Compare commits

..
Author SHA1 Message Date
Garry TanandClaude Fable 5 d1f03cb346 test(dream): conform dream-dir-source-stamp to canonical PGLite isolation pattern
check:test-isolation R3/R4 flagged the new test file: engine was created
in beforeEach (outside beforeAll) and never disconnected in afterAll.
Switch to the canonical shared-engine pattern (beforeAll create,
beforeEach resetPgliteState, afterAll disconnect) per
test/helpers/reset-pglite.ts.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-07-22 10:47:02 -07:00
Garry TanandClaude Fable 5 e41e3948cd fix(autopilot): close the engine on SIGTERM/SIGINT instead of hard-exiting (#1872)
systemctl stop (SIGTERM) previously hard-exited autopilot without ever
closing the engine. On PGLite the cycle steps run INLINE in the autopilot
process, so a mid-write exit kills WASM Postgres with the WAL dirty and
can corrupt the brain.

Now both exit paths close the engine first:
- autopilot's own shutdown() (SIGINT + internal stops like max_crashes /
  cycle-failure-cap) aborts the in-flight inline cycle via an
  AbortController threaded into runCycle, drains it briefly, and awaits
  engine.disconnect() before process.exit(0).
- process-cleanup's SIGTERM handler (installed at cli.ts module load,
  exits within its 3s cleanup deadline) reaches the same closeEngine via
  a registered 'autopilot-engine-close' cleanup callback.

PGLite's disconnect() drains the pending query and checkpoints before
closing; a second call is a no-op, so both paths firing is safe.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-07-21 15:12:48 -07:00
edad6b1d5f fix(dream): stamp path-derived sources so --dir runs land cycle freshness (#1869)
gbrain dream --dir <path> (and the configured sync.repo_path fallback)
never wrote last_source_cycle_at / last_full_cycle_at because runCycle's
stamp gate reads opts.sourceId and dream only set it from --source.
Doctor's cycle_freshness stayed perpetually stale on path-scoped brains.

Fix at the command level: dream derives the source id from the resolved
brain dir via resolveSourceForDir (now exported from cycle.ts) and passes
it as opts.sourceId. runCycle's stamp/lock semantics are untouched, so
legacy global callers (autopilot-global-maintenance runs GLOBAL_PHASES
with a brainDir and no sourceId) cannot falsely stamp per-source
freshness — the flaw that sank the runCycle-wide variant in PR #2549.
A derived match on an archived source is skipped (mirrors the explicit
--source archived guard).

Takeover of #2549.

Co-authored-by: javieraldape <javieraldape@users.noreply.github.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-07-21 15:12:48 -07:00
9 changed files with 255 additions and 72 deletions
+42 -1
View File
@@ -38,6 +38,7 @@ import { logSelfUpgrade } from '../core/audit/self-upgrade-audit.ts';
import { detectInstallMethod } from './upgrade.ts';
import { evaluateQuietHours } from '../core/minions/quiet-hours.ts';
import { inspectLock } from '../core/db-lock.ts';
import { registerCleanup } from '../core/process-cleanup.ts';
/**
* v0.37.7.0 #1162 — classify autopilot reconnect-loop errors.
@@ -433,6 +434,37 @@ export async function runAutopilot(engine: BrainEngine, args: string[]) {
let stopping = false;
let childSupervisor: ChildWorkerSupervisor | null = null;
// #1872: graceful engine shutdown. On PGLite the cycle steps run INLINE in
// this process, so a hard `process.exit` mid-write (systemctl stop →
// SIGTERM) kills WASM Postgres with the WAL dirty and can corrupt the
// brain. Two exit paths must both close the engine:
// - autopilot's own shutdown() below (owns SIGINT + internal stops like
// max_crashes / cycle-failure-cap), and
// - process-cleanup's SIGTERM handler (installed at cli.ts module load;
// it runs the cleanup registry with a 3s deadline and then exits) —
// which is why closeEngine is ALSO registered there.
