fix: goroutine leak during watch leases (kube)

[iiiceoo]

Description

In the implementation of Kubernetes API, leaseWatchChan did not close correctly after ctx cancellation:

flannel/pkg/subnet/kube/kube.go

Lines 482 to 484 in 148b6d6

	case <-ctx.Done():
	return context.Canceled
	}

Will cause the following loop to not end:

flannel/pkg/subnet/subnet.go

Lines 148 to 165 in 148b6d6

	for watchResults := range leaseWatchChan {
	for _, wr := range watchResults {
	var batch []lease.Event
	if len(wr.Events) > 0 {
	batch = lw.Update(wr.Events)
	} else {
	batch = lw.Reset(wr.Snapshot)
	}
	for i := range batch {
	log.Infof("Batch elem [%d] is { %#v }", i, batch[i])
	}
	if len(batch) > 0 {
	receiver <- batch
	}
	}
	}

Todos

• Tests
• Documentation
• Release note

Release Note

None required

Signed-off-by: iiiceoo [email protected]

[iiiceoo]

When the CI was executing, an error unrelated to this PR was reported (workflow), the key logs are as follows:

I0901 14:38:03.171316    4220 subnet_test.go:451] found key: /coreos.com/network/subnets/10.3.16.0-24
    subnet_test.go:462: Failed to renew lease: bad expiration; expected 2023-09-02 14:38:05 +0000 UTC, got 2023-09-02 14:38:00 +0000 UTC
...
--- FAIL: TestRenewLease (11.04s)

This should be due to the fact that the acceptableMargin is too short and kvApi.Get(...) takes more time than acceptableMargin/2. I push another commit and fixed it by the way.