[DRMAA-WG] Some thoughts on drmaa_wifaborted vs. DRMAA spec

[Piotr Domagalski]

Hi Hrabri,

On Wed, Aug 6, 2008 at 11:24 PM, Rajic, Hrabri <hrabri.rajic at intel.com> wrote:
> It seems that the SGE people are on vacation ( I am starting mine tomorrow ...).

Yes, apparently. I'm starting mine on Friday ;-)

The thing is that I'm rather confused about this drmaa_wifaborted
stuff and have no idea whether it's just something that should have
been changed long time ago in the specification or the SGE
implementation is buggy. As a matter of fact, as I've been involved
(at least partly) in implementations for LSF, PBS and did some fixes
for Condor,  I can tell that most implementations also behave just
like SGE, i.e. drmaa_wifaborted = true if job was terminated, whatever
the actual state.

But please correct me if I'm wrong. I haven't seen the source codes
for quite a long time.

Another question arises if the specification is actually how it was
meant to be. What is the output of wifexited, wifsignalled, wifaborted
for job that was terminated during execution? I have always assumed
that aborted = true. But if we assume that aborted = false in this
case (because it was terminated after having entered the running state
as spec says) then should it be signaled = true?

--
Piotr Domagalski