Zombie process accumulation when openab runs as PID 1

[agent-worker-use]

Description

When openab runs as PID 1 in a container, it does not reap orphaned child processes, leading to zombie process accumulation over time.

• openab runs as PID 1 (container entrypoint: openab run /etc/openab/config.toml)
• openab forks kiro-cli (ACP mode), which in turn forks worker processes (e.g., python3, pkill)
• When these worker processes terminate and their direct parent has already exited, they are reparented to PID 1 (openab)
• openab does not call wait() or waitpid(), causing these orphaned processes to remain as zombies (Z state)
• After 2 days of container runtime, 9 zombie processes accumulated, all with PPID=1
• Zombies consume no CPU or memory, but occupy PID table entries; long-running containers risk PID space exhaustion

Steps to Reproduce

1. Start openab as container PID 1 (openab run config.toml)
2. Trigger kiro-cli via ACP to execute shell subprocesses (e.g., subprocess.run() launching python3)
3. Allow worker processes to complete normally
4. Observe with ps aux — zombie processes appear with Z (defunct) state and PPID=1

Expected Behavior

Zombie processes should be reaped and removed from the process table. Suggested solutions:

1. SIGCHLD Handler: Register a SIGCHLD signal handler that calls waitpid(-1, WNOHANG) to reap zombies
2. Child Subreaper: Use prctl(PR_SET_CHILD_SUBREAPER) combined with reaper logic
3. Init Wrapper: Document the recommendation to use a dedicated init process (e.g., tini, dumb-init) as the container entrypoint

Environment

• openab version: 0.7.3
• Container runtime: Kubernetes pod (k3s)
• OS: Linux

Screenshots / Logs

$ ps -eo pid,ppid,stat,comm | grep Z
     65       1 Z    pkill
     67       1 Z    python3
     81       1 Z    python3
     87       1 Z    python3
    158       1 Z    python3
    436       1 Z    python3
    462       1 Z    python3
    533       1 Z    python3
    735       1 Z    pkill

This is a known issue with PID 1 processes in containers. The Go runtime does not automatically handle SIGCHLD, so explicit zombie reaping is required when running as PID 1.

[github-actions]

Thanks for the report! It looks like this issue was created without a template and is missing some required fields. Please add one of the following labels: bug, feature, documentation, or guidance, then edit this issue to include the matching template fields — the incomplete label will be removed automatically.

[agent-worker-use]

Closing to resubmit with the correct bug report template.