Upload blob async frozen when using memory stream, no error

[diegosasw]

I'm experiencing a problem with Azure.Storage.Blobs 12.4.0
No error, just frozen forever when trying to upload a stream (an in-memory json string) using Azurite.

It seems the server receives the attempt, but does nothing and the API does not throw any error either:

sqlserver_1  | 172.18.0.1 - - [24/Mar/2020:01:38:55 +0000] "PUT /devstoreaccount1/landing/test637206559996473201.json?st=2020-03-23T13%3A57%3A00Z&se=2020-03-25T13%3A57%3A37Z&sp=racwdl&sv=2018-03-28&sr=c&sig=RYDvl1aQeEIY3isg17LGGtx63Krum13gTXgn981cOco%3D HTTP/1.1" - -

Azurite is running, I can access with the Azure Blob Storage Explorer and generate a sas token uri pasted below.

My code:

var fileName = $"test{DateTime.UtcNow.Ticks}.json";
var fileStream = new MemoryStream();
fileStream.Write(Encoding.UTF8.GetBytes("{'foo': 'bar'}"));
var blobContainerClient = new BlobContainerClient("http://127.0.0.1:10000/devstoreaccount1/landing?st=2020-03-23T13%3A57%3A00Z&se=2020-03-25T13%3A57%3A37Z&sp=racwdl&sv=2018-03-28&sr=c&sig=RYDvl1aQeEIY3isg17LGGtx63Krum13gTXgn981cOco%3D");
await blobContainerClient.UploadBlobAsync(fileName, fileStream); //this gets stuck, no error, just frozen there

PS: The Azurite I use for the test is:

version: '3'
services:
    sqlserver:
        image: mcr.microsoft.com/azure-storage/azurite
        restart: always
        ports:
            - 10000:10000
            - 10001:10001
# Run
# docker-compose -f azurite.yml up -d 

Can anybody spot what the issue might be?

[diegosasw]

I've also tried with a real azure blob storage with a sas uri:

https://hidden.blob.core.windows.net/prueba?st=2020-03-23T14%3A24%3A00Z&se=2020-03-26T14%3A24%3A00Z&sp=racwdl&sv=2018-03-28&sr=c&sig=DLVa3Lbg8JKZlCGRSDs1M%2F69Cc4s9F7keFEE1roRScA%3D

same result, so it does not seem related to Azurite, it must be something in my code sample or with the Azure.Storage.Blobs 12.4.0

[diegosasw]

Just ran another test with a different stream. This time I read a PDF as a FileStream
and it works fine for both real Azure Blob Storage and Azurite.
So the problem is with the memory stream

var fileStream = new MemoryStream();
fileStream.Write(Encoding.UTF8.GetBytes("{'foo': 'bar'}"));

Not sure whether the stream is not properly created and opened to be read, but in any case maybe it should throw some error and not freeze.

The sample below works fine.

var fileName = $"test{DateTime.UtcNow.Ticks}.pdf";
var fileStream = new FileStream("TestSupport\\sample.pdf", FileMode.Open, FileAccess.Read);
var blobContainerClient = new BlobContainerClient("http://127.0.0.1:10000/devstoreaccount1/landing?st=2020-03-23T13%3A57%3A00Z&se=2020-03-25T13%3A57%3A37Z&sp=racwdl&sv=2018-03-28&sr=c&sig=RYDvl1aQeEIY3isg17LGGtx63Krum13gTXgn981cOco%3D");
await blobContainerClient.UploadBlobAsync(fileName, fileStream); //it works!

[ghost]

Thanks for the feedback! We are routing this to the appropriate team for follow-up. cc @xgithubtriage.

[DarrenClevenger]

I was also having this issue. I found that I had to reset the position to 0 to get this to work.

using var ms = new MemoryStream(); 
ms.Write(ASCIIEncoding.UTF8.GetBytes(jsonData));  
ms.Position = 0;
client.UploadBlob(path, ms);

[francescocristallo]

I was also having this issue. I found that I had to reset the position to 0 to get this to work.

using var ms = new MemoryStream(); 
ms.Write(ASCIIEncoding.UTF8.GetBytes(jsonData));  
ms.Position = 0;
client.UploadBlob(path, ms);

I have the same problem with the DataLake Gen2 APIs.
Resetting the position to 0 worked for me.

[pakrym]

Seems like a bug, when the stream position is not reset the expected behavior should be uploading an empty blob, not hanging.

