feat: graceful shutdown, disconnect-driven cancellation, and queue acceptance coverage
All checks were successful
Quality gates / quality (pull_request) Successful in 1m58s

- server.mjs tracks in-flight chats: browser disconnects abort the
  queued turn (releasing its slot), and SIGTERM/SIGINT cancels every
  in-flight turn so no Hermes subprocess is orphaned.
- vision-service forwards a per-request AbortSignal combined with the
  provider deadline via AbortSignal.any.
- acceptance suites cover overload fallback, urgent bypass under
  saturation, disconnect cleanup, recovery, and shutdown orphan checks.
- zero-call invariant tests prove urgent text never reaches Hermes or
  the vision provider under load.
This commit is contained in:
Timmy 2026-08-22 23:01:20 +00:00
parent 5dcaaae4d6
commit 9118b69057
7 changed files with 397 additions and 6 deletions

View File

@ -4,12 +4,14 @@
"private": true, "private": true,
"type": "module", "type": "module",
"scripts": { "scripts": {
"test": "node --test tests/domain.test.js tests/analysis.test.js tests/vision-service.test.js tests/vision-config.test.js tests/hermes-agent-service.test.js tests/agent-gateway.acceptance.test.js tests/staging-health.test.js tests/service-worker-runtime.test.js tests/training-ingest.test.js tests/ci-workflow.test.js tests/product-decisions.test.js tests/release-demo.test.js tests/selfhost-bootstrap.test.js tests/staging-config.test.js", "test": "node --test tests/domain.test.js tests/analysis.test.js tests/vision-service.test.js tests/vision-config.test.js tests/hermes-agent-service.test.js tests/agent-queue.test.js tests/agent-turn-timeout.test.js tests/inference-queue.test.js tests/inference-queue-cancellation.test.js tests/inference-zero-call.test.js tests/agent-gateway.acceptance.test.js tests/staging-health.test.js tests/service-worker-runtime.test.js tests/training-ingest.test.js tests/ci-workflow.test.js tests/product-decisions.test.js tests/release-demo.test.js tests/selfhost-bootstrap.test.js tests/staging-config.test.js",
"test:ui": "node tests/ui.acceptance.mjs", "test:ui": "node tests/ui.acceptance.mjs",
"test:photo": "node tests/photo-first.acceptance.mjs", "test:photo": "node tests/photo-first.acceptance.mjs",
"test:sleek": "node tests/sleek-chat.acceptance.mjs", "test:sleek": "node tests/sleek-chat.acceptance.mjs",
"test:queue-acceptance": "node tests/inference-queue.acceptance.mjs",
"test:shutdown-acceptance": "node tests/server-shutdown.acceptance.mjs",
"test:staging-smoke": "node tests/staging.acceptance.mjs", "test:staging-smoke": "node tests/staging.acceptance.mjs",
"check:syntax": "node --check app.js && node --check server.mjs && node --check service-worker.js && node --check src/analysis.js && node --check src/domain.js && node --check src/hermes-agent-service.js && node --check src/vision-config.js && node --check src/vision-service.js && node --check scripts/record_release_demo.mjs && node --check tests/staging.acceptance.mjs && bash -n scripts/bootstrap_selfhost_smolvlm.sh && bash -n scripts/run_selfhost_smolvlm.sh && python3 -m py_compile scripts/ingest_training_photo.py scripts/build_release.py scripts/deploy_staging.py", "check:syntax": "node --check app.js && node --check server.mjs && node --check service-worker.js && node --check src/analysis.js && node --check src/domain.js && node --check src/hermes-agent-service.js && node --check src/inference-queue.js && node --check src/vision-config.js && node --check src/vision-service.js && node --check scripts/record_release_demo.mjs && node --check tests/inference-queue.acceptance.mjs && node --check tests/server-shutdown.acceptance.mjs && node --check tests/staging.acceptance.mjs && bash -n scripts/bootstrap_selfhost_smolvlm.sh && bash -n scripts/run_selfhost_smolvlm.sh && python3 -m py_compile scripts/ingest_training_photo.py scripts/build_release.py scripts/deploy_staging.py",
"check:diff": "bash scripts/check_diff.sh", "check:diff": "bash scripts/check_diff.sh",
"start": "node server.mjs" "start": "node server.mjs"
}, },

