Observations on updating to MLflow 2.8

[amotl]

About

MLflow 2.8.0 was released three days ago. Dependabot told us about it on behalf of GH-42, but the software tests are failing on CI.

Summary

This is a summary about the failing test cases.

FAILED tests/test_tracking.py::TestSqlAlchemyStore::test_log_batch_params_max_length_value - AssertionError: Regex pattern did not match.
 Regex: 'exceeded length'
 Input: "Changing param values is not allowed. Params were already logged='[{'key': 'long param', 'old_value': 
FAILED tests/test_tracking.py::TestSqlAlchemyStore::test_log_param_max_length_value - AssertionError: Regex pattern did not match.
 Regex: 'exceeded length'
 Input: "Changing param values is not allowed. Param with key='blahmetric' was already logged with 
======= 2 failed, 136 passed, 3 skipped, 6 warnings in 279.01s (0:04:39) =======

-- https://github.com/crate-workbench/mlflow-cratedb/actions/runs/6689895692/job/18174227007?pr=42#step:6:1116

[amotl]

About the error shared above, I only have been able to discover those resources, which are still talking about MLflow 1.x.

• https://github.com/mlflow/mlflow/issues/7046
• https://github.com/mlflow/mlflow/issues/7152

[amotl]

Aha. A few numbers had to be increased, to raise the same expected error here as beforehand, i.e. it was just about fixing the test cases.

• https://github.com/mlflow/mlflow/blob/v2.8.0/tests/store/tracking/test_sqlalchemy_store.py#L1117-L1126
• https://github.com/mlflow/mlflow/blob/v2.8.0/tests/store/tracking/test_sqlalchemy_store.py#L2729-L2737

[amotl]

Problem

FAILED tests/test_tracking_issues.py::test_log_model_twice - mlflow.exceptions.MlflowException: (crate.client.exceptions.ProgrammingError) ColumnUnknownException[Column storage_location unknown]

-- https://github.com/crate-workbench/mlflow-cratedb/actions/runs/6721143498/job/18266209126?pr=51#step:6:1058

Thoughts

Column storage_location unknown

That's probably just a missing DDL detail, not carried over properly from the SQLAlchemy-based definition into our SQL-based definition.

Solution

Has been addressed by adding storage_location to the SQL DDL.

[amotl]

We added support for MLflow 2.8, including corresponding adjustments, and documentation.

• GH-51
• GH-57

[amotl]

mlflow-cratedb 2.8.0 has been released.