[Paraview-developers] communication errors

Tue Aug 23 03:10:31 EDT 2011

Just to say that I picked up the gatekeeper changes from Friday and everything seems to be behaving normally now. Rolling back the progress related changes has solved the wrong tag messages and lockups and I'm able to work again.

thanks

JB

-----Original Message-----
From: paraview-developers-bounces at paraview.org [mailto:paraview-developers-bounces at paraview.org] On Behalf Of Biddiscombe, John A.
Sent: 19 August 2011 21:17
To: Utkarsh Ayachit
Cc: paraview-developers at paraview.org
Subject: Re: [Paraview-developers] communication errors

Utkarsh,

They're called from a slot using Qt::QueuedConnection - I didn't quite understand if that's the good one or not.

If the progress stuff is being backed out, then fine. I'll merge as soon as I see the changes.

JB

-----Original Message-----
From: Utkarsh Ayachit [mailto:utkarsh.ayachit at kitware.com] 
Sent: 19 August 2011 16:03
To: Biddiscombe, John A.
Cc: paraview-developers at paraview.org
Subject: Re: [Paraview-developers] communication errors

John,

Are those methods called in an "on-idle" callback or in a slot
connected using Qt::QueuedConnection? If so, yes, that'd explain it.
And yes checking avoiding calling those methods when
IsInPorgressEvents() is true would be the solution.

We are disabling the progress change that started causing this issues
for 3.12 release, just to give us more time to track down any similar
issues before it gets included in a release. The topic will be merged
in today.

