Hi, we came to prefect in search of an orchestration framework. Our problem is pretty simple still, but it could grow in complexity, hence why we are exploring prefect. In essence, what we would need is: • to schedule execution of multiple flow runs • to have tasks defined (with dependencies between them) on each flow run Since we would like to deploy this on Azure container group, I could think that minimally we would need this setup: • one container instance for the prefect server • one container instance for the prefect agent (defaulting to local run, without external "workers") The question then is about the choice of database. Our setup is simple, so sqlite would suffice in theory, but the question is then where is this database file stored, because it might be beneficial to store it outside the container filesystem (and mount a volume for instance), so that it would be persistent to the rest of the infrastructure. Does anyone have recommendation / tried to achieve something simple with prefect (but in production)? At the moment the prefect server would really just service one application, so I am wondering if this is a stretch of how to use the prefect framework. We were looking for something simpler than airflow, but it looks like this might not necessarily be the case. Thanks in advance for any feedback!

Hi Emile - are you looking at doing self-hosted, or prefect cloud with a free workspace for this? That would dictate the necessity for sqlite, but yes, sqlite would be local to the installation

Hi Christopher! Thank you for your replies. Yes, we would be looking into our own deployment on Azure. Why would that "dictate the location for sqlite"? So finally last week I tested locally with docker compose creating the following 2-container setup: - a docker container running the server, based only on the image

prefecthq/prefect:2-python3.10

with entrypoint:

["prefect", "orion", "start"]

- a docker container running the agent, based on the same image

prefecthq/prefect:2-python3.10

, but with some extra care that feels in part "cheating": - we add the flow and task files to the base image - we use a shell script as entrypoint, so that we can run multiple commands, namely 1) run a deployment python script (which calls

Deployment.build_from_flow()

) 2) runs the agent This setup (with only these two containers) was defaulting to a SQLite database, and I could mount the file outside of the containers as you mentioned its location. I see no issue in theory to deploy this pattern on Azure Container Group, with 2 containers within the Container Group. However, it does feel a bit "hacky". I'd be glad to have your feedback. PS: if anyone is interested, I could also post the exact configuration for reference.