Merge pull request #95276 from jayunit100/95216_ginkgo

Have udp tests return idiomatic ginkgo errors, update reporting of mutating parts

test/e2e/common/networking.go

@@ -103,7 +103,10 @@ var _ = ginkgo.Describe("[sig-network] Networking", func() {
 		framework.ConformanceIt("should function for node-pod communication: http [LinuxOnly] [NodeConformance]", func() {
 			config := e2enetwork.NewCoreNetworkingTestConfig(f, true)
 			for _, endpointPod := range config.EndpointPods {
-				config.DialFromNode("http", endpointPod.Status.PodIP, e2enetwork.EndpointHTTPPort, config.MaxTries, 0, sets.NewString(endpointPod.Name))
+				err := config.DialFromNode("http", endpointPod.Status.PodIP, e2enetwork.EndpointHTTPPort, config.MaxTries, 0, sets.NewString(endpointPod.Name))
+				if err != nil {
+					framework.Failf("Error dialing HTTP node to pod %v", err)
+				}
 			}
 		})
 
@@ -117,7 +120,10 @@ var _ = ginkgo.Describe("[sig-network] Networking", func() {
 		framework.ConformanceIt("should function for node-pod communication: udp [LinuxOnly] [NodeConformance]", func() {
 			config := e2enetwork.NewCoreNetworkingTestConfig(f, true)
 			for _, endpointPod := range config.EndpointPods {
-				config.DialFromNode("udp", endpointPod.Status.PodIP, e2enetwork.EndpointUDPPort, config.MaxTries, 0, sets.NewString(endpointPod.Name))
+				err := config.DialFromNode("udp", endpointPod.Status.PodIP, e2enetwork.EndpointUDPPort, config.MaxTries, 0, sets.NewString(endpointPod.Name))
+				if err != nil {
+					framework.Failf("Error dialing UDP from node to pod: %v", err)
+				}
 			}
 		})
 	})

test/e2e/framework/network/utils.go

@@ -380,17 +380,20 @@ func (config *NetworkingTestConfig) GetHTTPCodeFromTestContainer(path, targetIP
 	return code, nil
 }
 
-// DialFromNode executes a tcp or udp request based on protocol via kubectl exec
+// DialFromNode executes a tcp/udp curl/nc request based on protocol via kubectl exec
 // in a test container running with host networking.
-// - minTries is the minimum number of curl attempts required before declaring
+// - minTries is the minimum number of curl/nc attempts required before declaring
-//   success. Set to 0 if you'd like to return as soon as all endpoints respond
+//   success. If 0, then we return as soon as all endpoints succeed.
-//   at least once.
+// - There is no logical change to test results if faillures happen AFTER endpoints have succeeded,
-// - maxTries is the maximum number of curl attempts. If this many attempts pass
+//   hence over-padding minTries will NOT reverse a successful result and is thus not very useful yet
-//   and we don't see all expected endpoints, the test fails.
+//   (See the TODO about checking probability, which isnt implemented yet).
-// maxTries == minTries will confirm that we see the expected endpoints and no
+// - maxTries is the maximum number of curl/echo attempts before an error is returned.  The
-// more for maxTries. Use this if you want to eg: fail a readiness check on a
+//   smaller this number is, the less 'slack' there is for declaring success.
-// pod and confirm it doesn't show up as an endpoint.
+// - if maxTries < expectedEps, this test is guaranteed to return an error, because all endpoints wont be hit.
-func (config *NetworkingTestConfig) DialFromNode(protocol, targetIP string, targetPort, maxTries, minTries int, expectedEps sets.String) {
+// - maxTries == minTries will return as soon as all endpoints succeed (or fail once maxTries is reached without
+//   success on all endpoints).
+//   In general its prudent to have a high enough level of minTries to guarantee that all pods get a fair chance at receiving traffic.
+func (config *NetworkingTestConfig) DialFromNode(protocol, targetIP string, targetPort, maxTries, minTries int, expectedEps sets.String) error {
 	var cmd string
 	if protocol == "udp" {
 		cmd = fmt.Sprintf("echo hostName | nc -w 1 -u %s %d", targetIP, targetPort)
@@ -408,6 +411,7 @@ func (config *NetworkingTestConfig) DialFromNode(protocol, targetIP string, targ
 	eps := sets.NewString()
 
