openclaw - 💡(How to fix) Fix Gateway crashes from unhandled promise: 'Agent listener invoked outside active run' [1 participants]

Bug Report

Version: 2026.4.5 (3e72c03) Node: v25.6.1 OS: Linux 6.17.13-2-pve (x64), LXC container

Description

The gateway process crashes with an unhandled promise rejection when an orphaned exec subprocess writes to stdout/stderr after the agent run has ended. This causes the gateway to exit with status 1, and systemd restarts it — which can create a crash loop if the orphan process writes repeatedly.

Reproduction

1. Start the gateway with any agent that has exec-based tools
2. A cron job or session spawns an exec child process
3. The gateway restarts (SIGUSR1 for config change, or crashes for another reason)
4. The exec child process survives the restart (gets reparented to PID 1)
5. When the orphan process writes to its stdout/stderr pipe, the agent event handler fires Agent.processEvents outside of an active run
6. Unhandled promise rejection → Node.js exits with code 1

Stack Trace

[openclaw] Unhandled promise rejection: Error: Agent listener invoked outside active run
    at Agent.processEvents (node_modules/@mariozechner/pi-agent-core/src/agent.ts:533:10)
    at node_modules/@mariozechner/pi-agent-core/src/agent.ts:380:21
    at Object.onUpdate (node_modules/@mariozechner/pi-agent-core/src/agent-loop.ts:539:7)
    at emitUpdate (dist/exec-defaults-uj0McX2k.js:1524:8)
    at handleStdout (dist/exec-defaults-uj0McX2k.js:1546:4)
    at Object.onSupervisorStdout [as onStdout] (dist/exec-defaults-uj0McX2k.js:1610:3)
    ...
    at Socket.<anonymous> (dist/exec-defaults-uj0McX2k.js:568:4)

Also observed with handleStderr path.

Observed Impact

• 10 crashes in ~10 minutes when crash loop occurs
• Gateway becomes unresponsive during rapid restart cycles
• All channels (Telegram, Discord) lose connectivity

Expected Behavior

Either:

1. Kill child processes on SIGUSR1/restart — the gateway should terminate all exec subprocess children before restarting in-process
2. Handle gracefully — catch the "outside active run" error and ignore it instead of letting it become an unhandled rejection
3. Both — kill children AND handle the edge case defensively

Workaround

A startup hook that finds orphaned processes with open FDs to .jsonl session files and kills them before the new gateway instance starts processing.

Environment

• Running in LXC container on Proxmox
• Multiple agents (ada, k2, cora, winston, synergy) + 4 tool agents
• 13 cron jobs, several of which use exec tools
• KillMode=control-group in systemd service (doesn't help for SIGUSR1 in-process restarts)