The workaround was simply to convert the cluster to use PVC-provided storage by deleting the StatefulSet without cascading and rolling over the cluster one node at a time.

 agree. We will do this.

Could you please share the workaround?

We are able to work around the inability to restore backups to emptyDir-based clusters. It might just be nice to document the restriction and fail with a more specific error indicating the restriction directly.

Hello , thank you for submitting this.

Looks like the problem with restore has deeper roots. When we perform the restoration in PXC Operator (in Mongo it works differently), we create a separate Pod, which restores the data. Once restore is complete, we start PXC Pods.

But in the case of emptyDir it does not work, as it is cleaned up right after restore Pod is gone. PXC Pod starts, but data is not there already.

Could you please describe your use case for emptyDir in more details, so that we understand how to resolve it faster?

 Hello!

We would like to send you a gift for your activity! Please, send us your mailing address to community-team@percona.com