[R-sig-hpc] R error throwing and poe error catching

Hao Yu hyu at stats.uwo.ca
Thu Aug 8 23:19:59 CEST 2013


Jim,

Try to alter the behave of error in R. Run at the beginning

options(error=quote(assign(".mpi.err", FALSE, envir = .GlobalEnv)))

Check Rmpi's Rprofile usage.

Hao


Jim Gattiker wrote:
> I'm having trouble with running R under "poe" on an iDataPlex (NCAR's
> yellowstone). The purpose of running this way is to enable an interactive
> master instance of R that uses mpi_spawn to launch slaves, in Rmpi. In
> this
> environment it seems I must use PE_MPI and POE.
>
> The problem is, it seems poe catches even warning messages from R (e.g., a
> simple syntax error), and the entire environment crashes out. This happens
> in base R, it's not an issue of parallel resources. This is very
> inconvenient for EDA, when a typo causes hard exit.
>
> My launch looks like:
>
> bsub -Is -q xxx -n 4 -W 1:00 -P yyy tcsh
>
> and then
>
> poe R --no-save -procs 4 -world_sizes 1
>
> It runs, and is fully functional. But, I don't have to do anything exotic
> to get a hard error:
>
>> m
> m
> Error: object 'm' not found
> Execution halted
>
>
> ...and that's it. My interpretation is POE catches the error and doesn't
> know how to handle it.
>
> If anyone has any ideas ... I see two directions for solutions:
> 1) avoid POE, which means get an interactive master set up with a slave
> pool up without spawn(), which means launching them at startup. I've
> tried;
> it doesn't seem to be possible to both launch jobs on all PEs and have
> (just) one remain interactive.
> 2) figure out how to get poe to not trap R's own errors. Maybe this isn't
> an HPC specific problem, if the issue is getting R to not throw its errors
> outside its environment. Not sure how to proceed with that either.
>
> If a specific question will help: why does this error get outside of R,
> and
> why is this normally OK?
>
>      chrs,
>        --jim
>
> 	[[alternative HTML version deleted]]
>
> _______________________________________________
> R-sig-hpc mailing list
> R-sig-hpc at r-project.org
> https://stat.ethz.ch/mailman/listinfo/r-sig-hpc
>


-- 
Department of Statistics & Actuarial Sciences
Office Phone#:(519)-661-3622
Fax Phone#:(519)-661-3813
The University of Western Ontario
London, Ontario N6A 5B7
http://www.stats.uwo.ca/yu



More information about the R-sig-hpc mailing list