Hello, has anyone had issues with continuous deployment? I'm using docker to deploy dagster/dagit and whenever I have a job running and I redeploy the container, it looks like the job misses connection with the run and stays there forever, and I need to go and force terminate it. Is this expected behaviour? Should I schedule my deployment for when I have no jobs running?

Hi Dinis - if you use the docker run launcher, each run will happen in its own container that won’t be interrupted when you redeploy

I'm currently using docker compose to lauch dagit and daemon as such:

docker compose up -d --no-deps --build

https://docs.dagster.io/deployment/guides/docker#launching-runs-in-containers shows how to launch each run in its own container

Wow, did not know about this! Have been trying to configure it now. But I'm reaching an error at the moment with this

docker.errors.APIError: 400 Client Error for <http+docker://localhost/v1.41/containers/01fa7c4a19109e3f428ccfabce6e2ab0bcdeb66fd99f77dc67ffaeb23382dd01/start>: Bad Request ("failed to create shim task: OCI runtime create failed: runc create failed: unable to start container process: exec: "dagster": executable file not found in $PATH: unknown")

Any idea on the problem?

That error makes me think that the image doesn’t have the dagster package installed in it

Well only if my image is not being used properly.. I'm defining the same image as the one used to launch dagit and daemon, providing the network in which they are configured in compose

run_launcher:
  module: dagster_docker
  class: DockerRunLauncher
  config:
    image: custom_dagster
    network: docker-dagster-network

What ENTRYPOINT/CMD does the image have?

Your questions were step by step leading me to find the problem. I don't define any entrypoint in the Dockerfile, only in the docker-compose.yml. But in my Dockerfile I'm using conda. The way I set up the image, every command needs to have the prefix:

conda run -n dagster_env [my_command]

(In the docker-compose as well) So now, I just assumed that the Docker Launcher tries to run with entrypoint

python something ...

which won't work with my setup. I've updated my Dockerfile in order to don't need my previous prefix, and it works