I'm working through various CI/CD methods for deploying Dagster through Docker. At some point in the process, I envision the containers will have to be rebuilt using the

docker compose

command. My question is - what would prevent this from essentially killing any actively running jobs/materializations when the containers are restarted? How can I make sure the CI/CD action waits until all actively running Dagster processes have stopped. Just trying to think through how I can prevent a user from pushing a code change that might result in breaking a long running job or corrupt some asset due to the container being stopped mid-process.

Hi AJ - the way that we recommend handling this in Dagster is to use the DockerRunLauncher that launches each run in its own container, and to use a unique image tag every time that you make a new change via CI/CD. If your CI/CD updates the image tag of your code container to refer to a new tag, that won't disrupt any runs that are currently happening using the old tag - they'll be unaffected by the redeploy.

@daniel awesome - I will take a look at the docs to learn how to implement. Thank you, sir!