Merge pull request #93837 from jayunit100/DialFromContainerB

intra-pod test improvements for bug-triage and node-level reporting

test/e2e/common/networking.go

@@ -18,6 +18,7 @@ package common
 
 import (
 	"github.com/onsi/ginkgo"
+	v1 "k8s.io/api/core/v1"
 	"k8s.io/apimachinery/pkg/util/sets"
 	"k8s.io/kubernetes/test/e2e/framework"
 	e2enetwork "k8s.io/kubernetes/test/e2e/framework/network"
@@ -28,6 +29,45 @@ var _ = ginkgo.Describe("[sig-network] Networking", func() {
 
 	ginkgo.Describe("Granular Checks: Pods", func() {
 
+		checkNodeConnectivity := func(config *e2enetwork.NetworkingTestConfig, protocol string, port int) {
+			// breadth first poll to quickly estimate failure.
+			failedPodsByHost := map[string][]*v1.Pod{}
+			// First time, we'll quickly try all pods, breadth first.
+			for _, endpointPod := range config.EndpointPods {
+				framework.Logf("Breadth first check of %v on host %v...", endpointPod.Status.PodIP, endpointPod.Status.HostIP)
+				if err := config.DialFromTestContainer(protocol, endpointPod.Status.PodIP, port, 1, 0, sets.NewString(endpointPod.Name)); err != nil {
+					if _, ok := failedPodsByHost[endpointPod.Status.HostIP]; !ok {
+						failedPodsByHost[endpointPod.Status.HostIP] = []*v1.Pod{}
+					}
+					failedPodsByHost[endpointPod.Status.HostIP] = append(failedPodsByHost[endpointPod.Status.HostIP], endpointPod)
+					framework.Logf("...failed...will try again in next pass")
+				}
+			}
+			errors := []error{}
+			// Second time, we pass through pods more carefully...
+			framework.Logf("Going to retry %v out of %v pods....", len(failedPodsByHost), len(config.EndpointPods))
+			for host, failedPods := range failedPodsByHost {
+				framework.Logf("Doublechecking %v pods in host %v which werent seen the first time.", len(failedPods), host)
+				for _, endpointPod := range failedPods {
+					framework.Logf("Now attempting to probe pod [[[ %v ]]]", endpointPod.Status.PodIP)
+					if err := config.DialFromTestContainer(protocol, endpointPod.Status.PodIP, port, config.MaxTries, 0, sets.NewString(endpointPod.Name)); err != nil {
+						errors = append(errors, err)
+					} else {
+						framework.Logf("Was able to reach %v on %v ", endpointPod.Status.PodIP, endpointPod.Status.HostIP)
+					}
+					framework.Logf("... Done probing pod [[[ %v ]]]", endpointPod.Status.PodIP)
+				}
+				framework.Logf("succeeded at polling %v out of %v connections", len(config.EndpointPods)-len(errors), len(config.EndpointPods))
+			}
+			if len(errors) > 0 {
+				framework.Logf("pod polling failure summary:")
+				for _, e := range errors {
+					framework.Logf("Collected error: %v", e)
+				}
+				framework.Failf("failed,  %v out of %v connections failed", len(errors), len(config.EndpointPods))
+			}
+		}
+
 		// Try to hit all endpoints through a test container, retry 5 times,
 		// expect exactly one unique hostname. Each of these endpoints reports
 		// its own hostname.
@@ -39,9 +79,7 @@ var _ = ginkgo.Describe("[sig-network] Networking", func() {
 		*/
 		framework.ConformanceIt("should function for intra-pod communication: http [NodeConformance]", func() {
 			config := e2enetwork.NewCoreNetworkingTestConfig(f, false)
-			for _, endpointPod := range config.EndpointPods {
+			checkNodeConnectivity(config, "http", e2enetwork.EndpointHTTPPort)
-				config.DialFromTestContainer("http", endpointPod.Status.PodIP, e2enetwork.EndpointHTTPPort, config.MaxTries, 0, sets.NewString(endpointPod.Name))
-			}
 		})
 
 		/*
@@ -52,9 +90,7 @@ var _ = ginkgo.Describe("[sig-network] Networking", func() {
 		*/
 		framework.ConformanceIt("should function for intra-pod communication: udp [NodeConformance]", func() {
 			config := e2enetwork.NewCoreNetworkingTestConfig(f, false)
-			for _, endpointPod := range config.EndpointPods {
+			checkNodeConnectivity(config, "udp", e2enetwork.EndpointUDPPort)
-				config.DialFromTestContainer("udp", endpointPod.Status.PodIP, e2enetwork.EndpointUDPPort, config.MaxTries, 0, sets.NewString(endpointPod.Name))
-			}
 		})
 
 		/*

test/e2e/framework/network/utils.go

@@ -167,17 +167,17 @@ type NetexecDialResponse struct {
 }
 
 // DialFromEndpointContainer executes a curl via kubectl exec in an endpoint container.
-func (config *NetworkingTestConfig) DialFromEndpointContainer(protocol, targetIP string, targetPort, maxTries, minTries int, expectedEps sets.String) {
+func (config *NetworkingTestConfig) DialFromEndpointContainer(protocol, targetIP string, targetPort, maxTries, minTries int, expectedEps sets.String) error {
-	config.DialFromContainer(protocol, echoHostname, config.EndpointPods[0].Status.PodIP, targetIP, EndpointHTTPPort, targetPort, maxTries, minTries, expectedEps)
+	return config.DialFromContainer(protocol, echoHostname, config.EndpointPods[0].Status.PodIP, targetIP, EndpointHTTPPort, targetPort, maxTries, minTries, expectedEps)
 }
 
