add flag for max backoff duration

Author: judemars

cmd/gce-pd-csi-driver/main.go

@@ -43,6 +43,7 @@ var (
 	metricsPath          = flag.String("metrics-path", "/metrics", "The HTTP path where prometheus metrics will be exposed. Default is `/metrics`.")
 
 	errorBackoffInitialDurationMs = flag.Int("backoff-initial-duration-ms", 200, "The amount of ms for the initial duration of the backoff condition for controller publish/unpublish CSI operations. Default is 200.")
+	errorBackoffMaxDurationMs     = flag.Int("backoff-max-duration-ms", 300000, "The amount of ms for the max duration of the backoff condition for controller publish/unpublish CSI operations. Default is 300000 (5m).")
 	extraVolumeLabelsStr          = flag.String("extra-labels", "", "Extra labels to attach to each PD created. It is a comma separated list of key value pairs like '<key1>=<value1>,<key2>=<value2>'. See https://cloud.google.com/compute/docs/labeling-resources for details")
 
 	attachDiskBackoffDuration = flag.Duration("attach-disk-backoff-duration", 5*time.Second, "Duration for attachDisk backoff")
@@ -123,7 +124,9 @@ func handle() {
 		if err != nil {
 			klog.Fatalf("Failed to get cloud provider: %v", err)
 		}
-		controllerServer = driver.NewControllerServer(gceDriver, cloudProvider, time.Duration(*errorBackoffInitialDurationMs)*time.Millisecond)
+		initialBackoffDuration := time.Duration(*errorBackoffInitialDurationMs) * time.Millisecond
+		maxBackoffDuration := time.Duration(*errorBackoffMaxDurationMs) * time.Microsecond
+		controllerServer = driver.NewControllerServer(gceDriver, cloudProvider, initialBackoffDuration, maxBackoffDuration)
 	} else if *cloudConfigFilePath != "" {
 		klog.Warningf("controller service is disabled but cloud config given - it has no effect")
 	}

pkg/gce-pd-csi-driver/controller.go

@@ -37,11 +37,6 @@ import (
 	gce "sigs.k8s.io/gcp-compute-persistent-disk-csi-driver/pkg/gce-cloud-provider/compute"
 )
 
-const (
-	errorBackoffInitialDuration = 200 * time.Millisecond
-	errorBackoffMaxDuration     = 5 * time.Minute
-)
-
 type GCEControllerServer struct {
 	Driver        *GCEDriver
 	CloudProvider gce.GCECompute
@@ -1596,7 +1591,7 @@ func pickRandAndConsecutive(slice []string, n int) ([]string, error) {
 	return ret, nil
 }
 
-func newCsiErrorBackoff(initialDuration time.Duration) *csiErrorBackoff {
+func newCsiErrorBackoff(initialDuration, errorBackoffMaxDuration time.Duration) *csiErrorBackoff {
 	return &csiErrorBackoff{flowcontrol.NewBackOff(initialDuration, errorBackoffMaxDuration)}
 }
 

pkg/gce-pd-csi-driver/controller_test.go

@@ -2149,6 +2149,8 @@ type backoffTesterConfig struct {
 }
 
 func newFakeCsiErrorBackoff(tc *clock.FakeClock) *csiErrorBackoff {
+	errorBackoffInitialDuration := 200 * time.Millisecond
+	errorBackoffMaxDuration := 5 * time.Minute
 	return &csiErrorBackoff{flowcontrol.NewFakeBackOff(errorBackoffInitialDuration, errorBackoffMaxDuration, tc)}
 }
 

pkg/gce-pd-csi-driver/gce-pd-driver.go

@@ -149,13 +149,13 @@ func NewNodeServer(gceDriver *GCEDriver, mounter *mount.SafeFormatAndMount, devi
 	}
 }
 
-func NewControllerServer(gceDriver *GCEDriver, cloudProvider gce.GCECompute, errorBackoffInitialDuration time.Duration) *GCEControllerServer {
+func NewControllerServer(gceDriver *GCEDriver, cloudProvider gce.GCECompute, errorBackoffInitialDuration, errorBackoffMaxDuration time.Duration) *GCEControllerServer {
 	return &GCEControllerServer{
 		Driver:        gceDriver,
 		CloudProvider: cloudProvider,
 		seen:          map[string]int{},
 		volumeLocks:   common.NewVolumeLocks(),
-		errorBackoff:  newCsiErrorBackoff(errorBackoffInitialDuration),
+		errorBackoff:  newCsiErrorBackoff(errorBackoffInitialDuration, errorBackoffMaxDuration),
 	}
 }
 

pkg/gce-pd-csi-driver/gce-pd-driver_test.go

@@ -16,6 +16,7 @@ package gceGCEDriver
 
 import (
 	"testing"
+	"time"
 
 	gce "sigs.k8s.io/gcp-compute-persistent-disk-csi-driver/pkg/gce-cloud-provider/compute"
 )
@@ -43,7 +44,10 @@ func initBlockingGCEDriver(t *testing.T, cloudDisks []*gce.CloudDisk, readyToExe
 func initGCEDriverWithCloudProvider(t *testing.T, cloudProvider gce.GCECompute) *GCEDriver {
 	vendorVersion := "test-vendor"
 	gceDriver := GetGCEDriver()
-	controllerServer := NewControllerServer(gceDriver, cloudProvider, errorBackoffInitialDuration)
+	errorBackoffInitialDuration := 200 * time.Millisecond
+	errorBackoffMaxDuration := 5 * time.Minute
+
+	controllerServer := NewControllerServer(gceDriver, cloudProvider, errorBackoffInitialDuration, errorBackoffMaxDuration)
 	err := gceDriver.SetupGCEDriver(driver, vendorVersion, nil, nil, controllerServer, nil)
 	if err != nil {
 		t.Fatalf("Failed to setup GCE Driver: %v", err)