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

Author: yguo0905

cmd/node_problem_detector.go

@@ -26,6 +26,7 @@ import (
 	"github.com/golang/glog"
 	"github.com/spf13/pflag"
 
+	"k8s.io/apimachinery/pkg/util/wait"
 	"k8s.io/node-problem-detector/cmd/options"
 	"k8s.io/node-problem-detector/pkg/custompluginmonitor"
 	"k8s.io/node-problem-detector/pkg/problemclient"
@@ -95,7 +96,25 @@ func main() {
 		startHTTPServer(p, npdo)
 	}
 
+	// This function may be blocked (until a timeout occurs) before
+	// kube-apiserver becomes ready.
+	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: timed out on waiting for kube-apiserver to return the node object: %v", err)
+	}
+
 	if err := p.Run(); err != nil {
 		glog.Fatalf("Problem detector failed with error: %v", err)
 	}
 }
+
+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
+	})
+}

cmd/options/options.go

@@ -20,6 +20,7 @@ import (
 	"flag"
 	"fmt"
 	"os"
+	"time"
 
 	"net/url"
 
@@ -46,6 +47,12 @@ type NodeProblemDetectorOptions struct {
 	ServerPort int
 	// ServerAddress is the address to bind the node problem detector server.
 	ServerAddress 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
 
 	// application options
 
@@ -65,6 +72,8 @@ func (npdo *NodeProblemDetectorOptions) AddFlags(fs *pflag.FlagSet) {
 		[]string{}, "List of paths to custom plugin monitor config files, comma separated.")
 	fs.StringVar(&npdo.ApiServerOverride, "apiserver-override",
 		"", "Custom URI used to connect to Kubernetes ApiServer")
+	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/problemclient/fake_problem_client.go

@@ -21,7 +21,7 @@ import (
 	"reflect"
 	"sync"
 
-	"k8s.io/api/core/v1"
+	v1 "k8s.io/api/core/v1"
 )
 
 // FakeProblemClient is a fake problem client for debug.
@@ -92,3 +92,7 @@ func (f *FakeProblemClient) GetConditions(types []v1.NodeConditionType) ([]*v1.N
 // Eventf does nothing now.
 func (f *FakeProblemClient) Eventf(eventType string, source, reason, messageFmt string, args ...interface{}) {
 }
+
+func (f *FakeProblemClient) GetNode() (*v1.Node, error) {
+	return nil, fmt.Errorf("GetNode() not implemented")
+}

pkg/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)