SUBMARINE-1023. Submarine SDK sqlalchemy store (model registry)

[KUAN-HSUN-LI]

What is this PR for?

• Implement the model registry SQL method in Python SDK
• Apply sqlalchemy mypy checks
• Replace submarine tracking_uri with db_uri

What type of PR is it?

[Feature]

Todos

What is the Jira issue?

https://issues.apache.org/jira/browse/SUBMARINE-1023

How should this be tested?

All of the tests are provided in submarine-sdk/pysubmarine/tests/store/model_registry/test_sqlalchemy_store.py

Screenshots (if appropriate)

Questions:

• Do the license files need updating? No
• Are there breaking changes for older versions? No
• Does this need new documentation? No

[KUAN-HSUN-LI]

@pingsutw @jeff-901 Can you help me review the code? Thanks

[pingsutw]

Do we need to put these functions here? https://github.com/apache/submarine/blob/master/submarine-sdk/pysubmarine/submarine/store/abstract_store.py

[KUAN-HSUN-LI]

model registry including registered_model, registered_model_tag, model_version, model_version_tag four tables.
tracking including param, metric, experiment three tables.
In my opinion, it is more clear to put these functions in different directories. In this PR, I only worked on the model registry directory maybe I should also refactor the tracking directory and implement the experiment table.

[pingsutw]

ditto.
https://github.com/apache/submarine/blob/master/submarine-sdk/pysubmarine/submarine/store/sqlalchemy_store.py

[pingsutw]

What's different between the tags in model version and registered model?

[KUAN-HSUN-LI]

registered model tag focuses on what is the goal of this model or the category of this model. model version tag focuses on what this model perform or feature of this model. I think it is a little confused in these two parts. Hopefully, this explanation can help.

[KUAN-HSUN-LI]

@pingsutw @jeff-901 Thanks for your review. I have replaced registered model with model container and replaced model version with model metadata. Additionally, I have fixed serval bugs.

[jeff-901]

model_version forget to change to model metadata

[KUAN-HSUN-LI]

I think this function name is fine.

[jeff-901]

ok

[jeff-901]

LGTM

[KUAN-HSUN-LI]

@pingsutw @jeff-901 I am so sorry for changing the table name in the database again. I think the previous names are the best. I will explain it as follow:
workflow of model registry:

1. save models in s3 bucket (unregistered model)
2. choose the best model and registry it (registered model, version 1)
3. train the model with different parameters or datasets and save models in s3 bucket (unregistered model)
4. Again, choose the best model and registry it (registered model, version 2)

Summary:

• The table is storing registered models so the name registered_model is more proper.
• The version of the registered model with the same name will define a different version which will be stored in another table. It is also proper to name the table model_version

We also need two tag tables.
example: