openclaw - 💡(How to fix) Fix Cron/manual runs queue behind healthy long-lived cron sessions under light load

Summary

The cron execution lane appears to serialize all cron/manual runs behind long-lived isolated cron sessions, so fresh runs can wait tens of seconds even when queue depth is effectively one active run plus at most 0-2 queued items.

Symptoms

On a local 2026-05-27 gateway log under otherwise light load:

• cron: queued manual run waiting for an execution slot for cla-work-loop with waitMs=33344, queuedAhead=2
• another cla-work-loop manual run with waitMs=27326, queuedAhead=1, then skipped as already-running
• jarl-work-loop manual run with waitMs=13233, queuedAhead=0
• paired diagnostic warnings: lane wait exceeded: lane=cron waitedMs=...

At the same time, adjacent isolated cron sessions were simply long-lived, not obviously stuck. Recent durations from the same instance:

• cla-work-loop: 96s, 172s, 186s, 308s
• omnifocus-replan-morning: 209s, 213s, 254s, 330s
• mail-inbox-triage: occasional 350s run

So one active long cron run is enough to delay unrelated cron/manual work materially.

Why this looks like a real bug

The waits are not just cosmetic queue chatter from a large backlog. The log shows queuedAhead=0 still waiting 13s for a slot, and queuedAhead=1..2 waiting 27-33s. That makes cron/manual timing drift noticeable even under light load.

Expected

Either:

1. cron work should have enough execution capacity/isolation that one healthy long-lived run does not starve unrelated due/manual runs for tens of seconds, or
2. the product should make that serialization/throughput limit explicit and provide supported knobs to separate heavy cron lanes from routine cron/manual dispatch.

Observed log lines

2026-05-27T02:24:16Z cron: queued manual run waiting for an execution slot {jobId=0b9583..., waitMs=33344, queuedAhead=2}
2026-05-27T02:24:16Z lane wait exceeded: lane=cron waitedMs=33344 queueAhead=2
2026-05-27T02:24:26Z cron: queued manual run waiting for an execution slot {jobId=0b9583..., waitMs=27326, queuedAhead=1}
2026-05-27T02:24:26Z cron: queued manual run skipped before execution {jobId=0b9583..., reason=already-running}
2026-05-27T02:24:26Z cron: queued manual run waiting for an execution slot {jobId=d154ec5..., waitMs=13233, queuedAhead=0}
2026-05-27T02:24:26Z lane wait exceeded: lane=cron waitedMs=13233 queueAhead=0

Environment

• OpenClaw stable, local macOS deployment
• isolated cron sessions
• multiple agents, but the problematic queueing happened with only a few concurrent cron/manual runs rather than a large burst