Protocol failures

[alranel]

Hello, using RabbitMQ server 2.7.1 I am getting failures when enabling heartbeat. If I set to, say, 30 seconds, my consumer daemon dies with the following message:

Bad frame read. at /opt/ed-test/bin/../../common/lib/ED/App/Daemon/MessageConsumer.pm line 98.

I tried to capture the traffic using Wireshark, and it appears that the failure happens when the server sends the third heartbeat while connection is inactive.

I forked Net::RabbitMQ from GitHub to contribute a patch, but then I found more problems. It appears that when using Net::RabbitMQ, the server sends a malformed Channel.Open-Ok response. The RabbitMQ Tracer dies on it:

1326806094072: <Tracer-0> ch#0 <- {#method<connection.start>(version-major=8, version-minor=0, server-properties={product=RabbitMQ, information=Licensed under the MPL.  See http://www.rabbitmq.com/, platform=Erlang/OTP, capabilities={}, copyright=Copyright (C) 2007-2011 VMware, Inc., version=2.7.1}, mechanisms=PLAIN AMQPLAIN, locales=en_US), null, "[B@67006d75"}
1326806094074: <Tracer-0> ch#0 -> {#method<connection.start-ok>(client-properties={}, mechanism=PLAIN, response=guestguest, locale=en_US), null, "[B@67006d75"}
1326806094074: <Tracer-0> ch#0 <- {#method<connection.tune>(channel-max=0, frame-max=131072, heartbeat=0), null, "[B@67006d75"}
1326806094108: <Tracer-0> ch#0 -> {#method<connection.tune-ok>(channel-max=0, frame-max=131072, heartbeat=30), null, "[B@67006d75"}
1326806094109: <Tracer-0> ch#0 -> {#method<connection.open>(virtual-host=/, capabilities=, insist=true), null, "[B@67006d75"}
1326806094149: <Tracer-0> ch#0 <- {#method<connection.open-ok>(known-hosts=), null, "[B@67006d75"}
1326806094188: <Tracer-0> ch#1 -> {#method<channel.open>(out-of-band=), null, "[B@67006d75"}
1326806094231: <Tracer-0> uncaught java.io.EOFException
    at java.io.DataInputStream.readInt(DataInputStream.java:392)
    at com.rabbitmq.client.impl.ValueReader.readBytes(ValueReader.java:86)
    at com.rabbitmq.client.impl.ValueReader.readLongstr(ValueReader.java:103)
    at com.rabbitmq.client.impl.ValueReader.readLongstr(ValueReader.java:111)
    at com.rabbitmq.client.impl.MethodArgumentReader.readLongstr(MethodArgumentReader.java:73)
    at com.rabbitmq.client.impl.AMQImpl$Channel$OpenOk.<init>(AMQImpl.java:600)
    at com.rabbitmq.client.impl.AMQImpl.readMethodFrom(AMQImpl.java:3242)
    at com.rabbitmq.client.impl.CommandAssembler.consumeMethodFrame(CommandAssembler.java:93)
    at com.rabbitmq.client.impl.CommandAssembler.handleFrame(CommandAssembler.java:158)
    at com.rabbitmq.client.impl.AMQCommand.handleFrame(AMQCommand.java:87)
    at com.rabbitmq.tools.Tracer$DirectionHandler.doFrame(Tracer.java:331)
    at com.rabbitmq.tools.Tracer$DirectionHandler.run(Tracer.java:343)
    at java.lang.Thread.run(Thread.java:636)

and Wireshark/tshark shows a Malformed Packet error:

  0.000980    127.0.0.1 -> 127.0.0.1    AMQP Protocol-Header 
  0.001148    127.0.0.1 -> 127.0.0.1    AMQP Connection.Start 
  0.001303    127.0.0.1 -> 127.0.0.1    AMQP Connection.Start-Ok 
  0.001405    127.0.0.1 -> 127.0.0.1    AMQP Connection.Tune 
  0.001511    127.0.0.1 -> 127.0.0.1    AMQP Connection.Tune-Ok 
  0.031825    127.0.0.1 -> 127.0.0.1    AMQP Connection.Open 
  0.031946    127.0.0.1 -> 127.0.0.1    AMQP Connection.Open-Ok 
  0.032313    127.0.0.1 -> 127.0.0.1    AMQP Channel.Open 
  0.032770    127.0.0.1 -> 127.0.0.1    AMQP [Malformed Packet]
  0.032822    127.0.0.1 -> 127.0.0.1    AMQP Tx.Select 
  0.032886    127.0.0.1 -> 127.0.0.1    AMQP Tx.Select-Ok 

This does not happen when using other client libraries (such the Java client library), so I suspect there's something wrong in what the Net::RabbitMQ library sends, that causes the server to send a malformed answer.

I tried to check Net::RabbitMQ source code, but I was unable to find what's wrong. To recap, two issues:

• Channel.Open-Ok from server is malformed (but it isn't with other client libraries)
• enabling heartbeats causes the app to die with "Bad frame read"

I'm now reverting to my own Net::RabbitMQ::Java, but I'd love to use Net::RabbitMQ instead, so let me know your thoughts and how can I help about this issue.