[
https://issues.apache.org/jira/browse/QPID-6651?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14639049#comment-14639049
]
Robbie Gemmell commented on QPID-6651:
--------------------------------------
I had a quick play with this. I was able to make C Messenger send+recv examples
do what was described against the Java broker (though I had to use slightly
different size of messages), but not the C++ broker. I expected that when I
started, both due to what was reported here, and some stuff I have been looking
at this week which would likely explain the difference anyway...
Proton defaults to setting the max frame size value to its largest value
(either by specifying that directly, or not specifying a value and getting the
default). The C++ broker doesnt seem to set a specific limit itself so it gets
that default behaviour, which means that proton-based clients which do the same
will not actually split messages into multiple frames when sending to the C++
broker. Similarly, that means C++ broker wont split outgoing messages into
multiple frames for proton-based clients that havent secified a particular
frame size limit.
The Java broker on the other hand is setting the max frame size in its Open
frame to 32768, and so proton-based clients will fragment messages sent to the
Java broker if they would exceed that. The broker also similarly fragments
messages it is sending to proton-based clients though the client doesnt
strictly require it.
I agree with Robs observation that this could point to an issue with proton-c.
I tried to prove that by hacking proton-j (since I dont know where to do the
same in proton-c) such that the new JMS client using it would fragment messages
being sent to the C++ broker despite the brokers advertised max frame size not
actually requiring it. I expected that to fail but it didnt, the broker
received the multiple frames for the message and sent it back out (in 1 frame)
without issue, to both the JMS client and the C Messenger recv example.
I'm away for the next few days but I'll post on the proton list to point out
this seems like an proton-c issue (beign honest im about at the limit of my
proton-c debugging capability here anyway hehe).
Robbie
> Proton recv issue connecting to QPID Java Broker
> ------------------------------------------------
>
> Key: QPID-6651
> URL: https://issues.apache.org/jira/browse/QPID-6651
> Project: Qpid
> Issue Type: Bug
> Components: C++ Broker, Java Broker
> Affects Versions: 0.32
> Environment: Mac OS X 10.10.4, Java 1.8.0_45
> Centos 7, Java 1.8.0_45 (VM)
> Reporter: Philip Preston
> Attachments: qpid.log
>
>
> I have been testing the broker based functionality that is being added to
> OpenMAMA (www.openmama.org) through use of the proton-c (0.8) and have come
> across an issue I have been able to reproduce with the {{send}} / {{recv}}
> proton example applications.
> The issue is the the {{recv}} application is not able to receive large
> messages from the send application through the Java QPID Broker, but sending
> the same message through the C++ QPID Broker it works fine.
> This is reproducible as follows:
> * Java QPID Broker is setup relatively vanilla - simply switching the
> authentication to Anonymous, and adding a new topic exchange called "MAMA"
> (setup on Linux or Mac OS X)
> * C++ QPID Broker is built with AMQP_1 support and ran with
> {{--topic-patterns MAMA.*}} (setup on Linux)
> I run the {{send}} application as follows:
> {noformat}
> ./send -a topic://10.211.55.14/MAMA/PHIL `for i in {1..32640}; do echo -n
> "x" ; done`
> {noformat}
> This is just generating a large message to send. On the receive side it is
> ran as follows:
> {noformat}
> ./recv topic://10.211.55.14/MAMA/PHIL
> {noformat}
> In the case when the number of characters sent is 32640 the message works
> fine. If I increase the characters to 32641 (or more) then the {{recv}}
> application has issues. The issue is different on Mac vs Linux so:
> *Mac*
> {noformat}
> Assertion failed: (msg && bytes && size), function pn_message_decode, file
> /Users/ppreston/Developer/Libraries/Source/qpid-proton/proton-c/src/message/message.c,
> line 656.
> Abort trap: 6
> {noformat}
> *Linux*
> {noformat}
> [0x10a3ab0]:ERROR amqp:session:invalid-field sequencing error, expected
> delivery-id 2, got 1
> {noformat}
> If I run the exact same test to the C++ QPID Broker (0.32) there is no issue
> at all. Certainly with the Linux error message something is getting out of
> sync with large messages, which are fragmented?
> I have attached the Java Broker Log (ran on Linux) for a run of the above
> where a Mac {{recv}} and Linux {{recv}} are running, and a message is send
> from Linux {{send}}.
> Let me know if any extra information required.
> Thanks
> Phil
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
