Le 17/07/2018 à 01:08, PiBa-NL a écrit :
Hi List,
With a build of 1.8.12 (and the 1.9 snapshot of 20180623 ) im getting
the 'old' haproxy process take up 100% cpu usage when using 3 threads in
the config and reloading with -sf parameter. I'm using FreeBSD.. (It
also happens with the 14-7 snapshot.)
It seems to happen after 1 thread quits, one of the others gets out of
control.
Most of the time it happens after the first reload.:
haproxy -f /var/etc/haproxy/haproxy.cfg -D
haproxy -f /var/etc/haproxy/haproxy.cfg -D -sf 19110
The main features i use are: ssl offloading / lua / threads
Only a little to no traffic passing through though, im seeing this
behavior also on my in-active production node, the strange part sofar
though is that i could not reproduce it yet on my test machine.
If someone has got a idea on how to patch or what direction to search
for a fix i'm happy to try.
If there is nothing obvious that can be spotted with the info from the
stack-traces of both 1.8 and 1.9 below ill try and dig further tomorrow
:).Thanks in advance for anyone's time :).
Hi Pieter,
This is a problem with the sync point... again. Could you try to revert
the following commit please ?
* ba86c6c25 MINOR: threads: Be sure to remove threads from
all_threads_mask on exit
This fix was probably added too quickly. I suspect a problem when a
thread exits, removing its tid_bit from all_threads_mask while some
others are looping in the sync point.
Regards,
--
Christopher Faulet
