Skip to content

feat(ha): promote a Standby to Active on Active-hub failure - #417

Open
sumanthd032 wants to merge 32 commits into
kubeslice:masterfrom
sumanthd032:feat/297-promotion
Open

feat(ha): promote a Standby to Active on Active-hub failure#417
sumanthd032 wants to merge 32 commits into
kubeslice:masterfrom
sumanthd032:feat/297-promotion

Conversation

@sumanthd032

@sumanthd032 sumanthd032 commented Aug 5, 2026

Copy link
Copy Markdown

Description

Implements failover (ADR #293): detecting the Active hub is gone and promoting the Standby.

  • Elector retains the last successfully-read remote Lease, so a dead API server (not just a dead pod) is detectable.
  • Two pre-promotion guards (self-health, final bounded re-read) with metrics.
  • Promotion sequence: stop and confirm the mirror, take the Lease, publish status.activeController, open the write fence, kick a reconcile of existing objects. Idempotent, bounded by --ha-promotion-dial-timeout / --ha-promotion-grace-period.
  • Emits a PromotedToActive event.

This branch merges feat/297-active-controller (#414) on top of #416. Its diff includes #414's commits until that merges. New commits here start at 825f7531 through d1d33fc (7 commits plus the merge).

Part of #297

How Has This Been Tested?

go test -race -count=1 ./pkg/ha/..., including a concurrency test for the promote-once-only latch. Live-tested against a Kind hub with its API server shut down: staleness detected in ~21s against a 20s budget.

Checklist:

  • The title of the PR states what changed and the related issues number (used for the release note).
  • Does this PR requires documentation updates?
  • I've updated documentation as required by this PR.
  • I have performed a self-review of my own code.
  • I have commented my code, particularly in hard-to-understand areas.
  • I have tested it for all user roles.
  • I have added all the required unit test cases.

Does this PR introduce a breaking change for other components like worker-operator?

No. Promotion only runs under --ha-mode=standby.


Introduce pkg/ha, the foundation for cross-cluster (Active/Standby)
high availability per ADR kubeslice#293 (issue kubeslice#294).

- HAMode (active|standby|standalone) with fail-safe parsing: empty or
  unknown input maps to standalone so a misconfig never disables writes.
- Lease helpers over coordination.k8s.io/v1: acquire/renew (bumping
  leaseTransitions on takeover), get, and a leaseDuration+padding
  staleness check.
- ClusterLeaderElector: IsLeader() reads an atomic flag kept current by
  background loops, so it is cheap enough to call at the top of every
  Reconcile and always reflects live leadership. StartLeaseRenewal
  (Active) renews the local Lease and releases leadership once the renew
  deadline is exceeded (natural fencing). WatchRemoteLease (Standby)
  reads the Active's Lease and logs staleness but does not promote;
  promotion is issue kubeslice#297.

Standalone is the default and is always the leader, preserving today's
single-hub behaviour (no regression).

Unit tests are race-clean and cover leadership by mode, renew-deadline
loss, lease staleness, and the standby-never-promotes boundary.

vendor: add controller-runtime fake client + interceptor packages
(test-only) via go mod vendor.

Signed-off-by: Sumanth D <sumanthd032@gmail.com>
Wire pkg/ha into the controller so only the Active hub writes (issue kubeslice#294).

- Add a LeaderElector field to all nine reconcilers and a per-call guard
  at the top of every Reconcile: a Standby logs "standby mode, skipping
  reconcile" and returns without writing. The guard is nil-safe, so a
  reconciler built without an elector keeps today's behaviour.
- main.go: add --ha-mode, --ha-identity, --ha-active-kubeconfig,
  --ha-lease-duration, --ha-renew-deadline, --ha-retry-period and
  --ha-padding-seconds; construct the elector (building a remote client
  from the mounted Active kubeconfig in standby mode); start
  StartLeaseRenewal (active) or WatchRemoteLease (standby) and pass the
  shared signal-handler context to the manager.
- Add coordination.k8s.io/leases RBAC for the Lease.

--ha-mode=standalone is the default and is always the leader, so existing
single-hub deployments are unaffected (no regression). The existing
--leader-elect (in-cluster pod election) is left untouched.

A controller test asserts the Standby skips and logs on every call.

vendor: add go.uber.org/zap/zaptest/observer (test-only).
Signed-off-by: Sumanth D <sumanthd032@gmail.com>
The HA leader-election Lease lives in the controller's own namespace and
is already covered by the existing leader-election Role's leases grant
(config/rbac/leader_election_role.yaml). The kubebuilder marker added a
redundant cluster-wide grant and left the generated manifests out of sync
(make manifests was not run). Replace it with a note pointing at the role
that provides the permission, keeping markers and manifests consistent.

Signed-off-by: Sumanth D <sumanthd032@gmail.com>
…hutdown

- acquireOrRenewLease rounds LeaseDurationSeconds up to whole seconds and
  clamps to a minimum of 1, so a sub-second --ha-lease-duration is not
  truncated to 0 (invalid, and skews staleness checks).
- StartLeaseRenewal and WatchRemoteLease return nil instead of ctx.Err() on
  context cancellation, so a graceful shutdown is not logged as an error.

Signed-off-by: Sumanth D <sumanthd032@gmail.com>
The Lease namespace defaulted to a hard-coded constant, but the controller
runs in a namespace injected at runtime via KUBESLICE_CONTROLLER_MANAGER_NAMESPACE
(downward API), and the leader-election Role that grants leases is namespaced to
that deploy namespace. Deploying into any other namespace would create the Lease
where the controller has no leases permission, so the Active could not renew it
and would fence itself permanently.

Add --ha-lease-namespace defaulting to KUBESLICE_CONTROLLER_MANAGER_NAMESPACE so
the Lease lands in the controller's own namespace. Empty (local runs) falls back
to the pkg/ha default.

Signed-off-by: Sumanth D <sumanthd032@gmail.com>
NewClusterLeaderElector fell back to the hard-coded DefaultLeaseNamespace
whenever LeaseNamespace was empty, independent of main.go's flag default.
Since the controller already exposes KUBESLICE_CONTROLLER_MANAGER_NAMESPACE
to represent its actual runtime namespace, check that env var first so the
package resolves correctly on its own, not only by accident of how main.go
wires the --ha-lease-namespace flag default. Only falls back to
DefaultLeaseNamespace when the env var is unset too (e.g. running outside
a pod). Regression tests included.