View File

@ -39,6 +39,34 @@ function rejectCrossSite(req){const site=String(req.headers['sec-fetch-site']||'
function agentCookie(token){const secure=agentConfig.publicOrigin.startsWith('https://')?'; Secure':'';return `timmy_agent=${encodeURIComponent(token)}; HttpOnly; SameSite=Strict; Path=${appRoot}; Max-Age=86400${secure}`} function agentCookie(token){const secure=agentConfig.publicOrigin.startsWith('https://')?'; Secure':'';return `timmy_agent=${encodeURIComponent(token)}; HttpOnly; SameSite=Strict; Path=${appRoot}; Max-Age=86400${secure}`}
function sendAgentError(res,error){const status=error instanceof AgentGatewayError?error.status:503;const message=error instanceof AgentGatewayError?error.message:'Hermes is temporarily unavailable. Your local journal still works.';return sendJson(res,status,{error:message})} function sendAgentError(res,error){const status=error instanceof AgentGatewayError?error.status:503;const message=error instanceof AgentGatewayError?error.message:'Hermes is temporarily unavailable. Your local journal still works.';return sendJson(res,status,{error:message})}
// Graceful shutdown: SIGTERM must cancel in-flight Hermes turns (killing their
// child processes) instead of abandoning them as orphan subprocesses.
const inFlightChats = new Set();
let shuttingDown = false;
function trackChat(res, controller) {
if (shuttingDown) controller.abort(new Error('Timmy is shutting down. Your journal still works — try again shortly.'));
const entry = { res, controller };
inFlightChats.add(entry);
// A vanished client must release its queue slot and stop any in-flight
// Hermes turn instead of consuming inference capacity silently.
res.on('close', () => {
inFlightChats.delete(entry);
if (!res.writableEnded) {
try { controller.abort(new Error('The browser closed the request. Your journal still works.')); } catch { /* already aborted */ }
}
});
return entry;
}
for (const signalName of ['SIGTERM', 'SIGINT']) {
process.on(signalName, () => {
shuttingDown = true;
for (const { controller } of inFlightChats) {
try { controller.abort(new Error('Server is shutting down.')); } catch { /* already aborted */ }
}
process.exit(0);
});
}
http.createServer(async(req,res)=>{ http.createServer(async(req,res)=>{
try{ try{
const url=new URL(req.url,'http://localhost'); const url=new URL(req.url,'http://localhost');
@ -62,7 +90,7 @@ http.createServer(async(req,res)=>{
try{rejectCrossSite(req);const payload=await readJson(req,4096);const result=await agentService.unlock({origin:requestOrigin(req),accessCode:String(payload?.accessCode||'')});res.setHeader('set-cookie',agentCookie(result.cookieToken));return sendJson(res,200,result.public)}catch(error){return sendAgentError(res,error)} try{rejectCrossSite(req);const payload=await readJson(req,4096);const result=await agentService.unlock({origin:requestOrigin(req),accessCode:String(payload?.accessCode||'')});res.setHeader('set-cookie',agentCookie(result.cookieToken));return sendJson(res,200,result.public)}catch(error){return sendAgentError(res,error)}
} }
if(appPath==='/api/agent/chat'&&req.method==='POST'){ if(appPath==='/api/agent/chat'&&req.method==='POST'){
try{rejectCrossSite(req);const payload=await readJson(req,128*1024);return sendJson(res,200,await agentService.chat({origin:requestOrigin(req),cookieToken:cookie(req,'timmy_agent'),payload}))}catch(error){return sendAgentError(res,error)} try{rejectCrossSite(req);const payload=await readJson(req,128*1024);const chatAbort=new AbortController();trackChat(res,chatAbort);return sendJson(res,200,await agentService.chat({origin:requestOrigin(req),cookieToken:cookie(req,'timmy_agent'),payload,signal:chatAbort.signal}))}catch(error){return sendAgentError(res,error)}
} }
if(appPath.startsWith('/api/'))return sendJson(res,404,{error:'Not found'}); if(appPath.startsWith('/api/'))return sendJson(res,404,{error:'Not found'});
if(req.method!=='GET'&&req.method!=='HEAD'){res.writeHead(405,{'allow':'GET, HEAD'});return res.end()} if(req.method!=='GET'&&req.method!=='HEAD'){res.writeHead(405,{'allow':'GET, HEAD'});return res.end()}

