openclaw - 💡(How to fix) Fix v2026.4.29: idle gateway burns 14% CPU, causes Telegram timeouts and event loop delays [2 comments, 3 participants]

Description

After upgrading from v2026.4.15 to v2026.4.29, the gateway burns 14% CPU constantly on its main event loop thread even when completely idle (zero active agent runs).

Symptoms

• Main thread (LWP == PID) consumes 14% CPU 24/7
• Event loop delays spike to 7-9 seconds (logged as liveness warning: event_loop_delay)
• Telegram fetch operations timeout with ETIMEDOUT (event loop too busy to service HTTP responses)
• eventLoopUtilization hovers at 22-92%

Environment

• OpenClaw v2026.4.29 (npm global install, upgraded from v2026.4.15)
• Node 22.22.2
• Ubuntu 24.04, 16GB RAM, ~10GB free
• 5 agents configured (Cosmo, Snowden, Sirius, Delta, Gamma)
• All agents idle during testing

Reproduction

1. Install [email protected]
2. Start gateway with 4-5 agents configured
3. Wait for agents to finish startup
4. Observe top/ps: main thread burns 14% CPU indefinitely with zero agent activity
5. Observe journalctl: periodic liveness warnings and Telegram fetch timeouts

Downgrade fixes it

Reverting to [email protected] eliminates the CPU burning. The same agent config works with <1% idle CPU on v2026.4.15.

Thread analysis

ps -p <pid> -L -o pid,lwp,%cpu,stat,time,args | sort -k3 -rn | head -3
  PID   LWP %CPU STAT     TIME
 5349  5349 14.2 Rsl  00:09:47  (main event loop thread)
 5349  5355  0.4 Ssl  00:00:17  (worker thread)

The main event loop thread is in Runnable state (R) and accumulating CPU time at ~14%/sec steadily, suggesting a tight loop or excessive recurring work in the v2026.4.29 event loop.