CHANGES.md

Changelog

in progress

2026-05-14 v3.12.0

• Updated to MLflow 3.12.0
• Model: Added tables guardrails and guardrail_configs
• OCI: Updated MLflow server to use Python 3.14 and Debian 13 (trixie)

2026-04-24 v3.11.1

• Updated to MLflow 3.11.1
• Model: Updated tables trace_info, trace_metrics, trace_tags, trace_request_metadata
• Model: Added tables assessments, budget_policies, endpoints, endpoint_bindings, endpoint_model_mappings, endpoint_tags, issues, spans, span_metrics
• Model: Renamed field names that contain dots
• SQL: Relaxed OBJECT(DYNAMIC) attribute access behaviour where attribute does not exist yet
• SQL: Adjusted percentile aggregation and time bucketing expression to render the percentile and floor clauses literally
• Documentation: Added setup for Docker Compose or Podman Compose

2026-03-30 v3.10.2

• Amalgamation: Started supporting huey jobs that are invoked in subprocesses
• Model: Added tables jobs, online_scoring_configs, scorers, scorer_versions
• Server: Switched from gunicorn to uvicorn as suggested

2026-03-29 v3.10.1

• Updated to MLflow 3.10.1
• Model: Added table workspaces and column workspace
• MLflow API: Renamed artifact_path to name

2026-03-29 v3.9.0

• Updated to MLflow 3.9.0

2026-03-29 v3.8.1

• Updated to MLflow 3.8.1

2026-03-28 v3.5.1

• Updated to MLflow 3.5.1
• Dropped support for Python 3.9
• Model: Added SQL DDL for tables evaluation_datasets, evaluation_dataset_records, evaluation_dataset_tags
• Amalgamation: Made sure SQLAlchemy is patched first, before importing any other modules

2026-03-28 v3.3.2

• Updated to MLflow 3.3.2.
• Model: Added SQL DDL for tables webhooks and webhook_events

2026-03-27 v3.1.4

• Updated to MLflow 3.1.4.
• Model: Added step column to inputs table "step" BIGINT DEFAULT '0' NOT NULL
• Model: Added SQL DDL for tables logged_models, logged_model_metrics, logged_model_param, and logged_model_tags

2026-03-26 v2.22.4

• Updated to MLflow 2.22.4.
• Dropped support for Python 3.8
• Added support for Python 3.12 and 3.13
• OCI: Updated server to Python 3.13 (Debian 12 bookworm)
• OCI: Updated runtime to Debian 12 bookworm

2024-06-25 v2.14.1

• Started using more SQLAlchemy patches and polyfills from sqlalchemy-cratedb
• Updated to MLflow 2.14.1. See release notes for MLflow 2.14.1.

2024-06-18 v2.14.0

• Remove patch for SQLAlchemy Inspector's get_table_names. Use sqlalchemy-cratedb>=0.37 instead, which includes the patch.
• Update to MLflow 2.14.0. See release notes for MLflow 2.14.0.

2024-06-11 v2.13.2

• Dependencies: Migrate from crate[sqlalchemy] to sqlalchemy-cratedb
• Update to MLflow 2.13.2. See release notes for MLflow 2.13.2.

2024-06-04 v2.13.1

• Update to MLflow 2.13.1. See release notes for MLflow 2.13.1.

2024-05-21 v2.13.0

• Update to MLflow 2.13.0. See release notes for MLflow 2.13.0.

2024-05-17 v2.12.2

• Update to MLflow 2.12.2. See release notes for MLflow 2.12.2.

2024-05-07 v2.12.1

• Update to MLflow 2.12.1. See release notes for MLflow 2.12.1.
• Update to PyCaret 3.3.2. See release notes for PyCaret 3.3.1, PyCaret 3.3.2.
• Chore: Update to most recent mypy, pyproject-fmt, ruff, and sqlparse
• CI: Run tests on Python 3.10 and 3.11

2024-04-10 v2.11.3

• Update to MLflow 2.11.3. See release notes for MLflow 2.11.0, MLflow 2.11.1, MLflow 2.11.2, MLflow 2.11.3.
• Fix incompatibility with Python 3.11.9

2024-02-13 v2.10.2

• Update to MLflow 2.10.2

2024-01-30 v2.10.0

• Update to MLflow 2.10.0

2024-01-25 v2.9.2

• Update to MLflow 2.9.2
• Update to Python 3.11 across the board

2023-11-24 v2.8.1

• Update to MLflow 2.8.1

2023-11-24 v2.8.0

• Update to MLflow 2.8.0

  Attention: Please note when updating from a previous version: The database model changed slightly, adding the storage_location column to the model_versions table. In order to accommodate the update, run this SQL DDL command, for example using crash.

  ALTER TABLE mlflow.model_versions ADD COLUMN storage_location TEXT NULL;

  Note that it is always advised to create backups of your database content. This is an excellent opportunity to do that.

• Fix uniqueness constraints

  • mlflow.server.auth.db.models.SqlUser.username.
  • m.s.a.d.m.SqlExperimentPermission: "experiment_id", "user_id"
  • m.s.a.d.m.SqlRegisteredModelPermission: "name", "user_id"

• Fix OCI build re. psutil package on aarch64.

• Optimize OCI image sizes.

• Add example experiment program tracking_pycaret.py, and a corresponding test case. Thanks, @andnig.

• Examples: Use refresh_table to synchronize CrateDB write operations.

2023-11-01 v2.7.1

• Fix uniqueness constraint with SqlRegisteredModel.name. Thanks, @andnig.
• Downgrade to Python 3.10. A few packages like PyCaret are not ready for Python 3.11 yet. Thanks, @andnig.

2023-10-11 v0.2.0

• Update to MLflow 2.7
• Improve table_exists() in example_merlion.py
• SQLAlchemy: Use server-side now() function for "autoincrement" columns
• Use SQLAlchemy patches and polyfills from cratedb-toolkit
• Update and improve documentation

2023-09-12 v0.1.1

• Documentation: Improve "Container Usage" page
• Documentation: Update README with real pip install command

2023-09-12 v0.1.0

• Initial thing, proof-of-concept
• Add software tests
• CLI: Add mlflow-cratedb cratedb --version command
• Project: Add versioningit, for effortless versioning
• Add patch for SQLAlchemy Inspector's get_table_names
• Reorder CrateDB SQLAlchemy Dialect polyfills
• Add example experiment program tracking_merlion.py, and corresponding tests
• Add example program tracking_dummy.py, and improve test infrastructure
• Documentation: Add information about how to connect to CrateDB Cloud
• CI: Add GHA workflows to build and publish OCI container images to GHCR
• Tests: Enable code coverage tracking
• Fix SQL DDL files, and add missing columns to make the Models tab load in the UI, see GH-17. Thanks, @hammerhead.