// closeEngine aborts the in-flight inline cycle (runCycle checks the
// signal between phases and threads it into phase sub-work), gives it a
// short bounded window to wind down, then disconnects. PGLite's
// disconnect() drains the pending query and checkpoints before closing;
// a second call is a no-op (disconnect snapshots + nulls the handle), so
// both paths firing is safe.
const shutdownAbort = new AbortController();
let inflightInlineCycle: Promise<unknown> | null = null;
const closeEngine = async () => {
shutdownAbort.abort(new Error('autopilot shutdown'));
if (inflightInlineCycle) {
// ponytail: 2s cap keeps us inside process-cleanup's 3s deadline; a
// between-phase abort resolves instantly, a mid-phase one may not.
await Promise.race([
inflightInlineCycle.catch(() => { /* cycle errors already logged by the loop */ }),
new Promise((r) => setTimeout(r, 2_000)),
]);
}
try { await engine.disconnect(); } catch { /* best-effort */ }
};
const deregisterEngineClose = registerCleanup('autopilot-engine-close', closeEngine);
if (spawnManagedWorker) {
const cliPath = resolveGbrainCliPath();
// Cgroup-aware auto-sized RSS watchdog cap (issue #1678). The old flat
@@ -520,6 +552,10 @@ export async function runAutopilot(engine: BrainEngine, args: string[]) {
childSupervisor.killChild('SIGKILL');
}
}
// #1872: abort the in-flight inline cycle and close the engine BEFORE
// process.exit — a hard exit mid-write corrupts PGLite's WASM Postgres.
await closeEngine();
deregisterEngineClose();
try { unlinkSync(lockPath); } catch { /* already gone */ }
process.exit(0);
};
@@ -1008,16 +1044,21 @@ export async function runAutopilot(engine: BrainEngine, args: string[]) {
// path's phase set). Now both converge on the same primitive.
try {
const { runCycle } = await import('../core/cycle.ts');
const report = await runCycle(engine, {
// #1872: track the promise so closeEngine can drain it on shutdown,
// and pass the abort signal so the cycle winds down between phases.
const cyclePromise = runCycle(engine, {
brainDir: repoPath,
// Autopilot daemon path: pulls by default (matches
// pre-v0.17 autopilot behavior). CLI dream defaults false
// for cron safety; that choice is scoped to dream only.
pull: true,
signal: shutdownAbort.signal,
yieldBetweenPhases: async () => {
await new Promise(r => setImmediate(r));
},
});
inflightInlineCycle = cyclePromise;
const report = await cyclePromise.finally(() => { inflightInlineCycle = null; });
// Only 'failed' (every attempted phase failed) trips the autopilot
// circuit breaker. 'partial' means at least one phase warned or
// failed while others ran — that's a soft signal, not a fatal
+23 -3
View File
@@ -26,6 +26,7 @@
import type { BrainEngine } from '../core/engine.ts';
import {
runCycle,
resolveSourceForDir,
ALL_PHASES,
type CyclePhase,
type CycleReport,
@@ -380,9 +381,9 @@ Options:
--source <id> Scope the cycle to one source so doctor's
cycle_freshness check sees a fresh stamp on
completion. Without this, gbrain dream's
timestamp never lands and federated brains
see "stale cycle" forever.
completion. When omitted, gbrain derives the
source from --dir / the configured checkout
when it matches a source's local_path (#1869).
--source-id <id> Alias for --source. Matches the v0.37.7.0+
naming used by import/extract/graph-query.
@@ -634,6 +635,25 @@ export async function runDream(engine: BrainEngine | null, args: string[]): Prom
);
process.exit(1);
}
// #1869: a path-scoped run (--dir, or the configured sync.repo_path) whose
// directory matches a registered source's local_path IS that source's cycle
// — derive the source id so runCycle writes last_source_cycle_at /
// last_full_cycle_at on success and doctor's cycle_freshness check stops
// reading perpetually stale. Explicit --source still wins (resolved above).
