TARGETARCH was referenced but never declared, so every platform in a
multi-arch buildx run received an amd64 binary. The arm64 image
crashed on Raspberry Pi nodes with an exec format error.
go.mod requires go >= 1.26.0 since the dependency updates, and the
golang alpine images set GOTOOLCHAIN=local, so the 1.25 builder failed
at go mod download.
Mirror update events were filtered out, so an in-place edit of a
mirrored copy was invisible to the controller. The periodic resync did
not catch it either: NeedsSync only compares the source content hash
with the recorded annotation, and tampered mirror data leaves both
unchanged.
Changes:
- The mirror watch now passes real update events (resourceVersion
changed, object is a managed mirror) and maps them to the source.
Informer resync replays keep an unchanged resourceVersion and stay
filtered.
- reconcileMirror detects in-place drift by rebuilding the expected
mirror in memory and comparing content hashes. Metadata differences
never count as drift, so the controller's own writes converge in one
no-op pass.
- The resync-period default moves from 10m to 1h. Watches now cover
source changes, mirror deletion, mirror drift, and namespace changes,
so the resync is only a safety net for missed watch events. The chart
default moves with it.
- Add predicate and tamper-repair regression tests; drift repair is
verified to make no direct apiserver reads.
The mirror reconcile path issued one direct apiserver GET per target
namespace on every pass. With N sources and M target namespaces each
resync produced N*M GET requests. On a 65-namespace cluster this
generated 63 GET/s and overloaded one control-plane node.
Changes:
- reconcileMirror now trusts the informer cache. Content drift is
detected via the source hash annotation. A stale resourceVersion
makes Update return a conflict, which requeues the source.
- The verify-source-freshness flag now defaults to false. The Helm
chart passes the value explicitly, so the chart default of false
matches the binary behaviour.
- Reconcile performs one namespace LIST per pass. Target resolution
and orphan cleanup share the result.
- Add load regression tests that count direct API reads in the
reconcile hot path.
Correct two wrong CLI flag defaults (leader-elect, verify-source-freshness) and add missing flags/Helm values. Remove documented-but-nonexistent annotations (excluded-namespaces, namespace-pattern) that silently did nothing, document the real exclude/paused and glob targeting, fix the mirror ownership-check example, and add the missing charts/kubemirror/README.md that README linked to.
kubemirror overrides controller names to <Kind>.<version>.<group> (e.g. Secret.v1.), but the dashboards, alerts, and recording rules filtered controller=~"secret|configmap" (the controller-runtime default name), so every panel and alert matched nothing. Switch all selectors to Secret.*|ConfigMap.* and document the naming convention.
The startup log hardcoded version="dev", so release builds always reported dev. Use the ldflags-set Version variable. Also modernise interface{} to any.
exclude="true" opts a resource out of mirroring and tears down its existing mirrors; paused="true" freezes mirrors in place (no updates, no cleanup). Both respected by the source and namespace reconcilers. Also dedup target-namespace resolution via delegation, remove 8 unused annotation constants and their misleading godoc, and silence gosec G101 false positives on base64 test fixtures.
Replace the triplicated ownership-verified delete guard with deleteOwnedMirror, and hand-built controller-name strings with gvkControllerName. Use constants.Domain/ControllerName over magic strings and maps.Copy/Clone for plain copy loops. Behaviour-preserving; tests green.
Discloses the single anonymous adoption ping sent on startup and points
users to the upstream README section for full opt-out instructions
instead of duplicating the table here.
Send a single fire-and-forget ping at startup to help track adoption
and version spread. No persistent identifiers are collected.
Adds main.Version var (defaulting to "dev") so the existing goreleaser
ldflags injection (-X main.Version={{.Version}}) now binds to a real
symbol.
Opt out via any of:
DO_NOT_TRACK=1
OSS_TELEMETRY_DISABLED=1
KUBEMIRROR_DISABLE_TELEMETRY=1
All registration goes through SetupWithManagerForResourceType (eager and
lazy paths in main.go); the legacy single-resource SetupWithManager and
its corev1 import were dead code that misled readers about how the
controller wires up.
M7: extractUnstructuredContent only hashed 'spec' when present, dropping
all other top-level content fields. Resources with both spec and data
(or any non-spec content) silently drifted until the next 10m resync.
