Repository navigation
Description
Summary
Four user-visible features have no end-to-end coverage at all. Each is already implemented and documented; none requires new infrastructure, and at MAX_PARALLELISM=5 with 38 playbooks (~8 waves) the four additional playbooks cost roughly one extra wave (~2-4 minutes), not four.
Verified absent from test-e2e/test_*.yml: no kind: Migration, no copyMethod: Clone, no paused, and no negative/error-recovery scenario. (external is a provider extension point, so leave that out of scope.)
1. spec.paused
No playbook sets paused: true on an RS/RD.
Assert: no sync occurs while paused (status does not advance, no mover Job is created), and unpausing resumes on the next trigger without a duplicate/renamed mover Job.
2. copyMethod: Clone
Unit/envtest tests cover CopyMethodClone; e2e does not (only Snapshot, and Direct in the restic playbooks).
Assert: data lands on the destination, the generated PVC uses a PVC data source, and the temporary source clone is cleaned up.
The Kind fixture's csi-driver-host-path (v1.18.0, hack/setup-kind-cluster.sh:245-251) advertises CLONE_VOLUME, so this should not need new storage infrastructure — confirm on one supported version first.
3. CLI migration (data movement)
kubectl volsync migration create / migration rsync / migration delete are exercised only as Go helpers; migration_rsync.go runs a local rsync over SSH to the destination Service. There is no kind: Migration playbook anywhere.
Assert: create produces the destination PVC + rsync RD; migration rsync moves data and the destination PVC compares equal; delete removes resources.
Note: this is the only e2e case that requires local rsync in the runner; keep the write-set small (one file plus one directory).
4. Negative path / error reporting
Nothing asserts what happens when a sync fails (bad repository credentials, unreachable destination).
Assert: the source reports Synchronizing=False with the mover-error reason introduced by #2188 (or the current reason if that lands later), and the controller does not loop creating jobs.
Cheapest variant: restic with a Secret containing invalid credentials and a manual trigger.
Acceptance criteria
- One playbook per scenario, tagged consistently with the existing
e2e/ mover tags, using existing roles (create_namespace,compare_pvc_data,pvc_has_data,dump_logs). - Each verifies data where applicable and status/conditions (do not re-implement those assertions; depend on the shared status role).
- Total added wall-clock: ≤1 wave at
MAX_PARALLELISM=5. - All four tagged so they can be excluded from the old-Kubernetes smoke subset if a feature is version-gated (e.g. Clone on 1.20).
(AI generated)
Metadata
Metadata
Assignees
Labels
Type
Projects
- StatusShow more project fieldsNo status
Activity