Always Restart Policy

The always restart policy unconditionally restarts a container whenever it stops, for any reason — including a deliberate, explicit docker stop — and even restarts it automatically if the Docker daemon itself restarts, making it the most aggressive of the available restart policies.

Configuring always

docker run -d --restart=always --name myapp myapp:1.0

Why always Restarts Even After a Deliberate Stop

Unlike unless-stopped, the always policy does not remember or respect a deliberate stop — if the Docker daemon restarts (a host reboot, a daemon upgrade) while a container that was explicitly stopped is configured with always, that container will be started again automatically.

docker run -d --restart=always --name myapp myapp:1.0
docker stop myapp
sudo systemctl restart docker
docker ps --filter "name=myapp"

Surprisingly, myapp appears running again after the daemon restart, despite having been explicitly stopped beforehand — this is the key distinguishing behavior of always versus unless-stopped.

Why This Behavior Can Be Surprising or Undesirable

An operator who deliberately stopped a container, intending for it to remain stopped, might be surprised to find it running again after an unrelated daemon restart — this is precisely the scenario unless-stopped exists to avoid.

docker run -d --restart=unless-stopped --name myapp myapp:1.0

For most production services, unless-stopped is generally the more predictable and commonly preferred choice for exactly this reason.

When always Might Still Be the Right Choice

For a container that genuinely should never remain stopped under any circumstances — perhaps a critical piece of infrastructure where even a deliberate stop should be quickly reversed by automation — always's unconditional behavior might be intentional and appropriate.

docker update --restart=always myapp

Why Understanding the always Policy Matters

Recognizing the specific, sometimes surprising difference between always and unless-stopped — particularly around how each treats a deliberate stop — is essential for choosing the restart policy that actually matches the intended operational behavior for a given container.