X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/c76e5bc93b066d93a668c4e00b52aa550028e1f4..be28c5f528a93ee32eef4c1dc2d0872cb718b29f:/lib/cloud/azure/azure.go

diff --git a/lib/cloud/azure/azure.go b/lib/cloud/azure/azure.go
index d2dbde96d7..7f949d9bdb 100644
--- a/lib/cloud/azure/azure.go
+++ b/lib/cloud/azure/azure.go
@@ -36,21 +36,23 @@ import (
 var Driver = cloud.DriverFunc(newAzureInstanceSet)
 
 type azureInstanceSetConfig struct {
-	SubscriptionID               string
-	ClientID                     string
-	ClientSecret                 string
-	TenantID                     string
-	CloudEnvironment             string
-	ResourceGroup                string
-	ImageResourceGroup           string
-	Location                     string
-	Network                      string
-	NetworkResourceGroup         string
-	Subnet                       string
-	StorageAccount               string
-	BlobContainer                string
-	DeleteDanglingResourcesAfter arvados.Duration
-	AdminUsername                string
+	SubscriptionID                 string
+	ClientID                       string
+	ClientSecret                   string
+	TenantID                       string
+	CloudEnvironment               string
+	ResourceGroup                  string
+	ImageResourceGroup             string
+	Location                       string
+	Network                        string
+	NetworkResourceGroup           string
+	Subnet                         string
+	StorageAccount                 string
+	BlobContainer                  string
+	SharedImageGalleryName         string
+	SharedImageGalleryImageVersion string
+	DeleteDanglingResourcesAfter   arvados.Duration
+	AdminUsername                  string
 }
 
 type containerWrapper interface {
@@ -373,6 +375,13 @@ func (az *azureInstanceSet) setup(azcfg azureInstanceSetConfig, dispatcherID str
 	return nil
 }
 
+func (az *azureInstanceSet) cleanupNic(nic network.Interface) {
+	_, delerr := az.netClient.delete(context.Background(), az.azconfig.ResourceGroup, *nic.Name)
+	if delerr != nil {
+		az.logger.WithError(delerr).Warnf("Error cleaning up NIC after failed create")
+	}
+}
+
 func (az *azureInstanceSet) Create(
 	instanceType arvados.InstanceType,
 	imageID cloud.ImageID,
@@ -410,7 +419,7 @@ func (az *azureInstanceSet) Create(
 		Tags:     tags,
 		InterfacePropertiesFormat: &network.InterfacePropertiesFormat{
 			IPConfigurations: &[]network.InterfaceIPConfiguration{
-				network.InterfaceIPConfiguration{
+				{
 					Name: to.StringPtr("ip1"),
 					InterfaceIPConfigurationPropertiesFormat: &network.InterfaceIPConfigurationPropertiesFormat{
 						Subnet: &network.Subnet{
@@ -437,7 +446,11 @@ func (az *azureInstanceSet) Create(
 	var storageProfile *compute.StorageProfile
 
