[Freeswitch-users] high pdd

Sat Aug 30 00:32:20 MSD 2014

Thanks all.  I put in a ticket and attached the dump images and CPU performance as requested.

I put it under mod_managed but now I think I should of put it under something else because the ticket went to Mike.

https://jira.freeswitch.org/browse/FS-6776

I’m also working on trying Debian as suggested however the mod_managed code needs to be changed. That will take some time.

Thanks everyone for helping.

Andre

From: freeswitch-users-bounces at lists.freeswitch.org [mailto:freeswitch-users-bounces at lists.freeswitch.org] On Behalf Of Michael Jerris
Sent: Thursday, August 28, 2014 11:10 AM
To: FreeSWITCH Users Help
Subject: Re: [Freeswitch-users] high pdd

Most likely it is neither of these.  I'm seriously doubting that the mutexes are causing problems unless you are actually seeing multiple calls blocking each other, which i don't think is happening here.  I think you need to be looking at what is using the most cpu.  Have you looked at running in the profiler?

On Aug 27, 2014, at 6:10 PM, Dave Kompel <drk at drkngs.net <mailto:drk at drkngs.net> > wrote:

Does anyone still have a dump with the stack trace of those threads in them. The more I think about this, the more it “feels” like it’s not the mutex code in libapr, but what is calling it. Can someone post one level up in the stacktrace?

--Dave

From:  <mailto:freeswitch-users-bounces at lists.freeswitch.org> freeswitch-users-bounces at lists.freeswitch.org [ <mailto:freeswitch-users-bounces at lists.freeswitch.org> mailto:freeswitch-users-bounces at lists.freeswitch.org] On Behalf Of Andre
Sent: Wednesday, August 27, 2014 5:34 PM
To: 'FreeSWITCH Users Help'
Subject: Re: [Freeswitch-users] high pdd

Thanks I’ll do that in the morning.

We still will try Travis APR code but I’m not sure how to compile it. Any hits would be grateful.  I’ve read the read me files but I still get errors :/

Also, I’ve installed Debian as suggested and hope to test soon.

Andre

From:  <mailto:freeswitch-users-bounces at lists.freeswitch.org> freeswitch-users-bounces at lists.freeswitch.org [ <mailto:freeswitch-users-bounces at lists.freeswitch.org> mailto:freeswitch-users-bounces at lists.freeswitch.org] On Behalf OfMoishe Grunstein
Sent: Wednesday, August 27, 2014 5:22 PM
To: FreeSWITCH Users Help
Subject: Re: [Freeswitch-users] high pdd

You should put this in a Jira, this thread is getting very complex to track.

From:  <mailto:freeswitch-users-bounces at lists.freeswitch.org> freeswitch-users-bounces at lists.freeswitch.org [ <mailto:freeswitch-users-bounces at lists.freeswitch.org> mailto:freeswitch-users-bounces at lists.freeswitch.org] On Behalf OfAndrew
Sent: Wednesday, August 27, 2014 4:25 PM
To: 'FreeSWITCH Users Help'
Subject: Re: [Freeswitch-users] high pdd

Hi,

I know Andre is trying on UNIX but wanted to share what I found..

I too have the same issue. But I am using Master x64 embedded.

After 2600 threads the program just keeps adding a ton threads (very fast) and wireshark shows

Lots of re invites.  My magic number is 1300 ports(bridge).  CPS does not seem to matter but its more ports/Threads.

Not sure but when doing a dump I see EnterCriticalSection(&cond->csection); is being pointed to in thread_cond.c

static APR_INLINE apr_status_t _thread_cond_timedwait(apr_thread_cond_t *cond,

                                                      apr_thread_mutex_t *mutex,

                                                      DWORD timeout_ms )

{

    DWORD res;

    apr_status_t rv;

    unsigned int wake = 0;

    unsigned long generation;

    EnterCriticalSection(&cond->csection);

    cond->num_waiting++;

    generation = cond->generation;

    LeaveCriticalSection(&cond->csection);

    apr_thread_mutex_unlock(mutex);

    do {

        res = WaitForSingleObject(cond->semaphore, timeout_ms);

        EnterCriticalSection(&cond->csection);

        if (cond->num_wake) {

            if (cond->generation != generation) {

                cond->num_wake--;

                cond->num_waiting--;

                rv = APR_SUCCESS;

                break;

            } else {

                wake = 1;

            }

        }

        else if (res != WAIT_OBJECT_0) {

            cond->num_waiting--;

            rv = APR_TIMEUP;

            break;

        }

        LeaveCriticalSection(&cond->csection);

        if (wake) {

            wake = 0;

            ReleaseSemaphore(cond->semaphore, 1, NULL);

        }

    } while (1);

    LeaveCriticalSection(&cond->csection);

    apr_thread_mutex_lock(mutex);

    return rv;

