Use "app" selector consistently in integration tests.

Also improves reliability of CapturePodLogs.

Signed-off-by: Thomas Hallgren <[email protected]>

integration_test/colliding_mounts_test.go

@@ -155,7 +155,7 @@ func (s *mountsSuite) Test_CollidingMounts() {
 			defer itest.TelepresenceOk(ctx, "leave", "hello")
 			require.Contains(stdout, "Using Deployment hello")
 			if i == 0 {
-				s.CapturePodLogs(ctx, "app=hello", "traffic-agent", s.AppNamespace())
+				s.CapturePodLogs(ctx, "hello", "traffic-agent", s.AppNamespace())
 			} else {
 				// Mounts are sometimes slow
 				dtime.SleepWithContext(ctx, 3*time.Second)

integration_test/env_interpolate_test.go

@@ -20,7 +20,7 @@ func (s *connectedSuite) Test_PrefixInterpolated() {
 
 	itest.TelepresenceOk(ctx, "intercept", "--mount", "false", svc)
 	defer itest.TelepresenceOk(ctx, "leave", svc)
-	out, err := s.KubectlOut(ctx, "get", "pod", "-o", "json", "-l", "service="+svc)
+	out, err := s.KubectlOut(ctx, "get", "pod", "-o", "json", "-l", "app="+svc)
 	rq.NoError(err)
 
 	var pods core.PodList

integration_test/gather_logs_test.go

@@ -137,7 +137,7 @@ func (s *connectedSuite) TestGatherLogs_OnlyMappedLogs() {
 		10*time.Second,
 		2*time.Second,
 	)
-	s.CapturePodLogs(ctx, fmt.Sprintf("app=%s", svc), "traffic-agent", otherOne)
+	s.CapturePodLogs(ctx, svc, "traffic-agent", otherOne)
 	itest.TelepresenceDisconnectOk(ctx)
 
 	itest.TelepresenceOk(ctx, "connect", "--namespace", otherTwo, "--manager-namespace", s.ManagerNamespace())
@@ -150,7 +150,7 @@ func (s *connectedSuite) TestGatherLogs_OnlyMappedLogs() {
 		10*time.Second,
 		2*time.Second,
 	)
-	s.CapturePodLogs(ctx, fmt.Sprintf("app=%s", svc), "traffic-agent", otherTwo)
+	s.CapturePodLogs(ctx, svc, "traffic-agent", otherTwo)
 	itest.TelepresenceOk(ctx, "leave", svc)
 
 	bothNsRx := fmt.Sprintf("(?:%s|%s)", otherOne, otherTwo)

integration_test/headless_test.go

@@ -26,7 +26,7 @@ func (s *connectedSuite) Test_SuccessfullyInterceptsHeadlessService() {
 	require := s.Require()
 	stdout := itest.TelepresenceOk(ctx, "intercept", "--mount", "false", svc, "--port", strconv.Itoa(svcPort))
 	require.Contains(stdout, "Using StatefulSet echo-headless")
-	s.CapturePodLogs(ctx, "service=echo-headless", "traffic-agent", s.AppNamespace())
+	s.CapturePodLogs(ctx, "echo-headless", "traffic-agent", s.AppNamespace())
 
 	defer itest.TelepresenceOk(ctx, "leave", "echo-headless")
 

integration_test/intercept_flags_test.go

@@ -1,6 +1,7 @@
 package integration_test
 
 import (
+	"context"
 	"encoding/json"
 	"fmt"
 	"os"
@@ -60,48 +61,58 @@ func (s *interceptFlagSuite) Test_ContainerReplace() {
 	ctx := s.Context()
 	require := s.Require()
 
-	n1 := "container_replaced"
+	const (
+		n1 = "container_replaced"
+		c1 = "hello-container-1"
+		n2 = "container_kept"
+		c2 = "hello-container-2"
+	)
+
 	localPort1, cancel1 := itest.StartLocalHttpEchoServer(ctx, n1)
 	defer cancel1()
 
