Hi!

Do the queues get flushed after some time (some minutes, hours)?

How do you generate the TCP traffic? I once used sipp and had similar problems, but suspected sipp to be the bad guy.

If a sending queue does not get flushed, this means the remote peer of this TCP connection has set the TCP windows size to zero (meaning that the receiving buffer on the remote peer is full). can you check the receiving buffer at the remote peer?

regards
klaus



serega wrote:
Hello,

I'm sorry for a big lag. I test system yesterday on other linux
kernel (2.6) and obtain the same results. In this case i suppose, that this problem don't bases on system socket implementation. I
tested openser on other numbers of tcp workers and obtain next
results: 20 children: 2 sockets don't flushed (all had data on
receive and send queues), cpu usage is above 48% on each of 2 openser
processes, one of them invokes poll function in tls_server.c, other
trying to write into tls socket. 1 child: 1 socket don't flushed
(receive queue not empty, send queue empty), cpu usage is 90% on one
process. 100 children: no problems, all work correctly 2 children: no
problems.


Hello,

On 09/29/06 16:06, serega wrote:
Hi Klaus and all other who read this.

It's not a problem of tcp socket - yesterday i rewrite gateway
config for fowarding directed to extgateway requests to gateway
itself. In tcp this problems not occured, but after i switched on
tls, i see overfilled queues again. In that case i haven't
extgateway, communications with remote sockets and etc. Did
anybody test openser whith big payload? I think this problem
occurs because in my case i have big traffic and several process
that simultaniously use tls. Did anybody have the same problem?

If you increase the number of the TLS/TCP workers do you get same results? In case of TLS, there is more CPU used because of encryption/decryption. How is the CPU usage in teh system while you
are testing?

As Klaus said, the management of socket queue is done by OS.

Cheers, Daniel


Hi Serega!

I'm not sure - just a guess. openser uses "worker threads".
This worker threads get a SIP message from the listener thread,
process them, and then they send them.

If for some reason the sending fails - e.g. the gateway sends
via TCP/TLS to extgateway and extgateway's receiving queue is
full. Then gateway's sending buffer gets full and the threads
wait until they can send. Thus, as the threads are all busy,
also the receiving queue of gateway gets full. I guess after
some time the threads give up sending and will read again.

The question is: why does the sending buffer gets full? Where
is gateway sending the messages to?

The sending queue must not get flushed. The queue is in the OS,
not in openser. As TCP guarantees that there is no
loss/reordering the TCP stack must not flush the queue.

regards klaus

serega wrote:

Hi all.

I have next problem while using openser with tls. System
consists of sip statefull server (in next time, simply,
server) and stateless sip gateway (gateway). Server used for
connect to jabber server. Server contains rewrote jabber
module logic and also use presence module. Gateway used to
connect to other sip gateway (extgateway) via tls protocol.
Gateway use 1.1 openser version and doesn’t contain changed code. When server together with gateway restarts and server
in its database contains above 260 subscriptions (in
watcherinfo table) i have error. Socket used by gateway to
connect with extgateway contains in receive and sent queues a
lot of data (above 50kByte on each sides). This data never
flushed out. This happens because after restart sip server
through gateway send notifications to extgateway. When I
attached using gdb to process that send data, I saw that it was in infinity loop because tls library returns
SSL_ERROR_WANT_WRITE and I think it’s correct because we have
overfilled send queue. In this case I don’t interest why
receive queue not empty (I think it happens because sending
process have got block on socket). But I don’t understand why
sent queue not flushed. I test this behavior using tcp – all
was correct. Socket not closed by other side because this
status can be unchanged above one day. Anybody can help me?
Why this happens?

I use openssl-0.9.8c and redhat os (Linux xdevel1
2.4.21-4.ELsmp #1 SMP Fri Oct 3 17:52:56 EDT 2003 i686 i686
i386 GNU/Linux). Computer have too Pentium 4 processors).


Gateway.cfg (sip.qa.hbex.com – gateway address):

# # $Id: router_qa.cfg,v 1.1 2006/08/02 18:14:34 ilya Exp $ #
# simple quick-start config script #

# ----------- global configuration parameters ------------------------

debug=9 # debug level (cmd line: -dddddddddd) log_facility=LOG_LOCAL0 fork=yes log_stderror=yes # (cmd
line: -E)


check_via=no    # (cmd. line: -v) dns=no          # (cmd. line:
-r) rev_dns=no # (cmd. line: -R) children=20 fifo="/tmp/openser_fifo" server_signature=no

fifo_db_url="mysql://root:@localhost/openser"

listen = udp:sip.qa.hbex.com:5060 #incoming sip server
address listen = tcp:sip.qa.hbex.com:5060

#tls address disable_tls = 0 listen =
tls:sip.qa.hbex.com:5061 listen = udp:sip.qa.hbex.com:5061

tls_certificate = "/ home/inop/ex.com-cert.pem"
tls_private_key = "/home/inop/ex.com-privkey.pem" tls_ca_list
= "/home/inop/ex.com-calist.pem"
tls_require_client_certificate=0 tls_verify_client=0
tls_verify_server=0