}

From:  <mailto:freeswitch-users-bounces at lists.freeswitch.org> freeswitch-users-bounces at lists.freeswitch.org [ <mailto:freeswitch-users-bounces at lists.freeswitch.org> mailto:freeswitch-users-bounces at lists.freeswitch.org] On Behalf OfAnthony Minessale
Sent: Wednesday, August 20, 2014 6:13 PM
To: FreeSWITCH Users Help
Subject: Re: [Freeswitch-users] high pdd

 <https://github.com/traviscross/apr> https://github.com/traviscross/apr

On Wed, Aug 20, 2014 at 4:44 PM, Andre < <mailto:andretodd at verizon.net> andretodd at verizon.net> wrote:

Thanks Anthony,

I can provide the Dump for you it’s 8gb unzipped so I’ll compress it. The tool I use is Visual studio but I’m sure there are other tools too.

I can run a the profiler tool in visual studio to find the CPU.

Also how do I get the modified copy of the APR?

Thanks again guys.

Andre

From:  <mailto:freeswitch-users-bounces at lists.freeswitch.org> freeswitch-users-bounces at lists.freeswitch.org [mailto: <mailto:freeswitch-users-bounces at lists.freeswitch.org> freeswitch-users-bounces at lists.freeswitch.org] On Behalf OfAnthony Minessale
Sent: Wednesday, August 20, 2014 5:26 PM

To: FreeSWITCH Users Help
Subject: Re: [Freeswitch-users] high pdd

The cond waits are all of your session threads in bypass media mode asleep in the state machine.

Its hard to tell with that blurry screen shot what else is going on.  Is there a way to produce that data in a way we could browse it?

You should be looking for threads in the sofia sip stack and the ones doing originate if you are worried about pdd.  Also which threads are costing the most CPU.  If conditionals are implemented in windows as waitforsingleobject i have no idea if that's as efficient as it is in linux.

You could try forklifting in our modified copy of APR that has been updated to work with their latest release and see if anything has improved in the windows implementations of conditionals.  I recall Jeff had to backport in at least one patch for that over the years.

On Wed, Aug 20, 2014 at 4:11 PM, Andre < <mailto:andretodd at verizon.net> andretodd at verizon.net> wrote:

Ok I’ll try that next. I’ll create xml dialplan to see if that makes a difference.

One note, when I didn’t load the deck in my managed code my dial plan did return a 503 in 0 – 20 milliseconds. Not sure if that matters since no bridge was ever called.

Thanks again,

Andre

From:  <mailto:freeswitch-users-bounces at lists.freeswitch.org> freeswitch-users-bounces at lists.freeswitch.org [mailto: <mailto:freeswitch-users-bounces at lists.freeswitch.org> freeswitch-users-bounces at lists.freeswitch.org] On Behalf OfMichael Jerris
Sent: Wednesday, August 20, 2014 5:02 PM

To: FreeSWITCH Users Help
Subject: Re: [Freeswitch-users] high pdd

I would test this without any of your managed code at all, to confirm that it is not the bottleneck.

On Aug 20, 2014, at 4:28 PM, Andre < <mailto:andretodd at verizon.net> andretodd at verizon.net> wrote:

I added the new code but it doesn’t seem to fix it.  :/

From:  <mailto:freeswitch-users-bounces at lists.freeswitch.org> freeswitch-users-bounces at lists.freeswitch.org [ <mailto:freeswitch-users-bounces at lists.freeswitch.org> mailto:freeswitch-users-bounces at lists.freeswitch.org] On Behalf Of Andre
Sent: Wednesday, August 20, 2014 3:37 PM
To: 'FreeSWITCH Users Help'
Subject: Re: [Freeswitch-users] high pdd

Thanks I’ll work on it right a way.

From:  <mailto:freeswitch-users-bounces at lists.freeswitch.org> freeswitch-users-bounces at lists.freeswitch.org [ <mailto:freeswitch-users-bounces at lists.freeswitch.org> mailto:freeswitch-users-bounces at lists.freeswitch.org] On Behalf Of Peter Olsson
Sent: Wednesday, August 20, 2014 3:30 PM
To: FreeSWITCH Users Help
Subject: Re: [Freeswitch-users] high pdd

It seems it might be stuck in the condition code for Windows.

Try this timing patch for Windows, and see if it makes any difference. I've tested in on lower load, but it would be interesting how it performs in this scenario.

 <https://jira.freeswitch.org/secure/attachment/20613/windows-waitable-timer-diff.txt> https://jira.freeswitch.org/secure/attachment/20613/windows-waitable-timer-diff.txt

2014-08-20 21:13 GMT+02:00 Andre < <mailto:andretodd at verizon.net> andretodd at verizon.net>:

I will, need to get hardware and figure out how to use mono.

In the meantime, I’ve attached some of the call traces.   