-	n2 := "container_kept"
 	localPort2, cancel2 := itest.StartLocalHttpEchoServer(ctx, n2)
 	defer cancel2()
 
 	tests := []struct {
-		name      string
-		iceptName string
-		replace   bool
-		localPort int
-		port      uint16
+		name         string
+		iceptName    string
+		appContainer string
+		replace      bool
+		localPort    int
+		port         uint16
 	}{
 		{
-			name:      n1,
-			iceptName: n1,
-			replace:   true,
-			localPort: localPort1,
-			port:      80,
+			name:         n1,
+			iceptName:    n1,
+			appContainer: c1,
+			replace:      true,
+			localPort:    localPort1,
+			port:         80,
 		},
 		{
-			name:      n2,
-			iceptName: n2,
-			replace:   false,
-			localPort: localPort2,
-			port:      81,
+			name:         n2,
+			iceptName:    n2,
+			appContainer: c2,
+			replace:      false,
+			localPort:    localPort2,
+			port:         81,
 		},
 		{
-			name:      "container_replace_kept",
-			iceptName: n1,
-			replace:   false,
-			localPort: localPort1,
-			port:      80,
+			name:         "container_replace_kept",
+			iceptName:    n1,
+			appContainer: c1,
+			replace:      false,
+			localPort:    localPort1,
+			port:         80,
 		},
 		{
-			name:      "container_kept_replaced",
-			iceptName: n2,
-			replace:   true,
-			localPort: localPort2,
-			port:      81,
+			name:         "container_kept_replaced",
+			iceptName:    n2,
+			appContainer: c2,
+			replace:      true,
+			localPort:    localPort2,
+			port:         81,
 		},
 	}
 
@@ -116,8 +127,13 @@ func (s *interceptFlagSuite) Test_ContainerReplace() {
 			}
 			args = append(args, "--port", fmt.Sprintf("%d:%d", tt.localPort, tt.port), "--output", "json", "--detailed-output", "--workload", s.serviceName, tt.iceptName)
 			jsOut := itest.TelepresenceOk(ctx, args...)
+			agentCaptureCtx, agentCaptureCancel := context.WithCancel(ctx)
+			s.CapturePodLogs(agentCaptureCtx, s.serviceName, "traffic-agent", s.AppNamespace())
+
 			defer func() {
+				agentCaptureCancel()
 				itest.TelepresenceOk(ctx, "leave", tt.iceptName)
+				s.CapturePodLogs(ctx, s.serviceName, tt.appContainer, s.AppNamespace())
 				s.Eventually(func() bool {
 					out, err := itest.Output(ctx, "curl", "--silent", "--max-time", "1", iputil.JoinHostPort(s.serviceName, tt.port))
 					if err != nil {

integration_test/intercept_localhost_test.go

@@ -40,7 +40,7 @@ func (s *interceptLocalhostSuite) SetupSuite() {
 	s.Require().NoError(err)
 	dlog.Infof(ctx, "ip: %s: route: %s", s.defaultRoute.LocalIP, s.defaultRoute)
 	s.port, s.cancelLocal = itest.StartLocalHttpEchoServerWithHost(ctx, s.ServiceName(), s.defaultRoute.LocalIP.String())
-	s.CapturePodLogs(ctx, "app=echo", "traffic-agent", s.AppNamespace())
+	s.CapturePodLogs(ctx, "echo", "traffic-agent", s.AppNamespace())
 }
 
 func (s *interceptLocalhostSuite) TearDownSuite() {

integration_test/intercept_mount_test.go

@@ -57,7 +57,7 @@ func (s *interceptMountSuite) SetupSuite() {
 	port, s.cancelLocal = itest.StartLocalHttpEchoServer(ctx, s.ServiceName())
 	stdout := itest.TelepresenceOk(ctx, "intercept", s.ServiceName(), "--mount", s.mountPoint, "--port", strconv.Itoa(port))
 	s.Contains(stdout, "Using Deployment "+s.ServiceName())
-	s.CapturePodLogs(ctx, "app=echo", "traffic-agent", s.AppNamespace())
+	s.CapturePodLogs(ctx, "echo", "traffic-agent", s.AppNamespace())
 }
 
 func (s *interceptMountSuite) TearDownSuite() {
@@ -191,7 +191,7 @@ func (s *singleServiceSuite) Test_NoInterceptorResponse() {
 	}, 10*time.Second, time.Second)
 
 	time.Sleep(2000 * time.Millisecond) // List is really fast now, so give the mount some time to become effective
-	s.CapturePodLogs(ctx, "app="+s.ServiceName(), "traffic-agent", s.AppNamespace())
+	s.CapturePodLogs(ctx, s.ServiceName(), "traffic-agent", s.AppNamespace())
 
 	mountPoint := filepath.Join(nwd, "rel-dir")
 	st, err := os.Stat(mountPoint)

integration_test/itest/cluster.go

@@ -47,6 +47,7 @@ import (
 	"github.com/telepresenceio/telepresence/v2/pkg/maps"
 	"github.com/telepresenceio/telepresence/v2/pkg/proc"
 	"github.com/telepresenceio/telepresence/v2/pkg/shellquote"
+	"github.com/telepresenceio/telepresence/v2/pkg/slice"
 	"github.com/telepresenceio/telepresence/v2/pkg/version"
 )
 