 	filterCmd := fmt.Sprintf("%s | grep -v '^\\s*$'", cmd)
+	framework.Logf("Going to poll %v on port %v at least %v times, with a maximum of %v tries before failing", targetIP, targetPort, minTries, maxTries)
 	for i := 0; i < maxTries; i++ {
 		stdout, stderr, err := config.f.ExecShellInPodWithFullOutput(config.HostTestContainerPod.Name, filterCmd)
 		if err != nil || len(stderr) > 0 {
@@ -424,8 +428,8 @@ func (config *NetworkingTestConfig) DialFromNode(protocol, targetIP string, targ
 
 		// Check against i+1 so we exit if minTries == maxTries.
 		if eps.Equal(expectedEps) && i+1 >= minTries {
-			framework.Logf("Found all expected endpoints: %+v", eps.List())
+			framework.Logf("Found all %d expected endpoints: %+v", eps.Len(), eps.List())
-			return
+			return nil
 		}
 
 		framework.Logf("Waiting for %+v endpoints (expected=%+v, actual=%+v)", expectedEps.Difference(eps).List(), expectedEps.List(), eps.List())
@@ -435,7 +439,7 @@ func (config *NetworkingTestConfig) DialFromNode(protocol, targetIP string, targ
 	}
 
 	config.diagnoseMissingEndpoints(eps)
-	framework.Failf("Failed to find expected endpoints:\nTries %d\nCommand %v\nretrieved %v\nexpected %v\n", maxTries, cmd, eps, expectedEps)
+	return fmt.Errorf("failed to find expected endpoints, \ntries %d\nCommand %v\nretrieved %v\nexpected %v", maxTries, cmd, eps, expectedEps)
 }
 
 // GetSelfURL executes a curl against the given path via kubectl exec into a

test/e2e/network/networking.go

@@ -210,18 +210,29 @@ var _ = SIGDescribe("Networking", func() {
 		ginkgo.It("should function for node-Service: http", func() {
 			config := e2enetwork.NewNetworkingTestConfig(f, true, false)
 			ginkgo.By(fmt.Sprintf("dialing(http) %v (node) --> %v:%v (config.clusterIP)", config.NodeIP, config.ClusterIP, e2enetwork.ClusterHTTPPort))
-			config.DialFromNode("http", config.ClusterIP, e2enetwork.ClusterHTTPPort, config.MaxTries, 0, config.EndpointHostnames())
+			err := config.DialFromNode("http", config.ClusterIP, e2enetwork.ClusterHTTPPort, config.MaxTries, 0, config.EndpointHostnames())
+			if err != nil {
+				framework.Failf("failed dialing endpoint, %v", err)
+			}
 			ginkgo.By(fmt.Sprintf("dialing(http) %v (node) --> %v:%v (nodeIP)", config.NodeIP, config.NodeIP, config.NodeHTTPPort))
-			config.DialFromNode("http", config.NodeIP, config.NodeHTTPPort, config.MaxTries, 0, config.EndpointHostnames())
+			err = config.DialFromNode("http", config.NodeIP, config.NodeHTTPPort, config.MaxTries, 0, config.EndpointHostnames())
+			if err != nil {
+				framework.Failf("failed dialing endpoint, %v", err)
+			}
 		})
 