View File

@ -7,18 +7,22 @@ function providerEndpoint(baseUrl) {
return `${url.toString().replace(/\/$/, '')}/chat/completions`; return `${url.toString().replace(/\/$/, '')}/chat/completions`;
} }
export async function analyzePhoto({ payload, fetchImpl = fetch, config }) { export async function analyzePhoto({ payload, fetchImpl = fetch, config, signal }) {
const photo = validatePhotoPayload(payload); const photo = validatePhotoPayload(payload);
if (!config?.model) throw new Error('AI analysis is not configured.'); if (!config?.model) throw new Error('AI analysis is not configured.');
const endpoint = providerEndpoint(config.baseUrl); const endpoint = providerEndpoint(config.baseUrl);
// The provider call honours both its own deadline and a per-request
// cancellation so abandoned clients stop consuming inference capacity.
const timeoutSignal = AbortSignal.timeout(config.requestTimeoutMs || 60_000);
const combinedSignal = signal ? AbortSignal.any([timeoutSignal, signal]) : timeoutSignal;
const response = await fetchImpl(endpoint, { const response = await fetchImpl(endpoint, {
method: 'POST', method: 'POST',
headers: { headers: {
'content-type': 'application/json', 'content-type': 'application/json',
authorization: `Bearer ${config.apiKey || 'local-proxy'}`, authorization: 'Bearer ' + String(config.apiKey || 'local-proxy'),
}, },
body: JSON.stringify(buildVisionRequest({ imageDataUrl: photo.imageDataUrl, model: config.model })), body: JSON.stringify(buildVisionRequest({ imageDataUrl: photo.imageDataUrl, model: config.model })),
signal: AbortSignal.timeout(config.requestTimeoutMs || 60_000), signal: combinedSignal,
}).catch(() => { throw new Error('AI analysis is temporarily unavailable. Continue manually.'); }); }).catch(() => { throw new Error('AI analysis is temporarily unavailable. Continue manually.'); });
if (!response.ok) throw new Error('AI analysis is temporarily unavailable. Continue manually.'); if (!response.ok) throw new Error('AI analysis is temporarily unavailable. Continue manually.');
let data; let data;

20
tests/fixtures/slow-hermes.mjs vendored Executable file
View File

@ -0,0 +1,20 @@
#!/usr/bin/env node
// Slow Hermes fixture for bounded-queue acceptance tests. Normal prompts
// answer immediately; prompts containing HOLD block until a release file
// appears inside the agent workdir (the value after `--in`), so tests can
// saturate the queue deterministically without touching shared /tmp state.
import { existsSync } from 'node:fs';
import { appendFileSync } from 'node:fs';
const args = process.argv.slice(2);
const promptIndex = args.indexOf('-q');
const prompt = promptIndex >= 0 ? args[promptIndex + 1] : '';
const workdirIndex = args.indexOf('--in');
const workdir = workdirIndex >= 0 ? args[workdirIndex + 1] : '/tmp';
appendFileSync(`${workdir}/fixture-pids.log`, `${process.pid}\n`);
process.stdout.write('session_id: slow_fixture_2026\n');
if (/HOLD/.test(prompt)) {
const releaseFile = `${workdir}/hermes-release`;
while (!existsSync(releaseFile)) await new Promise(resolve => setTimeout(resolve, 20));
}
process.stdout.write('Hermes fixture answered the bounded request.\n');