<image001.jpg><image002.jpg>

<image003.jpg>

From:  <mailto:freeswitch-users-bounces at lists.freeswitch.org> freeswitch-users-bounces at lists.freeswitch.org [mailto: <mailto:freeswitch-users-bounces at lists.freeswitch.org> freeswitch-users-bounces at lists.freeswitch.org] On Behalf OfBrian West
Sent: Wednesday, August 20, 2014 3:05 PM
To: FreeSWITCH Users Help
Subject: Re: [Freeswitch-users] high pdd

Try the same thing on Debian 7  if possible.

On Wed, Aug 20, 2014 at 12:35 PM, Andre < <mailto:andretodd at verizon.net> andretodd at verizon.net> wrote:

Hi, I’m using the master. Should I try 1.4.7?

I’ve done a dump after things started to go bad/ backed up.

I see 8193 threads in _thread_cond_timedwait. Looks like its on line 81 of thread_cond.c  res = WaitForSingleObject(cond->semaphore, timeout_ms);

The value of timeout_ms is  <tel:2016850392> 2016850392.

Not sure if this means anything.

Thanks for the support guys.

Andre

From:  <mailto:freeswitch-users-bounces at lists.freeswitch.org> freeswitch-users-bounces at lists.freeswitch.org [mailto: <mailto:freeswitch-users-bounces at lists.freeswitch.org> freeswitch-users-bounces at lists.freeswitch.org] On Behalf OfBrian West
Sent: Wednesday, August 20, 2014 11:57 AM

To: FreeSWITCH Users Help
Subject: Re: [Freeswitch-users] high pdd

1.2.X is EOL, 

1.4.7 is the current release.

We have releases and tags in git that no longer carry the 'stable' moniker starting with 1.4.  

On Sun, Aug 17, 2014 at 1:39 PM, Andre Demattia < <mailto:andretodd at verizon.net> andretodd at verizon.net> wrote:

Thanks, I'm using the newest master and the last stable 1.2.23
We are willing to pay to get this fixed.
Andre

  _____  

From:  <mailto:anthony.minessale at gmail.com> Anthony Minessale
Sent: ‎8/‎17/‎2014 12:30 PM

To:  <mailto:freeswitch-users at lists.freeswitch.org> Freeswitch-users
Subject: Re: [Freeswitch-users] high pdd

We dont have a lot of tools to troubleshoot windows.  You should mention windows in your subject to have a better chance to get the attention of our windows users.

You should consider installing debian 7 on the same hardware and perform similar tests to help determine if your issue is o/s specific or not.

You also have not indicated the revision of FS.  If you have not worked with latest master, you should be also to rule out improvements.

On Aug 17, 2014 9:54 AM, "Andre Demattia" < <mailto:andretodd at verizon.net> andretodd at verizon.net> wrote:

Yes that's off and I believe all the databases are off too. No authentication or registration is enabled.  The sizes of the database are 0 until I call the shutdown command then its about 90k.  I have the DBS in Ra drive too.

Windows seems pretty good until the 3000 sessions per min mark. Then I just get long pdd and the threads get backed up then no more sessions can be created. 

Funny thing is I has 10 profiles and the same results. I also tried two consoles with the same results.

FreeSWITCH is great.

I'd be grateful for help. I've worked on this for 2 years and hate to give up. 

If any one from  <mailto:consulting at freeswitch.org> consulting at freeswitch.org is willing to help I'd appreciate that.

Thanks again,
Andre

  _____  

From:  <mailto:anthony.minessale at gmail.com> Anthony Minessale
Sent: ‎8/‎17/‎2014 10:13 AM
To:  <mailto:freeswitch-users at lists.freeswitch.org> Freeswitch-users
Subject: Re: [Freeswitch-users] high pdd

Did you comment out manage-presence ?  Also we have no real idea how well FS performs in windows under extreme load.

On Aug 17, 2014 5:42 AM, "Andre Demattia" < <mailto:andretodd at verizon.net> andretodd at verizon.net> wrote:

Thanks Ken, I'll try the consulting again.
Andre

  _____  

From:  <mailto:krice at freeswitch.org> Ken Rice
Sent: ‎8/‎17/‎2014 1:19 AM
To:  <mailto:freeswitch-users at lists.freeswitch.org> FreeSWITCH Users Help
Subject: Re: [Freeswitch-users] high pdd

if the last post does not work contact  <mailto:consulting at freeswitch.org> consulting at freeswitch.org for pro help from one of the developers... thats about as far as the list will take you

Ken

Sent from my iPad

On Aug 16, 2014, at 15:23, "Andre" < <mailto:andretodd at verizon.net> andretodd at verizon.net> wrote:

Going with the idea it’s IO related I did a db_cache status. I just disabled limit in my code by removing it but limit should be HASH not a database however I stills how a database.

Below is the results of the db cache status. I’m not real sure what its saying. I should have all registration and limit disabled and should not be using any sql database from my understanding.

Anyone know what the results mean? I’ve only tested 2 calls on this status.

[The entire original message is not included.]

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.freeswitch.org/pipermail/freeswitch-users/attachments/20140829/1d88bcf6/attachment-0001.html 