[Paraview-developers] BUG?: Where can I get "mpirun" other than the folder of a released ParaView?

[Biddiscombe, John A.]

Utkarsh,


I spoke too soon. There are several bugs. I cannot launch jobs on linux either. This morning I removed my patches to be 100% certain it wasn't anything I was doing that broke things and I posted my earlier message before retesting.

in pqServerLauncher lines 643 to 651, I had to comment out the block.
Reason :
on the Cray and the viz cluster I launch my pvservers using a script which generates a slurm batch script using the params I pass in (there's some xml on the wikie I pusted last year which describes the sequence, this is now slurm instead of PBS, but it's the same). the script is submitted to slurm and returns almost immediately so these lines return QProcess - finished (!=running), and abort the connection. When I comment these out, the server connection resumes as usual and all is well.

As for the pvserver launch that you just mentioned regarding args etc. My experience was that about 4 weeks ago (?? but I'm not certain) I switched to the master branch and all my windows stuff stopped working (which is where I do daily debugging), I went back to my own private branch, which was rebased onto 3.12 Tag and it worked. I believe 3.12 was ok, but somewhere in the last month or so something got broken. This was one of the (many) reasons why I was so frustrated last week and came to the conclusion that nobody else uses paraview in parallel. I apologise profusely for not fixing all these issues myself, but I can't use all the Gerrit tools and suchlike so I no longer make contributions other than by ranting on the mailing lists :(

JB

From: Utkarsh Ayachit [mailto:utkarsh.ayachit at kitware.com]
Sent: 10 February 2012 17:15
To: Biddiscombe, John A.
Cc: ParaView Developers
Subject: Re: [Paraview-developers] BUG?: Where can I get "mpirun" other than the folder of a released ParaView?

Robert did some initial digging and we have starting point: If mpiexec is launched with full pvserver path, for some reason it ends up passing wrong arguments to pvserver and the launch of pvserver fails. I have some idea why that may be happening, but to confirm, do you know if this problem existed in 3.12 as well?

Utkarsh


On Fri, Feb 10, 2012 at 11:00 AM, Biddiscombe, John A. <biddisco at cscs.ch<mailto:biddisco at cscs.ch>> wrote:
Utkarsh

Don't ruin your weekend on my behalf. It's not a show stopper, I'm using the cluster and the cray most of the time, but for debugging filters locally in parallel, visual studio is unbeatable, and the fact that my regular routine on windows is broken is mostly annoying.

I'll have another go at finding out what's wrong and hopefully fix it myself.

JB


From: Utkarsh Ayachit [mailto:utkarsh.ayachit at kitware.com<mailto:utkarsh.ayachit at kitware.com>]
Sent: 10 February 2012 16:39
To: Biddiscombe, John A.

Cc: ParaView Developers
Subject: Re: [Paraview-developers] BUG?: Where can I get "mpirun" other than the folder of a released ParaView?

I'll look into it this weekend. That means I will have to touch a Windows machine (alas!) but I'll blame you for it :).

Utkarsh
On Fri, Feb 10, 2012 at 10:15 AM, Biddiscombe, John A. <biddisco at cscs.ch<mailto:biddisco at cscs.ch>> wrote:

>

By the dialog, you mean the warning about "StereoCapable" window, right?

<



Correct.



Just FYI.



Regarding the mpi connection to pvservers. I tried a few more experiments. Last night I manually started servers, then used reverse-connection local and it worked. Today I tried the same with my own build and it worked, but when I use the GUI, it does not work. I tried using the GUI from the downloaded binary and it also does not work. pvservers are correctly created, but the GUI does not connect to them.

Using reverse connection from the GUI, it also shows this error

[cid:image001.png at 01CCE819.103EA870]





Which is exactly the one I sometimes see from my own compilation (which uses mpich2 not openmpi).



The code in the pqServerLauncher was modified quite a lot recently and I can't quite locate where the launch is failing. I've tested this on several windows machines and none of them work unless I manually start servers. I'm sorry I can't find what I'm doing wrong as I'm sure you wouldn't release an RC2 unless you'd tried it yourselves and it worked, but if there's anything I can test, I will do. My own suspicion is that somehow the args to the pvserver launch commands are getting lost so the connection is not being correctly initiated, but everything looks normal under the debugger. I ought to point out that I've been using the same settings for years and only when I switched to master (and now to 3.14-RC2) a while back did this problem arise. However I seem to be the only person with this trouble :(



Linux works fine. Only windows is affected. Can anyone out there test and see if their gui launched servers work ok?



Thanks



JB




-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://public.kitware.com/pipermail/paraview-developers/attachments/20120210/441b3e38/attachment-0001.htm>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: image001.png
Type: image/png
Size: 10284 bytes
Desc: image001.png
URL: <http://public.kitware.com/pipermail/paraview-developers/attachments/20120210/441b3e38/attachment-0001.png>