Signed-off-by: Sumanth D <sumanthd032@gmail.com>
StartLeaseRenewal/WatchRemoteLease returning nil (not ctx.Err()) on
context cancellation was fixed in 0da6d82 but never actually got a
regression test. Also add coverage for: the mode-guard no-op branches,
WatchRemoteLease failing fast without a remote client, getLease/
checkRemoteLeaseOnce propagating a missing-Lease error instead of
reporting fresh, renewOnce keeping leadership on a transient failure
within renewDeadline, and setLeader logging Acquired/Lost exactly once
per transition (F4 in 294-evaluation.md).

No production code changes.

Signed-off-by: Sumanth D <sumanthd032@gmail.com>
…uard, ownerRef strip, status mirroring)

Part of kubeslice#295 — the write path RemoteSyncer's workqueue will drive
(next commit) to mirror KubeSlice CRDs from the Active hub onto the
Standby.

- MirroredResource + CRDMirrorSet (pkg/ha/mirror_set.go): the hub-side
  resource table (Project, Cluster, SliceConfig, ServiceExportConfig,
  SliceQoSConfig, VpnKeyRotation, WorkerSliceConfig, WorkerSliceGateway,
  WorkerServiceImport, plus core Namespace). Deliberately does not match
  issue kubeslice#295's own CRD table, which names Slice/SliceGateway/
  ServiceExport — none of those types exist in this repo; they are
  worker-cluster data-plane CRDs owned by the separate worker-operator
  repo, irrelevant to hub-to-hub mirroring.

- mirrorCreateOrUpdate/mirrorDelete (pkg/ha/mirror.go): strip
  resourceVersion/uid/managedFields/finalizers (+ ownerReferences only
  for VpnKeyRotation, the one mirrored type that carries one) before
  writing, label mirrored objects ha.kubeslice.io/synced-from=active,
  and only ever overwrite or delete a target object that already
  carries that label — the conflict guard that keeps the syncer off
  anything the Standby's own reconcilers or an operator created
  directly, including pre-existing namespaces like kube-system/default
  now that Namespace is an ordinary mirrored type rather than a
  special-cased cold-start step.

- Every mirrored type has a status subresource. A plain Update() never
  touches .status once one is registered, so mirrorCreateOrUpdate
  always follows up with an explicit Status().Update() when the source
  object has a non-empty status — matching this repo's own
  UpdateStatus/CleanupUpdateStatus convention. Regression-tested.

Fake-client unit tests cover create, update-of-existing, the conflict
guard on both update and delete, delete-idempotent-on-NotFound,
StripOwnerRefs true/false, and status mirroring.

Signed-off-by: Sumanth D <sumanthd032@gmail.com>
Part of kubeslice#295. Registered via prometheus.NewHistogramVec/NewCounterVec
+ prometheus.MustRegister in pkg/ha's own init(), the same
self-registration idiom metrics/prometheus.go already uses for
KubeSliceEventsCounter — not routed through
metrics.StartMetricsCollector, whose default labels are slice-specific
and don't apply to a cross-cluster mirror.

- ha_sync_lag_seconds{kind,operation}: time.Now() minus
  CreationTimestamp for creates; minus first-enqueue time for
  update/delete (RemoteSyncer's workqueue coalesces repeated events for
  the same object, so there's no single "delivery time" once a retry
  has backed off a few times — first-enqueue is the more useful number
  to alert on, since it reflects total time since the triggering
  change).
- ha_sync_errors_total{kind,operation}: counts mirror failures. The
  syncer keeps running and retries via its workqueue on every
  increment; this metric never indicates a crash.

vendor: add prometheus/client_golang/prometheus/testutil (test-only)
via go mod vendor, used to assert metric samples/labels directly.

Signed-off-by: Sumanth D <sumanthd032@gmail.com>
Part of kubeslice#295. RemoteSyncer runs only in standby mode: builds a
controller-runtime cache.Cache against the Active hub's rest.Config,
registers one informer per CRDMirrorSet entry, and starts a small
worker pool draining a rate-limited workqueue.TypedRateLimitingInterface
[syncKey] — the same primitive controller-runtime's own Controller
uses internally (internal/controller.Controller).

Informer callbacks (handlersFor) only enqueue a syncKey; no mirror
logic runs on the informer's own goroutine. A burst of Update events
for the same object coalesces into one queued item, and a worker
determines the real action at dequeue time by re-reading the Active
cache (found -> mirrorCreateOrUpdate, NotFound -> mirrorDelete) — the
same way a Reconcile call would.

This is the commit that satisfies issue kubeslice#295's own acceptance
criterion that the syncer "retries without crashing": on any mirror
failure, processOnce calls queue.AddRateLimited instead of dropping
the key. The concrete failure mode this fixes: a namespaced object
(e.g. a new Project's Cluster) created on Active after the Standby's
initial sync has already completed previously had no path to retry —
nothing fires again for an object that didn't change on Active once
its first mirror attempt failed on "namespace not found". Now it just
gets retried a few seconds later once the Namespace mirror (an
ordinary CRDMirrorSet row, no special-casing needed) has landed.

Start(ctx) delegates its blocking wait directly to
remoteCache.Start(ctx), which itself blocks on <-ctx.Done() and
returns nil — giving the "return nil, not ctx.Err(), on graceful
shutdown" contract StartLeaseRenewal/WatchRemoteLease already use, for
free.

remoteGetFunc is a small seam (defaulted to a real cache.Cache.Get in
the constructor) so the retry engine's tests exercise real workqueue
backoff/redelivery behaviour without a *rest.Config or live cluster.

Signed-off-by: Sumanth D <sumanthd032@gmail.com>
Part of kubeslice#295. Constructs ha.RemoteSyncer alongside the existing
ClusterLeaderElector, reusing the same remote *rest.Config and local
client the elector already builds rather than loading the Active
kubeconfig twice — hoisted the standby block's remoteCfg to an outer
remoteHACfg variable so both consumers can see it.