[amishra-dev]

@kasobol-msft can you please triage

[seanmcc-msft]

I believe we could fix this issue with:

if (content?.CanSeek)
{
    content.Position = 0;
}

In these places:

• https://github.com/Azure/azure-sdk-for-net/blob/master/sdk/storage/Azure.Storage.Blobs/src/PartitionedUploader.cs#L105
• https://github.com/Azure/azure-sdk-for-net/blob/master/sdk/storage/Azure.Storage.Files.DataLake/src/DataLakePartitionedUploader.cs#L103
• https://github.com/Azure/azure-sdk-for-net/blob/master/sdk/storage/Azure.Storage.Files.DataLake/src/DataLakeFileClient.cs#L1733
• https://github.com/Azure/azure-sdk-for-net/blob/master/sdk/storage/Azure.Storage.Blobs/src/BlockBlobClient.cs#L782
• https://github.com/Azure/azure-sdk-for-net/blob/master/sdk/storage/Azure.Storage.Files.Shares/src/ShareFileClient.cs#L3168

[pakrym]

Not sure if it's appropriate to rewind customers stream. Does MemoryStream.Read hang when it's at the end?

AFAIK it returns 0 immediately https://github.com/dotnet/runtime/blob/master/src/libraries/System.Private.CoreLib/src/System/IO/MemoryStream.cs#L354

[seanmcc-msft]

I'm seen MemoryStream.Read() hang when it's at the end. We can reproduce fairly easily by writing data to a memory stream, and then attempting to upload it without rewinding it.

[pakrym]

Just tried it,

using System;

namespace ms_read
{
    class Program
    {
        static void Main(string[] args)
        {
            var s = new System.IO.MemoryStream();
            var bytes = new byte[100];
            for (int i = 0; i < 10 ; i++)
            {
                Console.WriteLine(s.Read(bytes));
            }
        }
    }
}

Outputs:

D:\temp\ms-read> dotnet run
0
0
0
0
0
0
0
0
0
0

[seanmcc-msft]

        [Test]
        public async Task UploadAsync_MemoryStream()
        {
            // Arrange
            await using DisposingContainer test = await GetTestContainerAsync();

            var blockBlobName = GetNewBlobName();
            BlockBlobClient blob = InstrumentClient(test.Container.GetBlockBlobClient(blockBlobName));
            var data = GetRandomBuffer(Size);
            MemoryStream stream = new MemoryStream();
            stream.Write(data, 0, data.Length);

            // Act - hangs here.
            await blob.UploadAsync(content: stream);
        }

We end up calling _request.Content = Azure.Core.RequestContent.Create(body); when creating a request based on the stream.

[pakrym]

I'm not arguing that upload doesn't hang, just wonder if it hangs on memoryStream.Read or something else like our parallel upload logic.

[seanmcc-msft]

Try MemoryStream.Read() on a non-empty MemoryStream where offset == length.

[pakrym]

 var bytes = new byte[100];
            var s = new System.IO.MemoryStream(bytes);

            for (int i = 0; i < 5 ; i++)
            {
                Console.WriteLine($"{s.Position} {s.Length}");
                Console.WriteLine(s.Read(bytes));
            }

D:\temp\ms-read> dotnet run
0 100
100
100 100
0
100 100
0
100 100
0
100 100
0

[seanmcc-msft]

Hmm maybe we are seeking outside the range of the MemoryStream or something in our Upload logic.

[ANRCorleone]

I was also having this issue. I found that I had to reset the position to 0 to get this to work.

using var ms = new MemoryStream(); 
ms.Write(ASCIIEncoding.UTF8.GetBytes(jsonData));  
ms.Position = 0;
client.UploadBlob(path, ms);

I have the same problem with the DataLake Gen2 APIs.
Resetting the position to 0 worked for me.

Have the same issue - about an hour saved. Thank you for this!

[seanmcc-msft]

@pekspro suggested that we check the stream position in UploadAsync(), and throw an exception if Position != 0. I think this would be a good idea, to reduce the ambiguity around this situation.

@pakrym, would you be ok with this?

[pakrym]

No. Uploading a stream starting from a arbitrary position is a totally normal practice in .net and we shouldn't artificially limit it. In addition some streams don't track their position and throw instead.

I'm also worried that it would hide a real issue we have.

[seanmcc-msft]

I think we should do something to address this, the current user experience is non-optimal.

What if we check if stream.Position == stream.Length?

[pakrym]

We should find why we hang instead of exiting the upload loop when Stream.Read returns 0.

[seanmcc-msft]

This problem also occurs for small streams, that we attempt to upload in a single request. Is this an issue in other SDKs based on Core?

[pakrym]

No others that I know of.

[pekspro]

As I understand it, this is what happens:

In BlobRestClient there is this line:

_request.Headers.SetValue("Content-Length", 
          contentLength.ToString(System.Globalization.CultureInfo.InvariantCulture));

This setup the content-length of the request. This will be incorrect when Position in the stream is not zero.

Then in HttpClientTransport.ProcessAsync there is this line:

responseMessage = await _client.SendAsync(httpRequest, HttpCompletionOption.ResponseHeadersRead, message.CancellationToken)
                    .ConfigureAwait(false);

This will write the stream to the request. But since the expected length is incorrect it never completes.

I am sure it is more complicated than this, but this I what I found anyway :-)

I have created a PR #12901 that fixes, or at least reduces this problem.