Support waiting for kube-apiserver to be ready with timout during NPD startup

Author: yguo0905

cmd/options/options.go

@@ -20,6 +20,7 @@ import (
 	"flag"
 	"fmt"
 	"os"
+	"time"
 
 	"net/url"
 
@@ -49,6 +50,12 @@ type NodeProblemDetectorOptions struct {
 	EnableK8sExporter bool
 	// ApiServerOverride is the custom URI used to connect to Kubernetes ApiServer.
 	ApiServerOverride string
+	// APIServerWaitTimeout is the timeout on waiting for kube-apiserver to be
+	// ready.
+	APIServerWaitTimeout time.Duration
+	// APIServerWaitInterval is the interval between the checks on the
+	// readiness of kube-apiserver.
+	APIServerWaitInterval time.Duration
 
 	// prometheusExporter options
 	// PrometheusServerPort is the port to bind the Prometheus scrape endpoint. Use 0 to disable.
@@ -96,6 +103,8 @@ func (npdo *NodeProblemDetectorOptions) AddFlags(fs *pflag.FlagSet) {
 	fs.BoolVar(&npdo.EnableK8sExporter, "enable-k8s-exporter", true, "Enables reporting to Kubernetes API server.")
 	fs.StringVar(&npdo.ApiServerOverride, "apiserver-override",
 		"", "Custom URI used to connect to Kubernetes ApiServer. This is ignored if --enable-k8s-exporter is false.")
+	fs.DurationVar(&npdo.APIServerWaitTimeout, "apiserver-wait-timeout", time.Duration(5)*time.Minute, "The timeout on waiting for kube-apiserver to be ready. This is ignored if --enable-k8s-exporter is false.")
+	fs.DurationVar(&npdo.APIServerWaitInterval, "apiserver-wait-interval", time.Duration(5)*time.Second, "The interval between the checks on the readiness of kube-apiserver. This is ignored if --enable-k8s-exporter is false.")
 	fs.BoolVar(&npdo.PrintVersion, "version", false, "Print version information and quit")
 	fs.StringVar(&npdo.HostnameOverride, "hostname-override",
 		"", "Custom node name used to override hostname")

pkg/exporters/k8sexporter/k8s_exporter.go

@@ -25,6 +25,7 @@ import (
 	"github.com/golang/glog"
 
 	"k8s.io/apimachinery/pkg/util/clock"
+	"k8s.io/apimachinery/pkg/util/wait"
 
 	"k8s.io/node-problem-detector/cmd/options"
 	"k8s.io/node-problem-detector/pkg/exporters/k8sexporter/condition"
@@ -38,13 +39,23 @@ type k8sExporter struct {
 	conditionManager condition.ConditionManager
 }
 
-// NewExporterOrDie creates a exporter for Kubernetes apiserver exporting, panics if error occurs.
+// NewExporterOrDie creates a exporter for Kubernetes apiserver exporting,
+// panics if error occurs.
+//
+// Note that this function may be blocked (until a timeout occurs) before
+// kube-apiserver becomes ready.
 func NewExporterOrDie(npdo *options.NodeProblemDetectorOptions) types.Exporter {
 	if !npdo.EnableK8sExporter {
 		return nil
 	}
 
 	c := problemclient.NewClientOrDie(npdo)
+
+	glog.Infof("Waiting for kube-apiserver to be ready (timeout %v)...", npdo.APIServerWaitTimeout)
+	if err := waitForAPIServerReadyWithTimeout(c, npdo); err != nil {
+		glog.Warningf("kube-apiserver did not become ready: %v", err)
+	}
+
 	ke := k8sExporter{
 		client:           c,
 		conditionManager: condition.NewConditionManager(c, clock.RealClock{}),
@@ -91,3 +102,14 @@ func (ke *k8sExporter) startHTTPReporting(npdo *options.NodeProblemDetectorOptio
 		}
 	}()
 }
+
+func waitForAPIServerReadyWithTimeout(c problemclient.Client, npdo *options.NodeProblemDetectorOptions) error {
+	return wait.PollImmediate(npdo.APIServerWaitInterval, npdo.APIServerWaitTimeout, func() (done bool, err error) {
+		// If NPD can get the node object from kube-apiserver, the server is
+		// ready and the RBAC permission is set correctly.
+		if _, err := c.GetNode(); err == nil {
+			return true, nil
+		}
+		return false, nil
+	})
+}

pkg/exporters/k8sexporter/problemclient/problem_client.go

@@ -26,7 +26,7 @@ import (
 	typedcorev1 "k8s.io/client-go/kubernetes/typed/core/v1"
 	"k8s.io/kubernetes/pkg/api/legacyscheme"
 
-	"k8s.io/api/core/v1"
+	v1 "k8s.io/api/core/v1"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	"k8s.io/apimachinery/pkg/types"
 	"k8s.io/apimachinery/pkg/util/clock"
@@ -47,6 +47,9 @@ type Client interface {
 	SetConditions(conditions []v1.NodeCondition) error
 	// Eventf reports the event.
 	Eventf(eventType string, source, reason, messageFmt string, args ...interface{})
+	// GetNode returns the Node object of the node on which the
+	// node-problem-detector runs.
+	GetNode() (*v1.Node, error)
 }
 
 type nodeProblemClient struct {
@@ -79,7 +82,7 @@ func NewClientOrDie(npdo *options.NodeProblemDetectorOptions) Client {
 }
 
 func (c *nodeProblemClient) GetConditions(conditionTypes []v1.NodeConditionType) ([]*v1.NodeCondition, error) {
-	node, err := c.client.Nodes().Get(c.nodeName, metav1.GetOptions{})
+	node, err := c.GetNode()
 	if err != nil {
 		return nil, err
 	}
@@ -116,6 +119,10 @@ func (c *nodeProblemClient) Eventf(eventType, source, reason, messageFmt string,
 	recorder.Eventf(c.nodeRef, eventType, reason, messageFmt, args...)
 }
 
+func (c *nodeProblemClient) GetNode() (*v1.Node, error) {
+	return c.client.Nodes().Get(c.nodeName, metav1.GetOptions{})
+}
+
 // generatePatch generates condition patch
 func generatePatch(conditions []v1.NodeCondition) ([]byte, error) {
 	raw, err := json.Marshal(&conditions)