diff options
| author | Michael Goulish <mgoulish@apache.org> | 2011-03-30 19:11:09 +0000 |
|---|---|---|
| committer | Michael Goulish <mgoulish@apache.org> | 2011-03-30 19:11:09 +0000 |
| commit | cfbe48cd9b8432600864e89465c321020a8940cd (patch) | |
| tree | 3438ae6f8798e76d2ec2f3ba99d185ab8106ec7c /qpid/cpp/src/prof | |
| parent | 49d7dad9c00ff89ebfb61c5ce66c2e4300ccd846 (diff) | |
| download | qpid-python-cfbe48cd9b8432600864e89465c321020a8940cd.tar.gz | |
qpid-3171
The registration of the codec happens on a different thread from the
use of the codec. It is possible for the registration to occur after
the first attempted use. In my testing, this happened 3% of the time
-- 165 times out of 5000 tests -- when using RDMA transport, and 0 times
out of 5000 when using TCP. Which is why we didn't notice it earlier.
We have a function that tells when we are ready to encode --
CyrusSecurityLayer::canEncode. But it does not check the validity of
the codec pointer before using it, so it cores in this situation.
I believe simply checking that pointer is probably the best solution.
Introducing that check caused the crash not to show up in 10,000
trials. There were also no hangs.
git-svn-id: https://svn.apache.org/repos/asf/qpid/trunk@1087047 13f79535-47bb-0310-9956-ffa450edef68
Diffstat (limited to 'qpid/cpp/src/prof')
0 files changed, 0 insertions, 0 deletions