Merge pull request #552 from mcshooter/detectKubeProxyServiceProblem

Add healthChecker functionality for kube-proxy service

cmd/healthchecker/options/options.go

@@ -47,7 +47,7 @@ type HealthCheckerOptions struct {
 // AddFlags adds health checker command line options to pflag.
 func (hco *HealthCheckerOptions) AddFlags(fs *pflag.FlagSet) {
 	fs.StringVar(&hco.Component, "component", types.KubeletComponent,
-		"The component to check health for. Supports kubelet, docker and cri")
+		"The component to check health for. Supports kubelet, docker, kube-proxy, and cri")
 	// Deprecated: For backward compatibility on linux environment. Going forward "service" will be used instead of systemd-service
 	if runtime.GOOS == "linux" {
 		fs.MarkDeprecated("systemd-service", "please use --service flag instead")
@@ -73,8 +73,9 @@ func (hco *HealthCheckerOptions) AddFlags(fs *pflag.FlagSet) {
 // Returns error if invalid, nil otherwise.
 func (hco *HealthCheckerOptions) IsValid() error {
 	// Make sure the component specified is valid.
-	if hco.Component != types.KubeletComponent && hco.Component != types.DockerComponent && hco.Component != types.CRIComponent {
-		return fmt.Errorf("the component specified is not supported. Supported components are : <kubelet/docker/cri>")
+	if hco.Component != types.KubeletComponent && hco.Component != types.DockerComponent &&
+		hco.Component != types.CRIComponent && hco.Component != types.KubeProxyComponent {
+		return fmt.Errorf("the component specified is not supported. Supported components are : <kubelet/docker/cri/kube-proxy>")
 	}
 	// Make sure the service is specified if repair is enabled.
 	if hco.EnableRepair && hco.Service == "" {

config/windows-health-checker-kubeproxy.json

@@ -0,0 +1,34 @@
+{
+    "plugin": "custom",
+    "pluginConfig": {
+      "invoke_interval": "10s",
+      "timeout": "3m",
+      "max_output_length": 80,
+      "concurrency": 1
+    },
+    "source": "health-checker",
+    "metricsReporting": true,
+    "conditions": [
+      {
+        "type": "KubeProxyUnhealthy",
+        "reason": "KubeProxyIsHealthy",
+        "message": "kube-proxy on the node is functioning properly"
+      }
+    ],
+    "rules": [
+      {
+        "type": "permanent",
+        "condition": "KubeProxyUnhealthy",
+        "reason": "KubeProxyUnhealthy",
+        "path": "C:\\etc\\kubernetes\\node\\bin\\health-checker.exe",
+        "args": [
+          "--component=kube-proxy",
+          "--enable-repair=true",
+          "--cooldown-time=1m",
+          "--health-check-timeout=10s"
+        ],
+        "timeout": "3m"
+      }
+    ]
+  }
+  

pkg/healthchecker/health_checker_windows.go

@@ -68,14 +68,9 @@ func getRepairFunc(hco *options.HealthCheckerOptions) func() {
 func getHealthCheckFunc(hco *options.HealthCheckerOptions) func() (bool, error) {
 	switch hco.Component {
 	case types.KubeletComponent:
-		return func() (bool, error) {
-			httpClient := http.Client{Timeout: hco.HealthCheckTimeout}
-			response, err := httpClient.Get(types.KubeletHealthCheckEndpoint)
-			if err != nil || response.StatusCode != http.StatusOK {
-				return false, nil
-			}
-			return true, nil
-		}
+		return healthCheckEndpointOKFunc(types.KubeletHealthCheckEndpoint, hco.HealthCheckTimeout)
+	case types.KubeProxyComponent:
+		return healthCheckEndpointOKFunc(types.KubeProxyHealthCheckEndpoint, hco.HealthCheckTimeout)
 	case types.DockerComponent:
 		return func() (bool, error) {
 			if _, err := execCommand("docker.exe", "ps"); err != nil {
@@ -94,6 +89,18 @@ func getHealthCheckFunc(hco *options.HealthCheckerOptions) func() (bool, error)
 	return nil
 }
 
+// healthCheckEndpointOKFunc returns a function to check the status of an http endpoint
+func healthCheckEndpointOKFunc(endpoint string, timeout time.Duration) func() (bool, error) {
+	return func() (bool, error) {
+		httpClient := http.Client{Timeout: timeout}
+		response, err := httpClient.Get(endpoint)
+		if err != nil || response.StatusCode != http.StatusOK {
+			return false, nil
+		}
+		return true, nil
+	}
+}
+
 // execCommand creates a new process, executes the command, and returns the (output, error) from command.
 func execCommand(command string, args ...string) (string, error) {
 	cmd := util.Exec(command, args...)

pkg/healthchecker/types/types.go

@@ -30,12 +30,14 @@ const (
 	CmdTimeout                = 10 * time.Second
 	LogParsingTimeLayout      = "2006-01-02 15:04:05"
 
-	KubeletComponent  = "kubelet"
-	CRIComponent      = "cri"
-	DockerComponent   = "docker"
-	ContainerdService = "containerd"
+	KubeletComponent   = "kubelet"
+	CRIComponent       = "cri"
+	DockerComponent    = "docker"
+	ContainerdService  = "containerd"
+	KubeProxyComponent = "kube-proxy"
 
-	KubeletHealthCheckEndpoint = "http://127.0.0.1:10248/healthz"
+	KubeletHealthCheckEndpoint   = "http://127.0.0.1:10248/healthz"
+	KubeProxyHealthCheckEndpoint = "http://127.0.0.1:10256/healthz"
 
 	LogPatternFlagSeparator = ":"
 )