// Fixed here at the command level, NOT in runCycle's stamp gate, so legacy
// global callers (autopilot-global-maintenance runs GLOBAL_PHASES with a
// brainDir and no sourceId) can't falsely stamp per-source freshness.
// A derived match on an archived source is skipped silently (falls back to
// legacy unscoped behavior) — stamping it would mask staleness on restore,
// mirroring the explicit --source archived guard above.
if (resolvedSourceId === undefined && engine !== null && brainDir !== null) {
const derived = await resolveSourceForDir(engine, brainDir);
if (derived !== undefined) {
const src = await fetchSource(engine, derived);
if (src?.archived !== true) resolvedSourceId = derived;
}
}
// ─── issue #1678: bounded single-hold extract_atoms drain ──────────
if (opts.drain) {
if (engine === null) {
+1 -5
View File
@@ -22,7 +22,6 @@
*/
import { resolveRecipe } from './model-resolver.ts';
import { listRecipes } from './recipes/index.ts';
import { AIConfigError } from './errors.ts';
export interface ProviderCapabilities {
@@ -78,10 +77,7 @@ export function getProviderCapabilities(modelString: string): ProviderCapabiliti
if (!chat) {
throw new AIConfigError(
`Provider "${recipe.id}" does not offer a chat touchpoint.`,
// Computed from the registry so the hint can't drift into listing
// chat-less providers (the pre-fix list falsely included embedding-only
// recipes, sending users in circles — #1157).
`Known providers with chat: ${listRecipes().filter(r => r.touchpoints.chat).map(r => r.id).join(', ')}. Pick one for models.tier.subagent.`,
`Known providers with chat: openai, anthropic, google, openrouter, litellm-proxy, deepseek, groq, together, azure-openai, dashscope, minimax, zhipu, ollama, llama-server. Pick one for models.tier.subagent.`,
);
}
+4 -19
View File
@@ -1,10 +1,9 @@
import type { Recipe } from '../types.ts';
/**
* Zhipu AI (智谱AI) BigModel Open Platform. OpenAI-compatible /embeddings and
* /chat/completions endpoints at open.bigmodel.cn. Hosts embedding-2 (1024d),
* embedding-3 (Matryoshka up to 2048d), and the GLM chat family (glm-5.1 etc.)
* with native tool calling — usable for models.tier.subagent (#1157).
* Zhipu AI (智谱AI) BigModel Open Platform. OpenAI-compatible /embeddings
* endpoint at open.bigmodel.cn. Hosts embedding-2 (1024d) and embedding-3
* (Matryoshka up to 2048d).
*
* embedding-3 at 2048 dims exceeds pgvector's HNSW cap of 2000 — those
* brains fall back to exact vector scans (see
@@ -26,20 +25,6 @@ export const zhipu: Recipe = {
setup_url: 'https://open.bigmodel.cn/',
},
touchpoints: {
chat: {
// Informational list (openai-compat tier: assertTouchpoint doesn't
// enforce it), so newer GLM ids pass without a recipe edit.
models: ['glm-5.1', 'glm-4.6', 'glm-4.5'],
supports_tools: true,
// gbrain-side stable tool ids (v0.38 D11) decoupled the loop from
// Anthropic response formats; GLM tool calling is stable through the
// OpenAI-compat path, same as deepseek/groq.
supports_subagent_loop: true,
// Anthropic-style cache_control markers are not honored on the
// OpenAI-compat path — the loop runs hot (degraded:no_caching warn).
supports_prompt_cache: false,
max_context_tokens: 128000,
},
embedding: {
models: ['embedding-3', 'embedding-2'],
default_dims: 1024,
@@ -51,5 +36,5 @@ export const zhipu: Recipe = {
},
},
setup_hint:
'Get an API key at https://open.bigmodel.cn/, then `export ZHIPUAI_API_KEY=...`. Chat/subagent: use `zhipu:glm-5.1`.',
'Get an API key at https://open.bigmodel.cn/, then `export ZHIPUAI_API_KEY=...`',
};
+9 -1
View File
@@ -855,8 +855,16 @@ interface SyncPhaseResult extends PhaseResult {
* Resolve the source id for a brain directory by looking up the sources
* table. Returns undefined when no registered source matches (falls back
* to pre-v0.18 global config.sync.* keys).
