Re: GNU Parallel Bug Reports --memfree

[Ole Tange]

On Mon, Apr 25, 2016 at 2:52 PM,  <address@hidden> wrote:

> I parallelize every program within my pipeline separately and I use version
> 20160422 of parallel. In the case of base recalibrator parallelization the
> computations crashes without any ERROR info. I call parallel as follows:
> $pathToParallel --no-notice --joblog

Consider running '--bibtex' once.

> "/home/geryk/NGS_programs/GnuParallel/log" --jobs 0 --memfree $inputSize"m"
> baseRacalibrator_pass1 \
> ::: "address@hidden" ::: "$pathToProject" ::: "$deactivateFilters"
>
> where $inputSize is my estimate of memory requirements. If I add parameter
> --delay 15, the computations finishes succesfully.
> But if I decrease the parameter to --delay 10 computations crashes.
> In the second application of base recalibration (this program must be
> applied twice within pipeline) I must increase --delay to 30sec in order to
> successfully finish computations.
> It seems that the parameter --delay is hard to predict..
>
> Have you some interpretations of my observations?

You do not retry the jobs killed by too little memory.

Try adding --retries 100 and removing --delay.


/Ole