Merge pull request #38620 from wojtek-t/increase_wait_for_nodes_timeout

Automatic merge from submit-queue (batch tested with PRs 38617, 38620)

Increase timeout for waiting for nodes

hack/verify-flags/known-flags.txt

@@ -433,6 +433,7 @@ node-name
 node-os-distro
 node-path-override
 node-port
+node-schedulable-timeout
 node-startup-grace-period
 node-status-update-frequency
 node-sync-period

test/e2e/e2e.go

@@ -121,7 +121,7 @@ var _ = ginkgo.SynchronizedBeforeSuite(func() []byte {
 	// In large clusters we may get to this point but still have a bunch
 	// of nodes without Routes created. Since this would make a node
 	// unschedulable, we need to wait until all of them are schedulable.
-	framework.ExpectNoError(framework.WaitForAllNodesSchedulable(c, framework.NodeSchedulableTimeout))
+	framework.ExpectNoError(framework.WaitForAllNodesSchedulable(c, framework.TestContext.NodeSchedulableTimeout))
 
 	// Ensure all pods are running and ready before starting tests (otherwise,
 	// cluster infrastructure pods that are being pulled or started can block

test/e2e/empty.go

@@ -32,7 +32,7 @@ var _ = framework.KubeDescribe("Empty [Feature:Empty]", func() {
 		ns := f.Namespace.Name
 
 		// TODO: respect --allow-notready-nodes flag in those functions.
-		framework.ExpectNoError(framework.WaitForAllNodesSchedulable(c, framework.NodeSchedulableTimeout))
+		framework.ExpectNoError(framework.WaitForAllNodesSchedulable(c, framework.TestContext.NodeSchedulableTimeout))
 		framework.WaitForAllNodesHealthy(c, time.Minute)
 
 		err := framework.CheckTestingNSDeletedExcept(c, ns)

test/e2e/framework/test_context.go

@@ -67,6 +67,8 @@ type TestContextType struct {
 	GatherMetricsAfterTest            bool
 	// Currently supported values are 'hr' for human-readable and 'json'. It's a comma separated list.
 	OutputPrintType string
+	// NodeSchedulableTimeout is the timeout for waiting for all nodes to be schedulable.
+	NodeSchedulableTimeout time.Duration
 	// CreateTestingNS is responsible for creating namespace used for executing e2e tests.
 	// It accepts namespace base name, which will be prepended with e2e prefix, kube client
 	// and labels to be applied to a namespace.
@@ -187,6 +189,7 @@ func RegisterClusterFlags() {
 	flag.StringVar(&cloudConfig.ClusterTag, "cluster-tag", "", "Tag used to identify resources.  Only required if provider is aws.")
 	flag.IntVar(&TestContext.MinStartupPods, "minStartupPods", 0, "The number of pods which we need to see in 'Running' state with a 'Ready' condition of true, before we try running tests. This is useful in any cluster which needs some base pod-based services running before it can be used.")
 	flag.DurationVar(&TestContext.SystemPodsStartupTimeout, "system-pods-startup-timeout", 10*time.Minute, "Timeout for waiting for all system pods to be running before starting tests.")
+	flag.DurationVar(&TestContext.NodeSchedulableTimeout, "node-schedulable-timeout", 4*time.Hour, "Timeout for waiting for all nodes to be schedulable.")
 	flag.StringVar(&TestContext.UpgradeTarget, "upgrade-target", "ci/latest", "Version to upgrade to (e.g. 'release/stable', 'release/latest', 'ci/latest', '0.19.1', '0.19.1-669-gabac8c8') if doing an upgrade test.")
 	flag.StringVar(&TestContext.UpgradeImage, "upgrade-image", "", "Image to upgrade to (e.g. 'container_vm' or 'gci') if doing an upgrade test.")
 	flag.StringVar(&TestContext.PrometheusPushGateway, "prom-push-gateway", "", "The URL to prometheus gateway, so that metrics can be pushed during e2es and scraped by prometheus. Typically something like 127.0.0.1:9091.")

test/e2e/framework/util.go

@@ -176,10 +176,6 @@ const (
 
 	// TODO(justinsb): Avoid hardcoding this.
 	awsMasterIP = "172.20.0.9"
-
-	// Default time to wait for nodes to become schedulable.
-	// Set so high for scale tests.
-	NodeSchedulableTimeout = 4 * time.Hour
 )
 
 var (