qpid-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Aaron Fabbri <ajfab...@gmail.com>
Subject Re: segfault in Rdma broker
Date Thu, 28 Oct 2010 17:10:10 GMT
On Thu, Oct 28, 2010 at 8:16 AM, Andrew Stitcher <astitcher@redhat.com> wrote:
> On Wed, 2010-10-27 at 15:31 -0700, Aaron Fabbri wrote:
>> I recently updated to svn trunk latest and now my Rdma broker is crashing.
>>
>> 1. Is anyone else seeing this?
>
> Is this happening before connections, during, after?? What is your
> platform? What rdma hardware?

Before connections, on MT25208 DDR InfiniBand HCAs.

>>
>> 2. Should I open a bug.. and if so please give pointer to url and
>> anything else I'd need to know.
>
> If you open a bug make sure you a good way to replicate the bug (at
> least in your own testing).

I can reproduce it 100%.

>>
>> I'm running from gdb with this script
>>
>> # cat gdbscript
>> file /aafabbri/autotools_build/src/.libs/qpidd
>>
>> # get libraries loaded
>> break main
>> run
>> del 1
>>
>> define go
>>         run \
>>         --auth no \
>>         --mgmt-enable no \
>>         --load-module /aafabbri/autotools_build/src/.libs/rdma.so \
>>         --transport rdma \
>>         --worker-threads 4 \
>>         --log-to-stdout yes
>
> Can you add --no-module-dir to rule out any other modules from being
> loaded and confusing things.


Thanks!  That was the problem.  Perhaps someone I share the machines
with installed libs in the standard place.  I confirmed with strace
that qpid was opening two different rdma.so's:

[root@localhost build]# grep rdma.so /tmp/strace.out
11920 open("/root/aafabbri/apache_qpid/trunk/qpid/cpp/build/src/.libs/rdma.so",
O_RDONLY) = 5
11920 stat("/usr/local/lib/qpid/daemon/rdma.so",
{st_mode=S_IFREG|0755, st_size=978468, ...}) = 0
11920 open("/usr/local/lib/qpid/daemon/rdma.so", O_RDONLY) = 6

When I add --no-module-dir, everything works again, and strace only
shows a single open of rdma.so:

[root@localhost build]# grep rdma.so /tmp/strace2.out
11955 open("/root/aafabbri/apache_qpid/trunk/qpid/cpp/build/src/.libs/rdma.so",
O_RDONLY) = 5


>
>> end
>>
>> # LD_LIBRARY_PATH=/aafabbri/autotools_build/src/.libs gdb -x myscript
>> ...
>> (gdb) go
>> [Thread debugging using libthread_db enabled]
>> [New Thread 0x7ffff5af0710 (LWP 2850)]
>> 2010-10-27 15:28:32 notice SASL disabled: No Authentication Performed
>> 2010-10-27 15:28:32 notice SASL disabled: No Authentication Performed
>> 2010-10-27 15:28:32 notice Listening on TCP port 5672
>> 2010-10-27 15:28:32 notice Listening on TCP port 5672
>> 2010-10-27 15:28:32 notice Rdma: Listening on RDMA port 5672
>> 2010-10-27 15:28:32 notice Rdma: Listening on RDMA port 5672
>> 2010-10-27 15:28:32 notice Rdma: Listening on RDMA port 5672
>> 2010-10-27 15:28:32 notice Rdma: Listening on RDMA port 5672
>> 5672
>> 2010-10-27 15:28:32 notice Broker running
>> 2010-10-27 15:28:32 notice Broker running
>> [New Thread 0x7ffff4cd6710 (LWP 2851)]
>> [New Thread 0x7ffff42d5710 (LWP 2852)]
>> [New Thread 0x7ffff38d4710 (LWP 2853)]
>
> I'm a little suspicious of the duplicated lines here - is this just an
> artifact of your cut and paste? If not it might indicate that multiple
> rdma modules are being loaded.
>
> Andrew
>
>
>
> ---------------------------------------------------------------------
> Apache Qpid - AMQP Messaging Implementation
> Project:      http://qpid.apache.org
> Use/Interact: mailto:dev-subscribe@qpid.apache.org
>
>

---------------------------------------------------------------------
Apache Qpid - AMQP Messaging Implementation
Project:      http://qpid.apache.org
Use/Interact: mailto:dev-subscribe@qpid.apache.org


Mime
View raw message