Stateless clustering with garbage collection + deduplication

[zhming0]

Reading the document I have a question regarding this bit:

In a stateless clustering scheme, the image data is stored in the remote storage backend and the registry cache is disabled by turning off both deduplication and garbage collection.

Why do we need to turn off gc + dedupe? Is it because zot is not cluster aware so it may cause race conditions if we don't turn these feature off?

[rchincha]

@zhming0 Correct! With multiple writers, the race is between updates in the metadata cache (which could also be remote - dynamodb etc) and the actual storage.

[zhming0]

Thanks @rchincha! an extra question though, when would those dedupe + gc run? According to the doc and my observation, it seem they only run during startup?

[rchincha]

dedupe is run inline when you push, gc is triggered as a background job.