Starts remoteSyncer.Start(ctx) in its own goroutine alongside the
existing leaderElector.WatchRemoteLease(ctx) in the ha.ModeStandby
switch case; a no-op in any other mode, matching RemoteSyncer's own
mode check.

New flag: --ha-sync-workers (default ha.DefaultSyncWorkers), matching
the existing --ha-* flag style. --ha-sync-interval is intentionally
not added yet — it belongs to the periodic prune backstop (pkg/ha/
prune.go), which is out of scope for this PR and would otherwise be a
flag with no consumer.

Signed-off-by: Sumanth D <sumanthd032@gmail.com>
…mote read identity

Part of kubeslice#295. --ha-active-kubeconfig's RBAC scope on the Active cluster
has been undefined since kubeslice#294 introduced the flag — the dev demo uses
a full-admin kubeconfig, and config/rbac/leader_election_role.yaml
only ever granted configmaps/leases, nothing for the CRD/Namespace
reads RemoteSyncer now needs. The ADR (kubeslice#293) itself names this as an
unaddressed deployment-level boundary without resolving it.

config/ha/active-cluster-clusterrole.yaml: a read-only (get/list/watch)
ClusterRole covering Namespace plus every pkg/ha.CRDMirrorSet type,
plus a ClusterRoleBinding template (subject left as a placeholder —
it's deployment-specific, either a ServiceAccount for in-cluster
dialing or a client-cert User for a flattened kubeconfig, and can't be
hardcoded). config/ha/README.md explains this must be applied on the
Active cluster by whoever provisions it, not through this repo's own
deploy/kustomize flow — confirmed neither config/rbac/kustomization.yaml
nor config/default/kustomization.yaml reference this directory, so it
can never be accidentally auto-applied to the Standby's own cluster,
where it would be meaningless.

Also documents, ahead of time, that a later credential-mirroring PR
appending Secrets to this grant exposes every Secret in the project
namespaces on Active, not just the ones actually mirrored — RBAC can't
scope Secrets by .type — a real tradeoff to weigh when that lands.

Signed-off-by: Sumanth D <sumanthd032@gmail.com>
Start gave up permanently on the first GetInformer/AddEventHandler
error instead of retrying, unlike StartLeaseRenewal/WatchRemoteLease.
Split setup into registerInformers (retries with backoff) wrapping
registerInformersOnce (one attempt).

A naive retry would double-register handlers on resources that
already succeeded, since AddEventHandlerWithResyncPeriod isn't
idempotent. Added a handlerRegistered map so retries skip resources
already done.

CRDMirrorSet's Namespace entry had no filter, so it mirrored every
namespace on the Active hub, not just kubeslice ones — and
mirrorDelete's label-only guard meant an unrelated Active-side delete
could cascade-delete one of those on the Standby. Scoped the
Namespace informer to util.LabelsKubeSliceController via
cache.Options.ByObject.

Signed-off-by: Sumanth D <sumanthd032@gmail.com>
An Active-side object mid-Terminating (deletionTimestamp set, contents
still being garbage-collected) is delivered by the informer as an
ordinary Update, not a Delete yet. mirrorCreateOrUpdate's payload
carried deletionTimestamp straight through, and updating the Standby's
non-terminating mirror with it set fails real API server immutable-
field validation. Stripping it surfaced a second, related failure:
the status-mirror block was still copying a Terminating status onto
that now-non-terminating payload, which a real server also rejects
(status.Phase may only be Terminating if deletionTimestamp is set).

Both are stripped now: deletionTimestamp/deletionGracePeriodSeconds
unconditionally alongside the other identity fields already cleared
there, and status explicitly via delete(payload.Object, "status")
rather than relying on a real API server silently ignoring .status on
the main resource endpoint for subresource-registered types — a fake
client does not replicate that behavior, so the code's correctness
would otherwise depend on which client it's running against. The
Standby still converges correctly once Active reports NotFound and
mirrorDelete takes over; there's nothing useful to reflect about the
in-between Terminating state.

New tests: TestMirrorCreateOrUpdate_StripsDeletionTimestampFromTerminatingSource,
TestMirrorCreateOrUpdate_SkipsStatusMirrorWhenSourceIsTerminating.

Signed-off-by: Sumanth D <sumanthd032@gmail.com>
Dockerfile never copied pkg/ into the build context, so no image has
been buildable since pkg/ha was introduced — go build works directly,
docker build did not. Added COPY pkg/ pkg/ alongside the other source
directories.

config/rbac/role.yaml never granted namespaces/status: every other
CRDMirrorSet type already has an explicit <kind>/status rule, but
Namespace never needed one before it became a mirrored type with
explicit status mirroring. Under real RBAC this permanently fails the
Namespace status write. Added the +kubebuilder:rbac marker in main.go
next to the existing namespaces rule and regenerated via make
manifests.

config/ha/active-cluster-clusterrole.yaml never granted
coordination.k8s.io/leases: it was scoped only to what RemoteSyncer
itself reads (Namespace + CRDMirrorSet), but the same
--ha-active-kubeconfig identity is also used by WatchRemoteLease to
read the Active's own Lease directly. Under a least-privilege identity
using exactly this sample, the Standby could mirror correctly but
never observe Lease staleness at all. Added the leases rule and
updated the README to describe both consumers of this grant.

Signed-off-by: Sumanth D <sumanthd032@gmail.com>
Informers self-heal missed updates via periodic resync and the workqueue
owns retry-on-failure, but neither can remove a mirror whose Active-side
original was deleted while the Standby wasn't watching (e.g. between two
Standby runs): cold-start informers only deliver what currently exists,
so such an orphan would survive forever.

Add a prune loop to RemoteSyncer that periodically lists Standby objects
carrying the ha.kubeslice.io/synced-from label per mirrored type, diffs
them against the remote cache, and enqueues anything no longer present
on the Active hub onto the ordinary mirror workqueue. The worker re-reads
Active at dequeue time, so the existing conflict guard, NotFound->delete
semantics, and rate-limited retry apply unchanged, and no second write
path races the workers.