 		ginkgo.It("should function for node-Service: udp", func() {
 			config := e2enetwork.NewNetworkingTestConfig(f, true, false)
 			ginkgo.By(fmt.Sprintf("dialing(udp) %v (node) --> %v:%v (config.clusterIP)", config.NodeIP, config.ClusterIP, e2enetwork.ClusterUDPPort))
-			config.DialFromNode("udp", config.ClusterIP, e2enetwork.ClusterUDPPort, config.MaxTries, 0, config.EndpointHostnames())
+			err := config.DialFromNode("udp", config.ClusterIP, e2enetwork.ClusterUDPPort, config.MaxTries, 0, config.EndpointHostnames())
-
+			if err != nil {
+				framework.Failf("failed dialing endpoint, %v", err)
+			}
 			ginkgo.By(fmt.Sprintf("dialing(udp) %v (node) --> %v:%v (nodeIP)", config.NodeIP, config.NodeIP, config.NodeUDPPort))
-			config.DialFromNode("udp", config.NodeIP, config.NodeUDPPort, config.MaxTries, 0, config.EndpointHostnames())
+			err = config.DialFromNode("udp", config.NodeIP, config.NodeUDPPort, config.MaxTries, 0, config.EndpointHostnames())
+			if err != nil {
+				framework.Failf("failed dialing endpoint, %v", err)
+			}
 		})
 
 		ginkgo.It("should function for endpoint-Service: http", func() {
@@ -305,14 +316,15 @@ var _ = SIGDescribe("Networking", func() {
 			ginkgo.By(fmt.Sprintf("dialing(http) %v --> %v:%v (config.clusterIP)", config.TestContainerPod.Name, config.ClusterIP, e2enetwork.ClusterHTTPPort))
 			err := config.DialFromTestContainer("http", config.ClusterIP, e2enetwork.ClusterHTTPPort, config.MaxTries, 0, config.EndpointHostnames())
 			if err != nil {
-				framework.Failf("failed dialing endpoint, %v", err)
+				framework.Failf("failed dialing endpoint (initial), %v", err)
 			}
+			ginkgo.By("Deleting a pod which, will be replaced with a new endpoint")
 			config.DeleteNetProxyPod()
 
-			ginkgo.By(fmt.Sprintf("dialing(http) %v --> %v:%v (config.clusterIP)", config.TestContainerPod.Name, config.ClusterIP, e2enetwork.ClusterHTTPPort))
+			ginkgo.By(fmt.Sprintf("dialing(http) %v --> %v:%v (config.clusterIP) (endpoint recovery)", config.TestContainerPod.Name, config.ClusterIP, e2enetwork.ClusterHTTPPort))
 			err = config.DialFromTestContainer("http", config.ClusterIP, e2enetwork.ClusterHTTPPort, config.MaxTries, config.MaxTries, config.EndpointHostnames())
 			if err != nil {
-				framework.Failf("failed dialing endpoint, %v", err)
+				framework.Failf("failed dialing endpoint (recovery), %v", err)
 			}
 		})
 
@@ -321,39 +333,53 @@ var _ = SIGDescribe("Networking", func() {
 			ginkgo.By(fmt.Sprintf("dialing(udp) %v --> %v:%v (config.clusterIP)", config.TestContainerPod.Name, config.ClusterIP, e2enetwork.ClusterUDPPort))
 			err := config.DialFromTestContainer("udp", config.ClusterIP, e2enetwork.ClusterUDPPort, config.MaxTries, 0, config.EndpointHostnames())
 			if err != nil {
-				framework.Failf("failed dialing endpoint, %v", err)
+				framework.Failf("failed dialing endpoint (initial), %v", err)
 			}
+			ginkgo.By("Deleting a pod which, will be replaced with a new endpoint")
 			config.DeleteNetProxyPod()
 
-			ginkgo.By(fmt.Sprintf("dialing(udp) %v --> %v:%v (config.clusterIP)", config.TestContainerPod.Name, config.ClusterIP, e2enetwork.ClusterUDPPort))
+			ginkgo.By(fmt.Sprintf("dialing(udp) %v --> %v:%v (config.clusterIP) (endpoint recovery)", config.TestContainerPod.Name, config.ClusterIP, e2enetwork.ClusterUDPPort))
 			err = config.DialFromTestContainer("udp", config.ClusterIP, e2enetwork.ClusterUDPPort, config.MaxTries, config.MaxTries, config.EndpointHostnames())
 			if err != nil {
-				framework.Failf("failed dialing endpoint, %v", err)
+				framework.Failf("failed dialing endpoint (recovery), %v", err)
 			}
 		})
 
