kubevirt-dm: Basic controller implementation

[weshayutin]

Description:
Controller to watch kubevirt DataUploads (and eventually DataDownloads). For a new DU, this controller will identify the latest checkpoint for the VM from the BSL (if present) and validate it. Create a new VMBT if one is not present. Update the VMBT with the latest valid checkpoint from the BSL (if it isn't already there), create the temporary PVC, and finally create the VMB and update DU status.
Once the VMB is complete, create a datamover pod mounting the BSL. Once the datamover action is complete, clean up: delete PVC, maybe delete VMB (unless we want to keep it for debug purposes and delete VMB with the Velero backup), and move DU to completed

Subtasks:

DOD:

[shubham-pampattiwar]

Rough Implementation Plan

This controller watches DataUploads where Spec.DataMover = "kubevirt" and orchestrates the KubeVirt backup flow.

• Phase 1: Controller Scaffolding

• Create new controller package under internal/controller/ (e.g., kubevirt_datamover_controller.go)

• Implement Reconcile() watching DataUpload CRs

• Filter for DataUploads where Spec.DataMover == "kubevirt"

• Add controller to manager in cmd/main.go

• Define status conditions for tracking progress (Pending, InProgress, VMBackupCreated, Uploading, Completed, Failed)

• Add configuration for concurrency limits (concurrent-vm-backups, concurrent-vm-datauploads)

==========================================================

• Phase 2: VMBT/VMB Creation + Monitoring

• Extract VirtualMachine reference from DataUpload annotation

• Create temporary PVC for backup output

  • Determine PVC size (configurable via annotation/configmap, or use VM disk size as default)
  • Handle PVC creation failure / already exists

• Create or update VirtualMachineBackupTracker (VMBT)

  • Set source to VirtualMachine reference
  • Initially no checkpoint (forces full backup)

• Create VirtualMachineBackup (VMB) CR

  • Set source to VMBT
  • Set pvcName to temporary PVC
  • Set forceFullBackup if requested via backup annotation

• Watch VMB status until completion

  • Handle VMB failure (update DU status, cleanup)
  • Extract checkpoint ID and backup type (full/incremental) from VMB status

==========================================================

• Phase 3: Write to BSL (Datamover Pod)

• Get ObjectStore plugin from BSL configuration

• Launch datamover pod

  • Mount temporary PVC (contains qcow2 files)
  • Pass BSL credentials (Secret mount or ServiceAccount token)
  • Pass target path: -kubevirt-datamover/checkpoints////

• Pod uploads qcow2 files to BSL using https://github.com/vmware-tanzu/velero/blob/7442d20/pkg/plugin/velero/object_store.go

• Create/update checkpoints///index.json

  • Add new checkpoint entry with files, timestamp, type
  • Add Velero backup name to referencedBy array

• Create manifests//.json with checkpoint chain

• Create manifests//index.json if not exists

• Wait for pod completion, handle errors

==========================================================

• Phase 4: Read from BSL (Checkpoint Lookup for Incremental)

• Before creating VMBT, query BSL for existing checkpoints

  • Read checkpoints///index.json
  • Handle missing file (first backup for this VM)

• Validate checkpoint chain

  • Verify all qcow2 files in chain exist in BSL
  • If any missing, fall back to full backup

• Update VMBT with latest valid checkpoint

  • Set status.latestCheckpoint.name to enable incremental

• Handle checkpoint mismatch (VM restored from different backup, etc.)

==========================================================

• Phase 5: Cleanup & Completion

• Delete temporary PVC (and PV contents)

• Optionally delete VMB (configurable - keep for debugging?)

• Update DataUpload status to Completed

• Handle partial failures

  • If upload failed, mark DU as Failed with error message
  • Cleanup temp resources on failure

• Emit Kubernetes events for observability

==========================================================

Open Questions

• Where does this controller run? Separate deployment or in OADP operator pod?
• How to handle concurrent backups of the same VM? (need locking on index.json)