Merge pull request #1027 from dghubble/checkpointer-try-secure

pkg/checkpoint: Try kubelet secureClient, fallback to read-only

pkg/checkpoint/checkpoint.go

@@ -110,7 +110,7 @@ func (c *checkpointer) run() {
 	for {
 		time.Sleep(defaultPollingFrequency)
 
-		// We must use both the :10255/pods endpoint and CRI shim, because /pods
+		// We must use both the kubelet /pods endpoint and CRI shim, because /pods
 		// endpoint could have stale data. The /pods endpoint will only show the last cached
 		// status which has successfully been written to an apiserver. However, if there is
 		// no apiserver, we may get stale state (e.g. saying pod is running, when it really is

pkg/checkpoint/kubelet.go

@@ -2,6 +2,7 @@ package checkpoint
 
 import (
 	"fmt"
+	"time"
 
 	"github.com/golang/glog"
 	"k8s.io/api/core/v1"
@@ -49,9 +50,13 @@ func newKubeletClient(config *rest.Config) (*kubeletClient, error) {
 // localParentPods will retrieve all pods from kubelet api that are parents & should be checkpointed
 func (k *kubeletClient) localParentPods() map[string]*v1.Pod {
 	podList := new(v1.PodList)
-	if err := k.insecureClient.Get().AbsPath("/pods/").Do().Into(podList); err != nil {
-		// Assume there are no local parent pods.
-		glog.Errorf("failed to list local parent pods, assuming none are running: %v", err)
+	timeout := 15 * time.Second
+	if err := k.secureClient.Get().AbsPath("/pods/").Timeout(timeout).Do().Into(podList); err != nil {
+		glog.Errorf("failed to secure list local parent pods, fallback to insecure: %v", err)
+		if err := k.insecureClient.Get().AbsPath("/pods/").Timeout(timeout).Do().Into(podList); err != nil {
+			// Assume there are no local parent pods.
+			glog.Errorf("failed to insecure list local parent pods, assuming none are running: %v", err)
+		}
 	}
 	return podListToParentPods(podList)
 }