@ai-dossier/cli 0.92.1 → 0.93.0-next.714.1

package/dist/commands/sched.js

@@ -49,6 +49,7 @@ Object.defineProperty(exports, "__esModule", { value: true });
 exports.UPGRADE_RESTART_EXIT_CODE = void 0;
 exports.createBatchCapabilityRunner = createBatchCapabilityRunner;
 exports.reexecUpdatedCli = reexecUpdatedCli;
+exports.buildEngineDeps = buildEngineDeps;
 exports.registerSchedCommand = registerSchedCommand;
 const node_child_process_1 = require("node:child_process");
 const node_fs_1 = __importDefault(require("node:fs"));
@@ -73,9 +74,11 @@ const runstate_1 = require("../runstate");
 const runstate_stats_1 = require("../runstate-stats");
 const sched_alert_1 = require("../sched-alert");
 const sched_run_stats_1 = require("../sched-run-stats");
+const sched_service_1 = require("../sched-service");
+const sched_tick_runner_1 = require("../sched-tick-runner");
 const table_1 = require("../table");
 const batch_prep_1 = require("../usage/batch-prep");
-const sched_service_1 = require("./sched-service");
+const sched_service_2 = require("./sched-service");
 /**
  * Batch-worktree `ai-dossier cap run <id>` runner for the per-member
  * incremental gate (#523 AC2). `cap run`'s `task-failed` outcome — a
@@ -93,13 +96,17 @@ const sched_service_1 = require("./sched-service");
  */
 function createBatchCapabilityRunner(opts) {
     const timeoutMs = opts?.timeoutMs ?? batch_suite_runner_1.BATCH_SUITE_TIMEOUT_MS;
+    const inFlightDir = opts?.inFlightDir;
     return (worktree, capabilityId) => {
-        const run = (0, cap_envelope_1.spawnCapRun)(capabilityId, worktree, { timeoutMs });
+        const run = (0, cap_envelope_1.spawnCapRun)(capabilityId, worktree, {
+            timeoutMs,
+            ...(inFlightDir !== undefined ? { inFlightDir } : {}),
+        });
         const result = run.spawned;
         // Captured subprocess output — the evidence fallback on every path below.
         const captured = `${result.stdout ?? ''}${result.stderr ?? ''}`;
         if (result.error && !run.exitedWithVerdict) {
-            // #681: this runner's OWN timeout (spawnSync's, `BATCH_SUITE_TIMEOUT_MS`)
+            // #681: this runner's OWN timeout (`BATCH_SUITE_TIMEOUT_MS` — spawnSync's, or the tracked launcher's process-group kill with a synthetic ETIMEDOUT, #1233)
             // must be distinguishable from a genuine machinery failure — before, it
             // collapsed into the evidence-free `{outcome: 'automation-broken'}`
             // below, byte-for-byte the #679 shape. Emit the same
@@ -1233,11 +1240,24 @@ function registerStatusSubcommand(cmd) {
             // — it reads whatever `sched start` last cached rather than risking
             // a multi-second hang on an unreachable npm registry.
             const staleness = await (0, engine_version_1.checkEngineStaleness)({ noFetch: true });
+            // #1233: gates the engine launched and has not consumed — after an
+            // engine death this is where "gate pid N is still running" shows.
+            const gatesInFlight = (0, sched_1.listInFlight)((0, sched_1.inFlightDir)(store.dir));
             if (opts.json) {
-                console.log(JSON.stringify({ ...report, engine_staleness: staleness }, null, 2));
+                console.log(JSON.stringify({
+                    ...report,
+                    engine_staleness: staleness,
+                    gates_in_flight: gatesInFlight.map((g) => ({
+                        ...g.record,
+                        alive: g.alive,
+                        finished: g.finished,
+                    })),
+                }, null, 2));
             }
             else {
                 console.log(renderReport(report, staleness));
+                for (const g of gatesInFlight)
+                    console.log(`Gate in flight: ${(0, sched_1.describeInFlight)(g)}`);
             }
         }
         catch (err) {
@@ -1282,8 +1302,9 @@ function resumeBatchGate(opts) {
         exec: (0, sched_1.createExecFn)(sched_1.TEARDOWN_TIMEOUT_MS, {
             onError: (file, args, err) => process.stderr.write(`⚠ sched resume --batch: '${file} ${args.join(' ')}' failed: ${err.message}\n`),
         }),
-        runSuite: (0, batch_suite_runner_1.createBatchSuiteRunner)(config),
-        runCapability: createBatchCapabilityRunner(),
+        // #1233: adopt (never duplicate) a gate the engine already has in flight here.
