WIP Update gce operation timeout

Update gce operation timeout to 2 mins instead of 5 mins. The 90% of
attach/detach disk is less than 15 second, and create volume is less
than 1 min.

Reduce the timeout of volume operation so that it can recover from
previous operation quicker in case operation is dropped.

Author: jingxu97

deploy/kubernetes/base/controller/controller.yaml

@@ -32,7 +32,7 @@ spec:
             - "--enable-leader-election"
             - "--leader-election-type=leases"
             - "--leader-election-namespace=$(PDCSI_NAMESPACE)"
-            - "--timeout=250s"
+            - "--timeout=180s"
             - "--extra-create-metadata"
           # - "--run-controller-service=false" # disable the controller service of the CSI driver
           # - "--run-node-service=false"       # disable the node service of the CSI driver
@@ -52,7 +52,7 @@ spec:
             - "--metrics-address=:22012"
             - "--leader-election"
             - "--leader-election-namespace=$(PDCSI_NAMESPACE)"
-            - "--timeout=250s"
+            - "--timeout=90s"
           env:
             - name: PDCSI_NAMESPACE
               valueFrom:

pkg/gce-cloud-provider/compute/gce-compute.go

@@ -46,6 +46,28 @@ const (
 	GCEAPIVersionV1 GCEAPIVersion = "v1"
 	// Alpha key type
 	GCEAPIVersionBeta GCEAPIVersion = "beta"
+	// OpPollInterval is the ineterval between polling an operation
+	OpPollInterval = 2 * time.Second
+	// OpTimeout is the default total timeout for polling on an operation
+	OpTimeout = 2 * time.Minute
+	// defaultCallTimeout is the ddefault context timeout for creating/getting on an operation
+	defaultCallTimeout = 3 * time.Second
+	// Time interval for checking on an operation
+	AttachDiskInitialDelay = 6 * time.Second
+
+	InsertDiskInitialDelay = 3 * time.Second
+
+	// volumeAttachmentConsecutiveErrorLimit is the number of consecutive errors we will ignore when waiting for a volume to attach/detach
+	//volumeAttachmentStatusConsecutiveErrorLimit = 10
+
+	// Attach typically takes 2-5 seconds (average is 2). Asking before 2 seconds is just waste of API quota.
+	volumeAttachmentStatusInitialDelay = 2 * time.Second
+	// Detach typically takes 5-10 seconds (average is 6). Asking before 5 seconds is just waste of API quota.
+	volumeDetachmentStatusInitialDelay = 5 * time.Second
+	// After the initial delay, poll attach/detach with exponential backoff (2046 seconds total)
+	volumeAttachmentStatusPollDelay = 2 * time.Second
+	volumeAttachmentStatusFactor    = 2
+	volumeAttachmentStatusSteps     = 11
 )
 
 type GCECompute interface {
@@ -306,6 +328,11 @@ func ValidateDiskParameters(disk *CloudDisk, params common.DiskParameters) error
 	return nil
 }
 
+// ContextWithCallTimeout returns a context with a default timeout, used for generated client calls.
+func ContextWithCallTimeout() (context.Context, context.CancelFunc) {
+	return context.WithTimeout(context.Background(), defaultCallTimeout)
+}
+
 func (cloud *CloudProvider) InsertDisk(ctx context.Context, volKey *meta.Key, params common.DiskParameters, capBytes int64, capacityRange *csi.CapacityRange, replicaZones []string, snapshotID string, multiWriter bool) error {
 	klog.V(5).Infof("Inserting disk %v", volKey)
 
@@ -607,26 +634,35 @@ func (cloud *CloudProvider) AttachDisk(ctx context.Context, volKey *meta.Key, re
 		Type:       diskType,
 	}
 