@@ -460,61 +461,97 @@ func (s *cluster) RootdPProf() uint16 {
 }
 
 func (s *cluster) CapturePodLogs(ctx context.Context, app, container, ns string) {
-	var pods string
+	var pods []string
 	for i := 0; ; i++ {
-		var err error
-		pods, err = KubectlOut(ctx, ns, "get", "pods", "--field-selector", "status.phase=Running", "-l", app, "-o", "jsonpath={.items[*].metadata.name}")
-		if err != nil {
-			dlog.Errorf(ctx, "failed to get %s pod in namespace %s: %v", app, ns, err)
-			return
+		runningPods := RunningPods(ctx, app, ns)
+		if len(runningPods) > 0 {
+			if container == "" {
+				pods = runningPods
+			} else {
+				for _, pod := range runningPods {
+					cns, err := KubectlOut(ctx, ns, "get", "pods", pod, "-o", "jsonpath={.spec.containers[*].name}")
+					if err == nil && slice.Contains(strings.Split(cns, " "), container) {
+						pods = append(pods, pod)
+					}
+				}
+			}
 		}
-		pods = strings.TrimSpace(pods)
-		if pods != "" || i == 5 {
+		if len(pods) > 0 || i == 5 {
 			break
 		}
 		dtime.SleepWithContext(ctx, 2*time.Second)
 	}
-	if pods == "" {
-		dlog.Errorf(ctx, "found no %s pods in namespace %s", app, ns)
+
+	if len(pods) == 0 {
+		if container == "" {
+			dlog.Errorf(ctx, "found no %s pods in namespace %s", app, ns)
+		} else {
+			dlog.Errorf(ctx, "found no %s pods in namespace %s with a %s container", app, ns, container)
+		}
 		return
 	}
-
-	// Let command die when the pod that it logs die
-	ctx = context.WithoutCancel(ctx)
-
 	present := struct{}{}
 
 	// Use another logger to avoid errors due to logs arriving after the tests complete.
 	ctx = dlog.WithLogger(ctx, dlog.WrapLogrus(logrus.StandardLogger()))
-	dlog.Infof(ctx, "Capturing logs for pods %q", pods)
-	for _, pod := range strings.Split(pods, " ") {
-		if _, ok := s.logCapturingPods.LoadOrStore(pod, present); ok {
-			continue
-		}
-		logFile, err := os.Create(
-			filepath.Join(filelocation.AppUserLogDir(ctx), fmt.Sprintf("%s-%s.log", dtime.Now().Format("20060102T150405"), pod)))
-		if err != nil {
-			s.logCapturingPods.Delete(pod)
-			dlog.Errorf(ctx, "unable to create pod logfile %s: %v", logFile.Name(), err)
-			return
-		}
+	pod := pods[0]
+	key := pod
+	if container != "" {
+		key += "/" + container
+	}
+	if _, ok := s.logCapturingPods.LoadOrStore(key, present); ok {
+		return
+	}
 
-		args := []string{"--namespace", ns, "logs", "-f", pod}
-		if container != "" {
-			args = append(args, "-c", container)
+	logFile, err := os.Create(
+		filepath.Join(filelocation.AppUserLogDir(ctx), fmt.Sprintf("%s-%s.log", dtime.Now().Format("20060102T150405"), pod)))
+	if err != nil {
+		s.logCapturingPods.Delete(pod)
+		dlog.Errorf(ctx, "unable to create pod logfile %s: %v", logFile.Name(), err)
+		return
+	}
+
+	args := []string{"--namespace", ns, "logs", "-f", pod}
+	if container != "" {
+		args = append(args, "-c", container)
+	}
+	// Let command die when the pod that it logs die
+	cmd := Command(context.WithoutCancel(ctx), "kubectl", args...)
+	cmd.Stdout = logFile
+	cmd.Stderr = logFile
+	ready := make(chan struct{})
+	go func() {
+		defer func() {
+			_ = logFile.Close()
+			s.logCapturingPods.Delete(pod)
+		}()
+		err := cmd.Start()
+		if err == nil {
+			if container == "" {
+				dlog.Infof(ctx, "Capturing logs for pod %s", pod)
+			} else {
+				dlog.Infof(ctx, "Capturing logs for pod %s, container %s", pod, container)
+			}
+			close(ready)
+			err = cmd.Wait()
 		}
-		cmd := Command(ctx, "kubectl", args...)
-		cmd.Stdout = logFile
-		cmd.Stderr = logFile
-		go func(pod string) {
-			defer func() {
-				_ = logFile.Close()
-				s.logCapturingPods.Delete(pod)
-			}()
-			if err := cmd.Run(); err != nil {
-				dlog.Errorf(ctx, "log capture failed: %v", err)
+		if err != nil {
+			if container == "" {
+				dlog.Errorf(ctx, "log capture for pod %s failed: %v", pod, err)
+			} else {
+				dlog.Errorf(ctx, "log capture for pod %s, container %s failed: %v", pod, container, err)
 			}
-		}(pod)
+			select {
+			case <-ready:
+			default:
+				close(ready)
+			}
+		}
+	}()
+	select {
+	case <-ctx.Done():
+		dlog.Infof(ctx, "log capture for pod %s interrupted prior to start", pod)
+	case <-ready:
 	}
 }
 
