WAL overwrites across clusters with same name

[fcanovai]

The in-tree barman archiver prevents you from archiving wal files if the object stores already has wals in there generated by a different cluster with the same name.

The plugin instead allows doesn't have this protection and allows to overwrite the files, so there could be wal files from different sources.

[MRocholl]

Not an expert here. But I dont think this is a bug.
Correct me if I am wrong. But as far as I understood, this is not a check that exists on barman cloud by default.
It was probably introduced as a "dont shoot yourself in the foot" kind of precaution.

It would probably make sense to reintroduce the annotation to skip or enforce a emptyWallArchiveCheck.

[siegenthalerroger]

I'd definitely appreciate having a similar protection mechanism as I've run into this very issue with the zalando operator that has resulted in data loss incidents.

[jmealo]

I would like to bump this. I fear this will cause data loss for folks who thought they had their bases covered.

Relevant docs:
https://github.com/cloudnative-pg/cloudnative-pg/blob/954854cebc80666698df3ba3431dce033a3ec356/docs/src/recovery.md#restoring-into-a-cluster-with-a-backup-section

[leonardoce]

Hi! Can you please check if #458 fixed this issue?

[gbartolini]

Ideally, we should also provide the possibility to disable that. We had an annotation for the cluster, called cnpg.io/skipEmptyWalArchiveCheck. See https://cloudnative-pg.io/documentation/current/labels_annotations/#predefined-annotations

Maybe here it makes more sense to add the annotation to the object store?