Socket-mode outage: 43s agent turn blocks gateway event loop, drops all 12 Slack sockets (2026.4.27) | Friends of the Crustacean 🦞🤝 | Page 1

frozen terrace May 7, 2026, 1:22 PM

#

Hitting a recurring socket-mode outage on OpenClaw 2026.4.27. A single 43-second claude agent turn blocks the gateway event loop and drops all 12 Slack sockets in parallel — confirmed via eventLoopDelayMaxMs=42446 in the gateway's own diagnostic sampler at the moment of disconnect. When DNS is healthy the sockets reconnect; when DNS hiccups at the same time, channels stay stuck (we had an 8.5-hour outage on May 7). HTTP /health returns ok throughout, so it's invisible to standard liveness monitoring. Issue #58519 looks like the upstream tracker. Anyone seen a workaround that doesn't require waiting for an architectural fix? Considering a worker_threads patch on the chunk handler but want to know if it's been tried.

strange slateBOT May 7, 2026, 1:22 PM

#

To help others find answers, you can mark your question as solved via Right click solution message -> Apps -> ✅ Mark Solution

frozen terrace May 7, 2026, 1:33 PM

#

any thoughts on this one?

#

any thoughts on the original post above?

#

@junior pine

#Socket-mode outage: 43s agent turn blocks gateway event loop, drops all 12 Slack sockets (2026.4.27)