Utkarsh

On Fri, Aug 19, 2011 at 3:56 AM, Biddiscombe, John A. <biddisco at cscs.ch> wrote:
> Utkarsh,
>
> Are there any more patches waiting to go in? Out latest tests are still giving the wrong tag errors and I have a deadline to meet coming up. I must make a decision on whether to try to back port our stuff to 3.10 or persevere with 3.11. If you have more patches please tell!
>
> Also, Are there any things I must not do. For our live/steering work we call
>
>    this->Pipeline->UpdateVTKObjects();
>    this->Pipeline->UpdatePipeline();
>
> do we need to check for IsInProgressEvents() or something before calling this, or are the Update Requests added to a queue in any case.
>
> In several places we trigger updates - If I must wrap these in checks, then no problem, but I'm not sure what to check for.
>
> thanks
>
> JB
>
> -----Original Message-----
> From: Utkarsh Ayachit [mailto:utkarsh.ayachit at kitware.com]
> Sent: 29 July 2011 16:38
> To: Biddiscombe, John A.
> Cc: paraview-developers at paraview.org
> Subject: Re: [Paraview-developers] communication errors
>
> John, the meeting is at 2 pm today (NY time). The relevant fixes
> should be in by later tonight. If you still get the errors, we can
> debug it together. Just to give you a background of why these may be
> happening:
>
> Until recently, when progress events were received by the client, we
> never called "QApplication:processEvents()", we only "painted" the
> progress bar. That was incorrect. Cosnequently progress bar was not
> working on Macs for the longest time. This issue was resolved recently
> (that commit's in master already since it happened before the workflow
> change). However, it now means that if we have any pending slot
> triggers that do things like "gather information", for example (in
> short, those that may trigger server-side communication), these may
> get called while progress events are being processed. Progress events
> imply that server is already busy processing something e.g. rendering.
> If we start send another set of messages while server was busy in a
> certain task, we may end up send server messages that it wasn't
> expecting. That's what causes the WrongTag error messages.
>
> The solution has been to identify such slots and remove them. Most of
> these don't need to be "delayed". In other cases, they need to check
> to ensure no progress-events are pending i.e. the slot isn't being
> called during a progress event. If the update to master doesn't solve
> your problem I can show you examples of how both of these fixes are
> done in certain places in ParaView.
>
> Utkarsh
>
> On Fri, Jul 29, 2011 at 6:00 AM, Biddiscombe, John A. <biddisco at cscs.ch> wrote:
>> Utkarsh
>>
>> Any approximate idea of when this might make it into master. I cherry-picked 4 commits from next which stop the wrong tag messages, but the client still locks up, so paraview (client/server) is still essentially unusable for me. I do some more digging to try to see if any more patches might be required, but until it goes into master, I guess I won't be certain I've got what I need.
>>
>> ta
>>
>> JB
>>
>> -----Original Message-----
>> From: Utkarsh Ayachit [mailto:utkarsh.ayachit at kitware.com]
>> Sent: 28 July 2011 18:52
>> To: Biddiscombe, John A.
>> Cc: paraview-developers at paraview.org
>> Subject: Re: [Paraview-developers] communication errors
>>
>> NO! Switching to next is like working with the most-experimental
>> branch. All changes that don't cause dashboard issues are moved to
>> master every week (on Friday) in what we are calling "Gatekeeper
>> Meetings".
>>
>> Utkarsh
>>
>> On Thu, Jul 28, 2011 at 11:24 AM, Biddiscombe, John A. <biddisco at cscs.ch> wrote:
>>> Should I just switch to next instead of master?
>>>
>>>
>>>
>>> -----Original Message-----
>>> From: Utkarsh Ayachit [mailto:utkarsh.ayachit at kitware.com]
>>> Sent: 28 July 2011 16:12
>>> To: Biddiscombe, John A.
>>> Cc: paraview-developers at paraview.org
>>> Subject: Re: [Paraview-developers] communication errors
>>>
>>> "Wrong tag messages" started showing up as a consequence of a commit
>>> to fix the progress bar that made it to git-master before we switched
>>> to the new style worflow. That issue is currently being addressed on
>>> git-next and will make it to master (hopefull soon).
>>>
>>> Not sure what this is. Try attaching a debugger, but a break point in
>>> the error-macro and look at the call-stack. Does it show anything
>>> fishy?
>>>
>>> Utkarsh
>>>
>>> On Thu, Jul 28, 2011 at 7:06 AM, Biddiscombe, John A. <biddisco at cscs.ch> wrote:
>>>> I get this (error at end) a lot these days when I run in parallel. The server appears to keep on running (I'm using 64 cores on the viz cluster, client on a windows machine - but it still happens even when running only on the windows machine with 4 pvservers).
>>>>
>>>> I suspect it is something to do with my plugins - though this usually happens during an update when I rotate the model or something - sometimes during filter operation. (Note, not using our steering stuff, only standard MPI based modules for readers/filters etc)
>>>>
>>>> How can I debug this and find out what is going wrong? I suspect it's something I've done because it only happens after applying my filters - but I never use paraview for anything else anyway - so it's a self selecting case. I don';t know that it doesn't do this when I run paraview without my stuff (if you follow me).
>>>>
>>>> JB
>>>>
>>>>
>>>> ERROR: In ..\..\..\..\Code\pv-meshless\VTK\Common\vtkSocket.cxx, line 572
>>>> vtkClientSocket (00000000097394A0): Socket error in call to send. An established connection was aborted by the software in your host machine.
>>>>
>>>> ERROR: In ..\..\..\..\Code\pv-meshless\VTK\Parallel\vtkSocketCommunicator.cxx, line 716
>>>> vtkSocketCommunicator (0000000009711450): Could not send tag.
>>>>
>>>> (I also get a lot of 'received wrong tag messages' but not today, so I'm posting this one)
>>>>
>>>> --
>>>> John Biddiscombe,                            email:biddisco @ cscs.ch
>>>> http://www.cscs.ch/
>>>> CSCS, Swiss National Supercomputing Centre  | Tel:  +41 (91) 610.82.07
>>>> Via Cantonale, 6928 Manno, Switzerland      | Fax:  +41 (91) 610.82.82
>>>>
>>>>
>>>> _______________________________________________
>>>> Paraview-developers mailing list
>>>> Paraview-developers at paraview.org
>>>> http://public.kitware.com/mailman/listinfo/paraview-developers
>>>>
>>>
>>
>
_______________________________________________
Paraview-developers mailing list
Paraview-developers at paraview.org
http://public.kitware.com/mailman/listinfo/paraview-developers