Hey team, I am evaluating Dagster to be used within my org. I am currently exploring of having separate user code and dagit deployment by following this guide (https://docs.dagster.io/master/deployment/guides/kubernetes/customizing-your-deployment#separately-deploying-dagster-infrastructure-and-user-code) . One question, is there a way to add workspace or repository without having to configure the dagit deployment? It seems that dagit deployment need to know all repository servers beforehand.

the dagit deployment needs to know the user code deployments, but the repositories within the user code deploy can change freely

Thanks for the clarification. All pipelines within a repository will have same python environment and dependencies, that means if I am developing new pipeline with different set of dependencies. I’ll need to create new repository, am I correct on this one?

Technically a different “repository location” which is equivalent to a different entry in the

load_from:

list in the workspace A given “repository location” corresponds to one python environment. It can have N repositories within it, though the common case is 1.

@alex What would be the recommended approach if I have many pipelines with different dependencies? If I have 100 pipelines with different dependency, that means I need 100 repository location. I believe each repository location also translate to one user-code deployment.

yea if the dependencies conflict and need to be in different environments, than you will have the 100 user code deploys as you said.

I see. Can you point me to the interface exposed by user-code deployment to the dagit? I was wondering whether it’s possible to mimic user-code deployment.

its a grpc server - you can see the relevant code here https://github.com/dagster-io/dagster/tree/master/python_modules/dagster/dagster/grpc

Thanks @alex!