fix(cron): use agent-turn safety timeout for systemEvent jobs with wakeMode=now

When a cron job has sessionTarget="main", payload.kind="systemEvent", and
wakeMode="now", executeJobCore() calls runHeartbeatOnce() which blocks until
the full agent turn completes. The turn duration is unbounded (multi-step
tasks may take 30+ minutes), but resolveCronJobTimeoutMs() was returning
DEFAULT_JOB_TIMEOUT_MS (10 minutes) for all systemEvent jobs, causing
spurious "cron: job execution timed out" errors.

Fix: when wakeMode="now" + sessionTarget="main" + kind="systemEvent",
return AGENT_TURN_SAFETY_TIMEOUT_MS (60 minutes) instead of the short 10-minute default.

Fixes #50621
This commit is contained in:
HollyChou 2026-03-20 08:46:00 +08:00
parent 41eef15cdc
commit ec0a1d7797

View File

@ -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;
}