It is not clear how what sources you are using, what does it mean 'latest release tarball' -- version 4.3.3? Then did you take all the patches from master since version 4.3.0?
I tested master with 20000 registrations over TLS sent by sipp, but I couldn't spot any leak there.
Can you test with bare master branch?
Cheers, Daniel
On 18/11/15 02:16, Anthony Messina wrote:
Sorry for the delay, I just got home from my $PAYINGJOB. And thanks a lot for helping figure this out.
I build Kamailio RPMs from the latest release tarball, with the changes between the release and git master applied via patch, but here is the version output:
# kamailio -v version: kamailio 4.4.0-dev6 (x86_64/linux) e275bc flags: STATS: Off, USE_TCP, USE_TLS, USE_SCTP, TLS_HOOKS, USE_RAW_SOCKS, DISABLE_NAGLE, USE_MCAST, DNS_IP_HACK, SHM_MEM, SHM_MMAP, PKG_MALLOC, Q_MALLOC, F_MALLOC, TLSF_MALLOC, DBG_SR_MEMORY, USE_FUTEX, FAST_LOCK- ADAPTIVE_WAIT, USE_DNS_CACHE, USE_DNS_FAILOVER, USE_NAPTR, USE_DST_BLACKLIST, HAVE_RESOLV_RES ADAPTIVE_WAIT_LOOPS=1024, MAX_RECV_BUFFER_SIZE 262144, MAX_LISTEN 16, MAX_URI_SIZE 1024, BUF_SIZE 65535, DEFAULT PKG_SIZE 8MB poll method support: poll, epoll_lt, epoll_et, sigio_rt, select. id: e275bc compiled on 22:23:43 Nov 13 2015 with gcc 5.1.1
I am using tcpops functions as follows, though again, these were in use for quite some time before the issue appeared:
To update the tcp connection lifetime after successful auth: # User authenticated - update tcp_connection_lifetime if(proto!=UDP) tcp_set_connection_lifetime("3605");
and in the reply route:
onreply_route[MANAGE_REPLY] { xdbg("incoming reply\n");
if(status=~"[12][0-9][0-9]") route(NATMANAGE);#!ifdef WITH_TCPOPS if(proto!=UDP && status=="200") { if(is_method("INVITE")) { # enable on callee's connection tcp_keepalive_enable("60", "5", "5"); # enable on caller's connection if($avp(caller_conid)!=$null) tcp_keepalive_enable("$avp(caller_conid)", "60", "5", "2"); } if(is_method("BYE")) { tcp_keepalive_disable(); tcp_keepalive_disable("$avp(bye_conid)"); } }