sorry am scatter-brained today here is the correct link for my show-sess.out https://docs.google.com/open?id=0BzPvBvLIIq7NelNrbmRmY3BkdFE
On Fri, Dec 7, 2012 at 10:54 AM, Bryan Berry <[email protected]> wrote: > ugh, sorry I didn't mention that I ran > > rm haproxy-1.5-dev14 -rf > wget http://haproxy.1wt.eu/download/1.5/src/devel/haproxy-1.5-dev14.tar.gz > tar xvzf haproxy-1.5-dev14* && cd haproxy-1.5-dev14 > patch -p1 < 0001-BUG-MAJOR-raw_sock-must-check-error-code-on-hangup.patch > patch -p1 < > 0002-BUG-MAJOR-polling-do-not-set-speculative-events-on-E.patch > make TARGET=linux26 ARCH=x86_64 USE_LINUX_SPLICE=1 CPU=native > USE_VSYSCALL=1 \ USE_STATIC_PCRE=1 USE_OPENSSL=1 > make install > > I also did a hash of the new haproxy binary to ensure that is different > from the old one > > > On Fri, Dec 7, 2012 at 10:50 AM, Bryan Berry <[email protected]>wrote: > >> Unfortunately I am still seeing the same issue :( >> >> 168.100.2.181, 168.100.2.237, 168.100.2.195, 168.100.2.183 # these >> have been changed from originals >> >> here is my show-sess.out again >> https://docs.google.com/open?id=0BzPvBvLIIq7NLVpFRWtvOUxyZ0U >> >> >> On Fri, Dec 7, 2012 at 12:28 AM, Willy Tarreau <[email protected]> wrote: >> >>> Hi Bryan, >>> >>> here come two fixes that I have pushed to the Git tree. They will be >>> in friday morning's snapshot, but I'm attaching the patches. >>> >>> The bug is complex to reproduce, it requires a specific timing that I >>> can only get with a combination of two machines and a certain number >>> of concurrent connections. It manifests when an error is reported at >>> the same time as a clean connection close. Only the connection close >>> was handled, the error did not cause an abort of the connection. The >>> issue is that afterwards, the error flag was lost, and the polling >>> remained active, causing the loops you have noticed. These loops all >>> eventually terminate thanks to the timeouts in the configuration. >>> >>> The two patches address different aspects of the issue, the first one >>> being for a real bug and the second one more about a misdesign from me >>> which fuels the bug. >>> >>> I'd like to really thank you for the amount of precise information you >>> provided, that was really helpful, especially because I was suspecting >>> totally unrelated issues (checks) and would not have found without your >>> help. >>> >>> I'd be happy if you can test to confirm that the issue does not reappear >>> anymore. Once I get your go (and possibly other pending fixes that might >>> appear in between), I'll issue dev15 to avoid causing issues to other >>> users. >>> >>> Thanks! >>> Willy >>> >>> >> >
