Support dump controller/agent info when fail

Signed-off-by: Hang Yan <[email protected]>

Author: hangyan

pkg/antctl/raw/supportbundle/command.go

@@ -15,14 +15,11 @@
 package supportbundle
 
 import (
-	"antrea.io/antrea/pkg/util/compress"
 	"bufio"
 	"context"
 	"encoding/json"
 	"fmt"
-	"github.com/spf13/afero"
 	"io"
-	utilerror "k8s.io/apimachinery/pkg/util/errors"
 	"net"
 	"os"
 	"path"
@@ -31,6 +28,11 @@ import (
 	"sync"
 	"time"
 
+	"antrea.io/antrea/pkg/apis/crd/v1beta1"
+	"antrea.io/antrea/pkg/util/compress"
+	"github.com/spf13/afero"
+	utilerror "k8s.io/apimachinery/pkg/util/errors"
+
 	"github.com/cheggaaa/pb/v3"
 	"github.com/spf13/cobra"
 	"golang.org/x/sync/errgroup"
@@ -610,7 +612,7 @@ func controllerRemoteRunE(cmd *cobra.Command, args []string) error {
 
 	results := requestAll(agentClients, controllerClient, bar)
 	results = downloadAll(agentClients, controllerClient, dir, bar, results)
-	return processResults(k8sClientset, results, dir)
+	return processResults(antreaClientset, k8sClientset, results, dir)
 }
 
 func genErrorMsg(resultMap map[string]error) string {
@@ -624,7 +626,7 @@ func genErrorMsg(resultMap map[string]error) string {
 // processResults will output the failed nodes and their reasons if any. If no data was collected,
 // error is returned, otherwise will return nil. For failed nodes and controller, will also trying to get logs from
 // kubernetes api.
-func processResults(k8sClient kubernetes.Interface, resultMap map[string]error, dir string) error {
+func processResults(antreaClientset antrea.Interface, k8sClient kubernetes.Interface, resultMap map[string]error, dir string) error {
 	resultStr := ""
 	var failedNodes []string
 	allFailed := true
@@ -656,7 +658,7 @@ func processResults(k8sClient kubernetes.Interface, resultMap map[string]error,
 
 	// download logs from kubernetes api
 	if failedNodes != nil || controllerFail {
-		err := downloadLogsFromKubernetes(k8sClient, failedNodes, controllerFail, dir)
+		err := downloadPodInfoFromKubernetes(antreaClientset, k8sClient, failedNodes, controllerFail, dir)
 		if err != nil {
 			fmt.Println("Failed to download logs from kubernetes api: " + err.Error())
 		} else {
@@ -671,8 +673,18 @@ func processResults(k8sClient kubernetes.Interface, resultMap map[string]error,
 	}
 }
 
-// downloadLogsFromKubernetes will try to download logs from kubernetes api for failed nodes and controller.
-func downloadLogsFromKubernetes(k8sClient kubernetes.Interface, failedNodes []string, isControllerFail bool, dir string) error {
+// downloadLogsFromKubernetes will try to download pod logs from kubernetes api for failed nodes and controller, as well as controllerinfo/agentinfo...
+func downloadPodInfoFromKubernetes(antreaClientset antrea.Interface, k8sClient kubernetes.Interface, failedNodes []string, isControllerFail bool, dir string) error {
+	agentInfoList, err := antreaClientset.CrdV1beta1().AntreaAgentInfos().List(context.TODO(), metav1.ListOptions{ResourceVersion: "0"})
+	if err != nil {
+		return err
+	}
+
+	agentInfoMap := map[string]v1beta1.AntreaAgentInfo{}
+	for _, agentInfo := range agentInfoList.Items {
+		agentInfoMap[agentInfo.Name] = agentInfo
+	}
+
 	pods, err := k8sClient.CoreV1().Pods("kube-system").List(context.TODO(), metav1.ListOptions{
 		ResourceVersion: "0",
 		LabelSelector:   "app=antrea",
@@ -689,41 +701,67 @@ func downloadLogsFromKubernetes(k8sClient kubernetes.Interface, failedNodes []st
 	var errors []error
 
 	for _, pod := range pods.Items {
-		if pod.Labels["component"] == "antrea-controller" && isControllerFail {
-			err := downloadPodLogs(k8sClient, "controller", pod.Namespace, pod.Name, []string{"antrea-controller"}, dir)
+
+		tmpDir, err := afero.TempDir(defaultFS, "", "bundle_tmp_")
+		if err != nil {
 			errors = append(errors, err)
 			continue
 		}
+		defer defaultFS.RemoveAll(tmpDir)
+
+		if pod.Labels["component"] == "antrea-controller" && isControllerFail {
+			controllerInfo, err := antreaClientset.CrdV1beta1().AntreaControllerInfos().Get(context.TODO(), "antrea-controller", metav1.GetOptions{})
+			if err != nil {
+				errors = append(errors, err)
+				continue
+			}
+			data, err := yaml.Marshal(controllerInfo)
+			if err != nil {
+				errors = append(errors, err)
+				continue
+			}
+			err = afero.WriteFile(defaultFS, filepath.Join(tmpDir, "controllerinfo"), data, 0644)
+			errors = append(errors, err)
+
+			err = downloadPodLogs(k8sClient, "controller", pod.Namespace, pod.Name, []string{"antrea-controller"}, dir, tmpDir)
+			errors = append(errors, err)
+
+		}
 
 		if _, exist := failedNodesMap[pod.Spec.NodeName]; !exist {
 			continue
 		}
 
 		if pod.Labels["component"] == "antrea-agent" {
-			err := downloadPodLogs(k8sClient, "agent_"+pod.Spec.NodeName, pod.Namespace, pod.Name, []string{"antrea-agent", "antrea-ovs", "install-cni"}, dir)
+			if agentInfo, ok := agentInfoMap[pod.Spec.NodeName]; ok {
+				data, err := yaml.Marshal(agentInfo)
+				if err != nil {
+					errors = append(errors, err)
+					continue
+				}
+				err = afero.WriteFile(defaultFS, filepath.Join(tmpDir, "agentinfo"), data, 0644)
+				errors = append(errors, err)
+			}
+
+			err = downloadPodLogs(k8sClient, "agent_"+pod.Spec.NodeName, pod.Namespace, pod.Name, []string{"antrea-agent", "antrea-ovs", "install-cni"}, dir, tmpDir)
 			errors = append(errors, err)
+
 		}
 	}
 	return utilerror.NewAggregate(errors)
 }
 
-func downloadPodLogs(k8sClient kubernetes.Interface, comp string, namespace string, podName string, containers []string, dir string) error {
+func downloadPodLogs(k8sClient kubernetes.Interface, comp string, namespace string, podName string, containers []string, dir string, tmpDir string) error {
 	var errors []error
 
-	tmpDir, err := afero.TempDir(defaultFS, "", "bundle_tmp_")
-	if err != nil {
-		return err
-	}
-	defer defaultFS.RemoveAll(tmpDir)
-
 	for _, containerName := range containers {
 		containerDirName := containerName
 		if strings.HasPrefix(containerName, "antrea-") {
 			containerDirName = strings.ReplaceAll(containerName, "antrea-", "")
 		}
 
 		podLogDir := filepath.Join(tmpDir, "logs", containerDirName)
-		err = os.MkdirAll(podLogDir, 0755)
+		err := os.MkdirAll(podLogDir, 0755)
 		if err != nil {
 			return err
 		}