From be8930d6f9eb00ad124a67bc9219de4827ca836b Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Yasunori=20Morishima=EF=BC=88=E7=9B=9B=E5=B3=B6=E5=BA=B7?=
 =?UTF-8?q?=E5=BE=B3=EF=BC=89?= <fw_yasu11@hotmail.com>
Date: Mon, 2 Mar 2026 22:47:36 +0900
Subject: [PATCH] fix: clear stale runningAtMs in cron.run() before
 already-running check (#17949)

Add recomputeNextRunsForMaintenance() call in run() so that stale
runningAtMs markers (from a crashed Phase-1 persist) are cleared by the
existing normalizeJobTickState logic before the already-running guard.

Without this, a manual cron.run() could be blocked for up to
STUCK_RUN_MS (2 hours) even though no job was actually running.

Fixes #17554

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
---
 src/cron/service.issue-regressions.test.ts | 56 ++++++++++++++++++++++
 src/cron/service/ops.ts                    |  4 ++
 2 files changed, 60 insertions(+)

diff --git a/src/cron/service.issue-regressions.test.ts b/src/cron/service.issue-regressions.test.ts
index dd82b52ab..ba32e3fdf 100644
--- a/src/cron/service.issue-regressions.test.ts
+++ b/src/cron/service.issue-regressions.test.ts
@@ -8,6 +8,7 @@ import * as schedule from "./schedule.js";
 import { CronService } from "./service.js";
 import { createDeferred, createRunningCronServiceState } from "./service.test-harness.js";
 import { computeJobNextRunAtMs } from "./service/jobs.js";
+import { run } from "./service/ops.js";
 import { createCronServiceState, type CronEvent } from "./service/state.js";
 import {
   DEFAULT_JOB_TIMEOUT_MS,
@@ -1450,6 +1451,61 @@ describe("Cron issue regressions", () => {
     expect(startedAtEvents).toEqual([dueAt, dueAt + 50]);
   });
 
+  it("#17554: run() clears stale runningAtMs and executes the job", async () => {
+    const store = await makeStorePath();
+    const now = Date.parse("2026-02-06T10:05:00.000Z");
+    const staleRunningAtMs = now - 2 * 60 * 60 * 1000 - 1;
+
+    await fs.writeFile(
+      store.storePath,
+      JSON.stringify(
+        {
+          version: 1,
+          jobs: [
+            {
+              id: "stale-running",
+              name: "stale-running",
+              enabled: true,
+              createdAtMs: now - 3_600_000,
+              updatedAtMs: now - 3_600_000,
+              schedule: { kind: "at", at: new Date(now - 60_000).toISOString() },
+              sessionTarget: "main",
+              wakeMode: "now",
+              payload: { kind: "systemEvent", text: "stale-running" },
+              state: {
+                runningAtMs: staleRunningAtMs,
+                lastRunAtMs: now - 3_600_000,
+                lastStatus: "ok",
+                nextRunAtMs: now - 60_000,
+              },
+            },
+          ],
+        },
+        null,
+        2,
+      ),
+      "utf-8",
+    );
+
+    const enqueueSystemEvent = vi.fn();
+    const state = createCronServiceState({
+      cronEnabled: true,
+      storePath: store.storePath,
+      log: noopLogger,
+      nowMs: () => now,
+      enqueueSystemEvent,
+      requestHeartbeatNow: vi.fn(),
+      runIsolatedAgentJob: vi.fn().mockResolvedValue({ status: "ok", summary: "ok" }),
+    });
+
+    const result = await run(state, "stale-running", "force");
+    expect(result).toEqual({ ok: true, ran: true });
+    expect(enqueueSystemEvent).toHaveBeenCalledWith(
+      "stale-running",
+      expect.objectContaining({ agentId: undefined }),
+    );
+  });
+
   it("honors cron maxConcurrentRuns for due jobs", async () => {
     vi.useRealTimers();
     const store = await makeStorePath();
diff --git a/src/cron/service/ops.ts b/src/cron/service/ops.ts
index 2b7ebf57f..dd02ca4ab 100644
--- a/src/cron/service/ops.ts
+++ b/src/cron/service/ops.ts
@@ -341,6 +341,10 @@ export async function run(state: CronServiceState, id: string, mode?: "due" | "f
   const prepared = await locked(state, async () => {
     warnIfDisabled(state, "run");
     await ensureLoaded(state, { skipRecompute: true });
+    // Normalize job tick state (clears stale runningAtMs markers) before
+    // checking if already running, so a stale marker from a crashed Phase-1
+    // persist does not block manual triggers for up to STUCK_RUN_MS (#17554).
+    recomputeNextRunsForMaintenance(state);
     const job = findJobOrThrow(state, id);
     if (typeof job.state.runningAtMs === "number") {
       return { ok: true, ran: false, reason: "already-running" as const };