Fail-safe choices:
- The first pass waits for the remote cache to sync; an unsynced cache
  lists empty, which would otherwise read as "everything was deleted"
  and prune every mirror on the Standby.
- A failed list (remote or local) skips that kind for the round and
  increments ha_sync_errors_total{kind,"prune"} instead of pruning on
  partial information.

Configurable via --ha-sync-interval (default 60s).

Part of kubeslice#295

Signed-off-by: Sumanth D <sumanthd032@gmail.com>
…lure

Surface mirror failures as Kubernetes events on the Standby, attached to
the object that failed to sync, using the EventRecorder main.go already
builds for the reconcilers. The entry lives in
config/events/controller.yaml with the generated map and config-map
output from make generate-events committed alongside — RecordEvent
hard-fails for any EventName missing from the generated EventsMap, so
skipping that step would silently no-op the whole feature. A regression
test pins the entry's presence in the generated map (and that a missing
entry errors loudly), so an accidental revert of the generated code
fails in go test rather than at runtime.

One event per failure episode, not per retry attempt: NumRequeues is 0
only on the first failure since the last success, and early workqueue
backoff retries arrive milliseconds apart — although the recorder
aggregates repeats into one Event's Count, every call is still an
API-server write. ha_sync_errors_total continues to count every attempt.

The recorder is called directly rather than through util.RecordEvent:
that helper logs via util.CtxLogger, which panics on any context that
didn't pass through a reconciler's request-context setup — true for the
syncer's own context (main.go's signal-handler context). Caught by the
new event-emission test before it could crash a live Standby.

Part of kubeslice#295

Signed-off-by: Sumanth D <sumanthd032@gmail.com>
…Account, Role, RoleBinding)

Mirror the credential objects a promoted Standby needs to serve its
worker clusters without manual re-provisioning: worker-identity RBAC
(Role/RoleBinding/ServiceAccount — the only RBAC kinds
access_control_service.go ever creates; no ClusterRole or
ClusterRoleBinding exists to mirror, despite ADR Decision 6's broader
wording) and Secrets such as the gateway certificates the ovpn job
generates.

kubernetes.io/service-account-token Secrets are excluded: SA tokens are
signed by the issuing cluster's service-account key, so an Active-minted
token is cryptographically invalid on the Standby. Mirroring the
ServiceAccount is what matters — the Standby's own token controller
mints a locally-valid token for it.

Unlike the CRD set, these are core types that exist cluster-wide, so
every row is scoped in two layers:

- Server-side, the remote informers are scoped in cache.Options:
  ServiceAccount/Role/RoleBinding by the same
  util.LabelsKubeSliceController selector the Namespace informer already
  uses (util.GetOwnerLabel embeds that exact key/value pair on every
  credential object the controller creates), and Secret — which cannot
  be label-scoped, the cert-generator job creates its Secrets unlabeled
  — by a field selector excluding the SA-token type.
- Client-side, every row sets RequireMirroredNamespace: the object only
  mirrors if its namespace is in the remote cache's label-scoped
  Namespace view, i.e. a namespace the syncer itself mirrors. The
  boundary is deliberately NOT name-based: under the Helm chart's
  real-world --project-namespace-prefix ("kubeslice-"), the
  controller's own kubeslice-controller namespace matches the
  project-namespace naming pattern, and a prefix rule would have
  mirrored its webhook TLS key, image-pull credentials, and Helm
  release Secrets onto the Standby — found by running the Standby
  against a Helm-installed Active hub. The label boundary is the one
  ReconcileProjectNamespace actually maintains. A transient failure
  reading the namespace surfaces as an error (workqueue retry), never
  as a silent skip.

StripOwnerRefs is set on every row: ownerReferences resolve by UID,
which never survives a cross-cluster copy, and credential objects are
also written by actors outside this repo (token controller, cert job),
so the CRD set's audited only-VpnKeyRotation-needs-it reasoning cannot
hold here.

Verified live against the real two-hub Kind topology: worker SAs,
Roles, RoleBindings, and dashboard Secrets in both project namespaces
mirror with the sync label; SA-token Secrets and everything in
kubeslice-controller and kube-system stay off the Standby.

Part of kubeslice#295

Signed-off-by: Sumanth D <sumanthd032@gmail.com>
The prune loop so far only walked one direction — Standby mirrors whose
Active-side original disappeared. Walk the other direction too: an
Active-side object with no Standby mirror gets re-enqueued onto the
ordinary workqueue, where the worker re-reads Active and runs the full
Skip/namespace/conflict-guard chain.

Three real cases produce that state, none of which the forward pass or
the informers handle promptly:

- a mirror someone deleted directly on the Standby (previously healed
  only by the informer's resync period, default 10 minutes);
- an object whose RequireMirroredNamespace verdict was decided before
  its namespace's informer had delivered on cold start — a skip is
  terminal for that queue item, so without this it also waited for
  resync;
- a key stuck deep in rate-limiter backoff after repeated failures
  (observed live: a fixed failure cause still took minutes to heal
  because the next retry was scheduled ~164s out; enqueue bypasses the
  rate limiter's delay, so recovery lands within one prune tick).

Re-enqueueing is always safe: objects that should not mirror simply
no-op through the guards again. Verified live: a mirrored Secret
deleted directly on the Standby was re-created 15s later, within one
--ha-sync-interval, with zero errors logged.

Part of kubeslice#295

Signed-off-by: Sumanth D <sumanthd032@gmail.com>
…C sample

The Standby now mirrors ha.FullMirrorSet — CRDMirrorSet plus
CredentialMirrorSet. No new configuration: the mirrored-namespace gate
derives entirely from the label boundary the controller already
maintains, so the syncer needs no knowledge of the deployment's
--project-namespace-prefix.

config/ha/active-cluster-clusterrole.yaml gains read access to
secrets/serviceaccounts (core) and roles/rolebindings (rbac), and the
README's forward-looking credential-mirroring note becomes present-tense
documentation. The Secret grant's security tradeoff is stated where it
will be read rather than buried: RBAC cannot scope Secrets by .type or
namespace label, and a ClusterRole binding is cluster-wide, so the
Standby's identity can read every Secret on the Active hub — the
syncer's field selector and mirrored-namespace gate narrow what gets
copied, not what the identity could read. The narrower per-namespace
RoleBinding alternative is documented alongside, with its maintenance
cost. Grant coverage verified live via an impersonation can-i matrix:
get/list/watch allowed for every mirrored type plus the HA Lease,
writes and unrelated resources denied.

