Hi Michael, On Fri, Jun 05, 2020 at 02:56:49PM +0200, JOIGNY Michael @Neteven wrote: > Hi Haproxy list, Memory on our haproxy process rises gradually, due to the > pool trash which grows quite quickly, ~ 1.5GB in 24 hours for 40 backend. > show pool gives this kind of results : echo "show pools" | sudo -u nagios > socat stdio unix-connect:/var/run/haproxy/admin.sock ... Pool trash (16416 > bytes) : 79586 allocated (1306483776 bytes), 79586 used, 0 failures, 1 > users, @0x55e6af326b40=17
Oh that seriously looks like a leak! > ...The reload command and the kill -3 do not flush > the memory contained in the trash pool. Only a restart flush the entire > memory. Is this the expected behavior? This documentation > (https://cbonte.github.io/haproxy-dconv/2.1/management.html#6) discusses the > memory flush during a reload of the haproxy process : /"During a reload > operation, the process switched to the graceful stop state also > automatically performs some flushes after releasing any connection so that > all possible memory is released to save it for the new process."/ > Configuration : OS : Debian 10.4 Version : Haproxy 2.1.5-1 Do you have an > idea ? Regards. Not yet, you'll have to share more info like your config, whether or not your doing things in Lua, what's your typical workload etc, so that we have an idea where to look. Also if you're aware of a version that does not have this issue, it could help as well. What I'm suspecting is that an error path is wrong in a function calling alloc_trash_chunk(), we'll have to audit them a little bit :-/ Willy