View File

@ -0,0 +1,136 @@
import test from 'node:test';
import assert from 'node:assert/strict';
import { chmod, mkdtemp, rm, readFile, writeFile } from 'node:fs/promises';
import { spawn } from 'node:child_process';
import { tmpdir } from 'node:os';
import { join } from 'node:path';
import { constants as fsConstants } from 'node:fs';
import { fileURLToPath } from 'node:url';
const root = fileURLToPath(new URL('..', import.meta.url));
const fixture = fileURLToPath(new URL('./fixtures/slow-hermes.mjs', import.meta.url));
const origin = 'http://127.0.0.1:4187';
const accessCode = 'queue-acceptance-code-2026';
async function waitReady(child) {
const deadline = Date.now() + 10_000;
while (Date.now() < deadline) {
if (child.exitCode !== null) throw new Error(`server exited ${child.exitCode}`);
try { const response = await fetch(`${origin}/api/healthz`); if (response.ok) return; } catch {}
await new Promise(resolve => setTimeout(resolve, 50));
}
throw new Error('server did not become ready');
}
function post(path, body, { cookie = '', signal } = {}) {
return fetch(`${origin}${path}`, {
method: 'POST',
headers: {
'content-type': 'application/json',
origin,
'sec-fetch-site': 'same-origin',
...(cookie ? { cookie } : {}),
},
body: JSON.stringify(body),
signal,
}).then(
async response => ({ status: response.status, body: await response.json().catch(() => ({})) }),
error => ({ failed: true, reason: String(error?.name || error) }),
);
}
test('bounded queue end-to-end: disconnect cleanup, overload fallback, urgent bypass, recovery', async t => {
const workdir = await mkdtemp(join(tmpdir(), 'timmy-queue-accept-'));
const releaseFile = join(workdir, 'hermes-release');
const pidLog = join(workdir, 'fixture-pids.log');
await chmod(fixture, fsConstants.S_IRWXU);
const child = spawn(process.execPath, ['server.mjs'], {
cwd: root,
env: {
...process.env,
PORT: '4187',
TIMMY_AGENT_ENABLED: 'true',
TIMMY_AGENT_ACCESS_TOKEN: accessCode,
TIMMY_PUBLIC_ORIGIN: origin,
TIMMY_AGENT_WORKDIR: workdir,
TIMMY_HERMES_COMMAND: fixture,
TIMMY_AGENT_MAX_CONCURRENT_TURNS: '1',
TIMMY_AGENT_MAX_QUEUE_DEPTH: '2',
},
stdio: ['ignore', 'pipe', 'pipe'],
});
let serverLogs = '';
child.stderr.on('data', chunk => { serverLogs += String(chunk); });
t.after(async () => {
child.kill('SIGTERM');
await rm(workdir, { recursive: true, force: true }).catch(() => {});
await rm(releaseFile, { force: true }).catch(() => {});
});
await waitReady(child);
// Authenticate once and keep the cookie for the whole scenario.
const rawResponse = await fetch(`${origin}/api/agent/unlock`, {
method: 'POST',
headers: { 'content-type': 'application/json', origin, 'sec-fetch-site': 'same-origin' },
body: JSON.stringify({ accessCode }),
});
assert.equal(rawResponse.status, 200);
const browserCookie = rawResponse.headers.get('set-cookie').split(';', 1)[0];
// Fill the slot: this HOLD runs until the fixture release file appears.
const heldOne = post('/api/agent/chat', { message: 'HOLD one', ledger: [] }, { cookie: browserCookie });
await new Promise(resolve => setTimeout(resolve, 400));
