[Gretl-devel] Re: MPI bootstrap failures depending on number of iterations
Am 04.11.2021 um 08:42 schrieb Sven Schreiber:
Am 04.11.2021 um 00:38 schrieb Allin Cottrell:
> On Thu, 4 Nov 2021, Sven Schreiber wrote:
>> Am 03.11.2021 um 17:32 schrieb Riccardo (Jack) Lucchetti:
>>> Could it be that there is an extreme corner case, that occurs with
>>> very small probability, that makes some command fail with error under
>>> very extreme circumstances?
>>
>> However, when I run the thing sequentially with the same (high, but not
>> that high, actually) number of replications, there is no problem.
>> Therefore I believe it must be something more MPI-specific.
>
> Sven, why don't you send us (some of us, offlist if you wish) all
> that's needed to produce the problem (if only sometimes)?
>
> There are tools we can use to investigate this, valgrind in particular.
>
Right, will do that eventually. Another thing that occurred to me is
that the output from within an MPI block is notoriously problematic on
Windows, IIRC. So I will do some testing on Linux to see if I can get
some informative output from debugging statements or so.
Still debugging on Windows right now, I believe I have at least
identified one part of my problems:
I have GUI-gretl open and work in there (so gretlw32.exe or whatever),
with mpi blocks, and then in parallel (no pun intended) I execute
gretlcli.exe in batch mode from time to time. (In fact, in
package-building mode --makepkg.) Now, with gretlcli there's no mpi
involved, but it seems that somehow it causes subsequent trouble with
the mpi blocks in GUI-gretl. Maybe affecting the dotdir with some side
effects?
Maybe that already gives you some clue - I will keep investigating and
debugging.
thanks
sven