openclaw - 💡(How to fix) Fix Container accumulates zombie processes — needs tini or --init as PID 1 wrapper [1 comments, 2 participants]

Problem

When OpenClaw runs as PID 1 in a Docker container (the default), the Node.js process does not reap orphaned child processes. Node's default SIGCHLD handling only calls waitpid() for children it spawned via child_process — it does NOT perform the waitpid(-1) sweep that a proper init system does.

Every process spawned via the exec tool that exits becomes a zombie. Over a container's lifetime this accumulates hundreds of zombie entries in the process table.

Observed

• PID 1: node /app/node_modules/.bin/openclaw start
• After ~2 hours of normal operation: 148 zombie processes (bash, git, node, python, etc.)
• Zombie count grows monotonically — they are never reaped
• Chromium-based tools that exit as zombies poison lock files, preventing browser servers from restarting

Root Cause

Node.js is not a proper init system. When it runs as PID 1, orphaned children get reparented to it, but it never calls waitpid(-1, WNOHANG) to reap them.

Fix

tini is already installed in the container at /usr/local/bin/tini. It just needs to be used as the entrypoint wrapper.

Dockerfile:

ENTRYPOINT ["tini", "--"]
CMD ["node", "/app/node_modules/.bin/openclaw", "start"]

Or docker run:

docker run --init ...

Or cloud deployments (Render, Railway, Fly):

tini -- node /app/node_modules/.bin/openclaw start

tini is 30KB, adds zero overhead, and is the standard solution. Docker's own --init flag literally injects tini.

Impact

• Process table pollution (148+ zombies after 2 hours)
• Browser/Chromium lock files poisoned by zombie processes
• PDF generation breaks when browse server is wedged
• Eventually hits kernel PID limit on long-running containers

References

• https://blog.phusion.nl/2015/01/20/docker-and-the-pid-1-zombie-reaping-problem/
• https://github.com/krallin/tini
• https://docs.docker.com/engine/reference/run/#specify-an-init-process