Fixes kubeslice#295

Signed-off-by: Sumanth D <sumanthd032@gmail.com>
The signal a worker uses to find the Active hub after a failover, per ADR
kubeslice#293 Decision 7. Each hub writes this field about itself, on its own API
server, and only while it holds leadership.

A Standby's copy is populated by the state mirror from the Active, so it
names the Active rather than itself. That is what lets a worker watching
both hub endpoints resolve which one is Active by the rule "trust whichever
endpoint is reachable and reports an ActiveIdentity matching that
endpoint's own identity", without needing to know which role either hub
currently holds, and without inferring a death from a timeout.

LastUpdated is not in the ADR's YAML sketch. It is added deliberately:
Decision 7's open tie-break question needs a freshness signal if a
partition causes both hubs to self-declare at once, and comparing a
timestamp already on the object is cheaper than making the worker read
coordination.k8s.io Leases across clusters. StorageCapabilities.LastUpdated
in this same struct is existing precedent for the pattern.

The field is additive and omitempty throughout, so a non-HA deployment
never populates it and an existing worker sees no behaviour change.

The same types are being added to github.com/kubeslice/apis, which is what
worker-operator imports; this repo carries its own copy of them.

Note on the CRD manifest: only the activeController schema is included.
make manifests also rewrites the controller-gen version annotation in all
ten CRD files, because the committed manifests were generated with v0.19.0
while the Makefile pins v0.17.3. That pre-existing drift is left alone
rather than folded into this change.

Part of kubeslice#297

Signed-off-by: Sumanth D <sumanthd032@gmail.com>
…ship

ADR kubeslice#293 Decision 7 requires each hub to declare itself on its own API
server while it holds leadership, so a worker watching both hub endpoints
can tell which one is Active without knowing either hub's role.

Publishing only at promotion would leave a worker unable to identify the
Active before the first-ever failover, so this is a continuous loop rather
than a step in the promotion sequence. It is also standalone rather than
part of ClusterService.ReconcileCluster, because it has to converge
independently of reconciler traffic — and reconciler traffic is exactly
what is absent right after a promotion, when the write fence has just
opened but nothing has re-enqueued the pre-existing objects yet.
PublishOnce is exported so promotion can run one synchronous pass and not
tie failover latency to the tick.

Details worth calling out:

- The convergence check deliberately excludes LastUpdated. Including it
  would make every pass differ from itself and turn a convergence check
  into a write to every Cluster CR on every tick.

- The publisher refuses to write an empty endpoint or the shipped
  placeholder (https://controller.cisco.com:6443/), because advertising an
  unreachable address as the failover target is worse than advertising
  nothing. Refusing is not an error: a hub that cannot describe itself
  should keep reconciling.

- The placeholder literal is duplicated in pkg/ha rather than imported,
  because main.go overwrites service.ControllerEndpoint with the flag value
  at startup and the default is unrecoverable afterwards.
  TestPlaceholderMatchesServiceDefault fails if the two ever drift.

- An unreadable CA bundle is logged and publication continues without it.
  The endpoint and identity are what select a hub, and a worker that
  already pins the hub's CA does not need it republished.

- Nothing ever clears the field. A hub stops publishing only by losing
  leadership, which means it stopped renewing its Lease and is unreachable,
  so a worker cannot read the stale declaration anyway. Auto-demotion of a
  recovered hub is an explicit ADR non-goal (Decision 8), and LastUpdated
  is what lets a consumer prefer the fresher of two claims if it ever does
  see both.

The elector is taken as a narrow two-method interface so the publisher is
testable without a live Lease. 12 tests, covering the not-leader no-op, the
converged-pass-writes-nothing property, both endpoint refusals, CA bundle
encoding and absence, partial failure across clusters, and graceful
shutdown.

Part of kubeslice#297

Signed-off-by: Sumanth D <sumanthd032@gmail.com>
Adds --ha-self-ca-bundle-path (default the in-pod service account CA path)
and starts the publisher alongside the existing HA loops.

Two wiring decisions worth stating:

It is deliberately not started in standalone mode. Standalone is always the
leader, so the publisher would run and start writing status.activeController
on every existing non-HA deployment. Leaving the field absent there is what
keeps an existing worker's behaviour unchanged, which is the no-regression
guarantee HA is built on.

A Standby does start it. The publisher no-ops while the hub is not the
leader, so it costs one list per interval and needs no extra wiring when
promotion flips leadership in a later change.

It writes through localHAClient — the same direct, uncached client the
elector uses — rather than the manager's cached client, so it does not
depend on the manager cache having started.

Part of kubeslice#297

Signed-off-by: Sumanth D <sumanthd032@gmail.com>
Found in live testing against a Kind hub: a freshly started Active took 31
seconds to advertise itself, not the ~2 seconds intended.

Start ran its first pass immediately, but an Active does not hold its Lease
yet at that instant — acquisition lands a second or two later. So the first
pass saw IsLeader() false, skipped, and the next attempt was a full publish
interval away. Any worker booting inside that window could not identify the
hub.

Unit tests could not catch this: the test double is the leader from the
first call, so the race does not exist there. The fix is driven by the loop
now waiting on the short leadership interval whenever a pass found this hub
was not the leader, and on the publish interval only once it is. A
non-leader returns before touching the API server, so polling at 2s costs
nothing while idle — and it means a Standby also picks up leadership
promptly at promotion, independently of promotion remembering to call
PublishOnce.

The regression test then caught a second, narrower version of the same bug
in the first fix: choosing the wait from its own IsLeader() call meant
leadership arriving between the publish check and the wait check still cost
a full interval. publishOnce now reports whether it held leadership, and
the wait is chosen from what the pass actually did rather than from a
second read.

Verified live after the fix: published in 2s. resourceVersion held steady
across a full publish interval, so the convergence check still writes
nothing once converged.