*
* Exported for dream.ts (#1869): a `gbrain dream --dir <path>` run whose
* path matches a registered source's local_path is a per-source cycle in
* everything but name, so dream derives the source id up front and passes
* it as opts.sourceId — landing the freshness stamp without changing
* runCycle's stamp/lock semantics for legacy global callers (the
* autopilot-global-maintenance handler runs GLOBAL_PHASES with a brainDir
* and MUST NOT stamp per-source freshness; see rejected PR #2549).
*/
async function resolveSourceForDir(
export async function resolveSourceForDir(
engine: BrainEngine,
brainDir: string | null,
): Promise<string | undefined> {
-39
View File
@@ -69,45 +69,6 @@ describe('recipe: zhipu', () => {
expect(sql.toLowerCase()).toContain('hnsw');
});
test('chat touchpoint declares GLM models with tool + subagent-loop support (#1157)', () => {
const r = getRecipe('zhipu')!;
expect(r.touchpoints.chat).toBeDefined();
expect(r.touchpoints.chat!.models).toContain('glm-5.1');
expect(r.touchpoints.chat!.supports_tools).toBe(true);
expect(r.touchpoints.chat!.supports_subagent_loop).toBe(true);
expect(r.touchpoints.chat!.supports_prompt_cache).toBe(false);
});
test('zhipu:glm-5.1 passes the subagent capability gate (degraded:no_caching, not refused)', async () => {
// Pre-fix: getProviderCapabilities threw "does not offer a chat touchpoint"
// and classifyCapabilities returned 'unknown' → subagent submit refused.
const { getProviderCapabilities, classifyCapabilities } =
await import('../../src/core/ai/capabilities.ts');
const caps = getProviderCapabilities('zhipu:glm-5.1');
expect(caps.supportsToolCalling).toBe(true);
expect(classifyCapabilities('zhipu:glm-5.1')).toBe('degraded:no_caching');
});
test('no-chat-touchpoint error hint lists only providers that actually have chat', async () => {
// The hint is computed from the registry; every provider it names must
// really carry a chat touchpoint (pre-fix it hardcoded zhipu/dashscope/
// minimax, all embedding-only at the time).
const { getProviderCapabilities } = await import('../../src/core/ai/capabilities.ts');
const { listRecipes } = await import('../../src/core/ai/recipes/index.ts');
let hint = '';
try {
getProviderCapabilities('voyage:voyage-3');
throw new Error('expected AIConfigError for embedding-only provider');
} catch (e) {
hint = (e as { fix?: string }).fix ?? String(e);
}
const listed = hint.match(/chat: ([^.]+)\./)?.[1]?.split(', ') ?? [];
expect(listed.length).toBeGreaterThan(0);
const withChat = new Set(listRecipes().filter(r => r.touchpoints.chat).map(r => r.id));
for (const id of listed) expect(withChat.has(id)).toBe(true);
expect(listed).toContain('zhipu');
});
test('dimsProviderOptions threads dimensions for embedding-3 (Matryoshka)', async () => {
// Codex finding #1: Zhipu embedding-3 is Matryoshka 256-2048. Without
// `dimensions` on the wire, user-selected non-default dims are
@@ -0,0 +1,63 @@
/**
* #1872 — autopilot SIGTERM/SIGINT must close the engine before exit.
*
* On PGLite the cycle steps run INLINE in the autopilot process, so a hard
* `process.exit` mid-write (systemctl stop → SIGTERM) kills WASM Postgres
* with the WAL dirty and can corrupt the brain. Two exit paths must both
* close the engine:
*
* - autopilot's own shutdown() (owns SIGINT + internal stops like
* max_crashes / cycle-failure-cap), and
* - process-cleanup's SIGTERM handler (installed at cli.ts module load,
* which exits within its 3s cleanup deadline) — reached via the
* registered 'autopilot-engine-close' cleanup callback.
