On Mon, 19 Sep 2011 07:13:28 +0200, Gergely Nagy wrote:
Jakub Jankowski <shasta@toxcorp.com> writes:
I think I'm having a memleak in 3.3's HEAD. My production system, which receives ~300 messages/s throughout the day (so ~26.4 million messages daily) leaks ~700MB of memory per day.
I did some investigation, here are the details:
# syslog-ng -V syslog-ng 3.3.0beta2 Installer-Version: 3.3.0beta2 Revision: ssh+git://bazsi@git.balabit//var/scm/git/syslog-ng/syslog-ng-ose--mainline--3.3#no_branch#d6eea6a1130f8343b6600976d9f2319f9ed86c88
Could you try with a more recent git checkout? Bazsi fixed a couple of memory leaks since d6eea6a1130f8343b6600976d9f2319f9ed86c88, and they should all be gone with the most recent git HEAD (0a3d844ff94a14d770bcdfa993f02e87e58a81f2).
Actually, the code I run is from HEAD, despite syslog-ng -V output. This is because of me having difficulties building directly from git checkout, I took 3.3.0beta2 tarball and applied all the patches that went into Bazsi's tree since 3.3.0beta2 tag. So, for example: $ grep -B 4 'nv_table_clone.*value_len' lib/logmsg.c if (!log_msg_chk_flag(self, LF_STATE_OWN_PAYLOAD)) { NVTable *payload = self->payload; self->payload = nv_table_clone(payload, name_len + value_len + 2); $ Which, if I'm right, comes from 951659d2cdda1a522740881354de25dcbd4e42ce, newer than that is only 0a3d844ff94a14d770bcdfa993f02e87e58a81f2 that adds nothing but a comment.
(I'll have a look at the valgrind logs when time permits, thanks for investigating!)
I'd be grateful, thanks. And sorry for misleading version output. Cheers, -- Jakub Jankowski|shasta@toxcorp.com|http://toxcorp.com/ GPG: FCBF F03D 9ADB B768 8B92 BB52 0341 9037 A875 942D