Run containerized tasks with restart policies

The ease and speed of deploying containers in Azure Container Instances provides a compelling platform for executing run-once tasks like build, test, and image rendering in a container instance.

With a configurable restart policy, you can specify that your containers are stopped when their processes complete. Because Azure bills container instances by the second, you're only charged for the compute resources used while the container executing your task is running.

The examples presented in this article use the Azure CLI. You must have Azure CLI version 2.0.21 or greater installed locally.

Container restart policy

When you create a container group in Azure Container Instances, you can specify one of three restart policy settings.

Restart policy	Description
Always	Containers in the container group are always restarted. This policy is the default setting applied when no restart policy is specified at container creation.
Never	Containers in the container group are never restarted. The containers run at most once.
OnFailure	Containers in the container group are restarted only when the process executed in the container fails (when it terminates with a nonzero exit code). The containers are run at least once.

Specify a restart policy

How you specify a restart policy depends on how you create your container instances, such as with the Azure CLI, Azure PowerShell cmdlets, or in the Azure portal. In the Azure CLI, specify the --restart-policy parameter when you call az container create.

az container create \
    --resource-group myResourceGroup \
    --name mycontainer \
    --image mycontainerimage \
    --restart-policy OnFailure

Run to completion example

To see the restart policy in action, create a container instance from the Microsoft aci-wordcount image, and specify the OnFailure restart policy. This example container runs a Python script that, by default, analyzes the text of Shakespeare's Hamlet, writes the 10 most common words to STDOUT, and then exits.

Run the example container with the following az container create command:

az container create \
    --resource-group myResourceGroup \
    --name mycontainer \
    --image mcr.microsoft.com/azuredocs/aci-wordcount:latest \
    --restart-policy OnFailure

Azure Container Instances starts the container, and then stops it when its application (or script, in this case) exits. When Azure Container Instances stops a container whose restart policy is Never or OnFailure, the container's status is set to Terminated. You can check a container's status with the az container show command:

az container show \
    --resource-group myResourceGroup \
    --name mycontainer \
    --query containers[0].instanceView.currentState.state

Example output:

"Terminated"

Once the example container's status shows Terminated, you can see its task output by viewing the container logs. Run the az container logs command to view the script's output:

az container logs --resource-group myResourceGroup --name mycontainer

Output:

[('the', 990),
 ('and', 702),
 ('of', 628),
 ('to', 610),
 ('I', 544),
 ('you', 495),
 ('a', 453),
 ('my', 441),
 ('in', 399),
 ('HAMLET', 386)]

This example shows the output that the script sent to STDOUT. Your containerized tasks, however, might instead write their output to persistent storage for later retrieval. For example, to an Azure file share.

Next steps

Task-based scenarios, such as batch processing a large dataset with several containers, can take advantage of custom environment variables or command lines at runtime.

For details on how to persist the output of your containers that run to completion, see Mounting an Azure file share with Azure Container Instances.