Too strict invariant for celery deployment storage

[mrdrprofuroboros]

I'm trying to set up a dockerized celery deployment on a single machine:

• I made a doker-compose with rabbitmq, dagit master and dagster celery workers containers
• I had to use custom celery_config.yaml to overwrite broker url for internal docker network address
• But I still want to use filesystem storage through mapping all my containers to a single shared volume on disk

And when I try to launch my pipeline I get
dagster.check.CheckError: Invariant failed. Description: Must use S3 or GCS storage with non-local Celery broker: amqp://guest:guest@cube_rabbitmq:5672// and backend: rpc://

[mgasner]

it would be nice if our local filesystem-related failures were early -- right now, we will fail when a solid looks for an upstream intermediate. maybe we could have a sentinel file written to the filesystem that sub-executions look for (though they would need to know they were sub-executions which right now they don't). it's complicated b/c distributed disjoint executions without dependencies don't actually need a shared filesystem at all.