Restart Policies

no, on-failure, always, unless-stopped

The restart policy controls what Docker does when a container exits: no (default), on-failure (restart after a non-zero exit, optionally with a retry limit), always (always restart, including after the Docker daemon restarts) and unless-stopped (like always, but not if you stopped it deliberately). For long-running services on a server, unless-stopped is a common choice; for one-off jobs, leave it as no.

Choosing a restart policy

Match the policy to the workload.

policy           restarts after crash   after daemon restart   typical use
no               no                     no                     one-off tasks, migrations
on-failure[:N]   yes (non-zero exit)    yes                    workers that may crash
always           yes                    yes                    critical services
unless-stopped   yes                    yes, unless stopped    services on a single server

Watch for crash loops

A service restarting every few seconds hides a real bug; check docker compose ps and logs for repeated restarts.

Quick check: Which policy suits a database migration job?

• always
• no (run once and exit)
• unless-stopped
• on-failure with infinite retries for a job that succeeds once