@kognai/orchestrator-core 0.2.4 → 0.2.5

package/dist/lib/sprint-runner-engine.js

@@ -886,15 +886,52 @@ async function runSprintCycle(opts) {
     // overlapping orphans can't accumulate even if cron ever gets re-armed
     const orchestratorTimeoutMs = PER_RUN_HARD_TIMEOUT_MIN * 60 * 1000;
     log(`Spawning orchestrator with ${PER_RUN_HARD_TIMEOUT_MIN}-min hard timeout`);
-    const result = (0, child_process_1.spawnSync)('npx', ['ts-node', orchestratorPath, activePath], {
-        stdio: 'inherit',
-        cwd: ROOT,
-        env: { ...process.env },
-        timeout: orchestratorTimeoutMs,
-        killSignal: 'SIGKILL', // SIGTERM ignored when blocked on subprocess I/O
+    // TICKET-347: spawnSync's `timeout` only SIGKILLs the DIRECT child (`npx`).
+    // Its `ts-node` child + the orchestrator grandchildren (and everything THEY
+    // spawn) get orphaned and keep running — reparented to init. A live incident
+    // (2026-06-13) had a "timed-out" run keep executing for ~2h after the runner
+    // declared it failed, burning ~1.3M tokens + attempting wallet settlements.
+    // Fix: spawn DETACHED so the child leads its own process group, then SIGKILL
+    // the WHOLE group (negative pid) on timeout — taking the entire subtree down.
+    const result = await new Promise((resolveRun) => {
+        const child = (0, child_process_1.spawn)('npx', ['ts-node', orchestratorPath, activePath], {
+            stdio: 'inherit',
+            cwd: ROOT,
+            env: { ...process.env },
+            detached: true, // new process group (setsid) → killable as a unit
+        });
+        let timedOut = false;
+        let settled = false;
+        const finish = (status) => {
+            if (settled)
+                return;
+            settled = true;
+            clearTimeout(timer);
+            resolveRun({ status, timedOut });
+        };
+        const timer = setTimeout(() => {
+            timedOut = true;
+            // Negative pid = signal the entire process group (the detached subtree).
+            try {
+                if (child.pid)
+                    process.kill(-child.pid, 'SIGKILL');
+            }
+            catch { /* group already gone */ }
+            // Belt-and-braces: also target the direct child in case the group call missed.
+            try {
+                child.kill('SIGKILL');
+            }
+            catch { /* already dead */ }
+        }, orchestratorTimeoutMs);
+        child.on('error', (err) => { log(`Orchestrator spawn error: ${err.message}`); finish(1); });
+        child.on('exit', (code) => finish(code));
     });
     const elapsed = Math.round((Date.now() - start) / 60000);
-    const status = result.status === 0 ? '✅ Completed' : `❌ Failed (exit ${result.status})`;
+    const status = result.status === 0
+        ? '✅ Completed'
+        : result.timedOut
+            ? `⏱️ Killed — ${PER_RUN_HARD_TIMEOUT_MIN}-min hard timeout (process group SIGKILLed)`
+            : `❌ Failed (exit ${result.status})`;
     log(`Orchestrator finished: ${status} (${elapsed} min)`);
     // Founder directive 2026-05-26: on non-zero exit, write an incident record
     // and emit an event for CTO + CEO to investigate autonomously. The swarm

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@kognai/orchestrator-core",
-  "version": "0.2.4",
+  "version": "0.2.5",
   "description": "Kognai sovereign orchestrator — core engine (template-agnostic). Shared by all products (Kognai/coding, Voxight/market-intel, Invoica/fin-compliance); each supplies only its template. Replaces per-repo forks of orchestrate-agents-v2 / sprint-runner / lib.",
   "license": "MIT",
   "author": "SkinGem",