {
console.log("==> Dry run complete");
return;
}
await runDockerPreflight(baseEnv, {
cleanup: preflightCleanup,
enabled: preflightEnabled,
runTimeoutMs: preflightRunTimeoutMs,
});
await runForegroundGroup(
[
@@ -686,22 +987,29 @@ async function main() {
const options = {
...schedulerOptions,
failFast,
poolLabel: "main",
startStaggerMs: laneStartStaggerMs,
statusIntervalMs,
timeoutMs: laneTimeoutMs,
};
const failures = await runLanePool(lanes, baseEnv, logDir, parallelism, options);
const mainResult = await runLanePool(orderedLanes, baseEnv, logDir, parallelism, options);
const failures = [...mainResult.failures];
const allResults = [...mainResult.results];
await writeTimingStore(timingStore, mainResult.results);
if (failFast && failures.length > 0) {
await printFailureSummary(failures, tailLines);
process.exit(1);
}
console.log("==> Running provider-sensitive Docker tail lanes");
failures.push(
...(await runLanePool(tailLanes, baseEnv, logDir, tailParallelism, {
...options,
...tailSchedulerOptions,
})),
);
const tailResult = await runLanePool(orderedTailLanes, baseEnv, logDir, tailParallelism, {
...options,
...tailSchedulerOptions,
poolLabel: "tail",
});
failures.push(...tailResult.failures);
allResults.push(...tailResult.results);
await writeTimingStore(timingStore, tailResult.results);
if (failures.length > 0) {
await printFailureSummary(failures, tailLines);
process.exit(1);
@@ -712,6 +1020,7 @@ async function main() {
"pnpm test:docker:cleanup",
baseEnv,
);
await writeTimingStore(timingStore, allResults);
console.log("==> Docker test suite passed");
}