Data Model Changelog

Record changes to the ZooKeeper data model which require API version increases here.

When making a model change:

  • Increment the value of MODEL_API in model_api.py.

  • Update code to use the new API by default and add backwards-compatibility handling for older versions. This makes it easier to clean up backwards-compatibility handling in the future.

  • Make sure code that special cases model versions either references a model_api variable or has a comment like MODEL_API: > {version} so that we can grep for that and clean up compatability code that is no longer needed.

  • Add a test to test_model_upgrade.py.

  • Add an entry to this log so we can decide when to remove backwards-compatibility handlers.

Version 0

Prior Zuul version:

4.11.0

Description:

This is an implied version as of Zuul 4.12.0 to initialize the series.

Version 1

Prior Zuul version:

4.11.0

Description:

No change since Version 0. This explicitly records the component versions in ZooKeeper.

Version 2

Prior Zuul version:

5.0.0

Description:

Changes the sempahore handle format from <item_uuid>-<job_name> to a dictionary with buildset path and job name.

Version 3

Prior Zuul version:

5.0.0

Description:

Add a new SupercedeEvent and use that for dequeuing of superceded items from other pipelines. This only affects the schedulers.

Version 4

Prior Zuul version:

5.1.0

Description:

Adds QueueItem.dequeued_missing_requirements and sets it to True if a change no longer meets merge requirements in dependent pipelines. This only affects schedulers.

Version 5

Prior Zuul version:

5.1.0

Description:

Changes the result data attributes on Build from ResultData to JobData instances and uses the inline/offloading paradigm from FrozenJob. This affects schedulers and executors.

Version 6

Prior Zuul version:

5.2.0

Description:

Stores the complete layout min_ltimes in /zuul/layout-data. This only affects schedulers.

Version 7

Prior Zuul version:

5.2.2

Description:

Adds the blob store and stores large secrets in it. Playbook secret references are now either an integer index into the job secret list, or a dict with a blob store key. This affects schedulers and executors.