fix: increase volume cloning timeout and reduce time cost

kubernetes-sigs · Jan 31, 2024 · 5a3c0ff · 5a3c0ff
1 parent 9fe96e1
commit 5a3c0ff
Show file tree

Hide file tree

Showing 3 changed files with 22 additions and 16 deletions.
diff --git a/pkg/blob/blob.go b/pkg/blob/blob.go
@@ -171,6 +171,7 @@ type DriverOptions struct {
 	EnableAznfsMount                       bool
 	VolStatsCacheExpireInMinutes           int
 	SasTokenExpirationMinutes              int
+	WaitForAzCopyTimeoutMinutes            int
 }
 
 func (option *DriverOptions) AddFlags() {
@@ -188,6 +189,7 @@ func (option *DriverOptions) AddFlags() {
 	flag.BoolVar(&option.EnableAznfsMount, "enable-aznfs-mount", false, "replace nfs mount with aznfs mount")
 	flag.IntVar(&option.VolStatsCacheExpireInMinutes, "vol-stats-cache-expire-in-minutes", 10, "The cache expire time in minutes for volume stats cache")
 	flag.IntVar(&option.SasTokenExpirationMinutes, "sas-token-expiration-minutes", 1440, "sas token expiration minutes during volume cloning")
+	flag.IntVar(&option.WaitForAzCopyTimeoutMinutes, "wait-for-azcopy-timeout-minutes", 18, "timeout in minutes for waiting for azcopy to finish")
 }
 
 // Driver implements all interfaces of CSI drivers
@@ -226,6 +228,8 @@ type Driver struct {
 	azcopySasTokenCache azcache.Resource
 	// sas expiry time for azcopy in volume clone
 	sasTokenExpirationMinutes int
+	// timeout in minutes for waiting for azcopy to finish
+	waitForAzCopyTimeoutMinutes int
 	// azcopy for provide exec mock for ut
 	azcopy *util.Azcopy
 }
@@ -248,6 +252,7 @@ func NewDriver(options *DriverOptions, kubeClient kubernetes.Interface, cloud *p
 		mountPermissions:                       options.MountPermissions,
 		enableAznfsMount:                       options.EnableAznfsMount,
 		sasTokenExpirationMinutes:              options.SasTokenExpirationMinutes,
+		waitForAzCopyTimeoutMinutes:            options.WaitForAzCopyTimeoutMinutes,
 		azcopy:                                 &util.Azcopy{},
 		KubeClient:                             kubeClient,
 		cloud:                                  cloud,

diff --git a/pkg/blob/blob_test.go b/pkg/blob/blob_test.go
@@ -50,12 +50,13 @@ const (
 
 func NewFakeDriver() *Driver {
 	driverOptions := DriverOptions{
-		NodeID:                  fakeNodeID,
-		DriverName:              DefaultDriverName,
-		BlobfuseProxyEndpoint:   "",
-		EnableBlobfuseProxy:     false,
-		BlobfuseProxyConnTimout: 5,
-		EnableBlobMockMount:     false,
+		NodeID:                      fakeNodeID,
+		DriverName:                  DefaultDriverName,
+		BlobfuseProxyEndpoint:       "",
+		EnableBlobfuseProxy:         false,
+		BlobfuseProxyConnTimout:     5,
+		WaitForAzCopyTimeoutMinutes: 1,
+		EnableBlobMockMount:         false,
 	}
 	driver := NewDriver(&driverOptions, nil, &azure.Cloud{})
 	driver.Name = fakeDriverName
@@ -79,12 +80,13 @@ func TestNewFakeDriver(t *testing.T) {
 
 func TestNewDriver(t *testing.T) {
 	driverOptions := DriverOptions{
-		NodeID:                  fakeNodeID,
-		DriverName:              DefaultDriverName,
-		BlobfuseProxyEndpoint:   "",
-		EnableBlobfuseProxy:     false,
-		BlobfuseProxyConnTimout: 5,
-		EnableBlobMockMount:     false,
+		NodeID:                      fakeNodeID,
+		DriverName:                  DefaultDriverName,
+		BlobfuseProxyEndpoint:       "",
+		EnableBlobfuseProxy:         false,
+		BlobfuseProxyConnTimout:     5,
+		WaitForAzCopyTimeoutMinutes: 1,
+		EnableBlobMockMount:         false,
 	}
 	driver := NewDriver(&driverOptions, nil, &azure.Cloud{})
 	fakedriver := NewFakeDriver()

diff --git a/pkg/blob/controllerserver.go b/pkg/blob/controllerserver.go
@@ -60,8 +60,7 @@ const (
 	SPN                             = "SPN"
 	authorizationPermissionMismatch = "AuthorizationPermissionMismatch"
 
-	waitForCopyInterval = 5 * time.Second
-	waitForCopyTimeout  = 3 * time.Minute
+	waitForAzCopyInterval = 2 * time.Second
 )
 
 // CreateVolume provisions a volume
@@ -735,8 +734,8 @@ func (d *Driver) copyBlobContainer(req *csi.CreateVolumeRequest, accountSasToken
 		return fmt.Errorf("srcContainerName(%s) or dstContainerName(%s) is empty", srcContainerName, dstContainerName)
 	}
 
-	timeAfter := time.After(waitForCopyTimeout)
-	timeTick := time.Tick(waitForCopyInterval)
+	timeAfter := time.After(time.Duration(d.waitForAzCopyTimeoutMinutes) * time.Minute)
+	timeTick := time.Tick(waitForAzCopyInterval)
 	srcPath := fmt.Sprintf("https://%s.blob.%s/%s%s", accountName, storageEndpointSuffix, srcContainerName, accountSasToken)
 	dstPath := fmt.Sprintf("https://%s.blob.%s/%s%s", accountName, storageEndpointSuffix, dstContainerName, accountSasToken)