*
* Because the shutdown path is deep inside `runAutopilot()` (a long-running
* daemon loop that ends in process.exit), a behavioral test would have to
* spawn + signal a real daemon. Following the established precedent
* (test/autopilot-supervisor-wiring.test.ts, test/autopilot-fanout-wiring.test.ts),
* these static-shape regressions pin the load-bearing wiring instead.
*/
import { describe, expect, it } from 'bun:test';
import { readFileSync } from 'fs';
import { join } from 'path';
const AUTOPILOT_SRC = readFileSync(
join(import.meta.dir, '..', 'src', 'commands', 'autopilot.ts'),
'utf8',
);
describe('autopilot.ts graceful engine shutdown (#1872)', () => {
it('registers an engine-close callback in the process-cleanup registry (SIGTERM path)', () => {
// process-cleanup owns SIGTERM (installed at cli.ts:10) and hard-exits
// after its cleanup pass; without this registration the engine is never
// closed on `systemctl stop`.
expect(AUTOPILOT_SRC).toContain(
"import { registerCleanup } from '../core/process-cleanup.ts';",
);
expect(AUTOPILOT_SRC).toContain(
"registerCleanup('autopilot-engine-close', closeEngine)",
);
});
it('closeEngine aborts the in-flight inline cycle then disconnects the engine', () => {
// Abort first (runCycle checks the signal between phases and threads it
// into phase sub-work), bounded drain, then disconnect.
expect(AUTOPILOT_SRC).toMatch(
/const closeEngine = async \(\) => \{[\s\S]{0,900}shutdownAbort\.abort\([\s\S]{0,900}engine\.disconnect\(\)/,
);
});
it('the inline runCycle call carries the shutdown abort signal and is tracked as in-flight', () => {
// PGLite / --inline path: the cycle runs in-process, so shutdown must be
// able to (a) signal it to wind down and (b) await it before closing.
expect(AUTOPILOT_SRC).toMatch(/signal:\s*shutdownAbort\.signal/);
expect(AUTOPILOT_SRC).toMatch(/inflightInlineCycle\s*=\s*cyclePromise/);
});
it('shutdown() awaits closeEngine() before process.exit(0) (SIGINT + internal-stop path)', () => {
expect(AUTOPILOT_SRC).toMatch(
/await closeEngine\(\);[\s\S]{0,400}process\.exit\(0\)/,
);
});
});
+99
View File
@@ -0,0 +1,99 @@
/**
* #1869 — `gbrain dream --dir <path>` stamps cycle freshness when the path
* matches a registered source's local_path.
*
* Pre-fix, only `--source <id>` runs wrote last_source_cycle_at /
* last_full_cycle_at (runCycle's stamp gate reads opts.sourceId, and dream
* never derived one from --dir), so a path-scoped brain showed doctor's
* cycle_freshness as perpetually stale.
*
* The fix lives in dream.ts (derive the source id from the resolved brain
* dir via resolveSourceForDir), NOT in runCycle's stamp gate — a runCycle-
* wide change would make the autopilot-global-maintenance handler (global
* phases, brainDir set, no sourceId) falsely stamp per-source freshness
* (the #2194 poisoning class; see rejected PR #2549).
*
* Same real-PGLite/no-mocks discipline as test/dream.test.ts; same
* GBRAIN_HOME isolation as test/cycle-last-full-cycle-at.test.ts (the
* cycle's PGLite file lock lives under ~/.gbrain).
