`every` 定时任务防漂移落地手册（PR #14832）

解决“每次任务执行后都往后漂移几秒，累计后严重错点”的场景：把 nextRun 计算锚定到 lastRun 而非 endedAt。

GITHUBDiscovered 2026-02-13Author shtse8

Prerequisites

OpenClaw version includes PR #14832 or equivalent scheduler fix.
You have recurring jobs with `schedule.kind = every` and can inspect run history.

Steps

List all `every` jobs and record configured intervals versus observed trigger timestamps.
Upgrade to PR #14832 build and restart gateway during a low-risk window.
Run a controlled test: create a 60s `every` job with 5-10s execution duration and sample 10+ runs.
Verify scheduler snaps forward from `lastRunAt + everyMs` and does not accumulate end-time drift.
Keep alerting on abnormal drift as a regression guard.

Commands

openclaw gateway status

openclaw gateway restart

openclaw cron list

openclaw cron runs --job <jobId>

Verify

After 10+ cycles, trigger points remain aligned to expected interval boundaries, without +execution-time drift.

Caveats

This fix targets `every` schedules; cron expression jobs use a different next-run path（需验证）.
Long runtime beyond multiple intervals still causes snap-forward skips by design.

Source attribution

This tip is aggregated from community/public sources and preserved with attribution.