fix(queue): harden drain/abort/timeout race handling

- reject new lane enqueues once gateway drain begins
- always reset lane draining state and isolate onWait callback failures
- persist per-session abort cutoff and skip stale queued messages
- avoid false 600s agentTurn timeout in isolated cron jobs

Fixes #27407
Fixes #27332
Fixes #27427

Co-authored-by: Kevin Shenghui <shenghuikevin@github.com>
Co-authored-by: zjmy <zhangjunmengyang@gmail.com>
Co-authored-by: suko <miha.sukic@gmail.com>

src/cron/service.issue-regressions.test.ts

@@ -9,7 +9,7 @@ import { CronService } from "./service.js";
 import { createDeferred, createRunningCronServiceState } from "./service.test-harness.js";
 import { computeJobNextRunAtMs } from "./service/jobs.js";
 import { createCronServiceState, type CronEvent } from "./service/state.js";
-import { executeJobCore, onTimer, runMissedJobs } from "./service/timer.js";
+import { DEFAULT_JOB_TIMEOUT_MS, executeJobCore, onTimer, runMissedJobs } from "./service/timer.js";
 import type { CronJob, CronJobState } from "./types.js";
 
 const noopLogger = {
@@ -838,6 +838,58 @@ describe("Cron issue regressions", () => {
     expect(job?.state.lastStatus).toBe("ok");
   });
 
+  it("does not time out agentTurn jobs at the default 10-minute safety window", async () => {
+    const store = await makeStorePath();
+    const scheduledAt = Date.parse("2026-02-15T13:00:00.000Z");
+
+    const cronJob = createIsolatedRegressionJob({
+      id: "agentturn-default-safety-window",
+      name: "agentturn default safety window",
+      scheduledAt,
+      schedule: { kind: "at", at: new Date(scheduledAt).toISOString() },
+      payload: { kind: "agentTurn", message: "work" },
+      state: { nextRunAtMs: scheduledAt },
+    });
+    await writeCronJobs(store.storePath, [cronJob]);
+
+    let now = scheduledAt;
+    const deferredRun = createDeferred<{ status: "ok"; summary: string }>();
+    const runIsolatedAgentJob = vi.fn(async ({ abortSignal }: { abortSignal?: AbortSignal }) => {
+      const result = await deferredRun.promise;
+      if (abortSignal?.aborted) {
+        return { status: "error" as const, error: String(abortSignal.reason) };
+      }
+      now += 5;
+      return result;
+    });
+    const state = createCronServiceState({
+      cronEnabled: true,
+      storePath: store.storePath,
+      log: noopLogger,
+      nowMs: () => now,
+      enqueueSystemEvent: vi.fn(),
+      requestHeartbeatNow: vi.fn(),
+      runIsolatedAgentJob,
+    });
+
+    const timerPromise = onTimer(state);
+    let settled = false;
+    void timerPromise.finally(() => {
+      settled = true;
+    });
+
+    await vi.advanceTimersByTimeAsync(DEFAULT_JOB_TIMEOUT_MS + 1_000);
+    await Promise.resolve();
+    expect(settled).toBe(false);
+
+    deferredRun.resolve({ status: "ok", summary: "done" });
+    await timerPromise;
+
+    const job = state.store?.jobs.find((entry) => entry.id === "agentturn-default-safety-window");
+    expect(job?.state.lastStatus).toBe("ok");
+    expect(job?.state.lastError).toBeUndefined();
+  });
+
   it("aborts isolated runs when cron timeout fires", async () => {
     vi.useRealTimers();
     const store = await makeStorePath();

src/cron/service/timer.ts

@@ -36,6 +36,7 @@ const MIN_REFIRE_GAP_MS = 2_000;
  * from wedging the entire cron lane.
  */
 export const DEFAULT_JOB_TIMEOUT_MS = 10 * 60_000; // 10 minutes
+const AGENT_TURN_SAFETY_TIMEOUT_MS = 60 * 60_000; // 60 minutes
 
 type TimedCronRunOutcome = CronRunOutcome &
   CronRunTelemetry & {
@@ -52,7 +53,7 @@ function resolveCronJobTimeoutMs(job: CronJob): number | undefined {
       ? Math.floor(job.payload.timeoutSeconds * 1_000)
       : undefined;
   if (configuredTimeoutMs === undefined) {
-    return DEFAULT_JOB_TIMEOUT_MS;
+    return job.payload.kind === "agentTurn" ? AGENT_TURN_SAFETY_TIMEOUT_MS : DEFAULT_JOB_TIMEOUT_MS;
   }
   return configuredTimeoutMs <= 0 ? undefined : configuredTimeoutMs;
 }