*/
import { describe, test, expect, beforeAll, afterAll, beforeEach, afterEach } from 'bun:test';
import { mkdtempSync, rmSync } from 'fs';
import { join } from 'path';
import { tmpdir } from 'os';
import { PGLiteEngine } from '../src/core/pglite-engine.ts';
import { resetPgliteState } from './helpers/reset-pglite.ts';
import { runDream } from '../src/commands/dream.ts';
import { withEnv } from './helpers/with-env.ts';
let engine: PGLiteEngine;
let brainDir: string;
let gbrainHome: string;
beforeAll(async () => {
engine = new PGLiteEngine();
await engine.connect({});
await engine.initSchema();
}, 60_000);
afterAll(async () => {
await engine.disconnect();
});
beforeEach(async () => {
await resetPgliteState(engine);
brainDir = mkdtempSync(join(tmpdir(), 'gbrain-dream-stamp-'));
gbrainHome = mkdtempSync(join(tmpdir(), 'gbrain-dream-stamp-home-'));
}, 60_000);
afterEach(() => {
rmSync(brainDir, { recursive: true, force: true });
rmSync(gbrainHome, { recursive: true, force: true });
});
async function seedSource(id: string, archived = false): Promise<void> {
await engine.executeRaw(
`INSERT INTO sources (id, name, local_path, config, archived, created_at)
VALUES ($1, $2, $3, '{}'::jsonb, $4, NOW())`,
[id, id, brainDir, archived],
);
}
async function readLastFullCycleAt(sourceId: string): Promise<string | null> {
const rows = await engine.executeRaw<{ config: Record<string, unknown> | null }>(
`SELECT config FROM sources WHERE id = $1`,
[sourceId],
);
const raw = rows[0]?.config?.last_full_cycle_at;
return typeof raw === 'string' ? raw : null;
}
describe('gbrain dream --dir <path> freshness stamp (#1869)', () => {
test('--dir matching a source local_path stamps last_full_cycle_at', async () => {
await withEnv({ GBRAIN_HOME: gbrainHome }, async () => {
await seedSource('path-scoped');
expect(await readLastFullCycleAt('path-scoped')).toBeNull();
const report = await runDream(engine, ['--dir', brainDir, '--phase', 'lint', '--json']);
expect(report).toBeTruthy();
if (report) expect(['ok', 'clean']).toContain(report.status);
// Pre-fix this stays null forever: dream never passed a sourceId, so
// runCycle's stamp gate skipped the write.
expect(await readLastFullCycleAt('path-scoped')).not.toBeNull();
});
}, 60_000);
test('--dir matching an ARCHIVED source does not stamp it', async () => {
await withEnv({ GBRAIN_HOME: gbrainHome }, async () => {
await seedSource('mothballed', true);
const report = await runDream(engine, ['--dir', brainDir, '--phase', 'lint', '--json']);
expect(report).toBeTruthy();
// Stamping an archived source would mask data staleness when it is
// later restored (mirrors the explicit --source archived guard).
expect(await readLastFullCycleAt('mothballed')).toBeNull();
});
}, 60_000);
});
+14 -4
View File
@@ -562,12 +562,22 @@ describe('runDream — --source / --source-id (v0.41.13)', () => {
// ─── Back-compat: bare `gbrain dream` does NOT write per-source stamp ─
test('gbrain dream (no --source) leaves all sources untouched (back-compat regression)', async () => {
await seedSource('alpha');
await seedSource('beta');
test('gbrain dream (no --source) stamps only the source whose local_path matches --dir (#1869)', async () => {
// Pre-#1869 this asserted NO source was ever stamped without an explicit
// --source — which is exactly the bug: a path-scoped `gbrain dream --dir`
// run never landed a freshness stamp and doctor's cycle_freshness stayed
// stale forever. New truth: the source whose local_path matches the
// resolved brain dir is derived and stamped; unrelated sources stay
// untouched (cross-source isolation).
await seedSource('alpha'); // local_path = repo → derived + stamped
await engine.executeRaw(
`INSERT INTO sources (id, name, local_path, config, archived, created_at)
VALUES ($1, $2, $3, '{}'::jsonb, false, NOW())`,
['beta', 'beta', '/somewhere/else'],
);
const report = await runDream(engine, ['--dir', repo, '--phase', 'lint', '--json']);
expect(report).toBeTruthy();
expect(await readLastFullCycleAt('alpha')).toBeNull();
expect(await readLastFullCycleAt('alpha')).not.toBeNull();
expect(await readLastFullCycleAt('beta')).toBeNull();
}, 60_000);