// 1. Disconnect cleanup: this request waits in the queue. When its client
// vanishes, Timmy must drop it and free the queued place.
const disconnectController = new AbortController();
const doomed = post('/api/agent/chat', { message: 'HOLD doomed', ledger: [] }, { cookie: browserCookie, signal: disconnectController.signal });
await new Promise(resolve => setTimeout(resolve, 400));
disconnectController.abort();
const doomedOutcome = await doomed;
assert.equal(doomedOutcome.failed, true, 'abandoned client saw its request cancelled');
// If the freed place was reclaimed, both follow-up HOLDs are admitted
// (slot + queue of two minus one freed place). A sanitized 503 here would
// mean the cancelled client did not release capacity.
const heldTwo = post('/api/agent/chat', { message: 'HOLD two', ledger: [] }, { cookie: browserCookie });
const heldThree = post('/api/agent/chat', { message: 'HOLD three', ledger: [] }, { cookie: browserCookie });
await new Promise(resolve => setTimeout(resolve, 500));
// 2. Overload: everything is saturated now, so a further request must fail
// fast with the stable manual-fallback copy and zero provider contact.
const overloadResult = await post('/api/agent/chat', { message: 'overload probe', ledger: [] }, { cookie: browserCookie });
assert.equal(overloadResult.status, 503);
assert.match(overloadResult.body.error, /busy|try again|manually/i);
assert.doesNotMatch(overloadResult.body.error, /spawn|hermes|session|token|workdir|fixture/i);
// 3. Urgent bypass while saturated: deterministic safety answer, zero calls.
const urgentResult = await post('/api/agent/chat', { message: 'I have rectal bleeding', ledger: [] }, { cookie: browserCookie });
assert.equal(urgentResult.status, 200);
assert.equal(urgentResult.body.safetyOverride, true);
assert.match(urgentResult.body.reply, /medical help/i);
// 4. Recovery: releasing the fixture lets every surviving request complete.
await writeFile(releaseFile, 'go');
for (const [label, promise] of [['one', heldOne], ['two', heldTwo], ['three', heldThree]]) {
const settled = await Promise.race([
promise,
new Promise(resolve => setTimeout(() => resolve({ status: 'TEST-TIMEOUT' }), 12_000)),
]);
assert.equal(settled.status, 200, `${label} completed after recovery`);
assert.match(settled.body.reply, /bounded request/);
assert.doesNotMatch(JSON.stringify(settled.body), /slow_fixture_2026|session_id/);
}
// 5. No orphan subprocesses: every fixture PID recorded at spawn is gone.
await new Promise(resolve => setTimeout(resolve, 400));
const recordedPids = ((await readFile(pidLog, 'utf8').catch(() => '')) || '')
.split('\n').map(Number).filter(Boolean);
assert.equal(recordedPids.length, 3, `exactly three turns ran (saw ${recordedPids.length})`);
for (const pid of [...new Set(recordedPids)]) {
let alive = true;
try { process.kill(pid, 0); } catch { alive = false; }
assert.equal(alive, false, `fixture process ${pid} outlived its turn`);
}
assert.doesNotMatch(serverLogs, /access-code|secret|token/i);
}, { timeout: 40_000 });

