[CIG-LONG] [0]PETSC ERROR: User provided function() line 0 in unknown directory unknown file

Walter Landry walter at geodynamics.org
Mon Sep 26 11:09:59 PDT 2011


"Brandon, Mark" <mark.brandon at yale.edu> wrote:
> Walter,
> I have not figured out what caused the SIGHUP error. My first attempts
> were running in the background (&), and then I tried running using screen,
> and then nohup in the background.
> All of these failed. I don't know why. What I did find is that I was able
> to run to completion to two runs, each going for 50 hr cpu using gnome
> window run remotely in an NX client.
> I wish I could find what was causing the instability. I ran strace for the
> first run I did using gnome/nx and found nothing unusual. For the time
> being, I will run in the gnome/nx
> configuration. But, given my curiosity, I would like to figure out why the
> other configurations were giving me trouble. Any ideas?

This still sounds like an ssh issue.  If you run Gale in the
background, log out and then log back in, is Gale still running?

> Two other things:
> 1) I hope you put this exchange in CIG_LONG. I suspect that others get
> various PETSC errors, but I was only able to find one discussion about
> such errors in the CIG_LONG list serve.

I am copying this to the list.  I would prefer it if everyone would
send their questions to the list and then I would respond on the list.
It even says that in the manual, but most people still send messages
to me directly.

> 2) There are many features of Gale that remain incompletely documented. I
> have found it helpful to go to Underworld to look for more detailed
> documentation of the various routines and plugins for Gale. 

Is the problem that some things are documented, but it is unclear or
incomplete?  Or is it that some plugins and components are not
documented at all?  I try to document everything that I know works.
So usually, if something is not documented, it is because I do not
know if it works at all.

> I know that it is big job to generate a more detailed software
> documentation. A compromise would be to set up a wiki site that
> contained the manual and perhaps more, but also allow authenticated
> users to add to the site. This arrange seems to work well for the
> Paraview wiki.

There is working group area on the CIG website meant for things like
that, but no one has ever actually used it.  I am happy for you or
anyone else to use it.

Cheers,
Walter Landry
walter at geodynamics.org



"Brandon, Mark" <mark.brandon at yale.edu> wrote:
> -----Original Message-----
> From: Walter Landry <walter at geodynamics.org>
> Date: Wed, 21 Sep 2011 10:38:04 -0700
> To: Mark Brandon <mark.brandon at yale.edu>
> Subject: Re: [CIG-LONG] [0]PETSC ERROR: User provided function() line 0 in
> unknown directory unknown file
> 
>>"Brandon, Mark" <mark.brandon at yale.edu> wrote:
>>> Walter,
>>> Thanks for trying the run. I should have mentioned that I ran the input
>>> using a reduced form (by 10 in each direction). It ran just fine.
>>> My question is more general... Can Gale send a SIGHUP to PETSC? I
>>>suspect
>>> that the answer is no. My guess is that the problem has to do with the
>>> host, which forces a signal 1 for the process.
>>> I am running the same job in gnome window as a NX client. So far, all is
>>> well. I suspect that the problem is related to issues associated with
>>> running the job remotely via ssh. Running the job in a NX client make
>>> allow the process to run without interruption. I will provide you with
>>>an
>>> update when the job finished. I remain surprised however that the
>>>process
>>> was halted by a SIGHUP while running under nohup....
>>
>>Some ssh servers will automatically close the connection if there has
>>not been any activity.  Unless you run the code in the background,
>>even running it under nohup will kill it.  The larger runs take
>>longer, so that may be why you do not see a problem with smaller runs.
>>
>>In general, I have never seen Gale send a SIGHUP to itself.
>>
>>Cheers,
>>Walter Landry
>>walter at geodynamics.org


More information about the CIG-LONG mailing list