Part of kubeslice#297

Signed-off-by: Sumanth D <sumanthd032@gmail.com>
Signed-off-by: Sumanth D <sumanthd032@gmail.com>

# Conflicts:
#	main.go
…ed pod

checkRemoteLeaseOnce returned (false, err) when the read failed — reporting
"not stale". So only one failure mode was ever detectable: the controller
pod dying while its API server stayed up. An Active that lost its API
server, its node, or the whole cluster was invisible, forever. That is the
disaster this feature exists for, and it is the one case every demo so far
could not have caught, because they all killed the process.

The fix follows from noticing that the two cases are the same event. When
the pod dies, reads succeed and renewTime is frozen at T. When the API
server dies, reads fail, so the newest renewTime this hub has ever seen is
frozen at T. In both, the newest proof of life stops advancing; the
difference is at the transport layer, not in the meaning.

So the elector now retains the last successfully-read Lease. A successful
read replaces it; a failed read leaves it alone and logs. The verdict is
then a single isLeaseStale call against that retained view, which ages on
its own against a moving clock — covering both modes with one threshold, no
second timer, and the already-tested staleness helper doing the work.
Detection lands at roughly leaseDuration + padding + one poll in both.

checkRemoteLeaseOnce still never changes leadership. It reports candidacy;
the guards and the promotion sequence are separate commits, so "we think
the Active is gone" and "we took over" stay independently testable, and
kubeslice#294's tests asserting a Standby does not promote continue to hold.

⚠️ The nil check for the retained Lease is deliberately a separate
statement and must never be folded into the isLeaseStale call.
isLeaseStale(nil, ...) returns TRUE — correct for its original caller,
where a Lease absent from your own cluster should be created — but here it
would mean a Standby that has never once read the Active's Lease promotes
itself on its first tick. A broken kubeconfig, a missing RBAC grant or a
mistyped namespace would each become a guaranteed split brain.
TestNeverArmed_NeverBecomesCandidate fails if anyone merges the two
conditions.

That nil case doubles as the arming rule: never promote without having
proved, at least once, that the Active's Lease is reachable. It separates
"it worked, then it stopped" from "it never worked". The cost is real and
accepted: a Standby restarting during an outage can never arm and so will
not promote. That is the safer failure — a missed promotion is visible
downtime an operator can resolve, a false one is silent dual writes — and
it is now documented in the ADR beside the split-brain non-goal.

lastGoodRead is recorded but not used by the verdict, which anchors on the
Lease's own renewTime. It is carried so an optional local-only staleness
floor stays available without a redesign if clock skew between hubs ever
becomes a practical problem.

Part of kubeslice#297

Signed-off-by: Sumanth D <sumanthd032@gmail.com>
Staleness says the Active's newest proof of life has aged out. It does not
say that is why. These guards ask whether the observation is actually
evidence about the Active, and record every refusal.

They exist because the costs are not symmetric. A missed promotion is
downtime — visible, and an operator can force a takeover. A false promotion
is two hubs writing to their own copies of the world simultaneously, with
the mirror still running in one direction: objects overwriting each other,
prune's reverse diff resurrecting deletes, workers taking contradictory
instructions, and recovery by hand. Two bounded reads to avoid the second
is a good trade inside a budget that already tolerates leaseDuration +
padding.

Self-health. A dead Active API server, a partition between the hubs, and
this hub losing its own networking produce byte-identical observations:
reads of the remote Lease simply stop succeeding. Asking whether the local
API server still answers is the only cheap way to separate the last case,
and it is the most likely cause of a false promotion after outright
misconfiguration. It reads this hub's own Lease through the existing local
client, so it needs no new client and no new RBAC — the leader-election
Role already grants leases in the controller's own namespace.

⚠️ NotFound counts as HEALTHY here, and getting it backwards would block
every real first failover while looking entirely reasonable in review. On a
first-ever promotion no Lease exists on this hub yet, and NotFound means
the API server answered — precisely what is being tested. Only transport
errors, timeouts and server errors indicate an unhealthy self.
TestSelfHealthy_NotFoundCountsAsHealthy pins it.

Final dial. One fresh, bounded read at decision time. Worth stating plainly
what it does and does not buy: against an Active that is reachable but has
stopped renewing, it closes a real polling race, because the Active may
have renewed moments after the last poll. Against an unreachable Active it
buys almost nothing — it is the next failed read after a sustained run of
them. And it is NOT a split-brain guard: in a genuine partition it travels
the same broken path as every other read, fails identically, and the
Standby promotes anyway. Safety on that path comes from duration and from
the arming rule, not from this read. The ADR previously implied otherwise
and has been corrected.

Both reads are bounded by --ha-promotion-dial-timeout (5s). This is not
optional: main.go builds the remote client with a plain uncached client.New
and no timeout, so a dial to a black-holed API server blocks until the OS
TCP timeout — minutes, far outside the failover budget.

Aborting changes nothing but the attempt. The cached Lease is not cleared
and the elector is not disarmed, so the next tick re-evaluates from scratch
and a genuinely recovered Active heals the state on its next successful
read. Clearing it would be worse than useless: an already-gone Active can
never re-arm the elector, so one transient guard failure would leave a hub
that never fails over again. TestGuardsAbort_DoesNotDisarm covers it.

Metrics: ha_failover_total, and ha_promotions_aborted_total{reason} with
reasons self_unhealthy, lease_live and already_promoting. Without the
second, every guard is invisible in production — a hub that correctly
declines to take over looks identical to one that never noticed anything —
and these are the branches most worth demonstrating.

--ha-promotion-grace-period is defined here too, used by the next commit.
It is a sequencing budget for publishing status.activeController before the
write fence opens, and is deliberately distinct from --ha-padding-seconds,
which is a detection threshold. Issue kubeslice#297's --ha-promotion-grace is an
alias of the latter and is not implemented.

Part of kubeslice#297

Signed-off-by: Sumanth D <sumanthd032@gmail.com>
The sequence a Standby runs once detection says the Active's newest proof
of life has aged out. Two orderings in it are load-bearing, and both issue
kubeslice#297 and an earlier draft of ADR Decision 5 had them wrong.