-	op, err := cloud.service.Instances.AttachDisk(cloud.project, instanceZone, instanceName, attachedDiskV1).Context(ctx).Do()
+	callCtx, cancel := ContextWithCallTimeout()
+	defer cancel()
+
+	op, err := cloud.service.Instances.AttachDisk(cloud.project, instanceZone, instanceName, attachedDiskV1).Context(callCtx).Do()
 	if err != nil {
 		return fmt.Errorf("failed cloud service attach disk call: %v", err)
 	}
+	klog.V(5).Infof("Attaching disk %s operation id is %s", volKey, op.Name)
 	err = cloud.waitForZonalOp(ctx, op.Name, instanceZone)
 	if err != nil {
-		return fmt.Errorf("failed when waiting for zonal op: %v", err)
+		return fmt.Errorf("failed when waiting for zonal op %s on attaching volume %v: %v", op.Name, volKey, err)
 	}
 	return nil
 }
 
 func (cloud *CloudProvider) DetachDisk(ctx context.Context, deviceName, instanceZone, instanceName string) error {
-	klog.V(5).Infof("Detaching disk %v from %v", deviceName, instanceName)
-	op, err := cloud.service.Instances.DetachDisk(cloud.project, instanceZone, instanceName, deviceName).Context(ctx).Do()
+
+	callCtx, cancel := ContextWithCallTimeout()
+	defer cancel()
+
+	klog.V(5).Infof("Detaching disk %s from %v", deviceName, instanceName)
+	op, err := cloud.service.Instances.DetachDisk(cloud.project, instanceZone, instanceName, deviceName).Context(callCtx).Do()
 	if err != nil {
 		return err
 	}
+	klog.V(5).Infof("Detaching disk %s operation id is %s", deviceName, op.Name)
 	err = cloud.waitForZonalOp(ctx, op.Name, instanceZone)
 	if err != nil {
-		return err
+		return fmt.Errorf("failed when waiting for zonal op %s on detaching volume %s: %v", op.Name, deviceName, err)
 	}
 	return nil
 }
@@ -677,17 +713,45 @@ func (cloud *CloudProvider) getRegionalDiskTypeURI(region, diskType string) stri
 	return cloud.service.BasePath + fmt.Sprintf(diskTypeURITemplateRegional, cloud.project, region, diskType)
 }
 
+// SleepWithContext will wait for the timer duration to expire, or the context
+// is canceled. Which ever happens first. If the context is canceled the Context's
+// error will be returned.
+//
+// Expects Context to always return a non-nil error if the Done channel is closed.
+/*func SleepWithContext(ctx Context, dur time.Duration) error {
+	t := time.NewTimer(dur)
+	defer t.Stop()
+
+	select {
+	case <-t.C:
+		break
+	case <-ctx.Done():
+		return ctx.Err()
+	}
+
+	return nil
+}*/
+
 func (cloud *CloudProvider) waitForZonalOp(ctx context.Context, opName string, zone string) error {
 	// The v1 API can query for v1, alpha, or beta operations.
 	svc := cloud.service
 	project := cloud.project
-	return wait.Poll(3*time.Second, 5*time.Minute, func() (bool, error) {
-		pollOp, err := svc.ZoneOperations.Get(project, zone, opName).Context(ctx).Do()
+	timeout := OpTimeout
+	if deadline, ok := ctx.Deadline(); ok {
+		timeout = time.Until(deadline)
+		klog.Infof("Getting timeout from passed context %v", timeout)
+	}
+	callCtx, cancel := ContextWithCallTimeout()
+	defer cancel()
+
+	return wait.Poll(OpPollInterval, timeout, func() (bool, error) {
+		pollOp, err := svc.ZoneOperations.Get(project, zone, opName).Context(callCtx).Do()
 		if err != nil {
 			klog.Errorf("WaitForOp(op: %s, zone: %#v) failed to poll the operation", opName, zone)
 			return false, err
 		}
 		done, err := opIsDone(pollOp)
+		klog.V(4).Infof("checking op %s is done %t", opName, done)
 		return done, err
 	})
 }
@@ -743,7 +807,10 @@ func (cloud *CloudProvider) WaitForAttach(ctx context.Context, volKey *meta.Key,
 }
 
 func opIsDone(op *computev1.Operation) (bool, error) {
-	if op == nil || op.Status != operationStatusDone {
+	if op == nil {
+		return true, fmt.Errorf("operation is nil")
+	}
+	if op.Status != operationStatusDone {
 		return false, nil
 	}
 	if op.Error != nil && len(op.Error.Errors) > 0 && op.Error.Errors[0] != nil {