Through (reported by yours truly), we got restore job resource requirements defined via the restore manifest.

Except it doesn’t seem like they are respected, see for example a restore I’m running right now:

And here’s the associated job:

Relevant mismatch being:

So the CPU request is ignored.

Also just before that, I tried setting both requests.cpu and limits.cpu, which resulted in the restore job using limits.cpu as both request and limit.

After a quick glance at the operator code, this doesn’t make sense to me, since https://github.com/percona/percona-xtradb-cluster-operator/blob/c85a021f2a21441500b02a2c0b3d17e8a8b25996/pkg/pxc/backup/restore.go#L110 seems reasonable.

I have to assume that the source value is what is wrong by then, not sure how/why though.

For reference, this is how the manifest looks: