Add pkg/contextmanager

[ntnn]

Summary

For migrating logical clusters we need to break client watches.

While implementing this solution I remembered the inactive annotation on logical clusters, which suffers the same problem.

This is the feature cut out of the migration feature branch.

I think breaking the wildcard watches as well is a bit heavy handed but required, as a wildcard watch might target objects in a cluster being marked as inactive (or being migrated).

What Type of PR Is This?

/kind bug
/kind feature

Related Issue(s)

Fixes #

Release Notes

New logical cluster annotation `LogicalClusterInactiveAnnotationKey` (`core.kcp.io/inactive`) and a respecitve phase to mark logical clusters as inactive
Deprecated the internal annotation `internal.kcp.io/inactive`, replaced with `LogicalClusterInactiveAnnotationKey`
Active client watches will be cancelled on logical clusters that are marked as inactive
Active wildcard watches will be cancelled when a logical cluster is marked as inactive

[mjudeikis]

/lgtm
/approve

[kcp-ci-bot]

LGTM label has been added.

Details

Git tree hash: 01b795c8a893ae167a3b34e04c7fa586d541e9d3

[ntnn]

/retest

Flakes

[ntnn]

/test pull-kcp-test-e2e-multiple-runs
flake

[Copilot]

Pull request overview

This PR introduces a per-logical-cluster context management mechanism so the server can proactively cancel long-lived client requests (notably watches) when a logical cluster is marked inactive (and during certain lifecycle transitions like deleting). It adds a small generic context manager package, wires a new request filter into the main handler chain, and triggers cancellations from the LogicalCluster reconciler, with an accompanying e2e test update.

Changes:

• Add pkg/contextmanager to create/cancel contexts keyed by logical cluster (and wildcard).
• Add a new server filter to wrap each request with a per-cluster managed context, and cancel all managed contexts during server shutdown.
• Extend LogicalCluster phase reconciliation to set an Inactive phase and cancel per-cluster and wildcard contexts when the inactive annotation is applied; update e2e coverage for per-cluster watch termination.

Reviewed changes

Copilot reviewed 12 out of 12 changed files in this pull request and generated 5 comments.

Show a summary per file

File	Description
test/e2e/workspace/inactive_test.go	Opens a watch before inactivation and asserts it is terminated; verifies watches can be reopened after re-activation.
staging/src/github.com/kcp-dev/sdk/apis/core/v1alpha1/logicalcluster_types.go	Adds LogicalClusterPhaseInactive constant and documentation.
pkg/server/server.go	Adds a pre-shutdown hook to cancel all managed contexts.
pkg/server/filters/perclustercontext.go	New filter to derive request contexts from both request context and cluster-keyed contexts.
pkg/server/filters/inactivelogicalcluster.go	Comment update clarifying annotation-based early blocking.
pkg/server/controllers.go	Wires the shared ClusterContextManager into the logicalcluster controller.
pkg/server/config.go	Instantiates ClusterContextManager and inserts WithPerClusterContext into the handler chain.
pkg/reconciler/core/logicalcluster/logicalcluster_reconcile.go	Passes context manager into the phase reconciler.
pkg/reconciler/core/logicalcluster/logicalcluster_reconcile_phase.go	Cancels per-cluster + wildcard contexts when a cluster becomes inactive; adds phase transitions for Inactive.
pkg/reconciler/core/logicalcluster/logicalcluster_controller.go	Extends controller constructor/state to accept/store the context manager.
pkg/contextmanager/doc.go	Package documentation for the new context manager.
pkg/contextmanager/contextmanager.go	Implements the generic keyed context manager with cancellation support.

---

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

[Copilot]

Pull request overview

Copilot reviewed 19 out of 19 changed files in this pull request and generated 5 comments.

Comments suppressed due to low confidence (1)

pkg/contextmanager/rootCtx_test.go:177

• sync.WaitGroup does not have a Go method; this test won’t compile. Use wg.Add(1) before starting each goroutine and defer wg.Done() inside the goroutine instead of wg.Go(...).

	for range readers {
		wg.Go(func() {
			for {
				select {
				case <-stop:

[mjudeikis]

/lgtm
/approve

[kcp-ci-bot]

LGTM label has been added.

Details

Git tree hash: 2ec5764d2dbaaa70f83aa87b76dcb4fdfac9d4bc

[kcp-ci-bot]

[APPROVALNOTIFIER] This PR is APPROVED

This pull-request has been approved by: mjudeikis

The full list of commands accepted by this bot can be found here.

The pull request process is described here

Details

Needs approval from an approver in each of these files:

• OWNERS [mjudeikis]
• config/crds/OWNERS [mjudeikis]
• staging/src/github.com/kcp-dev/sdk/apis/OWNERS [mjudeikis]

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment

[ntnn]

/retest

Flake

[ntnn]

/retest

Flake