> > This came up briefly in the face to face and maybe if we have an agreed
> > list we can push we can get something done about them.
> >
> > To kick things off here's my top three:
> > [...]
>
> This falls under robustness, I guess: I'd like to see proper support for LFNs
> in the lcg_utils. At the moment, e.g. lcg_cp fails if the first SE in the
> replica list is off-line for some reason, even though there are other replicas
> of the same LFN available.
>
> The reply I got from the developers was along the lines of "not our problem,
> the VO needs to write a copy-script that loops over the replicas"
>
> My question why there was any LFN support at all in lcg_utils then if it
> didn't actually work was not answered :-)
Wholeheartedly agreed here. First hand user experience shows 30 to 40%
job-failure rate solely related to lcg_cp failure (despite the number of
replicas).
Of course the global computing models are "to send the jobs to
the data", but while all this machinery is allegedly put into place I
suppose there's abundance of users who get smacked in their face at their
first try simply because of this shortcoming of the middleware, as they
illude themselves the LFNs are actually what they are advertised for.
Or what was the reason to have LFNs at all?
cheers,
gianfranco
|