Now hashes every non-Kubernetes-managed top-level field, matching the
fields updateUnstructuredMirror copies.
M6: when a source has a transform annotation, also hash the source's
labels and annotations (filtered of kubemirror.raczylo.com/* keys to
avoid the controller's own bookkeeping churning the hash). Templates
read these via TransformContext; without this a label change wouldn't
re-render the transformed mirror.
H3: text/template.Execute is not context-aware, so applyTemplateRule's
timeout cancels the select but leaks the executor goroutine. Added a
process-wide semaphore (cap 64) so a runaway template can't spawn an
unbounded number of stuck goroutines on every reconcile.
M4: zap dev mode (DPanic-on-error, console output, stacktraces on
warning) was hardcoded on. Defaulted to production; --zap-devel flag
remains for opt-in.
M5: WaitForInitialDiscovery was anchored on context.Background() with
its own WithTimeout, so SIGTERM during startup couldn't abort the wait.
Now anchors on signalCtx.
H2: scanAndRegister held d.mu (write lock) across registerController
and registerMirrorControllerOnly. Those calls enter controller-runtime's
manager state machine, which takes its own internal locks and can block
on cache sync — holding our application-level write lock across them
is a latent deadlock the moment any reentrant access happens (health
checks reading GetRegisteredCount, factories that introspect state).
Restructured into three phases: snapshot work under RLock, perform
registrations with NO lock held, then commit results under Lock.
Registration step routed through funcs to keep tests honest about
the lock state at the moment of invocation.
H4: NamespaceReconciler.reconcileMirror builds an ad-hoc SourceReconciler
to delegate mirror creation. The previous version left APIReader and
CircuitBreaker as nil, which silently disabled freshness verification
on the namespace-driven path (a label change to a target namespace
would mirror cached, possibly stale source data) and bypassed circuit
breaker accounting for those reconciles.
Construction extracted into newSourceReconciler so the forwarding is
covered by a unit test that pins both fields by identity.
C2: updateLastSyncStatus wrote the sync-status annotation on every
successful reconcile. Because the source's watch predicate is the
'enabled' label (server-side filter), that Update fires a watch event
that re-enters Reconcile. With reconciled/error counts varying across
cycles, the value differs each time, so the API server bumps RV and
the loop never quiesces. Now skips the Update when the value matches
the existing annotation.
C3: NamespaceReconciler's happy-path returned RequeueAfter=3s
unconditionally. Every namespace in the cluster re-reconciled every
3 seconds forever, generating constant List calls per source kind.
Now returns ctrl.Result{}; cache-staleness windows are handled by
the manager's resync period and source freshness verification.
H1: --verify-source-freshness used to default to false, so any source
update whose annotation was still in the informer cache (5-20s lag)
would resolve the wrong target list. cleanupOrphanedMirrors then ran
against the stale list and missed orphans (manifested in e2e as
'Orphaned mirror in kubemirror-e2e-app-1 not deleted within timeout'
after target-namespaces was changed). Defaulting to true fixes the
race; the trade-off is one extra API read per stale-cache reconcile.
M2: ResolveTargetNamespaces glob branch checked filter.IsAllowed but
not the opt-out map, so a namespace labeled allow-mirrors=false would
still receive a mirror through patterns like 'app-*'. The 'all' branch
already had the guard; the glob branch now does too. Direct namespace
listings still bypass opt-out by design (explicit author intent).
C1: deleteAllMirrors used to issue a blind Delete on every namespace
matching the source name+GVK, which would destroy unrelated resources
(e.g. a 'default' SA, 'ca-bundle' ConfigMap) sharing the source name.
Now reads each candidate, verifies managed-by label and source-reference
annotation, and only deletes confirmed mirrors.
M1: BlacklistedSecretTypes was declared but never enforced. Enabling
mirroring on a service-account-token / bootstrap-token / helm release
Secret would mirror credentials cluster-wide. Now refused at Reconcile.
M3: deleteAllMirrors swallowed per-namespace errors and returned nil,
so callers removed the finalizer even on partial failure (orphans).
Errors are now joined and returned.
lukaszraczylo
