This is an application error, not a Tomcat bug. I have confirmed that this works as expected. The users mailing list is the place to seek further help.

Created attachment 27821 [details]
Test case

This the war file that contains a simple servlet filter that uses servelet 3.0 async api to echo the posted content. 

package test;

import java.io.IOException;
import java.io.Reader;

import javax.servlet.AsyncContext;
import javax.servlet.Filter;
import javax.servlet.FilterChain;
import javax.servlet.FilterConfig;
import javax.servlet.ServletException;
import javax.servlet.ServletRequest;
import javax.servlet.ServletResponse;
import javax.servlet.http.HttpServletRequest;

public class TestAsyncFilter implements Filter {

    @Override
    public void destroy() {

    }

    @Override
    public void doFilter(ServletRequest request, ServletResponse response, FilterChain chain) throws IOException,
        ServletException {
        HttpServletRequest httpServletRequest = (HttpServletRequest)request;
        String path = httpServletRequest.getServletPath();
        if ("POST".equalsIgnoreCase(httpServletRequest.getMethod()) && path.startsWith("/test")) {
            final AsyncContext asyncContext = request.startAsync();
            asyncContext.start(new Runnable() {

                @Override
                public void run() {
                    try {
                        readContent(asyncContext);
                    } catch (IOException e) {
                        e.printStackTrace();
                    }
                }
            });

        } else {
            chain.doFilter(request, response);
        }
    }

    private void readContent(AsyncContext asyncContext) throws IOException {
        HttpServletRequest httpServletRequest = (HttpServletRequest)asyncContext.getRequest();
        Reader reader = httpServletRequest.getReader();
        StringBuilder sb = new StringBuilder();
        char[] buf = new char[4096];
        while (true) {
            int len = reader.read(buf);
            if (len < 0) {
                break;
            } else {
                sb.append(buf, 0, len);
            }
        }
        System.out.println("Request: " + sb.toString());
        asyncContext.getResponse().getWriter().println(sb.toString());
        asyncContext.complete();
    }

    @Override
    public void init(FilterConfig config) throws ServletException {

    }

}


And the test client is:

package test;

import java.io.IOException;

import org.apache.http.HttpResponse;
import org.apache.http.client.HttpClient;
import org.apache.http.client.methods.HttpPost;
import org.apache.http.entity.StringEntity;
import org.apache.http.impl.client.DefaultHttpClient;
import org.apache.http.util.EntityUtils;

public class TestHttpClient {
    public static void main(String[] args) throws IOException {
        HttpClient client = new DefaultHttpClient();
        for (int i = 0; i < 5; i++) {
            HttpPost post = new HttpPost("http://localhost:8080/tomcat7-async/test");
            StringEntity entity = new StringEntity("Test String: " + i);
            entity.setChunked(true);
            post.setEntity(entity);
            HttpResponse response = client.execute(post);
            System.out.println("Response[" + i + "]: " + EntityUtils.toString(response.getEntity()));
        }
    }
}

I would like to reopen the bug with the test case to prove I was wrong :-).

Thanks,
Raymond

To reproduce the problem, deploy the WAR and run the test.TestHttpClient to send chunked post.

BTW, the same application works well under Jetty 8.x. 

I also tried to convert the filter to a async servlet and the same behavior happened. Only the first request was processed correctly and other ones received empty content.

Yep, I can reproduce this now with the additional information. Previously, it looked like the standard getParameter() confusion.

Thanks for the test case. That made tracking this down a lot easier than it could have been.

The problem has been fixed in trunk (8.0.x) and 7.0.x and will be included in 7.0.23 onwards.

I am still facing this issue. My test cases are failing randomly when I run my application with <async-supported>true</async-supported>.
Everything work fine when client is not using chunked encoding. 
Also, everything work fine when run with <async-supported>false</async-supported>

I have debugged the problem and this bug report helped me in finding root cause. I found that ChunkedInputFilter is getting reused even when keepalive=false. And my test cases start failing after that point.
So, we need to recycle it in that case. 

Issue was fixed by modifying the AbstractHttp11Processor file.
Replaced lines #1538 to #1544
          
            if (!keepAlive) {
                return SocketState.CLOSED;
            } else {
                getInputBuffer().nextRequest();
                getOutputBuffer().nextRequest();
                return SocketState.OPEN;
            }
With
            getInputBuffer().nextRequest();
            if (!keepAlive) {
                return SocketState.CLOSED;
            } else {
                getOutputBuffer().nextRequest();
                return SocketState.OPEN;
            }

I was testing against 7.0.26. Will test with trunk and submit patch if required.
We can not recycle output buffer at this stage as it has not completed flushing data.
Let me know if there is a better place to handle this. Will test it and then post a patch.

I don't have a simple test case to reproduce the issue.
Let me know what else I can do to help you fix this issue. 

Thanks,
Sudhan

Thanks for the research. That pointed me in the right direction. This was fixed in r1334787 for trunk and r1334790 for 7.0.x and will be included in 7.0.28 onwards.

If you are able to confirm that the fix works, that would be great.

Yes, It is fixed.
I have tested in 7.0.x. It is working fine.
I was getting some NPE's in InternalNioInputBuffer with my fix. That's why did not post patch.
Not getting any exception with latest code from 7.0.x.

