@claire @sandy quick question on https://github.com/dagster-io/dagster/pull/11900 if one of you is around. I've just upgraded to Dagster 1.1.818, but realized after a little while that we cannot start backfills anymore, for the following reason: • We deploy a user code deployment as a container on kubernetes with the Helm chart. This container can't access the PostgreSQL database for Dagster for two reasons: (i) network policies, (ii) our database connection string has no password and relies on AWS IAM for RDS databases (i.e., we have an IAM policy to grant the Dagit/Dagster daemon/etc. containers permissions to log in using roles for service accounts). This was fine before since it didn't need any such access. • With 1.1.18, when we attempt to start a backfill, Dagster seems to invoke

ExternalPartitionTags

on the gRPC server of the user code deployment, which fails when trying to create a Dagster instance as it cannot connect to the database, and prevents the backfill from starting. Is there a way to disable this, esp. as we're not using dynamic partitions, or would you recommend to make arrangements to also allow user code deployments to access the Dagster database? Thanks!

Hi @peay, sorry that recent changes caused the backfill feature to stop working for you. I think ideally we can avoid creating the instance at the grpc endpoint if it isn't used. Would you mind filing an issue for this with a full stack trace?

Sure, see https://github.com/dagster-io/dagster/issues/12440