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

[Utkarsh Ayachit]

John/Alan,

Here's another patch. I think I've addressed most of the issues that I
could reproduce on my Windows machine. My mpiexec kept on failing,
however. I'm not sure if my openmpi version was causing the issues or
something else. So I'd appreciate if you could test it out. (I
verified on Mac that the launching is working fine with
MPI/forward/reverse).

Alan, the missing PV_CLIENT_HOST is back.

Thanks,
Utkarsh

On Fri, Feb 10, 2012 at 9:33 PM, Utkarsh Ayachit
<utkarsh.ayachit at kitware.com> wrote:
> I'm working on it.
>
> Utkarsh
>
> On Fri, Feb 10, 2012 at 8:27 PM, Scott, W Alan <wascott at sandia.gov> wrote:
>> John/Utkarsh,
>> I also want to thank you (John) for finding that GUI client/server doesn't work!!  Because I build master so much, I always do so Linux - and then do a "manual "  (script based) connection.  When I tried testing the GUI connect, it also fails.  (This is Linux GUI, redsky cluster).
>>
>> My error on the client side says: " Server launched aborted before attempting to connect to it."
>>
>> I tried installing the patch that Utkarsh gives with this e-mail, and it seems to fix the issue listed above.  However, the server side then fails.  Looking at the new output for the connect that Utkarsh puts out, I get the following:
>>
>> Server launch command is : /usr/bin/xterm -e ssh redsky /..../bin/paraview_redsky_mesa_3.14.0_go 1 PV_CLIENT_HOST 60 FY127407 --connect-id=15873140
>>
>> Note that PV_CLIENT_HOST is a special word in ParaView 3.12.0.  It has been removed in 3.14.0, and needs to be restored.  Manually hacking the default_servers.pvsc  file to manually define this PV_CLIENT_HOST variable works correctly. The old code is in pqCommandServerStartup.cxx and uses a call to localHostName().
>>
>> Alan
>>
>>
>>
>> -----Original Message-----
>> From: paraview-developers-bounces at paraview.org [mailto:paraview-developers-bounces at paraview.org] On Behalf Of Utkarsh Ayachit
>> Sent: Friday, February 10, 2012 11:32 AM
>> To: Biddiscombe, John A.
>> Cc: ParaView Developers
>> Subject: [EXTERNAL] Re: [Paraview-developers] BUG?: Where can I get "mpirun" other than the folder of a released ParaView?
>>
>> Utkarsh
>>
>>> Alternatively, I can add a new xml attribute that says that the script
>>> blocks or doesn't. That was if the script is supposed to block and the
>>> launch indeed fails,
>>
>> I think I have a better fix. If the script exists with error code then we assume that launch failed, if the script exists without error code then we assume the server has launched and attempt to connect to it.
>> Does that sound good? That way no one has to change any existing pvsc files. (patch attached)/
>>
>>> But this has now gone, so I add a terminal window to paraview so that I can see the results. It'd be quite nice to have a tag to say "Send launch commands/output to output console" as well so that it was easier to see stuff.
>>
>> I've found that it's easier if the pvsc file uses a platform specific terminal like xterm or cmd to launch the script. That way things like entering passwords or such other things can also work seamlessly.
>> I've used elaborate scripts in the past (inspired by Burlen's scripts) that make it possible for the user to check jobs status among other things in this terminal window directly.
>>
>>> Attached patch: No. It doesn't work, I tried standard mpiexec with forward connect, also reverse. Both failed. Forward is just a lock-up --pvservers do start, but connection is not made. Reverse shows the error msg dialog I posted earlier. Thanks for trying.
>>
>> I'll keep working on it.
>>
>> Utkarsh
>>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 0001-Fixes-server-connection-issues-on-windows.patch
Type: application/octet-stream
Size: 6939 bytes
Desc: not available
URL: <http://public.kitware.com/pipermail/paraview-developers/attachments/20120210/75794f2e/attachment.obj>