openclaw - 💡(How to fix) Fix Performance regression: Control UI startup RPCs take 5-10s and can delay agent runs [1 comments, 2 participants]

Summary

On OpenClaw 2026.5.18, Control UI startup often spends 5-10 seconds on basic gateway RPCs (commands.list, models.list, sessions.list, chat.history, models.authStatus). During this time, gateway health can report high event-loop utilization / CPU, and sending a webchat message can be delayed long enough to trigger request timeout behavior.

This was observed after upgrading from 2026.5.7 to 2026.5.18.

Environment

• OpenClaw: 2026.5.18 (50a2481)
• macOS: 15.7.4
• Node: v22.22.0
• Codex CLI package: @openai/codex 0.131.0
• Codex app: /Applications/Codex.app, CFBundleShortVersionString=26.513.31313
• Main model: openai/gpt-5.5
• Runtime provider: openai-codex/gpt-5.5
• Telegram disabled for later testing: channels.telegram.enabled=false

Observed slow startup RPCs

After gateway restart and Control UI reload:

2026-05-19T12:17:33.283+09:00 info gateway/ws webchat connected conn=5c1e7ab1... client=openclaw-control-ui webchat vcontrol-ui
2026-05-19T12:17:34.195+09:00 info gateway/ws ⇄ res ✓ health 237ms cached=true
2026-05-19T12:17:34.645+09:00 info gateway/ws ⇄ res ✓ agents.list 685ms
2026-05-19T12:17:42.308+09:00 info gateway/ws ⇄ res ✓ commands.list 7506ms
2026-05-19T12:17:42.536+09:00 info gateway/ws ⇄ res ✓ models.list 7736ms
2026-05-19T12:17:42.719+09:00 info gateway/ws ⇄ res ✓ sessions.list 7919ms
2026-05-19T12:17:42.841+09:00 info gateway/ws ⇄ res ✓ chat.history 8042ms
2026-05-19T12:17:43.881+09:00 info gateway/ws ⇄ res ✓ models.authStatus 9080ms

After archiving a large old agent:main:main session and creating a new main session, the first load still showed high latency:

2026-05-19T12:28:27.127+09:00 info gateway/ws webchat connected conn=53dc15d3... client=openclaw-control-ui webchat vcontrol-ui
2026-05-19T12:28:28.697+09:00 info gateway/ws ⇄ res ✓ health 491ms cached=true
2026-05-19T12:28:29.389+09:00 info gateway/ws ⇄ res ✓ agents.list 1196ms
2026-05-19T12:28:38.296+09:00 info gateway/ws ⇄ res ✓ chat.history 8535ms
2026-05-19T12:28:38.726+09:00 info gateway/ws ⇄ res ✓ commands.list 8963ms
2026-05-19T12:28:38.952+09:00 info gateway/ws ⇄ res ✓ models.list 9187ms
2026-05-19T12:28:39.137+09:00 info gateway/ws ⇄ res ✓ sessions.list 9374ms
2026-05-19T12:28:40.444+09:00 info gateway/ws ⇄ res ✓ models.authStatus 10682ms

Health symptoms

At times during this behavior:

Gateway event loop: degraded reasons=event_loop_delay,event_loop_utilization max=9093ms p99=9093ms util=1 cpu=0.893

After things settled, health could return to ok, but initial Control UI load remained slow.

Impact

• The Control UI feels very slow after reload.
• Sending a message while these startup RPCs are still running can lead to webchat request timeout / aborted fallback chain.
• CLI agent runs work, but Control UI is unreliable or slow.

Expected behavior

Initial Control UI RPCs should not take 5-10 seconds under this configuration, and they should not delay agent requests enough to trigger webchat timeouts.

Notes

A previous large agent:main:main session entry was archived. It reduced one suspected factor, but did not fully explain the startup RPC latency. The session store was about 391KB before archiving, with the old main entry about 268KB; after archiving, slow Control UI startup still occurred.