@tyhld/conductor 0.12.69 → 0.12.70

package/dist/headless.js

@@ -19,7 +19,7 @@ import { createInterface } from 'node:readline';
 import { createWriteStream, existsSync, mkdirSync, readFileSync, readdirSync, unlinkSync, writeFileSync } from 'node:fs';
 import path from 'node:path';
 import os from 'node:os';
-import { RELAY_DELIVER_MAX, RELAY_FETCH_TIMEOUT_MS, RELAY_HEARTBEAT_INTERVAL_MS, RELAY_PENDING_TIMEOUT_MS, RELAY_RESTORE_ORPHANS_ENABLED, buildAnchor, buildSendText, canDeliver, completeWithReport, createRelayState, deliveryReport, fetchCommand, fetchCommandStatus, fetchControl, fetchOrphans, isTerminalCommandStatus, kindOf, markDelivered, noteCentralDown, noteCentralUp, noteReadiness, patchStatus, reconnectDelayMs, selectOrphanToRestore, sendHeartbeat, } from './relay.js';
+import { CENTRAL_BUSY_HOLD_MAX_MS, RELAY_DELIVER_MAX, RELAY_FETCH_TIMEOUT_MS, RELAY_HEARTBEAT_INTERVAL_MS, RELAY_RESTORE_ORPHANS_ENABLED, buildAnchor, buildSendText, canDeliver, completeWithReport, createRelayState, deliveryReport, fetchCommand, fetchCommandStatus, fetchControl, fetchOrphans, isTerminalCommandStatus, kindOf, markDelivered, noteCentralDown, noteCentralUp, noteReadiness, patchStatus, reconnectDelayMs, selectOrphanToRestore, sendHeartbeat, } from './relay.js';
 const log = (msg) => {
     console.log(`[headless] ${msg}`);
 };
@@ -471,6 +471,7 @@ export function spawnClaudeProcess(cfg, cmd, sessionId, attempt, isResume) {
         exitCode: null,
         exitSignal: null,
         stopped: false,
+        timedOut: false,
         eventBuffer: [],
         lastFlushAt: Date.now(),
         lastStopCheckAt: Date.now(),
@@ -626,6 +627,23 @@ async function checkRunningJob(cfg, state) {
     if (job.completed) {
         await flushEvents(cfg, job);
         clearActiveJob(cfg.site);
+        if (job.timedOut) {
+            // 上限時間で強制停止した → 管制へ理由つきで failed を記録する
+            const elapsed = Math.round((Date.now() - job.startedAt) / 1000);
+            const reason = `上限時間(${Math.round(CENTRAL_BUSY_HOLD_MAX_MS / 1000)}s)で強制停止 (実行${elapsed}s, exit=${job.exitCode}, signal=${job.exitSignal})`;
+            await sendCommandEvents(cfg, job.commandId, [{
+                    type: 'error',
+                    text: reason.slice(0, EVENT_TEXT_MAX),
+                    at: new Date().toISOString(),
+                    attempt: job.attempt,
+                }]);
+            await patchStatus(cfg, job.commandId, 'failed', reason.slice(0, EVENT_TEXT_MAX));
+            rs.abandoned.add(job.commandId);
+            state.job = null;
+            rs.lastHeartbeatAt = null;
+            log(`WARN site=${cfg.site} id=${job.commandId} ${reason}`);
+            return;
+        }
         if (job.stopped) {
             // stop_requested で止めた
             await sendCommandEvents(cfg, job.commandId, [{
@@ -719,10 +737,12 @@ async function checkRunningJob(cfg, state) {
             return;
         }
     }
-    // タイムアウト
-    if (now - job.startedAt > RELAY_PENDING_TIMEOUT_MS) {
+    // タイムアウト — 中央の上限（3時間）で強制停止する。
+    // ★RELAY_PENDING_TIMEOUT_MS(15分)は tmux の pending 解除用。画面なし方式では
+    //   作業中の職人そのものを殺すので、中央と同じ CENTRAL_BUSY_HOLD_MAX_MS を使う。
+    if (now - job.startedAt > CENTRAL_BUSY_HOLD_MAX_MS) {
         log(`WARN site=${cfg.site} id=${job.commandId} タイムアウト(${Math.round((now - job.startedAt) / 1000)}s) → SIGTERM`);
-        job.stopped = true;
+        job.timedOut = true;
         await stopJob(job);
         return;
     }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@tyhld/conductor",
-  "version": "0.12.69",
+  "version": "0.12.70",
   "description": "采配くん管制の見守りアプリ（conductor-agent）。各PCで常駐し、中央(devlog-tracker)へ定期的に生存報告(heartbeat)を送る常駐CLI。",
   "type": "module",
   "bin": {