Fix AWS eventual consistency of AttachDisk

AWS eventual consistency can go back in time. It can return that a volume
is detached and then that it is attached.

When this happens during attachment of the same volume to the same node,
but with a different device name, retry DescribeVolumes a few times before
reporting an error. 10 retries should be enough to get a consistent result.

Author: jsafrane

staging/src/k8s.io/legacy-cloud-providers/aws/aws.go

@@ -2105,17 +2105,18 @@ func (c *Cloud) applyUnSchedulableTaint(nodeName types.NodeName, reason string)
 
 // waitForAttachmentStatus polls until the attachment status is the expected value
 // On success, it returns the last attachment state.
-func (d *awsDisk) waitForAttachmentStatus(status string) (*ec2.VolumeAttachment, error) {
+func (d *awsDisk) waitForAttachmentStatus(status string, expectedDevice string) (*ec2.VolumeAttachment, error) {
 	backoff := wait.Backoff{
 		Duration: volumeAttachmentStatusPollDelay,
 		Factor:   volumeAttachmentStatusFactor,
 		Steps:    volumeAttachmentStatusSteps,
 	}
 
-	// Because of rate limiting, we often see errors from describeVolume
+	// Because of rate limiting, we often see errors from describeVolume.
+	// Or AWS eventual consistency returns unexpected data.
 	// So we tolerate a limited number of failures.
-	// But once we see more than 10 errors in a row, we return the error
-	describeErrorCount := 0
+	// But once we see more than 10 errors in a row, we return the error.
+	errorCount := 0
 
 	// Attach/detach usually takes time. It does not make sense to start
 	// polling DescribeVolumes before some initial delay to let AWS
@@ -2144,8 +2145,8 @@ func (d *awsDisk) waitForAttachmentStatus(status string) (*ec2.VolumeAttachment,
 					return false, err
 				}
 			}
-			describeErrorCount++
-			if describeErrorCount > volumeAttachmentStatusConsecutiveErrorLimit {
+			errorCount++
+			if errorCount > volumeAttachmentStatusConsecutiveErrorLimit {
 				// report the error
 				return false, err
 			}
@@ -2154,8 +2155,6 @@ func (d *awsDisk) waitForAttachmentStatus(status string) (*ec2.VolumeAttachment,
 			return false, nil
 		}
 
-		describeErrorCount = 0
-
 		if len(info.Attachments) > 1 {
 			// Shouldn't happen; log so we know if it is
 			klog.Warningf("Found multiple attachments for volume %q: %v", d.awsID, info)
@@ -2177,11 +2176,29 @@ func (d *awsDisk) waitForAttachmentStatus(status string) (*ec2.VolumeAttachment,
 		if attachmentStatus == "" {
 			attachmentStatus = "detached"
 		}
+		if attachment != nil && expectedDevice != "" {
+			device := aws.StringValue(attachment.Device)
+			if device != "" && device != expectedDevice {
+				// AWS eventual consistency can go back in time.
+				// For example, we're waiting for a volume to be attached as /dev/xvdba, but AWS can tell us it's
+				// attached as /dev/xvdbb, where it was attached before and it was already detached.
+				// Retry couple of times, hoping AWS starts reporting the right status.
+				klog.Warningf("Expected device %s %s, but found device %s %s", expectedDevice, status, device, attachmentStatus)
+				errorCount++
+				if errorCount > volumeAttachmentStatusConsecutiveErrorLimit {
+					// report the error
+					return false, fmt.Errorf("attachment of disk %q failed: requested device %q but found %q", d.name, expectedDevice, device)
+				}
+				return false, nil
+			}
+		}
+
 		if attachmentStatus == status {
 			// Attachment is in requested state, finish waiting
 			return true, nil
 		}
 		// continue waiting
+		errorCount = 0
 		klog.V(2).Infof("Waiting for volume %q state: actual=%s, desired=%s", d.awsID, attachmentStatus, status)
 		return false, nil
 	})
@@ -2321,7 +2338,7 @@ func (c *Cloud) AttachDisk(diskName KubernetesVolumeID, nodeName types.NodeName)
 		klog.V(2).Infof("AttachVolume volume=%q instance=%q request returned %v", disk.awsID, awsInstance.awsID, attachResponse)
 	}
 
-	attachment, err := disk.waitForAttachmentStatus("attached")
+	attachment, err := disk.waitForAttachmentStatus("attached", ec2Device)
 
 	if err != nil {
 		if err == wait.ErrWaitTimeout {
@@ -2341,6 +2358,7 @@ func (c *Cloud) AttachDisk(diskName KubernetesVolumeID, nodeName types.NodeName)
 		return "", fmt.Errorf("unexpected state: attachment nil after attached %q to %q", diskName, nodeName)
 	}
 	if ec2Device != aws.StringValue(attachment.Device) {
+		// Already checked in waitForAttachmentStatus(), but just to be sure...
 		return "", fmt.Errorf("disk attachment of %q to %q failed: requested device %q but found %q", diskName, nodeName, ec2Device, aws.StringValue(attachment.Device))
 	}
 	if awsInstance.awsID != aws.StringValue(attachment.InstanceId) {
@@ -2398,7 +2416,7 @@ func (c *Cloud) DetachDisk(diskName KubernetesVolumeID, nodeName types.NodeName)
 		return "", errors.New("no response from DetachVolume")
 	}
 
-	attachment, err := diskInfo.disk.waitForAttachmentStatus("detached")
+	attachment, err := diskInfo.disk.waitForAttachmentStatus("detached", "")
 	if err != nil {
 		return "", err
 	}