expected behavior of TCP keepalive

[alberanid]

Hi,
I have some systems where, apparently randomly (after many hours, usually) and without other network failures, the connection is lost and is re-established only when a re-register happens after the configured regint time. This leaves some minutes when the the system is not reachable.

The account:

"user" <sip:user@sip.example.net;transport=tcp>;auth_user=user;auth_pass=password;answermode=manual;audio_codecs=PCMA,PCMU;regint=600;pubint=300;video_codecs=h264;outbound="sip:sip.example.net;transport=tcp";sipnat=outbound

As you can see I'm using sipnat outbound.

In this case, looking at the code, I assument it should enable a TCP keepalive that if I've understood correctly, will publish a \r\n every 120 seconds (TCP_KEEPALIVE_INTVAL) and wait 10 seconds before closing a non-working connection:

re/src/sip/transp.c

Line 1829 in c7599c1

int sip_keepalive_tcp(struct sip_keepalive *ka, struct sip_conn *conn,

However, from a tcpdump I do not see any '\r\n' lines sent at any interval.

Should I look for something else? Is this a feature that requires server-side support? (we are currently using FlexiSIP)

Thanks for your help!

[sreimers]

Can you try to add a debug msg to the keepalive handler:

--- a/src/sip/transp.c
+++ b/src/sip/transp.c
@@ -316,6 +316,8 @@ static void conn_keepalive_handler(void *arg)
        mb.pos  = 0;
        mb.end  = 4;
 
+       DEBUG_WARNING("KEEPALIVE\n");
+
        err = tcp_send(conn->tc, &mb);
        if (err) {
                conn_close(conn, err);

You can do the same for sip_keepalive_tcp to see if the timer handler setup is working.

[alfredh]

in general it is probably better to use the mailing list for questions like this.

I think the sip outbound ping/pong feature is only enabled if supported by the SIP Server.
Please check in the source code.

You can also use a low registration interval to keep the TCP connection open:

;regint=60

please note that some networks (CGN) will kill random TCP connections at random times.
If you want this to work reliably you need to re-connection TCP from the client side.

[alberanid]

Agreed, thanks for the support; I'll let you know if we will detect anything meaningful.

[alberanid]

we put the debug lines in place (conn_keepalive_handler, sip_keepalive_tcp and also keepalive_handler in src/sipreg/reg.c); apparently, the keepalive are never sent.

Active timers:

Timers (5):
0x18fa188: th=0x769b5aa4 expire=249ms
0x18f6410: th=0x2a2f8 expire=57110ms
0x1902940: th=0x74a74830 expire=267310ms
0x1901338: th=0x76ea286c expire=537279ms
0x1902280: th=0x76e9da04 expire=897309ms

We rebuilt both libre and baresip.

Is there something else that is needed, in order to enable the keepalives?