Re: [Jack-Devel] jack 1.9.8 + many many connections = problems?

On 12/28/2011 11:27 AM, Robin Gareus wrote:
> On 12/28/2011 03:03 AM, Fernando Lopez-Lezcano wrote:
>> Hi all, has anyone out there experienced something similar to what I
>> describe below? Sorry for the length of the post.
>>
>> I'm really puzzled...
>>
>> (running jack 1.9.8, freshly built today, just in case - running with
>> 4096 total ports and 768 ports per client)
>>
>> I have a quite complex jack based system (OpenMixer in our Listening
>> Room driving 22 speakers plus 4 subs). Many programs, all interconnected
>> through jack and running under Linux (ie: jack itself, three instances
>> of supercollider, three instances of ambdec, two instances of
>> jconvolver, four internal 24 channel netjack1 clients, etc, etc).
>> Everything is run from inside the supercollider language and startup is
>> automatic from a boot of the workstation. All jack connections are
>> managed by writing them to a file and then running aj-snapshot.
>>
>> Lately one or two speakers are "silent" after startup.
>
> Does downgrading jackd to 1.9.7 help?

It was happening in 1.9.7 and I upgraded :-)

> Which version of aj-snapshot are you using?

Latest, 0.9.5...

> There's been some activity the the last weeks. e.g 4 weeks ago:
>    "graph reorder jackd bug workaround implemented"
> I don't know the details, but there's more:
> http://aj-snapshot.git.sourceforge.net/git/gitweb.cgi?p=aj-snapshot/aj-snapshot;a=log

Thanks for the pointer, perhaps I should try git just in case. I looked 
at the 0.9.5 source and it seemed fine (and what it is doing is not 
complicated at all, just make the connection and wait till it gets the 
callback before proceeding using mutex's - I even added delays after 
each connection with no change in behavior).

>> If I do _not_ run the aj-snapshot port connection process and I manually
>> send a signal (from jnoise) to the ports that are normally affected,
>> everything is fine, that is, in my tests I never managed to get the
>> system to fail in that state (ie: everything running, no ports connected).
>>
>>
>> It would seem that something bad is happening when all ports are being
>> connected and the end result is that inside jack some of the soundcard
>> ports seem to go nowhere.
>
> sounds like jackd's internal connection state somehow becomes
> inconsistent. There was a bug in the dbus code a while ago (1.9.6): the
> dbus layer kept an independent list of the connections but that was
> fixed in 1.9.6+svn before 1.9.7.
>
> I can't see how aj-snapshot could be the culprit; but minimizing the
> number of tools involved makes it easy.

I don't think aj-snapshot is the culprit, it seems to me so far that 
aj-snapshot is triggering a problem inside jackd. But the clues are not 
clear at all.

>> The aj-snapshot process that connects everything takes a bit to run and
>> as it progresses jack reports more and more xruns until it finishes. It
>> would seem that something is running inside jack when ports are
>> connected that is proportional to the number of existing connections
>> (the graph reordering?, that would make sense - but it should not
>> generate xruns, right?).
>>
>> It also appears to be that if I minimize the number of ports I'm
>> connecting the problem goes away (but then of course the system is not
>> completely functional :-).
>>
>> Clues?
>> Ways to debug?
>
> pinpoint _when_ it goes wrong ->  that makes it easier to determine
> _where_ it goes wrong.
>
> 1)Try `jnoise` after startup ->  verify signals end up at speakers.
> 2)Launch all apps; walk the `jnoise` again.
> 3)run aj-connect.
> 4)`jnoise` again.

Yup, I was going to do something similar today from inside supercollider :-)
Thanks!
-- Fernando

1325104584.7919_0.ltw:2,a <4EFB7DB9.6050909 at ccrma dot stanford dot edu>