[asterisk-dev] [Code Review] Use kqueue(2) to reduce CPU consumption in *BSD and Mac OS X

[Tilghman Lesher]

> On 2010-03-09 14:15:36, Mark Michelson wrote:
> > /trunk/main/stdtime/localtime.c, line 343
> > <https://reviewboard.asterisk.org/r/543/diff/1/?file=8488#file8488line343>
> >
> >     There is a race condition here. It is possible that this call to ast_cond_signal can be made prior to the other thread reaching its call to ast_cond_wait.
> >     
> >     The way to ensure happens-before behavior is to use a combination of a condition and a boolean.
> >     
> >     Thread 1:
> >     
> >     ast_mutex_lock(lock);
> >     if (!boolean)
> >         ast_cond_wait(cond, lock);
> >     }
> >     ast_mutex_unlock(lock);
> >     
> >     Thread 2:
> >     
> >     ast_mutex_lock(lock);
> >     boolean = 1;
> >     ast_cond_signal(cond);
> >     ast_mutex_unlock(lock);
> >     
> >

Fixed in a different way -- by adding a lock on the mutex before sending the signal.


> On 2010-03-09 14:15:36, Mark Michelson wrote:
> > /trunk/main/stdtime/localtime.c, line 397
> > <https://reviewboard.asterisk.org/r/543/diff/1/?file=8488#file8488line397>
> >
> >     If I've read the code correctly, I believe that you will wait in this call to ast_cond_wait forever in most cases. This is because in kqueue_daemon, you only call ast_cond_signal in the case where a call to kqueue() returns a failure. It seems like ast_cond_signal should also be called in other circumstances as well.

Oops.  Fixed.


- Tilghman


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviewboard.asterisk.org/r/543/#review1643
-----------------------------------------------------------


On 2010-03-09 16:23:56, Tilghman Lesher wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviewboard.asterisk.org/r/543/
> -----------------------------------------------------------
> 
> (Updated 2010-03-09 16:23:56)
> 
> 
> Review request for Asterisk Developers.
> 
> 
> Summary
> -------
> 
> Currently, we use inotify(2), which is a Linux-only kernel API, to check as to whether timezone files have changed, which allows an admin to change the machine timezone on the fly, and we will reparse it.  The alternative currently is for us to sit in what is essentially a busy loop, checking whether a timezone file is changed.  For most situations, these files change very seldom, so this represents quite a bit of CPU consumption over the life of a process.
> 
> The BSDs (and Mac OS X) contain a similar interface, called kqueue(2).  This has been implemented for timezone files, which should conserve CPU usage while running Asterisk on those platforms.
> 
> 
> Diffs
> -----
> 
>   /trunk/configure.ac 251423 
>   /trunk/include/asterisk/autoconfig.h.in 251423 
>   /trunk/main/stdtime/localtime.c 251423 
> 
> Diff: https://reviewboard.asterisk.org/r/543/diff
> 
> 
> Testing
> -------
> 
> None yet; wanted to get some eyes on this before testing started.
> 
> 
> Thanks,
> 
> Tilghman
> 
>