+        runSuite: (0, batch_suite_runner_1.createBatchSuiteRunner)(config, { inFlightDir: (0, sched_1.inFlightDir)(store.dir) }),
+        runCapability: createBatchCapabilityRunner({ inFlightDir: (0, sched_1.inFlightDir)(store.dir) }),
     };
     try {
         const result = (0, sched_1.resumeBlockedGate)(deps, config, (0, sched_1.resolveDispatch)(config), opts.batch, new Date());
@@ -2048,6 +2069,90 @@ function reexecUpdatedCli(execveOverride) {
         return false;
     }
 }
+/**
+ * #1233: the off-thread tick runner for the continuous loop, or null for the
+ * inline tick (`DOSSIER_SCHED_INLINE_TICK=1`, or no compiled worker entry —
+ * running from sources). The inline fallback is announced: it is the mode in
+ * which a long gate freezes the heartbeat and defers signals.
+ */
+function resolveTickRunner(data) {
+    if (process.env[sched_tick_runner_1.INLINE_TICK_ENV] === '1') {
+        console.error(`⚠ sched: ${sched_tick_runner_1.INLINE_TICK_ENV}=1 — ticks run on the main thread; a long gate freezes the lease heartbeat and defers SIGTERM/SIGINT`);
+        return null;
+    }
+    const file = (0, sched_tick_runner_1.tickWorkerFile)();
+    if (file === null) {
+        console.error('⚠ sched: tick worker entry not found next to the CLI build — ticks run on the main thread');
+        return null;
+    }
+    return (0, sched_tick_runner_1.createWorkerTickRunner)(file, data);
+}
+/**
+ * The engine's dependencies for `sched start` — one place, so the inline tick
+ * and the tick worker (#1233, `sched-tick-worker.ts`) build identical engines.
+ */
+function buildEngineDeps(opts) {
+    const { store, journal, anchorRepo, getTickConfig } = opts;
+    const repoDir = opts.repoDir ?? process.cwd();
+    const gatesInFlightDir = (0, sched_1.inFlightDir)(store.dir);
+    return {
+        store,
+        journal,
+        groundTruth: (0, sched_1.createExecGroundTruth)(undefined, {
+            repoDir,
+            ...(anchorRepo !== undefined ? { repo: anchorRepo } : {}),
+        }),
+        ...(anchorRepo !== undefined ? { anchorRepo } : {}),
+        spawnDeps: (0, sched_1.createSpawnDeps)(repoDir),
+        now: () => new Date(),
+        repoDir,
+        teardownExec: (0, sched_1.createExecFn)(sched_1.TEARDOWN_TIMEOUT_MS, {
+            onError: (file, args, err) => process.stderr.write(`⚠ sched teardown: '${file} ${args.join(' ')}' failed: ${err.message}\n`),
+        }),
+        // #945: takeover rescue (git commit-tree + push of a dead agent's WIP), run
+        // OUTSIDE the state lock. Bounded like every other engine exec; failures
+        // are per-call diagnostics — a failed rescue never blocks the respawn.
+        rescueExec: (0, sched_1.createExecFn)(RESCUE_TIMEOUT_MS, {
+            // A push that wants credentials must fail, never hang on a prompt.