 // DialFromTestContainer executes a curl via kubectl exec in a test container.
-func (config *NetworkingTestConfig) DialFromTestContainer(protocol, targetIP string, targetPort, maxTries, minTries int, expectedEps sets.String) {
+func (config *NetworkingTestConfig) DialFromTestContainer(protocol, targetIP string, targetPort, maxTries, minTries int, expectedEps sets.String) error {
-	config.DialFromContainer(protocol, echoHostname, config.TestContainerPod.Status.PodIP, targetIP, testContainerHTTPPort, targetPort, maxTries, minTries, expectedEps)
+	return config.DialFromContainer(protocol, echoHostname, config.TestContainerPod.Status.PodIP, targetIP, testContainerHTTPPort, targetPort, maxTries, minTries, expectedEps)
 }
 
 // DialEchoFromTestContainer executes a curl via kubectl exec in a test container. The response is expected to match the echoMessage.
-func (config *NetworkingTestConfig) DialEchoFromTestContainer(protocol, targetIP string, targetPort, maxTries, minTries int, echoMessage string) {
+func (config *NetworkingTestConfig) DialEchoFromTestContainer(protocol, targetIP string, targetPort, maxTries, minTries int, echoMessage string) error {
 	expectedResponse := sets.NewString()
 	expectedResponse.Insert(echoMessage)
 	var dialCommand string
@@ -191,7 +191,7 @@ func (config *NetworkingTestConfig) DialEchoFromTestContainer(protocol, targetIP
 	} else {
 		dialCommand = fmt.Sprintf("echo%%20%s", echoMessage)
 	}
-	config.DialFromContainer(protocol, dialCommand, config.TestContainerPod.Status.PodIP, targetIP, testContainerHTTPPort, targetPort, maxTries, minTries, expectedResponse)
+	return config.DialFromContainer(protocol, dialCommand, config.TestContainerPod.Status.PodIP, targetIP, testContainerHTTPPort, targetPort, maxTries, minTries, expectedResponse)
 }
 
 // diagnoseMissingEndpoints prints debug information about the endpoints that
@@ -248,7 +248,8 @@ func makeCURLDialCommand(ipPort, dialCmd, protocol, targetIP string, targetPort
 // maxTries == minTries will confirm that we see the expected endpoints and no
 // more for maxTries. Use this if you want to eg: fail a readiness check on a
 // pod and confirm it doesn't show up as an endpoint.
-func (config *NetworkingTestConfig) DialFromContainer(protocol, dialCommand, containerIP, targetIP string, containerHTTPPort, targetPort, maxTries, minTries int, expectedResponses sets.String) {
+// Returns nil if no error, or error message if failed after trying maxTries.
+func (config *NetworkingTestConfig) DialFromContainer(protocol, dialCommand, containerIP, targetIP string, containerHTTPPort, targetPort, maxTries, minTries int, expectedResponses sets.String) error {
 	ipPort := net.JoinHostPort(containerIP, strconv.Itoa(containerHTTPPort))
 	cmd := makeCURLDialCommand(ipPort, dialCommand, protocol, targetIP, targetPort)
 
@@ -273,16 +274,19 @@ func (config *NetworkingTestConfig) DialFromContainer(protocol, dialCommand, con
 
 		// Check against i+1 so we exit if minTries == maxTries.
 		if (responses.Equal(expectedResponses) || responses.Len() == 0 && expectedResponses.Len() == 0) && i+1 >= minTries {
-			return
+			framework.Logf("reached %v after %v/%v tries", targetIP, i, maxTries)
+			return nil
 		}
 		// TODO: get rid of this delay #36281
 		time.Sleep(hitEndpointRetryDelay)
 	}
-
 	if dialCommand == echoHostname {
 		config.diagnoseMissingEndpoints(responses)
 	}
-	framework.Failf("Failed to find expected responses:\nTries %d\nCommand %v\nretrieved %v\nexpected %v\n", maxTries, cmd, responses, expectedResponses)
+	returnMsg := fmt.Errorf("did not find expected responses... \nTries %d\nCommand %v\nretrieved %v\nexpected %v", maxTries, cmd, responses, expectedResponses)
+	framework.Logf("encountered error during dial (%v)", returnMsg)
+	return returnMsg
+
 }
 
 // GetEndpointsFromTestContainer executes a curl via kubectl exec in a test container.
@@ -676,6 +680,7 @@ func (config *NetworkingTestConfig) setupCore(selector map[string]string) {
 
 	epCount := len(config.EndpointPods)
 	config.MaxTries = epCount*epCount + testTries
+	framework.Logf("Setting MaxTries for pod polling to %v for networking test based on endpoint count %v", config.MaxTries, epCount)
 }
 
 // setup includes setupCore and also sets up services