As the cache volume design https://github.com/vmware-tanzu/velero/blob/main/design/backup-repo-cache-volume.md, repository maintenance job also caches data, which is currently in the pod's root file system.
For some env with extreme small pod root file system, having repository maintenance jobs storing the cache into the root file system is still a problem, which causes the job killed because of running out of ephemeral disk space.