+            env: { GIT_TERMINAL_PROMPT: '0' },
+            onError: (file, args, err) => process.stderr.write(`⚠ sched rescue: '${file} ${args.slice(0, 2).join(' ')}…' failed: ${err.message}\n`),
+        }),
+        // #504: the ladder fences a superseded run before respawning its takeover.
+        // Its own exec rather than the ground-truth one: a fence is a WRITE, and
+        // borrowing `groundTruthExec` would file the only diagnostic for a failed write
+        // under `sched ground truth`, where nobody debugging a fence would look.
+        fencer: (0, sched_1.createExecRunFencer)((0, sched_1.createExecFn)(sched_1.FENCE_TIMEOUT_MS, {
+            onError: (file, args, err) => process.stderr.write(`⚠ sched fence: '${file} ${args.join(' ')}' failed: ${err.message}\n`),
+        }), { repoDir }),
+        // #840: a requeued parked batch member's first dispatch seeds its
+        // resume trail (a `setup done` milestone on its member branch) so the
+        // full-cycle gate resumes ON that branch. A write, so its own exec +
+        // diagnostic prefix, on the fence's budget (two short CLI calls).
+        resumeSeeder: (0, sched_1.createExecResumeSeeder)((0, sched_1.createExecFn)(sched_1.FENCE_TIMEOUT_MS, {
+            onError: (file, args, err) => process.stderr.write(`⚠ sched resume-seed: '${file} ${args.join(' ')}' failed: ${err.message}\n`),
+        }), { repoDir }),
+        // #523: batch git/milestone-CLI operations (worktree claim, commit-range
+        // recording, milestone posting, PR watch) and the aggregate suite runner
+        // that gates `executing → reviewing`. Reuses the same timeout as teardown
+        // (worktree/git ops). The cold-path warm-up install/build (#561) does
+        // NOT reuse this — see `batchWarmExec` below — a real `npm ci`+build
+        // routinely exceeds this budget.
+        batchExec: (0, sched_1.createExecFn)(sched_1.TEARDOWN_TIMEOUT_MS, {
+            onError: (file, args, err) => process.stderr.write(`⚠ sched batch: '${file} ${args.join(' ')}' failed: ${err.message}\n`),
+        }),
+        // #561: batch-setup's cold-path warm-up (install + build) on its own
+        // budget, matching `@ai-dossier/worktree-pool`'s own per-issue warm-up
+        // budget for the identical work (`WARM_COMMAND_TIMEOUT_MS`) rather than
+        // the git-op-tuned `batchExec` above.
+        batchWarmExec: (0, sched_1.createExecFn)(worktree_pool_1.WARM_COMMAND_TIMEOUT_MS, {
+            onError: (file, args, err) => process.stderr.write(`⚠ sched batch warm-up: '${file} ${args.join(' ')}' failed: ${err.message}\n`),
+        }),
+        // One config per tick: the suite runner reads the snapshot the tick started with.
+        // #1233: engine gates are tracked in the in-flight registry (adopted, never duplicated).
+        runBatchSuite: (0, batch_suite_runner_1.createBatchSuiteRunner)(getTickConfig, { inFlightDir: gatesInFlightDir }),
+        runBatchCapability: createBatchCapabilityRunner({ inFlightDir: gatesInFlightDir }),
+    };
+}
 function registerStartSubcommand(cmd) {
     cmd
         .command('start')
@@ -2057,6 +2162,7 @@ function registerStartSubcommand(cmd) {
         .option('--once', 'Run a single reconcile+refill tick and exit (cron-style)')
         .option('--auto-upgrade', 'After a completed tick, install and re-exec @ai-dossier/cli@latest when this engine is stale')
         .option('--alert-issue <n>', 'Comment on this issue when the engine restarts after a crash (#945); default $DOSSIER_SCHED_ALERT_ISSUE. The alert is always journaled and printed to stderr')
+        .option('--unsupervised', 'Acknowledge running without a supervisor (no `sched service install`) and silence the warning (#1233)')
         .option('--project <slug>', 'Project slug (default: owner-repo of the current directory)')
         .option('--json', 'Output tick results as JSON')
         .action(async (opts) => {
@@ -2119,7 +2225,7 @@ function registerStartSubcommand(cmd) {
         if (acquisition.reclaimed) {
             (0, sched_1.reportCrashRestart)(journalAtStart, acquisition.reclaimed, (0, sched_alert_1.createAlertNotifier)(project, () => (0, sched_1.resolveProjectRepo)(project, sched_1.defaultExec) ?? undefined, alertIssue, { stateDir: store.dir }), new Date(), previousStop !== null && previousStop.pid === acquisition.reclaimed.pid
                 ? previousStop
-                : null);
+                : null, { stateDir: store.dir });
         }
         try {
             let config;
@@ -2228,62 +2334,12 @@ function registerStartSubcommand(cmd) {
             if (anchorRepo === undefined && !(opts.once && opts.json)) {
                 console.log(`▶ sched anchor close: off — the current directory is not ${project}'s GitHub repository`);
             }
-            const deps = {
+            const deps = buildEngineDeps({
                 store,
                 journal: engineJournal,
-                groundTruth: (0, sched_1.createExecGroundTruth)(undefined, {
-                    repoDir: process.cwd(),
-                    ...(anchorRepo !== undefined ? { repo: anchorRepo } : {}),
-                }),
-                ...(anchorRepo !== undefined ? { anchorRepo } : {}),
-                spawnDeps: (0, sched_1.createSpawnDeps)(process.cwd()),
-                now: () => new Date(),
-                repoDir: process.cwd(),
-                teardownExec: (0, sched_1.createExecFn)(sched_1.TEARDOWN_TIMEOUT_MS, {
-                    onError: (file, args, err) => process.stderr.write(`⚠ sched teardown: '${file} ${args.join(' ')}' failed: ${err.message}\n`),
-                }),
-                // #945: takeover rescue (git commit-tree + push of a dead agent's WIP), run
-                // OUTSIDE the state lock. Bounded like every other engine exec; failures
-                // are per-call diagnostics — a failed rescue never blocks the respawn.
-                rescueExec: (0, sched_1.createExecFn)(RESCUE_TIMEOUT_MS, {
-                    // A push that wants credentials must fail, never hang on a prompt.
-                    env: { GIT_TERMINAL_PROMPT: '0' },
-                    onError: (file, args, err) => process.stderr.write(`⚠ sched rescue: '${file} ${args.slice(0, 2).join(' ')}…' failed: ${err.message}\n`),
-                }),
-                // #504: the ladder fences a superseded run before respawning its takeover.
-                // Its own exec rather than the ground-truth one: a fence is a WRITE, and
-                // borrowing `groundTruthExec` would file the only diagnostic for a failed write
-                // under `sched ground truth`, where nobody debugging a fence would look.
-                fencer: (0, sched_1.createExecRunFencer)((0, sched_1.createExecFn)(sched_1.FENCE_TIMEOUT_MS, {
-                    onError: (file, args, err) => process.stderr.write(`⚠ sched fence: '${file} ${args.join(' ')}' failed: ${err.message}\n`),
-                }), { repoDir: process.cwd() }),
-                // #840: a requeued parked batch member's first dispatch seeds its
-                // resume trail (a `setup done` milestone on its member branch) so the
-                // full-cycle gate resumes ON that branch. A write, so its own exec +
-                // diagnostic prefix, on the fence's budget (two short CLI calls).
-                resumeSeeder: (0, sched_1.createExecResumeSeeder)((0, sched_1.createExecFn)(sched_1.FENCE_TIMEOUT_MS, {
-                    onError: (file, args, err) => process.stderr.write(`⚠ sched resume-seed: '${file} ${args.join(' ')}' failed: ${err.message}\n`),
-                }), { repoDir: process.cwd() }),
-                // #523: batch git/milestone-CLI operations (worktree claim, commit-range
-                // recording, milestone posting, PR watch) and the aggregate suite runner
-                // that gates `executing → reviewing`. Reuses the same timeout as teardown
-                // (worktree/git ops). The cold-path warm-up install/build (#561) does
-                // NOT reuse this — see `batchWarmExec` below — a real `npm ci`+build
-                // routinely exceeds this budget.
