#494 dialog: statistics active_dialogs inaccurate on restart

[Aubrey Wells]

If OpenSER is restarted while calls are active, the statistics module stops accurately reporting the active_dialogs statistic. What appears happens is the counter resets to 0 on restart, but if there are active calls, the counter will start from 0. If a call hangs up, the counter appears to wrap around backwards to 4294967296.

There were 38 calls when I restarted OpenSER and this is what my counters get:

[root@sip-01-c61-56-atl ~]# openserctl fifo get_statistics active_dialogs
dialog:active_dialogs = 4294967293
[root@sip-01-c61-56-atl ~]# openserctl fifo dlg_list | grep dialog:: | wc -l
34

As you can see, 4 calls hung up during the restart. the dialog module knows that there are 34 calls, but statistics has wrapped backwards to 4294967293.

[Henning Westerholt]

dialog-wrap-fix.patch

[Henning Westerholt]

Logged In: YES
user_id=337916
Originator: NO

Hi,

i've attached a small patch, which adds a new statistics #define that checks against this conditions and change the dialog code to use this for relevant cases. The patch is for the trunk, but also apply on the 1.3 branch.

Would be nice if you could test, and somebody with a little bit more knowledge about the dialog module could review if this is the best way of fixing the problem.

Henning
File Added: dialog-wrap-fix.patch

[Aubrey Wells]

Logged In: YES
user_id=847429
Originator: YES

I'll try your patch in my maint window tomorrow. Thanks.

[Aubrey Wells]

Logged In: YES
user_id=847429
Originator: YES

Doesnt seem to have worked:

Before making a call:
[root@rolecall ~]# openserctl fifo get_statistics active_dialogs
dialog:active_dialogs = 0

Call up:
[root@rolecall ~]# openserctl fifo get_statistics active_dialogs
dialog:active_dialogs = 1

Restart openser with call active:
[root@rolecall ~]# service openser restart
Stopping openser: [ OK ]
Starting openser: [ OK ]
[root@rolecall ~]#

After restart, call still up:
[root@rolecall ~]# openserctl fifo get_statistics active_dialogs
dialog:active_dialogs = 0

Hung up call - counter has wrapped:
[root@rolecall ~]# openserctl fifo get_statistics active_dialogs
dialog:active_dialogs = 4294967295

[Henning Westerholt]

Logged In: YES
user_id=337916
Originator: NO

Hi,

ok, aparently the previous patch was not sufficient, sorry about that. Anyway i've ported a different fix to the trunk, see http://openser.svn.sourceforge.net/openser/?rev=4551&view=rev

Now the statistic counter should be properly initialized from the database content during startup. The patch also applies to the 1.3 branch.

Henning

[Aubrey Wells]

Logged In: YES
user_id=847429
Originator: YES

Will the patch apply cleanly with, or does it require the previous patch, or do I need to back out the first patch before applying this one?

[Henning Westerholt]

Logged In: YES
user_id=337916
Originator: NO

The previous patch is not needed anymore. If the new one apply correctly it will probably also work, if you don't like to backout the previous one. There more or less unrelated. I'll port tomorow some more bug fixes, so perhaps you want to wait for them too.

[Aubrey Wells]

Logged In: YES
user_id=847429
Originator: YES

Well, that didn't work. OpenSER now dumps core when the dialog module loads if there is anything in the dialog table. If I hang up and delete the entries in the dialog table, it will start fine.

Aug 5 21:36:50 [5829] INFO:snmpstats:mod_init: Starting up the SNMPStats Module
Aug 5 21:36:50 [5829] INFO:dialog:mod_init: Dialog module - initializing
Segmentation fault (core dumped)

[Aubrey Wells]

Logged In: YES
user_id=847429
Originator: YES

I'm trying to upload a core file, but sourceforge is not letting me. If you think a core file will be helpful, email me and I can send you ftp credentials to get it.

[Henning Westerholt]

Logged In: YES
user_id=337916
Originator: NO

Hi,

does it works also if you restart the server, or do you observe then a core dump again? E.g. do you get a core dump everytime? Perhaps you can attach the backtrace from the core dump, or send it to me via mail, if it contains sensitive informations, henning dot westerholt at 1und1 dot de

Henning

[Aubrey Wells]

Logged In: YES
user_id=847429
Originator: YES

It coredumps on startup if there are entries in the dialog db table. It will coredump every time. If i hang up running calls (or delete the entries from the dialog table by hand) the server will start up normally. I am emailing the core file now. I had to gzip it to get it to send.

[Ovidiu Sas]

Logged In: YES
user_id=1395524
Originator: NO

Hello Aubrey,

Please test the latest kamailio trunk version and see if the issue is still present (no need to apply any patch).

Regards,
Ovidiu Sas