The mirror is stopped, and confirmed stopped, before the write fence opens.
The most common trigger for a failover is the Active's controller pod dying
while its API server stays perfectly healthy — in which case the mirror's
informers are still live and still mirroring at the moment of promotion.
Every mirrored object carries the syncer's own label and the mirror's
conflict guard only skips objects WITHOUT it, so a fence opened first means
the mirror overwrites exactly the objects the new Active's reconcilers are
writing, while prune's reverse diff resurrects anything they delete. You
would promote into a hub fighting itself. A mirror that fails to stop
aborts the promotion rather than pressing on, because half-stopped is the
very state this step exists to prevent.

The reconcile kick runs after the fence opens, not before. The fence drops
requests rather than requeuing them, so a kick delivered while it was still
shut would be dropped without requeue — the exact failure the kick exists
to fix.

Steps 0 and 8 bracket the whole sequence with a promoting latch, and
IsLeader() reports false while it is held regardless of the leadership
flag. So the fence stays shut from the first step to the last, which gives
"the reconcilers are not live until promotion completes" real teeth without
inventing any external status surface.

Effects outside the elector are injected as PromotionHooks rather than
imported, so pkg/ha stays independent of the mirror, the publisher and the
manager, and so the entire sequence is testable without any of them. Nil
hooks are skipped: an elector with none still takes leadership correctly,
it just does it without the parts that make promotion safe and complete.

Failure handling follows what each step actually costs. A guard refusal is
a correct outcome, not an error, and leaves the hub a fenced, still-armed
Standby free to retry next tick. Publication is bounded by
--ha-promotion-grace-period and failing it still promotes, because a hub
that cannot describe itself is a better Active than no Active at all and
the publisher's own loop keeps retrying. Failing the kick likewise still
promotes; it costs pre-existing state staying unreconciled, not the
takeover.

mode becomes an atomic.Value. It is written by the promotion goroutine and
read by Mode(), StartLeaseRenewal and WatchRemoteLease from theirs, so a
plain field is a data race.

-race found a real bug here during development, and it is worth recording
because the shape recurs. Promotion was documented as once-only, but the
latch is released on success, so nothing actually stopped a second run —
and a second run is not merely redundant, it races the renewal loop the
first one started, which owns lastRenew and is actively writing the Lease.
An already-Active hub is already in the state promotion produces, so it now
short-circuits, with a re-check under the latch for two callers that both
passed the first check. TestPromote_IsOnceOnly asserts no hook runs twice.

Part of kubeslice#297

Signed-off-by: Sumanth D <sumanthd032@gmail.com>
Records a failover as a Kubernetes Event, so an operator finds out from
`kubectl get events` and not only from a log line.

The Lease is the involved object. It is the leadership record, there is
exactly one, and it lives in the controller's own namespace — and the
recorder derives an Event's namespace from the object it is attached to, so
passing the Lease is what puts the Event beside the controller that emitted
it.

⚠️ Deliberately not kubeslice-system. Issue kubeslice#297 asks for the Event there,
but that namespace does not exist on a hub: per ADR kubeslice#293 Decision 1 it is a
worker-cluster namespace, and the hub's is kubeslice-controller /
$KUBESLICE_CONTROLLER_MANAGER_NAMESPACE. What makes this worth a comment
rather than a silent correction is that a constant with exactly the wrong
meaning sits in the vendor tree — kubeslice-monitoring's
logger.ControlPlaneNamespace — waiting to be reached for by anyone
implementing the issue literally.

⚠️ recorder.RecordEvent is called directly, never util.RecordEvent. That
helper begins with util.CtxLogger(ctx), which nil-panics on any context
that has not been through PrepareKubeSliceControllersRequestContext — and
promotion runs on main.go's signal-handler context, which has not. The same
mistake crashed a live Standby during kubeslice#295, so it is called out here rather
than left to be rediscovered.

The event name is registered in config/events/controller.yaml and generated
into both events_generated.go and config/events/events_config_map.yaml.
RecordEvent hard-fails on an unregistered name, so a hand-written entry
without a `make generate-events` run would fail during a real failover;
a test asserts the generated entry exists and that the failure mode is loud
rather than silent.

Part of kubeslice#297

Signed-off-by: Sumanth D <sumanthd032@gmail.com>
Adds --ha-promotion-dial-timeout and --ha-promotion-grace-period, and
installs the promotion hooks on a Standby.

The mirror gets its own cancellable context, derived from the manager's,
so promotion can stop it without tearing down everything else that hangs
off the signal-handler context. StopMirror cancels it and then waits for
Start to return, which is the part that matters: RemoteSyncer.Start already
drains its workqueue and prune goroutine before returning, so returning
from it is a sufficient and already-correct barrier and there is nothing to
reimplement. Waiting is what stops a hub opening its write fence while the
mirror is still writing — and in the most common failover trigger, the
Active's pod dying while its API server stays up, the mirror is very much
still alive at that moment.

The publisher moves above the mode switch so it exists before the hooks
that reference it. Its PublishOnce becomes promotion step 7, which is what
keeps failover latency off the publisher's own tick.

The hooks are installed only in standby mode. An Active has nothing to
promote from, and standalone must stay untouched: it is always the leader,
so the whole HA path has to remain inert there.

The event recorder is passed by value, not address — EventRecorder is an
interface, and taking its address yields a pointer-to-interface that does
not satisfy it.

Part of kubeslice#297

Signed-off-by: Sumanth D <sumanthd032@gmail.com>
Found by live-testing an Active whose API server was shut down — the case
this feature exists for and the one no test had ever exercised.

The guards' reads were bounded; the periodic poll was not. The watch loop
calls checkRemoteLeaseOnce synchronously, so a read that blocks blocks the
loop, and while it is blocked no staleness is evaluated at all. main.go
builds the remote client with a plain uncached client.New and no timeout,
so an API server that accepts the connection and then stops answering
leaves the read hanging until the transport gives up.

Measured: a single read blocked for about twelve seconds, with no poll and
no staleness evaluation in the whole window, before the connection finally
broke. That was a graceful container shutdown, which at least tears the
connection down eventually. A powered-off node or a partition that drops
packets has nothing to break it, and the wait becomes the OS TCP timeout —
minutes. The failover budget would be blown by waiting rather than by
deciding, which is the failure mode the bounded guards were added to
prevent in the first place; the poll simply got missed.

