Per review feedback on #9791, the previous revision still let a
DataUpload with an empty velero.io/backup-name label fall through to
genConfigmap, creating a ConfigMap that deleteMovedSnapshots can never
match back to a snapshot. The CM is useless and only adds etcd churn.
Treat the missing-label case the same way as the foreign-owner case:
warn and skip the ConfigMap creation. Use a distinct warn message so
operators can tell the two misconfiguration classes apart in logs
(missing-label vs. owner mismatch from a captured velero namespace).
Test for the missing-label case is updated to assert no ConfigMap is
created and a warn is emitted. The warn assertion is generalized to
match the per-case message substring instead of a fixed string.
Signed-off-by: Christian Schlichtherle <christian@schlichtherle.de>
- added ClusterScopedFilterPolicy/NamespacedFilterPolicy
- added run time data structure, ResolvedResourceFilter and ResolvedNamespaceFilter
Signed-off-by: Adam Zhang <adam.zhang@broadcom.com>
Velero does not support self-protection: the velero namespace must
never be captured in a backup tarball. When it is, the tarball can
contain DataUpload CRs belonging to other backups, and the previous
revision of this change silently swallowed that case in the
DataUploadDeleteAction.
Per maintainer feedback, the action should make the misconfiguration
detectable rather than silent. Emit a warn-level log naming the
DataUpload, its owning backup-name label, and the executing backup,
and call out that the velero namespace should be excluded from
schedules. Continue to skip the snapshot-info ConfigMap creation so
that a mislabeled CM does not mask the real owning backup's snapshot
on deletion.
The test for the foreign-backup case now also asserts the warn is
emitted via a logrus test hook.
Signed-off-by: Christian Schlichtherle <cs@bsure-analytics.de>
When a backup tarball incidentally contains DataUpload CRs that belong to
a different backup (common when a schedule includes the velero namespace
where DataUploads live), DataUploadDeleteAction.Execute used to create a
"<du-name>-info" ConfigMap labeled with the *executing* backup's name
instead of the DataUpload's true owning backup. The ConfigMap is
created with Create-only semantics, so the wrong label is never
corrected.
deleteMovedSnapshots in the backup-deletion controller looks up these
ConfigMaps by velero.io/backup-name to discover which Kopia snapshots
to delete. With the wrong label, the real owning backup's expiry pass
finds no ConfigMaps for its DataUploads and silently leaves their Kopia
snapshots in object storage, leaking data over time.
Fix: in DataUploadDeleteAction.Execute, compare the DataUpload's
velero.io/backup-name label against input.Backup.Name (using
label.GetValidName to handle DNS-1035 truncation for long backup names).
If the label is present and differs, skip the DataUpload entirely; this
prevents the over-eager creation of misnamed ConfigMaps without changing
behavior for DataUploads that legitimately belong to the executing
backup, or for legacy DataUploads with no backup-name label.
Refs: #9472
Signed-off-by: Christian Schlichtherle <cs@bsure-analytics.de>
* Fix UT failures caused by client-go version bump.
* Some modifications to enhance the UT stability.
* Fix UT errors: non-constant format string in call to ...
* Fix linter issues.
Signed-off-by: Xun Jiang <xun.jiang@broadcom.com>
* Fix wildcard expansion when includes is empty and excludes has wildcards
When a Backup CR is applied via kubectl with empty includedNamespaces
and a wildcard in excludedNamespaces, ShouldExpandWildcards triggers
expansion. The empty includes expands to nil, but wildcardExpanded is
set to true, causing ShouldInclude to return false for all namespaces.
Populate expanded includes with all active namespaces when the original
includes was empty (meaning "include all") so that the wildcardExpanded
check does not falsely reject everything.
Signed-off-by: Joseph <jvaikath@redhat.com>
* Changelog
Signed-off-by: Joseph <jvaikath@redhat.com>
* Normalize empty includes to * instead of active namespaces list
This ensures consistent behavior between CLI and kubectl-apply paths
for Namespace CR inclusion when excludes contain wildcards.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Signed-off-by: Joseph <jvaikath@redhat.com>
* Move empty includes normalization to backup controller
Instead of normalizing empty IncludedNamespaces to ["*"] in the
collections layer's ExpandIncludesExcludes, do it earlier in
prepareBackupRequest. This ensures the spec is correct before any
downstream processing.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Signed-off-by: Joseph <jvaikath@redhat.com>
* Update TestProcessBackupCompletions for wildcard normalization
Add IncludedNamespaces: []string{"*"} to all expected BackupSpec
structs, reflecting the new prepareBackupRequest normalization.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Signed-off-by: Joseph <jvaikath@redhat.com>
* Add checks around empty includenamespaces
Signed-off-by: Joseph <jvaikath@redhat.com>
* gofmt
Signed-off-by: Joseph <jvaikath@redhat.com>
---------
Signed-off-by: Joseph <jvaikath@redhat.com>
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
klog v2 defaults -logtostderr to true, which silently ignores the
-stderrthreshold flag — all log levels are unconditionally sent to
stderr. This makes it impossible for log-aggregation systems to filter
by severity.
Bump klog to v2.140.0 and opt into the fixed behavior by setting
legacy_stderr_threshold_behavior=false and stderrthreshold=INFO (which
preserves current output while letting users override via CLI flags).
Ref: kubernetes/klog#212, kubernetes/klog#432
Signed-off-by: Pierluigi Lenoci <pierluigilenoci@gmail.com>