 		// Slow because we confirm that the nodePort doesn't serve traffic, which requires a period of polling.
 		ginkgo.It("should update nodePort: http [Slow]", func() {
 			config := e2enetwork.NewNetworkingTestConfig(f, true, false)
-			ginkgo.By(fmt.Sprintf("dialing(http) %v (node) --> %v:%v (nodeIP)", config.NodeIP, config.NodeIP, config.NodeHTTPPort))
+			ginkgo.By(fmt.Sprintf("dialing(http) %v (node) --> %v:%v (nodeIP) and getting ALL host endpoints", config.NodeIP, config.NodeIP, config.NodeHTTPPort))
-			config.DialFromNode("http", config.NodeIP, config.NodeHTTPPort, config.MaxTries, 0, config.EndpointHostnames())
+			err := config.DialFromNode("http", config.NodeIP, config.NodeHTTPPort, config.MaxTries, 0, config.EndpointHostnames())
-
+			if err != nil {
+				framework.Failf("Error dialing http from node: %v", err)
+			}
+			ginkgo.By("Deleting the node port access point")
 			config.DeleteNodePortService()
 
-			ginkgo.By(fmt.Sprintf("dialing(http) %v (node) --> %v:%v (nodeIP)", config.NodeIP, config.NodeIP, config.NodeHTTPPort))
+			ginkgo.By(fmt.Sprintf("dialing(http) %v (node) --> %v:%v (nodeIP) and getting ZERO host endpoints", config.NodeIP, config.NodeIP, config.NodeHTTPPort))
-			config.DialFromNode("http", config.NodeIP, config.NodeHTTPPort, config.MaxTries, config.MaxTries, sets.NewString())
+			err = config.DialFromNode("http", config.NodeIP, config.NodeHTTPPort, config.MaxTries, config.MaxTries, sets.NewString())
+			if err != nil {
+				framework.Failf("Error dialing http from node: %v", err)
+			}
 		})
 
 		// Slow because we confirm that the nodePort doesn't serve traffic, which requires a period of polling.
 		ginkgo.It("should update nodePort: udp [Slow]", func() {
 			config := e2enetwork.NewNetworkingTestConfig(f, true, false)
-			ginkgo.By(fmt.Sprintf("dialing(udp) %v (node) --> %v:%v (nodeIP)", config.NodeIP, config.NodeIP, config.NodeUDPPort))
+			ginkgo.By(fmt.Sprintf("dialing(udp) %v (node) --> %v:%v (nodeIP) and getting ALL host endpoints", config.NodeIP, config.NodeIP, config.NodeUDPPort))
-			config.DialFromNode("udp", config.NodeIP, config.NodeUDPPort, config.MaxTries, 0, config.EndpointHostnames())
+			err := config.DialFromNode("udp", config.NodeIP, config.NodeUDPPort, config.MaxTries, 0, config.EndpointHostnames())
+			if err != nil {
+				framework.Failf("Failure validating that nodePort service WAS forwarding properly: %v", err)
+			}
 
+			ginkgo.By("Deleting the node port access point")
 			config.DeleteNodePortService()
 
-			ginkgo.By(fmt.Sprintf("dialing(udp) %v (node) --> %v:%v (nodeIP)", config.NodeIP, config.NodeIP, config.NodeUDPPort))
+			ginkgo.By(fmt.Sprintf("dialing(udp) %v (node) --> %v:%v (nodeIP) and getting ZERO host endpoints", config.NodeIP, config.NodeIP, config.NodeUDPPort))
-			config.DialFromNode("udp", config.NodeIP, config.NodeUDPPort, config.MaxTries, config.MaxTries, sets.NewString())
+			err = config.DialFromNode("udp", config.NodeIP, config.NodeUDPPort, config.MaxTries, config.MaxTries, sets.NewString())
+			if err != nil {
+				framework.Failf("Failure validating that node port service STOPPED removed properly: %v", err)
+			}
 		})
 
 		// [LinuxOnly]: Windows does not support session affinity.