Thanks,
Sudhan

Got this issue again. :(
Earlier I was running with protocol="HTTP/1.1"
This time I got error with protocol="org.apache.coyote.http11.Http11NioProtocol"

Shall I reopen?

I was running single client. That is one request at a time.
Got exception in client and client closed socket.
Got following exception is server.

java.io.EOFException: Unexpected EOF read on the socket
	at org.apache.coyote.http11.InternalNioInputBuffer.readSocket(InternalNioInputBuffer.java:455)
	at org.apache.coyote.http11.InternalNioInputBuffer.fill(InternalNioInputBuffer.java:808)
	at org.apache.coyote.http11.InternalNioInputBuffer$SocketInputBuffer.doRead(InternalNioInputBuffer.java:833)
	at org.apache.coyote.http11.filters.ChunkedInputFilter.readBytes(ChunkedInputFilter.java:271)
	at org.apache.coyote.http11.filters.ChunkedInputFilter.parseCRLF(ChunkedInputFilter.java:355)
	at org.apache.coyote.http11.filters.ChunkedInputFilter.doRead(ChunkedInputFilter.java:147)
	at org.apache.coyote.http11.AbstractInputBuffer.doRead(AbstractInputBuffer.java:346)
	at org.apache.coyote.Request.doRead(Request.java:422)
	at org.apache.catalina.connector.InputBuffer.realReadBytes(InputBuffer.java:290)
	at org.apache.tomcat.util.buf.ByteChunk.substract(ByteChunk.java:431)
	at org.apache.catalina.connector.InputBuffer.read(InputBuffer.java:315)
	at org.apache.catalina.connector.CoyoteInputStream.read(CoyoteInputStream.java:200)

This is expected. But looks like filters were not recycled before next request.
Next request failed with

java.io.IOException: Invalid CRLF
	at org.apache.coyote.http11.filters.ChunkedInputFilter.parseCRLF(ChunkedInputFilter.java:366)
	at org.apache.coyote.http11.filters.ChunkedInputFilter.doRead(ChunkedInputFilter.java:147)
	at org.apache.coyote.http11.AbstractInputBuffer.doRead(AbstractInputBuffer.java:346)
	at org.apache.coyote.Request.doRead(Request.java:422)
	at org.apache.catalina.connector.InputBuffer.realReadBytes(InputBuffer.java:290)
	at org.apache.tomcat.util.buf.ByteChunk.substract(ByteChunk.java:431)
	at org.apache.catalina.connector.InputBuffer.read(InputBuffer.java:315)
	at org.apache.catalina.connector.CoyoteInputStream.read(CoyoteInputStream.java:200)

Re-open to investigate

Full stack traces please.

Rest of the stack trace is my application code.
I am starting async processing with

if (request.isAsyncSupported()) {
	final AsyncContext asyncCtx = request.startAsync(request, response);
	final Runnable command = new AsyncRequestProcessor(asyncCtx, this, this.actionObjects);
	Executor executor = (Executor)request.getServletContext().getAttribute(Constants.ASYNC_REQUEST_EXECUTOR);
	executor.execute(command);
}

Next line on stack trace is 
ServletUtils.getBytesOfStream(ServletUtils.java:425)

Code of same is
public static ByteArrayInputStreamExt getBytesOfStream(final Map<String, Object> helperMap, final InputStream stream, int len, boolean chunked) throws IOException {
	try {
		ByteArrayInputStreamExt bais = (ByteArrayInputStreamExt) helperMap.get(Constants.DATA);
		byte[] tempBuf = null;
		if (bais == null) {
			tempBuf = new byte[len];
			bais = new ByteArrayInputStreamExt(tempBuf);
		} else {
			tempBuf = bais.getBuffer();
		}
		int readSoFar = 0;
		int bytesRead = 0;
		while (len > readSoFar) {
			bytesRead = stream.read(tempBuf, readSoFar, (len - readSoFar));
			if (bytesRead == -1) {
				if (chunked) {
					if (readSoFar == 0) {
						bais.reset(0);
						return bais;
					} else {
						bais.reset(readSoFar);
						return bais;
					}
				}
				throw new IOException("Failed to read from inputstream");
			}
			readSoFar += bytesRead;
		}
		bais.reset(len);
		return bais;
	} catch (OutOfMemoryError th) {
		logger.error("Creating a byte array of length: " + len + " has resulted in OOM error: " + th.getMessage());
	}
	return null;
}

And this method is called with

ServletUtils.getBytesOfStream(request.getInputStream(), length);

This method is called in a loop reading 10MB at a time. For chunked, loop stops with EOF. The ByteArrayInputStreamExt handling is for reusing the byte[].

This happens only when client aborts.

Let me know if you need anymore info. Which class should I check if I have to debug this?

Thanks,
Sudhan

I think the issue is because of ChunkedInputFilter.needCRLFParse not being reset.
I think we need to do needCRLFParse = false; in ChunkedInputFilter.recycle().

That looks like the cause to me to. This has been fixed in trunk and 7.0.x and will be included in 7.0.28 onwards.

(In reply to comment #14)

Thank you. Fixed this in 6.0 as well, will be in 6.0.36.

(In reply to comment #14)

The needCRLFParse fix was backported to 5.5 in r1359748 and will be in 5.5.36.