Fix dumping logs for GCE scale tests

cmd/kops/toolbox_dump.go

@@ -210,7 +210,14 @@ func RunToolboxDump(ctx context.Context, f commandutils.Factory, out io.Writer,
 			return fmt.Errorf("adding key to SSH agent: %w", err)
 		}
 
-		dumper := dump.NewLogDumper(cluster.ObjectMeta.Name, sshConfig, keyRing, options.Dir)
+		// look for a bastion instance and use it if exists
+		bastionAddress := ""
+		for _, instance := range d.Instances {
+			if strings.Contains(instance.Name, "bastion") {
+				bastionAddress = instance.PublicAddresses[0]
+			}
+		}
+		dumper := dump.NewLogDumper(bastionAddress, sshConfig, keyRing, options.Dir)
 
 		var additionalIPs []string
 		var additionalPrivateIPs []string
@@ -224,7 +231,7 @@ func RunToolboxDump(ctx context.Context, f commandutils.Factory, out io.Writer,
 			}
 		}
 
-		if err := dumper.DumpAllNodes(ctx, nodes, additionalIPs, additionalPrivateIPs); err != nil {
+		if err := dumper.DumpAllNodes(ctx, nodes, options.MaxNodes, additionalIPs, additionalPrivateIPs); err != nil {
 			return fmt.Errorf("error dumping nodes: %v", err)
 		}
 

pkg/dump/dumper.go

@@ -34,11 +34,6 @@ import (
 	"k8s.io/klog/v2"
 )
 
-const (
-	// MaxNodesToDump is the maximum number of nodes to dump
-	MaxNodesToDump = 500
-)
-
 // logDumper gets all the nodes from a kubernetes cluster and dumps a well-known set of logs
 type logDumper struct {
 	sshClientFactory sshClientFactory
@@ -51,12 +46,15 @@ type logDumper struct {
 }
 
 // NewLogDumper is the constructor for a logDumper
-func NewLogDumper(clusterName string, sshConfig *ssh.ClientConfig, keyRing agent.Agent, artifactsDir string) *logDumper {
+func NewLogDumper(bastionAddress string, sshConfig *ssh.ClientConfig, keyRing agent.Agent, artifactsDir string) *logDumper {
 	sshClientFactory := &sshClientFactoryImplementation{
-		bastion:   "bastion." + clusterName,
 		keyRing:   keyRing,
 		sshConfig: sshConfig,
 	}
+	if bastionAddress != "" {
+		log.Printf("detected a bastion instance, with the address: %s", bastionAddress)
+		sshClientFactory.bastion = bastionAddress
+	}
 
 	d := &logDumper{
 		sshClientFactory: sshClientFactory,
@@ -106,9 +104,10 @@ func NewLogDumper(clusterName string, sshConfig *ssh.ClientConfig, keyRing agent
 // if the IPs are not found from kubectl get nodes, then these will be dumped also.
 // This allows for dumping log on nodes even if they don't register as a kubernetes
 // node, or if a node fails to register, or if the whole cluster fails to start.
-func (d *logDumper) DumpAllNodes(ctx context.Context, nodes corev1.NodeList, additionalIPs, additionalPrivateIPs []string) error {
+func (d *logDumper) DumpAllNodes(ctx context.Context, nodes corev1.NodeList, maxNodesToDump int, additionalIPs, additionalPrivateIPs []string) error {
 	var special, regular, dumped []*corev1.Node
 
+	log.Printf("starting to dump %d nodes fetched through the Kubernetes APIs", len(nodes.Items))
 	for i := range nodes.Items {
 		node := &nodes.Items[i]
 
