possible bug (deadlock) when posting batch request content over a certain length

[dan-drewes]

There seems to be a bug with posting batch request content over a certain length (greater than DEFAULT_PIPE_SIZE = 1024 of PipedInputStream).

Calling graphClient.getBatchRequestBuilder().post(batchRequestContent, null)) goes into a deadlock.

Could you please check this?

Expected behavior

Graph API executes the batch request and returns some kind of response.

Actual behavior

Deadlock in

PipedInputStream.awaitSpace() line: 273
PipedInputStream.receive(byte[], int, int) line: 231
PipedOutputStream.write(byte[], int, int) line: 149
ByteArrayOutputStream.writeTo(OutputStream) line: 167
BatchRequestContent.getBatchRequestContent() line: 177
CustomBatchRequestBuilder(BatchRequestBuilder).toPostRequestInformation(BatchRequestContent) line: 84
CustomBatchRequestBuilder(BatchRequestBuilder).post(BatchRequestContent, Map<String,ParsableFactory>) line: 49
CustomBatchRequestBuilder.post(BatchRequestContent, Map<String,ParsableFactory>) line: 41
...

when calling

final BatchResponseContent batchResponseContent = Objects.requireNonNull(
            graphClient.getBatchRequestBuilder().post(batchRequestContent, null));

Steps to reproduce the behavior

Maven pom.xml

<dependency>
	      <groupId>com.microsoft.graph</groupId>
	      <artifactId>microsoft-graph</artifactId>
	      <version>6.8.0</version>
</dependency>
<dependency>
	      <groupId>com.azure</groupId>
	      <artifactId>azure-identity</artifactId>
	      <version>1.12.1</version>
</dependency>

Code:

// Create the batch request content with the steps
final BatchRequestContent batchRequestContent = new BatchRequestContent(
		graphClient);

//add some request steps to increase content length
for (int i = 1; i <= 9; i++)
{
	// Use the Graph client to generate the requestInformation object for GET /me
	final RequestInformation meRequestInformation = graphClient.me()
			.toGetRequestInformation();

	// Add the requestInformation objects to the batch request content
	batchRequestContent
			.addBatchRequestStep(meRequestInformation);
}

try
{
	// Send the batch request content to the /$batch endpoint
	final BatchResponseContent batchResponseContent = Objects.requireNonNull(
			graphClient.getBatchRequestBuilder().post(batchRequestContent, null));

	System.out.println("response received");

}
catch (IOException e)
{
	e.printStackTrace();
}

[kbullock-re]

Also experiencing this issue. Any advice would be appreciated.

[cschenkeltherolf]

Also hitting this with BatchRequestContent.getBatchRequestContent(), same issue with the PipedInputStream. v3 was very much unaffected.

JavaDoc advises against using a PipedInputStream and PipedOutputStream on the same thread for this very reason.

https://docs.oracle.com/javase/8/docs/api/java/io/PipedInputStream.html

For my purposes, this was sufficient, as I'm writing the content of the BatchRequestContent to my own request container and emitting it through our own IO.

return outputStream.toString(StandardCharsets.UTF_8);

[petrhollayms]

@Ndiritu BatchRequestContent shall not be using PipedInputStream, could you please check?