View File

@ -0,0 +1,108 @@
import test from 'node:test';
import assert from 'node:assert/strict';
import { createHermesAgentService, resolveHermesAgentConfig } from '../src/hermes-agent-service.js';
import { analyzePhoto } from '../src/vision-service.js';
import { OverloadError, createInferenceQueue } from '../src/inference-queue.js';
const origin = 'http://127.0.0.1:4173';
const accessCode = 'test-agent-access-code-2026';
const configured = () => resolveHermesAgentConfig({
TIMMY_AGENT_ENABLED: 'true',
TIMMY_AGENT_ACCESS_TOKEN: accessCode,
TIMMY_PUBLIC_ORIGIN: origin,
TIMMY_AGENT_WORKDIR: '/tmp/timmy-agent-workspace',
});
function deferred() {
let resolve;
let reject;
const promise = new Promise((res, rej) => { resolve = res; reject = rej; });
return { promise, resolve, reject };
}
test('urgent messages intercept before Hermes even when the queue is saturated', async () => {
let clock = 0;
let hermesCalls = 0;
const gate = deferred();
const service = createHermesAgentService({
config: { ...configured(), maxQueueDepth: 0 },
randomToken: () => 'urgent-cookie',
now: () => clock,
runTurn: async () => {
hermesCalls += 1;
return gate.promise;
},
});
await service.unlock({ origin, accessCode });
// Saturate the single slot so any queued path would be rejected as overloaded.
const active = service.chat({ origin, cookieToken: 'urgent-cookie', payload: { message: 'normal question', ledger: [] } });
await new Promise(resolve => setImmediate(resolve));
assert.equal(hermesCalls, 1);
const urgent = await service.chat({ origin, cookieToken: 'urgent-cookie', payload: { message: 'I have rectal bleeding', ledger: [] } });
assert.equal(hermesCalls, 1, 'Hermes was called exactly once — only for the non-urgent request');
assert.equal(urgent.safetyOverride, true);
assert.match(urgent.reply, /medical help/i);
gate.resolve({ reply: 'late', sessionId: 'urgent_session_01' });
assert.match((await active).reply, /late/);
});
test('vision analysis is bounded by the shared queue and overload returns the manual-fallback copy', async () => {
let clock = 0;
let providerCalls = 0;
const gates = [];
const queue = createInferenceQueue({ concurrency: 1, maxQueueDepth: 1, requestTimeoutMs: 60_000, now: () => clock });
const visionFetch = async (url, options) => {
providerCalls += 1;
const gate = deferred();
gates.push(gate);
options.signal?.addEventListener('abort', () => gate.reject(new Error('aborted')));
return new Promise((resolve, reject) => {
gate.promise.then(() => resolve({ ok: true, json: async () => ({ choices: [{ message: { content: JSON.stringify({ isStool: true, bristolType: 4, color: 'brown', confidence: 0.8, imageQuality: 'good', observations: 'ok' }) } }] }) }), reject);
});
};
const payload = { imageDataUrl: 'data:image/jpeg;base64,YQ==', consent: true };
const config = { baseUrl: 'http://127.0.0.1:8645/v1', apiKey: 'test-key-not-real', model: 'm' };
const first = queue.run(signal => analyzePhoto({ payload, fetchImpl: visionFetch, config, signal }));
while (providerCalls < 1) await Promise.resolve();
const second = queue.run(signal => analyzePhoto({ payload, fetchImpl: visionFetch, config, signal }));
await Promise.resolve();
await assert.rejects(() => queue.run((signal) => analyzePhoto({ payload, fetchImpl: visionFetch, config, signal })), error => {
assert.ok(error instanceof OverloadError);
assert.match(error.message, /continue manually/i);
assert.doesNotMatch(error.message, /api|key|token|127\.0\.0\.1|fetch/i);
return true;
}, 'third request beyond depth is overloaded with sanitized copy');
assert.equal(providerCalls, 1, 'overloaded request never reached the provider');
gates[0].resolve('go');
const result = await Promise.race([first, new Promise(resolve => setTimeout(() => resolve('pending'), 100))]);
assert.notEqual(result, 'pending');
});
test('vision requests carry cancellation so disconnected clients release provider work', async () => {
let capturedSignal;
const controller = new AbortController();
const gate = deferred();
const fetchImpl = async (url, options) => {
capturedSignal = options.signal;
options.signal?.addEventListener('abort', () => gate.reject(new Error('aborted')));
return gate.promise;
};
const payload = { imageDataUrl: 'data:image/jpeg;base64,YQ==', consent: true };
const config = { baseUrl: 'http://127.0.0.1:8645/v1', apiKey: 'test-key-not-real', model: 'm' };
const attempt = analyzePhoto({ payload, fetchImpl, config, signal: controller.signal });
while (!capturedSignal) await Promise.resolve();
controller.abort(new Error('client disconnected'));
await assert.rejects(() => attempt, /continue manually|unavailable/i);
assert.equal(capturedSignal.aborted, true, 'the provider fetch observed the cancellation');
});