@@ -139,8 +138,8 @@ func (d *logDumper) DumpAllNodes(ctx context.Context, nodes corev1.NodeList, add
 	}
 
 	for i := range regular {
-		if len(dumped) >= MaxNodesToDump {
-			log.Printf("stopping dumping nodes: %d nodes dumped", MaxNodesToDump)
+		if len(dumped) >= maxNodesToDump {
+			log.Printf("stopping dumping nodes: %d nodes dumped", maxNodesToDump)
 			return nil
 		}
 		node := regular[i]
@@ -154,8 +153,8 @@ func (d *logDumper) DumpAllNodes(ctx context.Context, nodes corev1.NodeList, add
 
 	notDumped := findInstancesNotDumped(additionalIPs, dumped)
 	for _, ip := range notDumped {
-		if len(dumped) >= MaxNodesToDump {
-			log.Printf("stopping dumping nodes: %d nodes dumped", MaxNodesToDump)
+		if len(dumped) >= maxNodesToDump {
+			log.Printf("stopping dumping nodes: %d nodes dumped", maxNodesToDump)
 			return nil
 		}
 		err := d.dumpNotRegistered(ctx, ip, false)
@@ -166,8 +165,8 @@ func (d *logDumper) DumpAllNodes(ctx context.Context, nodes corev1.NodeList, add
 
 	notDumped = findInstancesNotDumped(additionalPrivateIPs, dumped)
 	for _, ip := range notDumped {
-		if len(dumped) >= MaxNodesToDump {
-			log.Printf("stopping dumping nodes: %d nodes dumped", MaxNodesToDump)
+		if len(dumped) >= maxNodesToDump {
+			log.Printf("stopping dumping nodes: %d nodes dumped", maxNodesToDump)
 			return nil
 		}
 		err := d.dumpNotRegistered(ctx, ip, true)

pkg/resources/gce/dump.go

@@ -63,6 +63,12 @@ func DumpManagedInstance(op *resources.DumpOperation, r *resources.Resource) err
 		klog.Warningf("instance %q not found", instance.Instance)
 	} else {
 		for _, ni := range instanceDetails.NetworkInterfaces {
+			if ni.NetworkIP != "" {
+				i.PrivateAddresses = append(i.PrivateAddresses, ni.NetworkIP)
+			}
+			if ni.Ipv6Address != "" {
+				i.PrivateAddresses = append(i.PrivateAddresses, ni.Ipv6Address)
+			}
 			for _, ac := range ni.AccessConfigs {
 				if ac.NatIP != "" {
 					i.PublicAddresses = append(i.PublicAddresses, ac.NatIP)

tests/e2e/kubetest2-kops/deployer/common.go

@@ -100,6 +100,8 @@ func (d *deployer) initialize() error {
 				d.SSHPublicKeyPath = publicKey
 			}
 			d.createBucket = true
+		} else if d.SSHPrivateKeyPath == "" && os.Getenv("KUBE_SSH_KEY_PATH") != "" {
+			d.SSHPrivateKeyPath = os.Getenv("KUBE_SSH_KEY_PATH")
 		}
 	}
 

tests/e2e/kubetest2-kops/deployer/deployer.go

@@ -68,6 +68,7 @@ type deployer struct {
 	ValidationWait     time.Duration `flag:"validation-wait" desc:"time to wait for newly created cluster to pass validation"`
 	ValidationCount    int           `flag:"validation-count" desc:"how many times should a validation pass"`
 	ValidationInterval time.Duration `flag:"validation-interval" desc:"time in duration to wait between validation attempts"`
+	MaxNodesToDump     string        `flag:"max-nodes-to-dump" desc:"max number of nodes to dump logs from, helpful to set when running scale tests"`
 
 	TemplatePath string `flag:"template-path" desc:"The path to the manifest template used for cluster creation"`
 

tests/e2e/kubetest2-kops/deployer/dumplogs.go

@@ -44,6 +44,10 @@ func (d *deployer) DumpClusterLogs() error {
 		"--private-key", d.SSHPrivateKeyPath,
 		"--ssh-user", d.SSHUser,
 	}
+
+	if d.MaxNodesToDump != "" {
+		args = append(args, "--max-nodes", d.MaxNodesToDump)
+	}
 	klog.Info(strings.Join(args, " "))
 	cmd := exec.Command(args[0], args[1:]...)
 	cmd.SetEnv(append(d.env(), "KOPS_TOOLBOX_DUMP_K8S_RESOURCES=1")...)