# ------------------ module loading ---------------------------------- loadmodule "/home/interop/openser/lib/openser/modules/rr.so" loadmodule "/home/interop/openser/lib/openser/modules/xlog.so" loadmodule "/home/interop/openser/lib/openser/modules/textops.so" loadmodule "/home/interop/openser/lib/openser/modules/maxfwd.so"
loadmodule "/home/interop/openser/lib/openser/modules/sl.so"
loadmodule "/home/interop/openser/lib/openser/modules/mysql.so" loadmodule "/home/interop/openser/lib/openser/modules/tm_unchanged.so" loadmodule "/home/interop/openser/lib/openser/modules/usrloc.so" loadmodule
"/home/interop/openser/lib/openser/modules/registrar.so"

# ----------------- setting module-specific parameters --------------- modparam("registrar", "default_expires", 120)
 modparam("registrar", "use_domain", 1)

modparam("usrloc", "use_domain", 1) modparam("usrloc",
"db_mode", 0) # Uncomment this if you want to use SQL
database # for persistent storage and comment the previous
line #modparam("usrloc", "db_mode", 2)

# add value to ;lr param to make some broken UAs happy
modparam("rr", "enable_full_lr", 1) modparam("maxfwd",
"max_limit", 10) # -------------------------  request routing
logic ------------------- # main routing logic route{ if
(!mf_process_maxfwd_header("10")) { sl_send_reply("483","To
Many Hops"); drop(); }; if (dst_port==5061) { if
(search("^To:[EMAIL PROTECTED]") ||
search("^To:[EMAIL PROTECTED]")) { # rewrite destination and
forward to jabber (sip server) route(1); return; };
sl_reply_error(); return; } if (method=="REGISTER") { xlog("XXX: saving location msg=$mb\n");
if(!save_noreply("location")) { log("XXX: Error saving
location!\n"); sl_reply_error(); } sl_send_reply("200","OK");
return; } if (search("^To:[EMAIL PROTECTED]") ||
search("^To:[EMAIL PROTECTED]")) { #forward to self route(3);
return; } ##forward to ext gateway route(2); }

route[1]{ if (method!="MESSAGE" && method!="SUBSCRIBE" && method!="NOTIFY") { log("XXX: Request not forwarded to sip server!\n"); sl_send_reply("202","Accepted"); return; } t_on_reply("1"); record_route();

if(!lookup("location")) { if (method!="SUBSCRIBE") {
log("XXX: only subscribes are processed for user that are not
registered\n"); sl_reply_error(); return; } subst_uri('/(.*)@(.*)/[EMAIL PROTECTED]/ig');
 if(!t_relay()) { log("XXX: error forwarding jabber01...\n");
 sl_reply_error(); return; } else {
sl_send_reply("200","OK"); return; } } else { log("XXX:
forwarding to the address of record...\n"); if(!t_relay()) {
log("XXX: error forwarding to address of record \n");
sl_reply_error(); return; } else { sl_send_reply("200",
"OK"); return; } }

}

# forwarding to external gateway route[2]{ log("XXX:
rewriting headers\n"); subst('/^(From:[EMAIL PROTECTED])#([EMAIL PROTECTED])[EMAIL PROTECTED](.*)/[EMAIL PROTECTED]/ig');


subst('/^(Contact:[EMAIL PROTECTED])#([EMAIL PROTECTED])[EMAIL 
PROTECTED](.*)/[EMAIL PROTECTED]/ig');



if(!forward("tls:43.123.141.166:3000")) { log("XXX: Error
forwarding to external gateway!\n"); sl_reply_error(); } else
{ sl_send_reply("200", "Accepted"); }; }

# forwarding to SELF route[3]{ log("XXX: rewriting
headers\n"); subst('/^(From:[EMAIL PROTECTED])#([EMAIL PROTECTED])[EMAIL PROTECTED](.*)/[EMAIL PROTECTED]/ig');


subst('/^(Contact:[EMAIL PROTECTED])#([EMAIL PROTECTED])[EMAIL 
PROTECTED](.*)/[EMAIL PROTECTED]/ig');



log("XXX: about to forward to self\n"); if(!forward("tcp:sip.qa.hbex.com:5061")) { log("XXX: Error
forwarding to self!\n"); sl_reply_error(); } else {
sl_send_reply("200", "Accepted"); }; }

# this is executed for replies onreply_route[1]{ subst('/^(To:[EMAIL PROTECTED])#([EMAIL PROTECTED])[EMAIL PROTECTED](.*)/[EMAIL PROTECTED]/ig');


subst('/^(Contact:[EMAIL PROTECTED])#([EMAIL PROTECTED])[EMAIL 
PROTECTED](.*)/[EMAIL PROTECTED]/ig');
 route(2); }



_______________________________________________ Users mailing
list [email protected]
http://openser.org/cgi-bin/mailman/listinfo/users



_______________________________________________ Users mailing
list [email protected] http://openser.org/cgi-bin/mailman/listinfo/users




_______________________________________________ Users mailing list [email protected] http://openser.org/cgi-bin/mailman/listinfo/users


_______________________________________________
Users mailing list
[email protected]
http://openser.org/cgi-bin/mailman/listinfo/users

Reply via email to