@@ -598,7 +635,7 @@ func (s *cluster) installChart(ctx context.Context, release bool, chartFilename
 	if err == nil {
 		err = RolloutStatusWait(ctx, nss.Namespace, "deploy/traffic-manager")
 		if err == nil {
-			s.self.CapturePodLogs(ctx, "app=traffic-manager", "", nss.Namespace)
+			s.self.CapturePodLogs(ctx, "traffic-manager", "", nss.Namespace)
 		}
 	}
 	return err
@@ -685,7 +722,7 @@ func (s *cluster) TelepresenceHelmInstall(ctx context.Context, upgrade bool, set
 	if err = RolloutStatusWait(ctx, nss.Namespace, "deploy/traffic-manager"); err != nil {
 		return err
 	}
-	s.self.CapturePodLogs(ctx, "app=traffic-manager", "", nss.Namespace)
+	s.self.CapturePodLogs(ctx, "traffic-manager", "", nss.Namespace)
 	return nil
 }
 

integration_test/itest/namespace.go

@@ -150,7 +150,7 @@ func (s *nsPair) RollbackTM(ctx context.Context) {
 	t := getT(ctx)
 	require.NoError(t, err)
 	require.NoError(t, RolloutStatusWait(ctx, s.Namespace, "deploy/traffic-manager"))
-	s.CapturePodLogs(ctx, "app=traffic-manager", "", s.Namespace)
+	s.CapturePodLogs(ctx, "traffic-manager", "", s.Namespace)
 }
 
 func (s *nsPair) AppNamespace() string {

integration_test/itest/traffic_manager.go

@@ -30,7 +30,7 @@ func (th *trafficManager) setup(ctx context.Context) bool {
 	if t.Failed() {
 		return false
 	}
-	th.CapturePodLogs(ctx, "app=traffic-manager", "", th.ManagerNamespace())
+	th.CapturePodLogs(ctx, "traffic-manager", "", th.ManagerNamespace())
 	return true
 }
 

integration_test/large_files_test.go

@@ -118,7 +118,7 @@ func (s *largeFilesSuite) createIntercepts(ctx context.Context) {
 			require.NotNil(info.Mount)
 			s.mountPoint[i] = info.Mount.LocalDir
 			s.NoError(itest.RolloutStatusWait(ctx, s.AppNamespace(), "deploy/"+svc))
-			s.CapturePodLogs(ctx, "service="+svc, "traffic-agent", s.AppNamespace())
+			s.CapturePodLogs(ctx, svc, "traffic-agent", s.AppNamespace())
 		}(i)
 	}
 	wg.Wait()

integration_test/multiport_test.go

@@ -152,7 +152,7 @@ func (s *connectedSuite) Test_UnnamedUdpAndTcpPort() {
 		itest.TelepresenceOk(ctx, "loglevel", "trace")
 		defer itest.TelepresenceOk(ctx, "loglevel", "debug")
 		itest.TelepresenceOk(ctx, "intercept", "--mount", "false", "--service", "echo-udp", "-p", fmt.Sprintf("%d:%s", localPort, svcPort), dep)
-		s.CapturePodLogs(ctx, "app="+dep, "traffic-agent", s.AppNamespace())
+		s.CapturePodLogs(ctx, dep, "traffic-agent", s.AppNamespace())
 		defer itest.TelepresenceOk(ctx, "leave", dep)
 
 		pingPong := func(conn net.Conn, msg string) {