Question

I’m using SQL Azure for Blob Meta Data Storage and Azure Blob Storage for the actual blobs. Blob Creation/Deletion is implemented by enlisting those operation in the ambient TransactionScope. Everything works fine so far but I’m wondering if anyone can recommend optimizations to the Delete-Operation (see source code below) that might get-rid of the requirement to download the blob contents in order to rollback.

public class CloudBlobDeletionEnlistment : CloudBlobBaseEnlistment,
  IEnlistmentNotification,
  IDisposable
{
  public CloudBlobDeletionEnlistment(Guid ownerId, string blobId, CloudBlobContainer container, Logger logger, IUserUploadActivity currentUploadActivity)
  {
    ctx = new Context { OwnerId = ownerId, BlobId = blobId, Container = container, Logger = logger, CurrentUploadActivity = currentUploadActivity };
  }

  public ~CloudBlobDeletionEnlistment()
  {
    Dispose(false);
  }

  public class Context
  {
    public Guid OwnerId;
    public string BlobId;
    public string ContentFileName;
    public string MimeType;
    public bool IsCompressed;
    public CloudBlobContainer Container;
    public Logger Logger;
    public IUserUploadActivity CurrentUploadActivity;
  }

  private readonly Context ctx;
  private CloudBlob blob;

  public void Prepare(PreparingEnlistment preparingEnlistment)
  {
    blob = ctx.Container.GetBlobReference(ctx.BlobId);

    // save backup information
    ctx.ContentFileName = Path.GetTempFileName();
    blob.DownloadToFile(ctx.ContentFileName);
    blob.FetchAttributes();
    ctx.MimeType = blob.Metadata[Constants.BlobMetaAttributeContentType];
    ctx.IsCompressed = bool.Parse(blob.Metadata[Constants.BlobMetaAttributeCompressed]);

    // delete it
    blob.DeleteIfExists();

    // done
    preparingEnlistment.Prepared();
  }

  public void Commit(Enlistment enlistment)
  {
    Cleanup();

    // done
    enlistment.Done();
  }

  public void Rollback(Enlistment enlistment)
  {
    if (blob != null)
    {
      try
      {
        blob.UploadFile(ctx.ContentFileName);
        blob.Metadata[Constants.BlobMetaAttributeContentType] = ctx.MimeType;
        blob.Metadata[Constants.BlobMetaAttributeCompressed] = ctx.IsCompressed.ToString();
        blob.SetMetadata();
      }

      finally
      {
        Cleanup();
      }
    }

    else Cleanup();

    // done
    enlistment.Done();
  }

  public void InDoubt(Enlistment enlistment)
  {
    Cleanup();

    enlistment.Done();
  }

  void Cleanup()
  {
    // delete the temporary file holding the blob content
    if (!string.IsNullOrEmpty(ctx.ContentFileName) && File.Exists(ctx.ContentFileName))
    {
      File.Delete(ctx.ContentFileName);
      ctx.ContentFileName = null;
    }
  }

  public void Dispose()
  {
    Dispose(true);
    GC.SuppressFinalize(this);
  }

  protected virtual void Dispose(bool disposing)
  {
    if (disposing)
    {
      // free managed resources
    }

    // free native resources if there are any.
    Cleanup();
  }
  #endregion
}

Answer 1

This doesn’t seem like a safe rollback mechanism to me – the upload could fail and if that occurs then your data consistency would be broken.

What about if you delete your blobs by putting their names into a ToBeDeleted table in the database – and then you set up some regular job deletes these blobs from time to time?