 	re := regexp.MustCompile(`^http(s?)://`)
-	if re.MatchString(string(imageID)) && az.blobcont != nil {
+	if re.MatchString(string(imageID)) {
+		if az.blobcont == nil {
+			az.cleanupNic(nic)
+			return nil, wrapAzureError(errors.New("Invalid configuration: can't configure unmanaged image URL without StorageAccount and BlobContainer"))
+		}
 		blobname = fmt.Sprintf("%s-os.vhd", name)
 		instanceVhd := fmt.Sprintf("https://%s.blob.%s/%s/%s",
 			az.azconfig.StorageAccount,
@@ -458,10 +471,17 @@ func (az *azureInstanceSet) Create(
 				},
 			},
 		}
-	} else if az.blobcont == nil {
+	} else {
+		id := to.StringPtr("/subscriptions/" + az.azconfig.SubscriptionID + "/resourceGroups/" + az.imageResourceGroup + "/providers/Microsoft.Compute/images/" + string(imageID))
+		if az.azconfig.SharedImageGalleryName != "" && az.azconfig.SharedImageGalleryImageVersion != "" {
+			id = to.StringPtr("/subscriptions/" + az.azconfig.SubscriptionID + "/resourceGroups/" + az.imageResourceGroup + "/providers/Microsoft.Compute/galleries/" + az.azconfig.SharedImageGalleryName + "/images/" + string(imageID) + "/versions/" + az.azconfig.SharedImageGalleryImageVersion)
+		} else if az.azconfig.SharedImageGalleryName != "" || az.azconfig.SharedImageGalleryImageVersion != "" {
+			az.cleanupNic(nic)
+			return nil, wrapAzureError(errors.New("Invalid configuration: SharedImageGalleryName and SharedImageGalleryImageVersion must both be set or both be empty"))
+		}
 		storageProfile = &compute.StorageProfile{
 			ImageReference: &compute.ImageReference{
-				ID: to.StringPtr("/subscriptions/" + az.azconfig.SubscriptionID + "/resourceGroups/" + az.imageResourceGroup + "/providers/Microsoft.Compute/images/" + string(imageID)),
+				ID: id,
 			},
 			OsDisk: &compute.OSDisk{
 				OsType:       compute.Linux,
@@ -469,8 +489,6 @@ func (az *azureInstanceSet) Create(
 				CreateOption: compute.DiskCreateOptionTypesFromImage,
 			},
 		}
-	} else {
-		return nil, wrapAzureError(errors.New("Invalid configuration: can't configure unmanaged image URL without StorageAccount and BlobContainer"))
 	}
 
 	vmParameters := compute.VirtualMachine{
@@ -483,7 +501,7 @@ func (az *azureInstanceSet) Create(
 			StorageProfile: storageProfile,
 			NetworkProfile: &compute.NetworkProfile{
 				NetworkInterfaces: &[]compute.NetworkInterfaceReference{
-					compute.NetworkInterfaceReference{
+					{
 						ID: nic.ID,
 						NetworkInterfaceReferenceProperties: &compute.NetworkInterfaceReferenceProperties{
 							Primary: to.BoolPtr(true),
@@ -512,18 +530,21 @@ func (az *azureInstanceSet) Create(
 
 	vm, err := az.vmClient.createOrUpdate(az.ctx, az.azconfig.ResourceGroup, name, vmParameters)
 	if err != nil {
-		if az.blobcont != nil {
+		// Do some cleanup. Otherwise, an unbounded number of new unused nics and
+		// blobs can pile up during times when VMs can't be created and the
+		// dispatcher keeps retrying, because the garbage collection in manageBlobs
+		// and manageNics is only triggered periodically. This is most important
+		// for nics, because those are subject to a quota.
+		az.cleanupNic(nic)
+
+		if blobname != "" {
 			_, delerr := az.blobcont.GetBlobReference(blobname).DeleteIfExists(nil)
 			if delerr != nil {
 				az.logger.WithError(delerr).Warnf("Error cleaning up vhd blob after failed create")
 			}
 		}
-		// Leave cleaning up of managed disks to the garbage collection in manageDisks()
 
-		_, delerr := az.netClient.delete(context.Background(), az.azconfig.ResourceGroup, *nic.Name)
-		if delerr != nil {
-			az.logger.WithError(delerr).Warnf("Error cleaning up NIC after failed create")
-		}
+		// Leave cleaning up of managed disks to the garbage collection in manageDisks()
 
 		return nil, wrapAzureError(err)
 	}
@@ -656,6 +677,10 @@ func (az *azureInstanceSet) manageDisks() {
 	}
 
 	for ; response.NotDone(); err = response.Next() {
+		if err != nil {
+			az.logger.WithError(err).Warn("Error getting next page of disks")
+			return
+		}
 		for _, d := range response.Values() {
 			if d.DiskProperties.DiskState == compute.Unattached &&
 				d.Name != nil && re.MatchString(*d.Name) &&