Skip to content

Backup providers

bret restores with a backup tool that you already run: Velero, VolSync or Kopiur. It never moves backup data itself. It creates the restore object of the tool, waits for it, then runs your check.

bret drives the restore object of one of three backup tools, all reading from S3
Velero VolSync Kopiur
Maturity in bret Supported Supported Alpha
Helm switch providers.velero.enabled, off by default providers.volsync.enabled, off by default providers.kopiur.enabled, on by default
Backup object bret reads Backup, phase Completed ReplicationSource, its status.lastSyncTime Snapshot, phase Succeeded
One backup holds Every PVC of the namespace One PVC One PVC
Restore object bret creates velero.io/v1 Restore ReplicationDestination with a manual trigger kopiur.home-operations.com/v1alpha1 Restore
What gets restored PVCs and PVs only, no other resources One PVC per source One PVC per source
Where the restored PVCs live Scratch namespace bret-scratch-<hash> The Restore namespace The Restore namespace
All PVCs from one point in time Yes, one Backup No, one sync per PVC No, one Snapshot per PVC
Restore an older backup by hand backupRef names any Backup restoreAsOf picks the newest restic snapshot at or before a time backupRef names any Snapshot
Scratch size Velero restores at the original size and storage class size or autoExpand size or autoExpand
Secrets for the check Copied by bret into the scratch namespace Read from the Restore namespace Read from the Restore namespace
Block mode volumes Not supported Not supported Not supported
Throughput limits Concurrency only, see throttling Mover CPU and memory only Kopia bandwidth and ops cap per repository
Tested with Velero 1.18.2 (chart 12.2.5), AWS plugin v1.14.4 VolSync 0.17.0 Kopiur 0.11.0

Block mode: the check pod mounts every restored volume as a file system, so a volumeMode: Block PVC cannot be checked.

Each provider has its own section with an overview, a walkthrough and its limitations: Velero, VolSync, Kopiur. Setup that applies to all three (CSI snapshots, S3, point in time, throttling) is on Shared setup.

Need Pick
Whole-namespace backups, many resource types, one tool for the cluster Velero
All PVCs of an app from one point in time, without quiescing the app Velero
Per-volume backups with restic, scheduled by the volume VolSync
Per-volume backups with Kopia, deduplication, and a repository you manage Kopiur

All three work with the same check. The check does not care which tool made the data.

A Restore or RestoreTest for a disabled provider stops with reason ProviderUnsupported.

Every walkthrough uses the same WordPress + MariaDB app, which has two PVCs:

  • html: the WordPress docroot, wp-config.php and wp-content. Owned by uid 33.
  • mysql: the MariaDB data dir. Owned by uid 999.

The example files are in examples/: examples/wordpress/ is the app and the check, examples/<tool>/ adds the backup setup, a Restore, a RestoreTest, and a negative Restore under negative/. The files were tested on a kind cluster with hack/check-test/operator-roundtrip.sh <tool> (see Test a check by hand).