novel sigil
OpenClaw 2026.5.2 (8b2a6e5), systemd gateway, Telegram channel.
Symptom:
Control UI intermittently shows:
sessions list failed: Error: gateway request timeout for sessions.list
Observed logs:
sessions.list 56625ms
sessions.list 91509ms
later sessions.list 117ms / 130ms after load clears
Liveness:
eventLoopDelayP99Ms=4794.1
eventLoopDelayMaxMs=12742.3
eventLoopUtilization=0.942
cpuCoreRatio=1.046
active=1 waiting=0 queued=1
Process:
main openclaw Node thread ~90% CPU
libuv/node worker threads idle
memory fine, no IO wait
Context:
active Telegram topic around 266k tokens / 27% context
trajectory file around 20.5 MB
session store before cleanup: 519 files, 70 .bak files
after sessions cleanup: sessions CLI improved from ~10.9s to ~3.8s
Behavior:
Closing the Control UI significantly improved liveness:
eventLoopUtilization dropped to ~0.298
eventLoopDelayP99Ms dropped to ~266ms
Also saw Telegram startup/network noise:
fetch-timeout api.telegram.org getMe
DNS-resolved IP unreachable
deleteWebhook failed but getWebhookInfo confirmed no webhook
Host had broken IPv6 route; IPv4 works. Added gai.conf IPv4 preference and NODE_OPTIONS=--dns-result-order=ipv4first.
Question:
Is there a 2026.5.2 regression or setting related to Control UI sessions polling / sessions.changed / sessions.list on large active sessions? Is there a recommended contextTokens/session compaction setting to avoid gateway event-loop stalls?