View File

@ -0,0 +1,93 @@
import test from 'node:test';
import assert from 'node:assert/strict';
import { chmod, mkdtemp, rm, readFile } from 'node:fs/promises';
import { spawn } from 'node:child_process';
import { tmpdir } from 'node:os';
import { join } from 'node:path';
import { constants as fsConstants } from 'node:fs';
import { fileURLToPath } from 'node:url';
const root = fileURLToPath(new URL('..', import.meta.url));
const fixture = fileURLToPath(new URL('./fixtures/slow-hermes.mjs', import.meta.url));
test('server shutdown cancels in-flight turns and leaves no orphan Hermes subprocesses', async () => {
const workdir = await mkdtemp(join(tmpdir(), 'timmy-shutdown-'));
await chmod(fixture, fsConstants.S_IRWXU);
const child = spawn(process.execPath, ['server.mjs'], {
cwd: root,
env: {
...process.env,
HOST: '127.0.0.1',
PORT: '4188',
TIMMY_AGENT_ENABLED: 'true',
TIMMY_AGENT_ACCESS_TOKEN: 'shutdown-accept-code-2026',
TIMMY_PUBLIC_ORIGIN: 'http://127.0.0.1:4188',
TIMMY_AGENT_WORKDIR: workdir,
TIMMY_HERMES_COMMAND: fixture,
},
stdio: ['ignore', 'pipe', 'pipe'],
});
let reportedOrigin = '';
child.stdout.on('data', chunk => {
const match = String(chunk).match(/listening on http:\/\/([^:]+):(\d+)/);
if (match && !reportedOrigin) reportedOrigin = `http://${match[1]}:${match[2]}`;
});
const deadline = Date.now() + 10_000;
while (!reportedOrigin && Date.now() < deadline && child.exitCode === null) {
await new Promise(resolve => setTimeout(resolve, 50));
}
assert.ok(reportedOrigin, 'server reported its port');
const unlockResponse = await fetch(`${reportedOrigin}/api/agent/unlock`, {
method: 'POST',
headers: { 'content-type': 'application/json', origin: reportedOrigin, 'sec-fetch-site': 'same-origin' },
body: JSON.stringify({ accessCode: 'shutdown-accept-code-2026' }),
});
assert.equal(unlockResponse.status, 200);
const browserCookie = unlockResponse.headers.get('set-cookie').split(';', 1)[0];
// Start a turn wedged on the fixture's HOLD gate.
fetch(`${reportedOrigin}/api/agent/chat`, {
method: 'POST',
headers: { 'content-type': 'application/json', origin: reportedOrigin, 'sec-fetch-site': 'same-origin', cookie: browserCookie },
body: JSON.stringify({ message: 'HOLD shutdown probe', ledger: [] }),
}).catch(() => {});
await new Promise(resolve => setTimeout(resolve, 600));
const pidLog = join(workdir, 'fixture-pids.log');
// Wait until the fixture for this run's HOLD turn is actually alive.
let lastPid = null;
const spawnDeadline = Date.now() + 8_000;
while (Date.now() < spawnDeadline) {
const recorded = ((await readFile(pidLog, 'utf8').catch(() => '')) || '')
.split('\n').map(Number).filter(Boolean);
const candidate = recorded.at(-1);
if (candidate) {
let alive = true;
try { process.kill(candidate, 0); } catch { alive = false; }
if (alive) {
lastPid = candidate;
break;
}
}
await new Promise(resolve => setTimeout(resolve, 100));
}
assert.ok(lastPid, 'fixture subprocess was spawned for the in-flight turn');
let aliveBefore = true;
try { process.kill(lastPid, 0); } catch { aliveBefore = false; }
assert.ok(aliveBefore, `fixture ${lastPid} is running before shutdown`);
// Shut the server down while the turn is still running.
child.kill('SIGTERM');
await new Promise((resolve, reject) => {
const timeout = setTimeout(() => reject(new Error('server did not exit after SIGTERM')), 5_000);
child.on('exit', code => { clearTimeout(timeout); resolve(code); });
});
// The in-flight fixture must be gone — no orphan subprocess may survive.
await new Promise(resolve => setTimeout(resolve, 500));
let aliveAfter = true;
try { process.kill(lastPid, 0); } catch { aliveAfter = false; }
assert.equal(aliveAfter, false, `fixture process ${lastPid} survived server shutdown`);
await rm(workdir, { recursive: true, force: true }).catch(() => {});
}, { timeout: 30_000 });