Docker Exec Background Process Detachment
SkillCloud & infraFix docker exec hanging when starting background processes (servers, daemons) inside containers. Use when: (1) docker exec never returns despite using & or nohup, (2) background server started in container causes docker exec to hang indefinitely, (3) env_startup_command in SWE-bench or similar frameworks times out, (4) setsid/disown needed for proper process detachment in Docker. Root cause: docker exec tracks ALL processes in the exec session, not just the top-level PID.
Available today. Use it from your connected AI after setup.
No other account needed.
Connect ahel once, and every AI you use reads what you have installed.
Then ask your AI: use the Docker Exec Background Process Detachment skill
What this skill tells your AI
The instructions your AI receives, as published by disentinel/grafema in .claude/skills/docker-exec-background-process/SKILL.md and read by ahel’s review.
Problem
When using docker exec to start a background process (like a database server) inside a
container, the docker exec command hangs indefinitely and never returns, even when using
&, nohup, or output redirection.
Context / Trigger Conditions
docker exec container bash -c "my-server &"never returnsdocker exec container bash -c "nohup my-server > /dev/null 2>&1 &"still hangs- Startup commands in frameworks (SWE-bench, mini-SWE-agent) time out at 300/600s
- Any scenario where
docker execneeds to start a long-running daemon and return
Root Cause
Docker exec tracks ALL processes started in the exec session, not just the top-level bash process. This is different from regular bash behavior:
&— creates background job but bash still tracks it in its job tablenohup— prevents SIGHUP but does NOT create a new session; docker still tracks it- Output redirection (
>/dev/null 2>&1) — prevents output but docker exec still waits for all session processes to terminate
Docker exec uses Linux process sessions/groups. All processes started within an exec call share the same session. Docker waits for the entire session to finish.
Solution
Use setsid + disown + stdin/stdout/stderr redirection:
docker exec container bash -c 'setsid /path/to/server arg1 arg2 </dev/null >/dev/null 2>&1 & disown && sleep 1 && echo "Ready"'
All four components are required:
| Component | Purpose |
|---|---|
setsid | Creates a NEW session (new session ID, detaches from exec session) |
</dev/null | Prevents stdin inheritance from docker exec |
>/dev/null 2>&1 | Prevents stdout/stderr from keeping pipe fds alive |
& disown | Removes job from bash's job table so bash doesn't wait for it |
Why each is necessary:
- Without
setsid: Process stays in docker exec's session → docker waits - Without
</dev/null: Process may hold stdin fd → docker waits - Without
>/dev/null 2>&1: Process may hold pipe fds (especially with| tail) → docker waits - Without
disown: Bash waits for background jobs before exiting → docker waits
Related Problem: Pipe FD Inheritance
A compounding issue occurs when using pipes in the startup command:
# THIS HANGS: tail waits for EOF, but server inherits the pipe fd
docker exec container bash -c 'my-server 2>&1 | tail -3'
If my-server spawns child processes with stdio: 'inherit' (common in Node.js), those
children inherit the pipe file descriptor. tail waits for ALL writers to close the pipe,
but the inherited fd in the child process keeps it open forever.
Fix: Don't pipe server output. Use >/dev/null 2>&1 or write to a log file:
# Instead of piping, redirect to file and read separately
docker exec container bash -c 'my-server > /tmp/server.log 2>&1 & sleep 2 && tail -3 /tmp/server.log'
Verification
After running the startup command:
docker execshould return promptly (within thesleepduration)docker exec container ps auxshould show the server process still running- The server should be accessible (check socket, port, etc.)
Example: Starting rfdb-server in SWE-bench
# mini-SWE-agent config
run:
env_startup_command: |
ln -sf /opt/grafema/node_modules/.bin/grafema /usr/local/bin/grafema && \
cp /opt/node_modules/@grafema/rfdb/prebuilt/linux-x64/rfdb-server /usr/local/bin/rfdb-server && \
chmod +x /usr/local/bin/rfdb-server && \
setsid /usr/local/bin/rfdb-server /testbed/.grafema/graph.rfdb \
--socket /testbed/.grafema/rfdb.sock </dev/null >/dev/null 2>&1 & disown && \
sleep 2 && echo "Server ready"
What Does NOT Work
# FAILS: nohup doesn't create new session
docker exec container bash -c 'nohup my-server &'
# FAILS: output redirection alone isn't enough
docker exec container bash -c 'my-server > /dev/null 2>&1 &'
# FAILS: disown alone isn't enough (still in same session)
docker exec container bash -c 'my-server & disown'
# FAILS: even all three without setsid
docker exec container bash -c 'nohup my-server > /dev/null 2>&1 & disown'
Notes
setsidis available on most Linux distributions (part of util-linux)- The
sleepafterdisowngives the server time to initialize before the script continues - If the server needs to be verified running, check its pidfile or socket after the sleep
- On macOS Docker Desktop, x86_64 emulation adds ~10x latency to all operations
- For SWE-bench/mini-SWE-agent,
env_startup_commandruns viadocker exec bash -c "...", so this pattern applies directly
Signals
- GitHub stars
- 36
- Forks
- 3
- Last commit
- Aug 2026
ahel review
S4info
community integration — published by disentinel, not docker
Automated review, not a security audit. Ruleset v1.
Advanced
- Catalog kind
- skill
- Gateway key
docker-exec-background-process- Source
- github.com/disentinel/grafema