[SR-Users] potential memory leak

[Dragos Oancea]

Hello


We've put Kamailio on a machine with a SIP-Proxy role, we replaced another SIP proxy implementation.
We started with the default 4Mb of PKG Memory and 1024 Mb of Shared Memory but we had messages like these in the logs:

2013-09-26 13:48:26,939  ERROR: app_perl [kamailioxs.xs:371]: pv_sprintf(): pv_sprintf: Memory exhausted!
2013-09-26 13:48:50,267  ERROR: app_perl [kamailioxs.xs:371]: pv_sprintf(): pv_sprintf: Memory exhausted!
2013-09-26 13:48:51,822  ERROR: <core> [data_lump.c:197]: insert_subst_lump_after(): ERROR: insert_new_lump_after: out of memory
2013-09-26 13:48:51,823  ERROR: <core> [parser/contact/contact.c:194]: parse_contacts(): parse_contacts(): No memory left
2013-09-26 13:48:51,823  ERROR: memcached [../../parser/../ut.h:736]: pkg_str_dup(): could not allocate private memory from available pool
2013-09-26 13:48:51,823  ERROR: <core> [data_lump.c:373]: anchor_lump(): ERROR: anchor_lump: out of memory
2013-09-26 13:48:51,823  ERROR: <core> [dset.c:525]: rewrite_uri(): No memory left to rewrite r-uri
2013-09-26 13:48:51,823  ERROR: <core> [action.c:810]: do_action(): ERROR: do_action: memory allocation  failure
2013-09-26 13:48:51,823  ERROR: <core> [msg_translator.c:2164]: build_res_buf_from_sip_req(): ERROR: build_res_buf_from_sip_req: out of memory  ; needs 357
2013-09-26 13:48:51,823  ERROR: <core> [msg_translator.c:2164]: build_res_buf_from_sip_req(): ERROR: build_res_buf_from_sip_req: out of memory  ; needs 376


Then we started kamailio like below, with 1024Mb of private memory (per process) :

/opt/kamailio/sbin/kamailio -f /opt/kamailio/etc/kamailio//kamailio.cfg -w /opt/kamailio/core_dir/corefiles -P /opt/kamailio/var/run//kamailio.pid -m 1024 -M 1024 -u kamailio -g kamailio


Then after about 4 days we realized that we have a steady increase of used memory ( in the output of free command the '-/+ buffers/cache:' row  ), and we simply restarted the kamailio process.

The machine has 12 Gb of RAM , we run with children=32 .  1gb of memory per process x 32 child processes means that we are over the physical capacity of the machine.
The question is , are we in any danger of running out of memory on this box where we only run kamailio ? We don't want to reach the point where the kernel starts killing processes.
Shall we suspect some memory leak in the app_perl module ? (we do some http queries over SSL from a perl script)


Thank you.

Regards,
Dragos Oancea
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.sip-router.org/pipermail/sr-users/attachments/20130930/74e4e27d/attachment.html>