openclaw - 💡(How to fix) Fix Bug: Zombie processes accumulate in worker containers after child process exits

Bug Description

After running HiClaw workers for a period of time, zombie processes accumulate in each worker container. Each worker container typically has 2 zombie processes:

1. [worker-entrypoi] - the worker entrypoint script
2. [hiclaw] - the hiclaw CLI process

Root Cause Analysis

The openclaw-gateway process inside worker containers spawns child processes (worker-entrypoint and hiclaw) but does not properly handle SIGCHLD signals to reap exited child processes.

Evidence

# Zombie process distribution (15 workers x 2 zombies each + 3 from manager)
$ ps aux | awk '$8=="Z"' | wc -l
32

# Each worker has 2 zombies:
$ docker exec hiclaw-worker-data-engineer ps aux | awk '$8=="Z"'
PID   PPID  STAT  COMMAND
143   0     Z     [worker-entrypoi]
199   0     Z     [hiclaw]

# Parent processes are openclaw-gateway:
$ ps -eo pid,ppid,stat,comm | awk '$3 ~ /^Z/ {print $2}' | sort | uniq -c
      2 163193  (openclaw-gateway)
      2 164766  (openclaw-gateway)
      ...

Impact

• Zombie processes do not consume CPU or memory
• They occupy entries in the process table (currently 32/4,194,304)
• Long-term accumulation could eventually exhaust the process table
• No functional impact at current levels

Workaround

Restarting worker containers clears the zombie processes temporarily.

Environment

• HiClaw Version: latest (hiclaw-worker:latest)
• OS: Ubuntu 26.04 LTS (host)
• Kernel: 7.0.0-15-generic

Suggested Fix

The openclaw-gateway process should:

1. Install a SIGCHLD signal handler
2. Call waitpid(-1, NULL, WNOHANG) in a loop to reap all exited children
3. Or use signal(SIGCHLD, SIG_IGN) to auto-reap children

Reproduction Steps

1. Start HiClaw with multiple workers
2. Wait for workers to execute tasks (which spawn child processes)
3. Observe zombie processes accumulating in worker containers