Why 'docker stop' before a reboot left my whole homelab down for two days

My server reboots itself automatically when updates need it (here’s why). An early version of that reboot script tried to be careful: it stopped all containers cleanly first, then rebooted.

The server came back up. None of the containers did. Not Jellyfin, not Home Assistant, and not DNS, which meant no device in the house could resolve websites until I changed them to a public DNS server by hand. It stayed like that for about two days.

What went wrong

Almost all my services use:

restart: unless-stopped

The name says it exactly: restart the container unless it was stopped. If you stop a container with docker stop (or docker compose stop), Docker remembers that you wanted it off, and keeps it off, even after a reboot.

So my “careful” script did this:

1. docker stop on every container → Docker records “the user stopped these”
2. reboot
3. Docker starts → sees that every container was deliberately stopped → leaves them all stopped

The fix: let the OS stop Docker

Just reboot. When the host shuts down, systemd stops the Docker service, and Docker stops the containers gracefully without marking them as user-stopped. On boot, everything with unless-stopped or always comes back.

The script now goes straight to:

sync
systemctl reboot

Giving containers time to shut down

The reason I’d added docker stop in the first place was to give databases time to shut down cleanly. The right tool for that is the daemon’s shutdown timeout in /etc/docker/daemon.json:

{
  "shutdown-timeout": 60
}

Docker then gives containers up to 60 seconds to exit when the daemon stops. Restart Docker after changing it.

If it happens to you

Bring every stack back up in one go:

for d in ~/dockerapps/*/; do (cd "$d" && docker compose up -d); done

Restart policies cheat sheet

Policy	After a crash	After a reboot	After you docker stop it, then reboot
no	stays down	stays down	stays down
always	restarts	starts	starts
unless-stopped	restarts	starts	stays down

unless-stopped is still the right default: it lets you turn things off on purpose. Just never “turn everything off on purpose” right before a reboot.

How to check what will come back after a reboot

Before you reboot, this lists every container with its restart policy and whether Docker considers it user-stopped:

docker ps -a --format '{{.Names}}' | while read c; do
  docker inspect "$c" --format '{{.Name}}  policy={{.HostConfig.RestartPolicy.Name}}  running={{.State.Running}}'
done

Anything with policy=unless-stopped running=false will stay down after the reboot. That’s what you want for retired projects, and a nasty surprise for anything you forgot about.

Testing it safely

You don’t need a real reboot to test restart behaviour. sudo systemctl restart docker stops and starts the daemon the same way a reboot does, so containers come back according to their policy. Do it at a quiet time, because everything restarts once.