[asterisk-dev] [Code Review] Append two container for dialogs to delete and for rtp timeout checks to replace all dialogs container in ao2 callback for dialogs_needdestroy
David Vossel
dvossel at digium.com
Mon Sep 13 12:13:47 CDT 2010
-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviewboard.asterisk.org/r/917/#review2714
-----------------------------------------------------------
I like the idea behind this patch. This could be a really cool optimization! Below are my comments after making my first pass on the code. Most of the comments have to do with the way the ao2 objects and containers work. Once we get some of that worked out I'll take a deeper look at this.
branches/1.6.2/channels/chan_sip.c
<https://reviewboard.asterisk.org/r/917/#comment5871>
It should not be necessary to ref the dialog before removing it from the dialogs container. Was this done because you saw a crash here? if so there is something else going on that needs to be investigated.
branches/1.6.2/channels/chan_sip.c
<https://reviewboard.asterisk.org/r/917/#comment5873>
What is lockowner and lockdialoglist for here?
branches/1.6.2/channels/chan_sip.c
<https://reviewboard.asterisk.org/r/917/#comment5872>
The same thing applies here to what I said about the dialog_unlink_needdestroy function. It should not be necessary to ref the dialog before unlinking it.
branches/1.6.2/channels/chan_sip.c
<https://reviewboard.asterisk.org/r/917/#comment5870>
Are we sure this function will never get called twice on the same dialog? If this does happen i think the dialog will be linked twice into the container.
branches/1.6.2/channels/chan_sip.c
<https://reviewboard.asterisk.org/r/917/#comment5874>
What happens on a re-invite. Will the dialog pointer be linked into this container twice?
branches/1.6.2/channels/chan_sip.c
<https://reviewboard.asterisk.org/r/917/#comment5875>
I don't understand the OBJ_UNLINK flags here. I understand this is the way the code was before your changes, but the callback never actually unlinks any of the dialogs by returning CMP_MATCH.
branches/1.6.2/channels/chan_sip.c
<https://reviewboard.asterisk.org/r/917/#comment5869>
What do we gain by running this sooner than 1000ms? If a Request or Response comes in, or some other event that requires the do_monitor loop it is supposed to break out early from this wait.
- David
On 2010-09-11 17:44:40, schmidts wrote:
>
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviewboard.asterisk.org/r/917/
> -----------------------------------------------------------
>
> (Updated 2010-09-11 17:44:40)
>
>
> Review request for Asterisk Developers.
>
>
> Summary
> -------
>
> Instead of iterate through all sip dialogs in the do_monitor function of chan_sip.c i created two separate containers which links to dialogs.
> The container dialogs_needdestroy links to dialogs which are marked as needdestroy, so they should be unrefered from the dialogs container.
> The container dialogs_rtpcheck links to dialogs which have active rtp, vrtp or trtp and rtptimeout, rtpholdtimeout or rtpkeepalive activated.
>
> normaly the container with needdestroy dialogs is empty so the callback takes only around 30 usec. even if there are more than 10 dialogs to be destroyed it takes around 200 usec.
> the callback for the rtpcheck tooks around 200 usec with 30 concurrent calls and 700 usec if the timeout is reached to do a softhangup.
>
> further i have found in check_rtp_timeout a possible locking situation. i am not sure if this would be a working solution cause it could happen that a call would be hung up several ms or also 1 sec later.
>
> i think after speeding the ao2 callback up, its not necessary to wait 1000 ms, with 100ms dialogs would be faster released, the timeout fires even shorter and scheduled events which has been scheduled in a very short future time would be fired faster. see change 40 for this.
>
> i am also not sure if its necessary to iterate through the containers on module_unload to unlink the entries. the dialogs itself would be destroyed after they are unrefered from the dialogs container. see change 42 for this.
>
>
>
> This addresses bug 17912.
> https://issues.asterisk.org/view.php?id=17912
>
>
> Diffs
> -----
>
> branches/1.6.2/channels/chan_sip.c 286373
>
> Diff: https://reviewboard.asterisk.org/r/917/diff
>
>
> Testing
> -------
>
> i have tested this with sipp and 500 concurrent calls all with rtptimeout 10 s and the active time was 10,9s per call.
> also the 20k peers register test run through without any problem. (with the astdbpatch)
>
> Messages Retrans Timeout Unexpected-Msg
> INVITE ----------> 500 0
> 100 <---------- 500 0 0
> 200 <---------- 500 0 0
> ACK ----------> 500 0
> Pause [ 40.0s] 500 500
> BYE ----------> 0 0 0
> 200 <---------- 0 0 0
> ------------------------------ Test Terminated --------------------------------
>
>
> ----------------------------- Statistics Screen ------- [1-9]: Change Screen --
> Start Time | 2010-09-11 23:12:32
> Last Reset Time | 2010-09-11 23:12:47
> Current Time | 2010-09-11 23:12:48
> -------------------------+---------------------------+--------------------------
> Counter Name | Periodic value | Cumulative value
> -------------------------+---------------------------+--------------------------
> Elapsed Time | 00:00:00:964 | 00:00:15:994
> Call Rate | 0.000 cps | 31.262 cps
> -------------------------+---------------------------+--------------------------
> Incoming call created | 0 | 0
> OutGoing call created | 0 | 500
> Total Call created | | 500
> Current Call | 0 |
> -------------------------+---------------------------+--------------------------
> Successful call | 0 | 0
> Failed call | 35 | 500
> -------------------------+---------------------------+--------------------------
> Call Length | 00:00:10:970 | 00:00:10:637
> ------------------------------ Test Terminated --------------------------------
>
>
> Thanks,
>
> schmidts
>
>
More information about the asterisk-dev
mailing list