Reduce GC e2e test flakiness

Increase GC wait timeout in a flaky e2e test. The test expects a GC operation to be performed within 30s, while in practice the operation often takes longer due to a delay between the enqueueing of the owner's delete operation and the GC's actual processing of that event. Doubling the time seems to stabilize the test. The test's assumptions can be revisited, and the processing delay under load can be investigated in the future.

Reduce GC e2e test flakiness
cc6ae8a8 · Dan Mace · 3a0d8f8f · cc6ae8a8
Commit cc6ae8a8 authored Jul 26, 2017 by Dan Mace
Hide whitespace changes
Inline Side-by-side

Showing with 5 additions and 1 deletion

garbage_collector.go test/e2e/apimachinery/garbage_collector.go +5 -1

No files found.
--- a/test/e2e/apimachinery/garbage_collector.go
+++ b/test/e2e/apimachinery/garbage_collector.go
@@ -535,7 +535,11 @@ var _ = SIGDescribe("Garbage collector", func() {
 		}
 		By("wait for the rc to be deleted")
 		// default client QPS is 20, deleting each pod requires 2 requests, so 30s should be enough
-		if err := wait.Poll(1*time.Second, 30*time.Second, func() (bool, error) {
+		// TODO: 30s is enough assuming immediate processing of dependents following
+		// owner deletion, but in practice there can be a long delay between owner
+		// deletion and dependent deletion processing. For now, increase the timeout
+		// and investigate the processing delay.
+		if err := wait.Poll(1*time.Second, 60*time.Second, func() (bool, error) {
 			_, err := rcClient.Get(rc.Name, metav1.GetOptions{})
 			if err == nil {
 				pods, _ := podClient.List(metav1.ListOptions{})