diff --git a/src/cron/service/timeout-policy.test.ts b/src/cron/service/timeout-policy.test.ts index 69ca6aa46c3..8aae440f505 100644 --- a/src/cron/service/timeout-policy.test.ts +++ b/src/cron/service/timeout-policy.test.ts @@ -23,11 +23,30 @@ function makeJob(payload: CronJob["payload"]): CronJob { } describe("timeout-policy", () => { - it("uses default timeout for non-agent jobs", () => { + it("uses default timeout for systemEvent jobs with wakeMode=next-heartbeat", () => { const timeout = resolveCronJobTimeoutMs(makeJob({ kind: "systemEvent", text: "hello" })); expect(timeout).toBe(DEFAULT_JOB_TIMEOUT_MS); }); + it("uses expanded safety timeout for systemEvent jobs targeting main with wakeMode=now", () => { + // wakeMode="now" + sessionTarget="main" blocks until the full agent turn completes + // via runHeartbeatOnce(), so it needs the same generous ceiling as agentTurn jobs. + const job: CronJob = { + id: "job-1", + name: "job", + createdAtMs: 0, + updatedAtMs: 0, + enabled: true, + schedule: { kind: "every", everyMs: 60_000 }, + sessionTarget: "main", + wakeMode: "now", + payload: { kind: "systemEvent", text: "check in" }, + state: {}, + }; + const timeout = resolveCronJobTimeoutMs(job); + expect(timeout).toBe(AGENT_TURN_SAFETY_TIMEOUT_MS); + }); + it("uses expanded safety timeout for agentTurn jobs without explicit timeout", () => { const timeout = resolveCronJobTimeoutMs(makeJob({ kind: "agentTurn", message: "hi" })); expect(timeout).toBe(AGENT_TURN_SAFETY_TIMEOUT_MS); diff --git a/src/cron/service/timeout-policy.ts b/src/cron/service/timeout-policy.ts index 7b03b8bda52..607a8fa8921 100644 --- a/src/cron/service/timeout-policy.ts +++ b/src/cron/service/timeout-policy.ts @@ -19,7 +19,25 @@ export function resolveCronJobTimeoutMs(job: CronJob): number | undefined { ? Math.floor(job.payload.timeoutSeconds * 1_000) : undefined; if (configuredTimeoutMs === undefined) { - return job.payload.kind === "agentTurn" ? AGENT_TURN_SAFETY_TIMEOUT_MS : DEFAULT_JOB_TIMEOUT_MS; + // agentTurn jobs get the large safety ceiling since turns can run for a long time. + if (job.payload.kind === "agentTurn") { + return AGENT_TURN_SAFETY_TIMEOUT_MS; + } + // systemEvent jobs targeting "main" with wakeMode="now" call runHeartbeatOnce(), + // which blocks until the full agent turn completes. The turn duration is + // unbounded, so apply the same generous ceiling as agentTurn jobs instead of + // the short DEFAULT_JOB_TIMEOUT_MS that was incorrectly timing out long-running + // main-session heartbeats. See: https://github.com/openclaw/openclaw/issues/50621 + if ( + job.payload.kind === "systemEvent" && + job.sessionTarget === "main" && + job.wakeMode === "now" + ) { + return AGENT_TURN_SAFETY_TIMEOUT_MS; + } + // All other systemEvent jobs (wakeMode !== "now") are fire-and-forget: + // they enqueue the event and return immediately, so the short default is fine. + return DEFAULT_JOB_TIMEOUT_MS; } return configuredTimeoutMs <= 0 ? undefined : configuredTimeoutMs; }