Retiring repository credentials while snapshotting results in stuck snapshot

[DaveCTurner]

Elasticsearch version (bin/elasticsearch --version): 7.10.0

Plugins installed: Cloud (incl. repository-s3)

JVM version (java -version): Bundled

OS version (uname -a if on a Unix-like system): Cloud

Description of the problem including expected versus actual behavior:

Access credentials to a snapshot repository were rotated while a snapshot was ongoing, and the old credentials were retired. The ongoing snapshot cannot make progress since it can no longer access the repository, but nor can it be cancelled since that also requires repository access. Meanwhile, the repository cannot be updated to use the new credentials because of the ongoing snapshot.

[elasticmachine]

Pinging @elastic/es-distributed (Team:Distributed)

[original-brownbear]

I looked into this now. I believe this is already fixed.

The ongoing snapshot cannot make progress since it can no longer access the repository, but nor can it be cancelled since that also requires repository access.

I don't think this is true and should never have been true either outside of bugs :)
If the credentials break it should be like any other exception during repo interaction. What seems to have happened in the linked SDH that motivated this is this:
The missing credentials threw a RuntimeException in a version that didn't have this fix #69110 -> leaked the listener -> snapshot never completed. With this fix in I don't think this requires any additional work so I'll close here. Please reopen if you think I missed anything.