Keep latest artifacts for the last successful jobs
Problem to Solve
expire_in function is great, but depending on how often your pipeline is built can be hard to tune. If you don't build frequently, setting a short
expire_in can result in your most recent artifact being deleted. Setting it too long can keep unneeded copies around. Over time this can use up a lot of disk space. Most of the time, people just want to ensure they always at least have the most recent artifact, whenever that was built.
Implement a default behavior for all projects to always keep the latest artifact for each successful job that was part of a successful pipeline (that produced artifacts) on any active (non-deleted) branch, merge request, or tag (i.e., the one downloaded via https://docs.gitlab.com/ce/user/project/pipelines/job_artifacts.html#downloading-the-latest-artifacts is exempt from any
expire_in policy). In other words, never let the most recent artifact expire that is the same one described at the docs link here:
- Artifacts for deleted branches follow
- All non-latest artifacts follow
- Only the latest artifact for a job on a non-deleted branch is kept
This would allow for setting a relatively aggressive expiration policy, while remaining confident you always have your most recent artifact for any active branches.
expire_in: could be set too short we would need to define a way to "lock" the latest artifacts to avoid that
ExpireBuildArtifactsWorker will delete it. For example add a new column
job_artifacts.locked (false by default). Then when a pipeline completes we will lock the latest artifacts in a moving window: lock last ones and unlock anything from previous pipelines. Finally the
ExpireBuildArtifactsWorker would need to be modified to filter out locked artifacts.
Another aspect to consider is to unlock artifacts when a branch is deleted or they would remain locked and never deleted.
Out of Scope for this MVC
Items not expected for this MVC issue:
- ability to keep latest N count of artifacts per job
- new checkbox in project's
Settings > CI/CD > General pipelinesfor
Keep latest job artifacts per ref
- allow artifacts to live longer than N days for compliance reasons
- aggressively remove artifacts which are not the latest even before they expire