Hi Pieter, On Wed, Jan 22, 2020 at 09:57:31PM +0100, PiBa-NL wrote: > Hi Olivier, > > Just to let you know, seems this commit has broken a few regtests: > http://git.haproxy.org/?p=haproxy.git;a=commit;h=477902bd2e8c1e978ad43d22dba1f28525bb797a > > https://api.cirrus-ci.com/v1/task/5885732300521472/logs/main.log > Testing with haproxy version: 2.2-dev1 > # top TEST reg-tests/mcli/mcli_show_info.vtc TIMED OUT (kill -9) > # top TEST reg-tests/mcli/mcli_show_info.vtc FAILED (10.044) signal=9 > # top TEST reg-tests/mcli/mcli_start_progs.vtc TIMED OUT (kill -9) > # top TEST reg-tests/mcli/mcli_start_progs.vtc FAILED (10.019) signal=9 > > Can reproduce it on my own FreeBSD machine as well, the testcase just sits > and waits.. until the vtest-timeout strikes. > > Do you need more info? If so what can i provide.?
We noticed it this evening, it just happens that these tests already used to fail on his machine so there was nothing alarming, but when tested on mine the same issues popped up and was new. Strangely, running the test by hand using socat works perfectly fine, and it seems that it only fails when run by the vtest CLI client. I can't grasp the difference between both at this point. I also noticed that after 10 connections to the master we can't create new ones, just as if something was not properly released and the connection wouldn't die... which is totally odd since it's only the connection setup which was cleaned up. Both of us gave up for today but we're still interested in a reliable reproducer outside of vtest in case you happen to stumble upon one (don't waste your time searching however). I suspect we've accidently uncovered a nasty corner case somewhere, which is good since the purpose of the code cleanup was precisely to make some complex bugs easier to spot and analyze :-) Thanks for your quick report! I hoped nobody else would notice in the mean time, and I was wrong :-) Cheers, Willy
