Bug

The gateway becomes unresponsive when handling multiple subagent completion announcements, causing cascading timeouts and failed deliveries. This is distinct from #7042 (which was hard crashes) - the gateway stays alive but stops processing requests effectively.

Symptoms

1. Subagent completion timeouts - repeated warnings:
   \
   [warn] Subagent announce completion direct announce agent call transient failure, retrying 2/4 in 5s: gateway timeout after 120000ms
   [warn] Subagent announce completion direct announce agent call transient failure, retrying 3/4 in 10s: gateway timeout after 120000ms
   \\

2. Provider profile failover - Anthropic requests time out, triggering failover:
   \
   Profile anthropic:manual timed out. Trying next account...
   embedded run failover decision: rotate_profile, failoverReason: timeout
   \\

3. Lane wait exceeded - sessions queue up waiting for processing:
   \
   lane wait exceeded: waitedMs=158509
   \\

4. Eventually the gateway stops responding to Telegram/channels entirely until restart.

Impact

• Overnight cron jobs (subagent builds) fail to deliver results
• Channels go silent even though gateway process is still running
• Watchdog wrapper doesn't help because process hasn't crashed

Environment

• OpenClaw \2026.4.5\
• Windows 10 (x64)
• Node.js v25.5.0
• Gateway as Windows Scheduled Task with watchdog wrapper
• Workload: overnight build cron spawning Sonnet subagents (30-60 min builds)

Reproduction

1. Configure overnight cron job that spawns subagent builds
2. Run 2-3 builds back-to-back (e.g. overnight build + dream cycle)
3. Gateway progressively slows, completion announcements start timing out
4. Eventually gateway becomes unresponsive to all channels

Logs

From \openclaw-2026-04-10.log:

• 08:05:56 - First subagent timeout (retry 2/4)
• 08:06:31 - Anthropic profile timeout, failover triggered
• 08:08:01 - Subagent timeout (retry 3/4)
• Pattern continues with increasing delays

Suggested Investigation

• Memory leak under sustained subagent load?
• WebSocket connection pool exhaustion?
• Anthropic API timeout handling blocking the event loop?

Workaround

Currently none - manual gateway restart required once symptoms appear. The watchdog from #7042 only helps with crashes, not unresponsive states.