-
Spike
-
Resolution: Unresolved
-
Major
-
None
-
None
-
False
-
False
Currently Quay allows only one manifest per tag. If a new version of the manifest is pushed to an existing tag we lose the old manifest. This can be a problem if a deployment is pulling that image by SHA- their image will no longer by available. Other registries keep these old manifests around so they can continue to be pulled even if the tag has been overwritten.
Note this is the 'inverse' way to implement the requirement stated in PROJQUAY-2443.
Task is to understand what the impact would be to our data model to no longer GC overwritten manifests. A few points to consider
- How does our data model and push/pull logic change to support this?
- How does our GC code need to change to support this?
- To avoid excessive manifests from filling up storage, can we cap the number of retained manifests by a configurable parameter?
- Can we feature flag this ability to let existing Quay customers continue to operate with existing semantics?
- Can we expose metrics to understand how many 'abandoned' manifests exist within a namespace/repo so that we can report on this as part of future showback/quota management features?
Deliverables:
- Basic write up on how we can address the points above
- Simple PoC to demonstrate an approach to implement this if possible (even a small subset of the functionality).
- is related to
-
PROJQUAY-3885 As a Quay user I want to see untagged artifacts in the UI
- New
-
PROJQUAY-2443 QBO: add ability to delete images inside Quay not referenced by other OpenShift objects
- Closed
- relates to
-
PROJQUAY-3887 Tagless image/artifact support
- New