Hi Daniel,

Tks for answering! Unfortunatly, core.tcp_list is only returning one connection.
But core.tcp_info reports 184 opened connections (same for "kamctl stats tcp")

sudo kamctl rpc core.tcp_list
{
  "jsonrpc":  "2.0",
  "result": {
    "id": 439290,
    "type": "TCP",
    "state":  "CONN_ACCEPT",
    "timeout":  2567,
    "lifetime": 3600,
    "ref_count":  1,
    "src_ip": "41.46.4.235",
    "src_port": 4957,
    "dst_ip": "91.121.30.149",
    "dst_port": 5060
  },
  "id": 9158
}

sudo kamctl rpc core.tcp_info
{
  "jsonrpc":  "2.0",
  "result": {
    "readers":  16,
    "max_connections":  50000,
    "max_tls_connections":  50000,
    "opened_connections": 184,
    "opened_tls_connections": 64,
    "write_queued_bytes": 0
  },
  "id": 9523
}

Did I missed something?


jack@sip:~$ /usr/sbin/kamailio -v
version: kamailio 5.2.2 (x86_64/linux)
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_URI_SIZE 1024, BUF_SIZE 65535, DEFAULT PKG_SIZE 8MB
poll method support: poll, epoll_lt, epoll_et, sigio_rt, select.
id: unknown
compiled with gcc 6.3.0

Regards
Aymeric

Le lun. 15 avr. 2019 à 09:10, Daniel-Constantin Mierla <miconda@gmail.com> a écrit :

Hello,

On 26.03.19 17:16, Aymeric Moizard wrote:
Hi Again,

Here is an issue with TCP connection being kept for more:

Yesterday, I have discovered that a User-Agent (<Avaya IP Phone 1120E (SIP1120e.04.04.30.00)> tried to register a lot. It was sending REGISTER over new established TCP socket *every 2 seconds*.

All the REGISTER was rejected with 401. (may be the device was misconfigured? or not receiving any of my answer? I can't tell)

NOTE: You can see the expires header was very large: 86400, ie: 24 hours...

I was checking the TCP/TLS connections on my server and discovered more than 1000 TCP established connection to that user/ip, and thus, I have tried to understand what happened.

Checking the logs, I received 4855 REGISTER from this device from "Mar 25 03:47:09" to "Mar 25 07:56:13" which is a rate of approx one new TCP connection every 2.5 seconds...

Today, I decided to check it again around 11am.

jack@sip:~$ sudo kamctl stats tcp
{
  "jsonrpc":  "2.0",
  "result": [
    "tcp:con_reset = 1857",
    "tcp:con_timeout = 35927",
    "tcp:connect_failed = 25",
    "tcp:connect_success = 2",
    "tcp:current_opened_connections = 2291",
    "tcp:current_write_queue_size = 0",
    "tcp:established = 80778",
    "tcp:local_reject = 0",
    "tcp:passive_open = 80776",
    "tcp:send_timeout = 2",
    "tcp:sendq_full = 0"
  ],
  "id": 7305
}

There was still A LOT of established connections. And the connections have been established more than 24 hours ago.

At 11H16:
$> lsof -n -l | grep kamailio | grep TCP | grep 41.234.242.69 | grep ESTA | wc -l
1161
At 11H22:
$> lsof -n -l | grep kamailio | grep TCP | grep 41.234.242.69 | grep ESTA | wc -l
1018
At 11H35:
$> lsof -n -l | grep kamailio | grep TCP | grep 41.234.242.69 | grep ESTA | wc -l
655
At 13H
$> lsof -n -l | grep kamailio | grep TCP | grep 41.234.242.69 | grep ESTA | wc -l
0

So the established connections are all gone now.

Between 11h16 and 11H35, I was seeing the server regularly sending [FIN, ACK] over each TCP established connection, with retransmissions for all of them. (no incoming trafic)

I do not have numbers/capture/stats, but I think that kamailio was already closing some
connection yesterday. I don't know when kamailio started to try closing those connections.

I'm now back with this status:

At 13pm:
jack@sip:~$ sudo kamctl stats tcp
{
  "jsonrpc":  "2.0",
  "result": [
    "tcp:con_reset = 1896",
    "tcp:con_timeout = 38042",
    "tcp:connect_failed = 26",
    "tcp:connect_success = 2",
    "tcp:current_opened_connections = 939",
    "tcp:current_write_queue_size = 0",
    "tcp:established = 81950",
    "tcp:local_reject = 0",
    "tcp:passive_open = 81948",
    "tcp:send_timeout = 2",
    "tcp:sendq_full = 0"
  ],
  "id": 12734
}

With around 155 registration entries using TCP and TLS in my location database.

As you can see, tcp:current_opened_connections = 939 is still pretty high compared to
my currently registred users.

I have "modparam("registrar", "max_expires", 86400)", because I'm keeping contact entries (even with TCP connection down) for push notifications.

I have "tcp_connection_lifetime=3600" configured.

Question 1

With "tcp_connection_lifetime=3600", I would expect kamailio to close the established connection after 3600 seconds without traffic. It is pretty obvious that no data has been exchanged over the 4855 established connection during a day.

Despite the issue with the Avaya phones is solved automatically after a day, I guess similar stuff or happening, at a different rate, for other users as well. (because  current_opened_connections is way higher than registred TCP/TLS users)


Yes, tcp connections should be closed if no traffic on them for the lifetime duration.



Question 2

I can list TLS connection with "kamctl rpc tls.list"
Can I get a similar list for TCP? (lsof returns a lot of duplicates...)


Yes, see:

http://www.kamailio.org/docs/docbooks/devel/rpc_list/rpc_list.html#core.tcp_list

Maybe you can compare what is listed by the rpc command to see what kamailio actually sees as active connections.

Cheers,
Daniel

-- 
Daniel-Constantin Mierla -- www.asipto.com
www.twitter.com/miconda -- www.linkedin.com/in/miconda
Kamailio World Conference - May 6-8, 2019 -- www.kamailioworld.com


--