Running Containers, Lifecycle and Resources

Key points

1. 1

   Arguments after the image name replace CMD and are appended to ENTRYPOINT; --entrypoint replaces ENTRYPOINT and clears the image CMD.

2. 2

   docker stop sends SIGTERM (or STOPSIGNAL), waits 10 s on Linux, then SIGKILL. Exit codes above 128 mean a signal: 137 = SIGKILL, 143 = SIGTERM.

3. 3

   PID 1 ignores signals it has no handler for, and shell-form CMD makes /bin/sh PID 1. Use exec form, exec in entrypoint scripts, or --init to reap zombies and forward signals.

4. 4

   Restart policies: on-failure[:n] only on non-zero exits; always restarts after a daemon restart even if stopped manually; unless-stopped does not. Unhealthy containers are not restarted by the standalone engine.

5. 5

   --memory is a hard cap enforced by the OOM killer (check State.OOMKilled); --memory-swap is memory plus swap; --cpus caps CPU while --cpu-shares only weights it under contention; --pids-limit stops fork bombs.

6. 6

   The writable layer survives stop and restart but not docker rm; docker cp works on stopped containers and docker commit never includes volumes.

7. 7

   Only stdout and stderr reach docker logs. json-file does not rotate by default; local does; dual logging keeps docker logs working with remote drivers.

Read the source

• Running containers
• Start containers automatically
• Resource constraints