search

Garden-AI / garden-backend

The backend service for thegardens.ai
MIT License
1 stars 0 forks source link

Upgrade ML Flow backend to a 2.x version #7

Closed WillEngler closed 1 year ago

WillEngler commented 1 year ago

We currently have ML Flow 1.22 on the backend, but as of late 2022, 2.x releases are available.

WillEngler commented 1 year ago

The error that was thwarting me when trying an official ML Flow image was that at task startup time, the task could not find the pymysql library. This is the driver referenced in our SQLAlchemy connection string (mysql+pymysql://${aws_rds_cluster.backend_store.master_username}). It seems like the getindata version bundles this and the official image doesn't. Maybe getting us to 2.x is as straightforward as building a wrapper container around the official image with the DB Driver we need installed.

BenGalewsky commented 1 year ago

Try this one https://hub.docker.com/r/evk02/mlflow

WillEngler commented 1 year ago

(building off our conversation in Gathertown): it looks like that one bundles the postgres driver in the outermost layer to make it work. (requirements link) That helps confirm our idea of what's going on here.

So to upgrade to 2.x we can either:

  1. Use some existing community container (like the exk02 one you linked) that installs the db wrapper and switch to a matching db (postgres in this case)
  2. make our own thin wrapper container

When I pick this up I'll start with (1)