[Asterisk-Dev] Unlocking the unlocked...

[Luigi Rizzo]

On Tue, Jun 28, 2005 at 08:14:06PM -0400, mattf wrote:
> Hello,
> 
> Happens to us every day for the last year and a half. It's a hard disk
> read/write speed issue for the recordings. We've keep our recording drives

Sorry but the problem is a deadlock-prone locking scheme.
all these hacks you suggest to reduce the length of the critical
section may only reduce contention in some cases, but won't
cure the problem that can only be solved moving to a different
locking scheme. We talked about this witk kpfleming a while ago.

cheers
luigi

> on separate physical disks from everything else and we try to keep that
> partition as defragged as possible and that does help reduce the
> channel_walk_locked warnings and the associated periodic recording audio
> skips.
> 
> I'm very interested in your experience with changing the number of retries
> and your results as we have not tried that yet. 
> 
> What audio format do you record in(wav or gsm)?
> What do you usually change the number of retries to and how are the results?
> 
> If this can help significantly, maybe a config variable in features.conf
> would be in order?
> 
> Thanks,
> 
> MATT---
> 
> -----Original Message-----
> From: Guillermo Winkler [mailto:gwinkler at inconcertcc.com]
> Sent: Tuesday, June 28, 2005 6:13 PM
> To: asterisk-dev at lists.digium.com
> Subject: [Asterisk-Dev] Unlocking the unlocked...
> 
> 
> Hi,
>     Under the following scenario:
> 
> 6000 outgoing calls/hour(SIP-->ZAP)
> 75% connected (25% is busy or no answer)
> 100% of connected calls beign recorded, when LINK event occurs.
> P4 2.8 HT.
> Krnl 2.4.21
> 
> We are having some "Avoid deadlock" issues(This is not happening without 
> recording)
> 
> As you can see in channel_walk_locked
> 
> if (retries < 10)
>      ast_log(LOG_DEBUG, "Avoiding initial deadlock for '%s'\n", l->name);
>     else
>      ast_log(LOG_WARNING, "Avoided initial deadlock for '%s', %d 
> retries!\n", l->name, retries);
>     ast_mutex_unlock(&chlock);
> 
> Ten 1ms retries are done to avoid deadlocks. But when recording is enabled, 
> the channel gets locked on start:
>  if (ast_monitor_start(c, format, fname, 1)) {
>   if (ast_monitor_change_fname(c, fname, 1)) {
> 
> Those ten retries are not enough in that case, using lock instead of 
> try_lock, or increasing the number of retries, solves the problem 
> temporarily.
> 
> I wanted to know if someone else is experimenting something similar?
> 
> I think ast_channel_walk_locked and ast_get_channel_by_name_locked could be 
> rewritten or a new set of functions be added , faster and more reliable (now
> 
> you don't know if the channel you're looking for is not there at all or a 
> deadlock(not so dead) issue on ANOTHER channel was the failure).
> 
> If you feel the same I would be glad to help.
> 
> Regards,
> Guillermo
> 
> 
> 
> 
> _______________________________________________
> Asterisk-Dev mailing list
> Asterisk-Dev at lists.digium.com
> http://lists.digium.com/mailman/listinfo/asterisk-dev
> To UNSUBSCRIBE or update options visit:
>    http://lists.digium.com/mailman/listinfo/asterisk-dev
> _______________________________________________
> Asterisk-Dev mailing list
> Asterisk-Dev at lists.digium.com
> http://lists.digium.com/mailman/listinfo/asterisk-dev
> To UNSUBSCRIBE or update options visit:
>    http://lists.digium.com/mailman/listinfo/asterisk-dev