Hey
I was testing the broker implementation that is currently being worked on in
OpenMAMA and came across an issue, which I have been able to recreate with
proton send / recv example apps.
The issue is seen with sending large payloads to Java QPID Broker using the
proton send application as follows:
./send -a topic://127.0.0.1/MAMA/PHIL <topic://127.0.0.1/MAMA/PHIL> <string
from file - in this case 47104 bytes>
And the receive as per below:
./recv topic://127.0.0.1/MAMA/PHIL <topic://127.0.0.1/MAMA/PHIL>
I am running the Java QPID Broker v0.32 running with java v1.8.0_45 - vanilla
setup (although I have ran this java broker on Centos 7 along with the clients
to be sure not a mac thing). The message makes it to the recv client and
throws an assertion:
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
I am running proton v0.8 (debug build). Running in lldb I can see whats
failing:
frame #0: 0x000000010002cef4
libqpid-proton.2.dylib`pn_message_decode(msg=0x0000000100204c00,
bytes=0x000000010101d800, size=0) + 36 at message.c:656
653
654 int pn_message_decode(pn_message_t *msg, const char *bytes, size_t size)
655 {
-> 656 assert(msg && bytes && size);
657
658 pn_message_clear(msg);
659
(lldb) p msg
(pn_message_t *) $0 = 0x0000000100204c00
(lldb) p bytes
(const char *) $1 = 0x000000010101d800 ""
(lldb) p size
(size_t) $2 = 0
(lldb)
When I inspect the messages in Wireshark I can see it flags malformed packet.
Following is from the send client to the broker.
Running the same clients to a C++ Qpid Broker v0.32, however, there are no
issues and the data is received in the recv client fine. The qpidd is running
vanilla with —topic-pattern parameter.
Have checked the following combinations with large payloads (ruling out
different MTU on loopback vs active interfaces)
send (mac) —> Java Broker (loopback mac) —> recv (mac)
: Fail
send (mac) —> Java Broker (vnic linux) —> recv (mac)
: Fail
send (linux) —> Java Broker (loopback linux) —> recv (linux)
: Fail
send (mac) —> C++ Broker (vnic linux) —> recv (mac)
: Pass
send (linux) —> C++ Broker (loopback linux) —> recv (linux)
: Pass
Wanted to check whether anyone else has experienced this? Something
configuration based? I am not that familiar with QPID brokers so not ruling
out something I haven’t setup in Java that’s not needed in C++.
Any help or guidance would be greatly appreciated before I go down a rabbit
hole.
Thanks
Phil
