What added advantage would using an Sql Alchemy database have, while I have an artifact destination for my MlFlow server?

naveen_marthala
Contributor

‎06-14-2022 11:47 PM

I am setting up mlflow server with Postgres and S3 on AWS ECS(or AWS EC2) for personal usage. I would like to know if using Postgres would actually give me any benefit?

as shown in scenario 5 in docs, I would like to set up server with proxied artifact storage, but without postgres in AWS RDS and also not in `--artifacts-only` and `--artifacts-only` modes as shown in scenario 6 in docs. So, if i launch a mlflow tracking server like this, will all the metadata that usually gets saved to an SQLAlchemy database(postgres in my case) be saved to the location of attribute `--artifacts-destination` as files?

I would like to do this to save costs from AWS RDS that runs Postgres.

If you aren't familiar with AWS ECS, it is a fully managed service (a simpler version of kubernetes) that runs container base applications. and EC2 is just user managed machines or servers.

Labels:

naveen_marthala
Contributor
In response to Retired_mod

‎06-23-2022 06:09 AM

Thanks for trying to help me @Kaniz Fatma​ .

My question was not to learn how to set up mlflow with postgres. But, to learn what additional advantage would using an sql-alchemy database like postgres would have, instead of storing all the metadata to files. that is, what advantage would I have, if I use an sql-alchemy for backend and not use one. please note that, the mlflow server will not be in '--artifacts-only' mode?