-                batchExec: (0, sched_1.createExecFn)(sched_1.TEARDOWN_TIMEOUT_MS, {
-                    onError: (file, args, err) => process.stderr.write(`⚠ sched batch: '${file} ${args.join(' ')}' failed: ${err.message}\n`),
-                }),
-                // #561: batch-setup's cold-path warm-up (install + build) on its own
-                // budget, matching `@ai-dossier/worktree-pool`'s own per-issue warm-up
-                // budget for the identical work (`WARM_COMMAND_TIMEOUT_MS`) rather than
-                // the git-op-tuned `batchExec` above.
-                batchWarmExec: (0, sched_1.createExecFn)(worktree_pool_1.WARM_COMMAND_TIMEOUT_MS, {
-                    onError: (file, args, err) => process.stderr.write(`⚠ sched batch warm-up: '${file} ${args.join(' ')}' failed: ${err.message}\n`),
-                }),
-                // One config per tick: the suite runner reads the snapshot the tick started with.
-                runBatchSuite: (0, batch_suite_runner_1.createBatchSuiteRunner)(() => tickConfig),
-                runBatchCapability: createBatchCapabilityRunner(),
-            };
+                anchorRepo,
+                getTickConfig: () => tickConfig,
+            });
             const describe = (result) => {
                 const parts = [];
                 if (result.spawned.length > 0)
@@ -2368,8 +2424,23 @@ function registerStartSubcommand(cmd) {
                 exitLogger.logNormalExit('once tick complete');
                 return;
             }
+            // #1233: an unsupervised engine dies for good the first time anything
+            // kills it (the 2026-09-29 engine sat dead for 15 days). Say so loudly.
+            if (!opts.unsupervised) {
+                const warning = (0, sched_service_1.unsupervisedWarning)(project, () => (0, sched_service_1.serviceStatus)(project, (0, sched_service_2.serviceIo)()));
+                if (warning !== null)
+                    console.error(warning);
+            }
             const interval = (engineConfig.reconcile_interval_ms ?? sched_1.DEFAULT_RECONCILE_INTERVAL_MS) / 1000;
             console.log(`▶ Scheduler engine running for ${project} (tick every ${interval}s, Ctrl-C to stop)`);
+            // #1233: run each tick in a worker thread so this event loop stays free
+            // during a long gate — the lease keeps its heartbeat and SIGTERM/SIGINT
+            // are handled (and journaled) instead of piling up behind a spawnSync.
+            const tickRunner = resolveTickRunner({
+                stateDir: store.dir,
+                repoDir: process.cwd(),
+                ...(anchorRepo !== undefined ? { anchorRepo } : {}),
+            });
             let upgradeHandoff = false;
             await (0, sched_1.runLoop)(deps, () => {
                 tickConfig = configReloader.current();
@@ -2393,7 +2464,8 @@ function registerStartSubcommand(cmd) {
                     upgradeHandoff = true;
                     stopping = true;
                 }
-            });
+            }, tickRunner !== null ? { runTick: (cfg) => tickRunner.runTick(cfg) } : {});
+            await tickRunner?.close();
             if (upgradeHandoff) {
                 exitLogger.logNormalExit('auto-upgrade handoff');
                 releaseLease();
@@ -2427,6 +2499,6 @@ function registerSchedCommand(program) {
     registerReprioritizeSubcommand(schedCmd);
     registerStartSubcommand(schedCmd);
     registerStatsSubcommand(schedCmd);
-    (0, sched_service_1.registerSchedServiceCommands)(schedCmd);
+    (0, sched_service_2.registerSchedServiceCommands)(schedCmd);
 }
 //# sourceMappingURL=sched.js.map