After the fix the same shutdown produces a regular two-second cadence with
one four-second gap where the read times out, and promotion lands at 21.1s
of staleness against a 20s budget.

A timed-out read is treated as exactly what it is — a failed read — so the
retained view is kept and continues to age. The verdict logic is unchanged.

Also adds tests for three promotion branches that were uncovered: the
concurrency latch under genuinely concurrent callers, failure to acquire
the Lease (the one step whose failure means the hub cannot lead, so it must
abort rather than open the fence), and failure to emit the event (a report,
not a step — a hub that took over but could not say so is still the
Active). promote() goes from 84% to 94% statement coverage.

Part of kubeslice#297

Signed-off-by: Sumanth D <sumanthd032@gmail.com>
An audit of the failure paths, rather than the happy one, turned up a hang
in the step most carefully reasoned about — and the reasoning is what
caused it.

Waiting indefinitely for the mirror to stop looked like the safe choice,
because proceeding without it is precisely the dual-writer state that step
exists to prevent. It is not safe. The watch loop calls promote
synchronously, so an unbounded wait on a mirror that never exits blocks the
loop: no further polls, no further staleness evaluation, no failover ever,
and nothing logged after "promotion sequence starting". Choosing "never
promote into a dual writer" that way silently buys "never promote at all",
which is strictly worse and completely invisible. Reproduced with a hook
that never returns; promote blocked forever.

Bounded by --ha-promotion-grace-period, expiry aborts the attempt loudly
and the next tick retries — so a merely slow mirror costs one tick, and a
genuinely stuck one is visible rather than mute. The hub is left exactly as
it was: fenced, still armed, free to try again.

The same bound now applies to the two steps that run after the fence opens.
A hang there cannot cost the failover, since leadership is already taken,
but it can cost the promotion ever finishing or reporting itself. The kick
is the one that will matter: it pushes into a channel per reconciled type,
and those are only drained once the manager is running, while main.go
starts this watch loop before mgr.Start. A kick arriving in that window has
nothing reading the other end.

Also adds an explicit precondition for a missing remote client. Without a
client to the Active there is no way to have observed it alive, so nothing
could justify concluding it is gone. It is unreachable from the watch loop,
which refuses to start without one and could not arm without one either,
but promote is a method and a caller reaching it another way crashed inside
the final dial instead of being told no.

Both flag descriptions were left inaccurate by these changes and by the
previous commit — the dial timeout now bounds every networked Lease read
including the periodic poll, and the grace period now bounds four
sequencing steps rather than one. Corrected.

Part of kubeslice#297

Signed-off-by: Sumanth D <sumanthd032@gmail.com>

Copilot AI left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pull request overview

This PR implements cross-cluster Active/Standby HA behavior for the KubeSlice controller, including detection of an unreachable Active hub and promotion of a Standby hub to Active, while adding supporting mirroring, metrics, events, and status publication needed for failover.

Changes:

  • Add Standby→Active promotion flow with pre-promotion guards, bounded waits/timeouts, and a promotion event/metric.
  • Add/extend Active→Standby mirroring (including prune anti-entropy, credential mirroring boundaries, and mirror failure events/metrics).
  • Wire HA mode, leader gating (write fence), remote syncer, and activeController status publication into main.go and reconcilers.

Reviewed changes

Copilot reviewed 45 out of 64 changed files in this pull request and generated 2 comments.

Show a summary per file
File Description
pkg/ha/promotion.go Implements Standby→Active promotion sequencing with hooks and bounded waits.
pkg/ha/promotion_guards.go Adds self-health + final bounded re-read guard logic to reduce false promotions.
pkg/ha/leader_elector.go Updates elector to retain last good remote Lease and trigger promotion on staleness.
pkg/ha/remote_syncer.go Adds/extends remote mirroring engine behavior, informer scoping, retries, events, and prune integration.
pkg/ha/prune.go Adds periodic diff-based pruning + reverse-diff re-enqueue to heal mirror drift.
pkg/ha/mirror*.go / pkg/ha/mirror_set.go Defines mirror sets, conflict guard, status mirroring, and credential mirroring rules.
pkg/ha/active_publisher.go Publishes status.activeController while leading; supports synchronous publish for promotion.
pkg/ha/metrics.go Adds HA sync/failover/abort metrics and registers them.
pkg/ha/*_test.go Adds extensive unit tests for promotion ordering, guards, mirroring, pruning, publisher, and metrics.
main.go Adds HA flags/config, wires elector/syncer/publisher, and gates reconcilers by leadership.
controllers/** Adds leader gating (write fence) at the start of reconcilers.
events/events_generated.go + config/events/* Registers HA-related events in generated map and config.
config/ha/* Adds Active-cluster RBAC manifest and guidance for Standby’s remote kubeconfig identity.
config/rbac/role.yaml Grants namespaces/status permissions needed for mirroring status updates.
Dockerfile Ensures pkg/ is included in the build context.
vendor/** + vendor/modules.txt Updates vendored dependencies to include new test/util packages used by HA tests.
Files not reviewed (1)
  • apis/controller/v1alpha1/zz_generated.deepcopy.go: Generated file

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Comment thread main.go
Comment on lines +572 to +576
leaderElector.SetPromotionHooks(ha.PromotionHooks{
// Cancel the mirror and WAIT for it to confirm it has stopped. The
// wait is the point: RemoteSyncer.Start already drains its workqueue
// and prune goroutine before returning, so returning from it is a
// sufficient and already-correct barrier. Without the wait, a hub can
Comment thread pkg/ha/promotion.go
Comment on lines +233 to +243
if e.hooks.KickReconcilers != nil {
kickCtx, cancel := context.WithTimeout(ctx, e.promotionGracePeriod)
err := e.hooks.KickReconcilers(kickCtx)
cancel()
if err != nil {
e.log.Errorw("could not re-enqueue objects after promotion; pre-existing state may "+
"stay unreconciled until the informer resync period", "error", err)
} else {
e.log.Infow("re-enqueued all reconciled types after promotion")
}
}
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants