[asterisk-dev] Asterisk 1.4.18-rc2 Now Available

Atis Lezdins atis at iq-labs.net
Wed Jan 30 10:16:53 CST 2008


On 1/30/08, Matt Florell <astmattf at gmail.com> wrote:
> On 1/30/08, Atis Lezdins <atis at iq-labs.net> wrote:
> > On 1/30/08, Brandon Kruse <bkruse at digium.com> wrote:
> > > > Atis Lezdins wrote:
> > > > Not that easy. Hour of running tests (got 2 such crashes so far). I'm
> > > > trying to get some valgrind log right now, probably i'll have them
> > > > tomorrow.
> > > >
> > > > Just posted it here, because maybe backtraces are helpful, and
> > > > somebody sees what to fix immediately.
> > >
> > >     It is helpful, and there would be a hack of a fix, but there is a deeper issue.
> > >     As file said, it could be memory corruption, because it is unlikely that the
> > >     spyee channel does not have the name field set in the ast_channel struct.
> > >
> >
> > Ok, i created an issue http://bugs.digium.com/view.php?id=11877
> >
> > When running valgrind - i often got "show channels" displaying nothing
> > or just few lines and then breaking (without total sum of channels).
> > I've seen this "show channels" behavior  under high load, within
> > previous versions. Does "show channels" have some kind of name
> > validity check?
> >
> > Regards,
> > Atis
>
> Hello,
>
> I can confirm this bug, although it is much better than previous 1.4.X
> releases. 1.4.11 was a mess in that respect, at least this RC is
> workable. I have been running 1.4.18rc2 in VICIDIAL performance
> testing mode for the last 9 hours with a total of 124959 IAX calls so
> far(over 35127 talk-minutes) all going through a series of meetme
> rooms, and aside from the growing load on the server(in spite of the
> number of concurrent calls staying consistent). I will post a pretty
> graph after the test finishes or Asterisk crashes.

I don't know if it's better, before audiohooks i got 1 chanspy crash
every 30 minutes, now every hour (testing environment of course).
Let's hope that overall stability increases (and that somebody can fix
this :)

>
> Back to the "show channels" behaviour, VICIDIAL uses it heavily and
> there have always been issues with the accuracy of the results of
> "show channels" going back to pre 1.0 Asterisk. I wrote in logic to
> the VICIDIAL applications to be able to tell when Asterisk was
> deciding just to not print out the Zap channels for instance in it's
> "show channels" output. As for the current 1.4.X releases, it seems
> that "show channels" is treated like several other elements in that it
> will be sacrificed to allow another task to finish(which also happens
> with DeadAGI [the 'h' exten] sometimes)

Well, correct display of all the channels is quite necessary for my frontend.
I have a proposal here (actually i have patch, but it's ugly) The
magical thing that would work for me is "realtime channels", every
channel change updates row in db, and db always have all the info that
"show channels" provides, without any locking. Works very nice :)
However it needs some architectural improvements, and i'm not ready to
work on that right now. If anybody would be interested, please contact
me.

Regards,
Atis

-- 
Atis Lezdins
VoIP Developer,
IQ Labs Inc.
atis at iq-labs.net
Skype: atis.lezdins
Cell Phone: +371 28806004
Work phone: +1 800 7502835



More information about the asterisk-dev mailing list