Summary

On macOS (LaunchAgent mode), restarting Gateway on v2026.4.26 can sometimes take ~3-4 minutes before the new process starts, even though normal startup is only a few seconds.

Environment

• OpenClaw: 2026.4.26 (be8c246)
• OS: macOS 15/26 (arm64)
• LaunchAgent label: ai.openclaw.gateway
• Active channel plugins include: openclaw-weixin, telegram

Observed behavior

A restart is requested (SIGTERM), but old process appears to linger for minutes before fresh startup begins.

Example timeline from logs:

• 12:57:37 [gateway] received SIGTERM; restarting
• 13:01:15 [gateway] loading configuration…
• 13:01:38 [gateway] ready

So total restart wall time is ~4 min, while internal startup path is fast:

• http server listening (...; 3.4s)

Additional signal

During startup there is also a silent gap before weixin provider init:

• 13:01:18.780 hooks loaded
• 13:01:38.209 weixin provider starting
  (~20s no logs)

Not sure if related, but the main pain is the pre-start delay between SIGTERM and fresh config load.

Also seen in same logs

Config warning for plugin manifest metadata:

• channel plugin manifest declares openclaw-weixin without channelConfigs metadata

(We tested adding channelConfigs: {} locally, but restart latency behavior still reproduced.)

Impact

Operationally painful for frequent restarts/debugging; perceived as "restart is stuck".

Request

Could you help check:

1. whether shutdown path in 2026.4.26 can block launchd supervisor restart;
2. whether plugin/channel teardown has a long timeout path without intermediate logs;
3. whether this is a known regression in 2026.4.26 and if a patch/workaround exists.

If useful, I can provide full redacted log slice around restart.