Bug: Cron scheduler timer never fires — every-type and all jobs affected

Environment

• OpenClaw version: 2026.2.6 (also reproduced on 2026.2.3-1)
• OS: Linux 6.14.0-37-generic (x64)
• Node: v25.5.0

Summary

The cron scheduler's timer mechanism is fundamentally broken in 2026.2.6. Two related but distinct issues:

1. every-type jobs: computeNextRunAtMs always returns a future timestamp, so runDueJobs never finds them due. This is the known bug from [Bug]: CRON jobs not executing at scheduled time with upgrade to 2026.2.3-1 #10812 / Cron scheduler advances nextRunAtMs but doesn't execute jobs #10655.

2. All job types (including cron expressions): Even after converting all every-type jobs to cron expressions as a workaround, the scheduler's onTimer callback never fires. The scheduler reports enabled: true with a valid nextWakeAtMs, but when that time passes, no jobs execute.

Steps to Reproduce

1. Configure several cron jobs (mix of every and cron expression types)
2. Observe that every-type jobs never execute (known issue)
3. Convert all every-type jobs to cron expressions
4. Restart gateway (both SIGUSR1 and full systemd restart attempted)
5. Run cron status — shows enabled: true, valid nextWakeAtMs
6. Wait past nextWakeAtMs — no jobs execute
7. cron runs shows no new entries after restart
8. cron run <jobId> --force works (bypasses due-check), confirming jobs themselves are valid

Root Cause Analysis

Issue 1: every-type computeNextRunAtMs

In gateway-cli-c_8Yf5s6.js (lines ~2925-2945), computeNextRunAtMs for every kind:

• Calculates intervals based on anchorMs and everyMs
• Always returns a future timestamp relative to Date.now()
• When onTimer → ensureLoaded(forceReload: true) → recomputeNextRuns is called, it recalculates all nextRunAtMs values
• This pushes every-type jobs perpetually into the future
• runDueJobs (which checks nextRunAtMs <= now) never finds them due

Issue 2: Timer never fires

After converting all jobs to cron expressions:

• cron status reports valid state with proper nextWakeAtMs
• Gateway logs show the cron module starting with different nextWakeAtMs values on restart
• But onTimer is never actually invoked when the wake time arrives
• No log entries for runDueJobs, armTimer, or any timer-firing events
• Gateway shows timeout errors on ws://127.0.0.1:18789 (60000ms) which may indicate a broken internal connection preventing scheduler operation

Workarounds Attempted

Only viable workaround: external system crontab calling openclaw cron run --force

Expected Behavior

1. every-type jobs should execute at their configured intervals
2. cron expression jobs should execute when their schedule matches
3. The scheduler timer should reliably fire at nextWakeAtMs

Related Issues

• [Bug]: CRON jobs not executing at scheduled time with upgrade to 2026.2.3-1 #10812 (Open) — every-type cron bug
• Cron scheduler advances nextRunAtMs but doesn't execute jobs #10655 (Closed) — related cron scheduling issue

Impact

• 12 cron jobs completely non-functional for 7+ hours
• Includes investment analysis, marketing automation, coding agent, and daily briefings
• All automated workflows halted; requires manual intervention with --force flag