[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: ROOT file_ids and Bundle_ids
> DMM:
>
> I expressed my concerns last month about faking Objectivity into
> generating db numbers as fileids for ROOT files, and I thought that
> Doug and Torre, among others, shared my concerns. By the way,
> implementing that deception may involve some subtleties.
>
> My understanding is that David Zimmerman is working on a method for
> generating consistent fileids for ROOT and Objectivity files, with
> a simple interface for use by Storage Manager components to translate
> from ids to names. If he does not have time for this, I will volunteer
> to do it.
For MDC2, it does not matter to the Storage Manager how file_ids are
assigned, only that they are 16 bit file_ids (similar to the Objectivity
file_ids). The actual file_name is needed by the cache manager and as long
as it has some way of getting it, that's OK. So, your plan is fine as long
as there is a way to get file_names from file_ids.
> > 2) All file names (including path_names if more that a single
> > directory is used) will be retrieved from the Objectivity catalog.
> >
>
> DMM:
>
> I would be happier if we could amend this to say "will be retrieved
> from the Objectivity federation."
That's fine.
> > 3) We assume files can be in different directories. However, the same
> > directory structure will exist beyond a root-path in HPSS as in Unix.
> > (e.g. suppose the Objectivity root-path is o1/o2/o3 and HPSS root-path
is
> > h1/h2, then a file in Objectivity directory o1/o2/03/x/y/z/file_name,
> > will be in the HPSS directory as h1/h2/x/y/z/file_name.) Thus, given
the
> > objectivity path, the HPSS path can be generated.
> >
>
> DMM:
>
> I'm not sure I understand this. Do you mean that I can have an
> arbitrary number of prefix pairs:
>
> o1/o2/o3 corresponds to HPSS h1/h2
> o1/o2/o4 corresponds to HPSS h1/h3
> o1/o5/06 corresponds to HPSS h1/h4
> o7/o8 corresponds to HPSS h5/h6/h7
> ...,
>
> or that I can have only one, so that the HPSS directory structure
> must (after substituting prefixes) precisely mirror the disk directory
> structure?
The latter is correct. i.e. all prefixes uare the same and the Objectivity
and HPSS subdirectories appended to the prefixes are the same. Thus, if the
prefix: o1/o2/o3 corresponds to HPSS h1/h2,
Then files:
o1/o2/o3/x/file1 corresponds to HPSS h1/h2/x/file1,
o1/o2/o3/x/file2 corresponds to HPSS h1/h2/x/file2, etc.
(where x can be sunstituted with any subdirectory path.)
> > 4) The QM will pass to the EI only the bundle_id and the set of
> > event_ids that qualify for this bundle. The EI will return this
> > bundle_id when releasing the bundle.
> >
>
> DMM:
>
> Can Alex or Henrik comment on this: How difficult would it be for MDC2
to
> support an interface that takes a bundle_id as input and returns the
> corresponding filelist (fileids or filenames)? If this is done, I think
> I can live with the bundle_id approach for the OOI in MDC2.
As Alex said: we can return file_ids for the bundle (or he can provide an
interface so the EI can get it), but the key point is that we need a unique
though temporary) id for the bundle returned to the QM when the EI is done
with that bundle.
Let us know how you decide to proceed with ROOT file_id assignment.
Thanks, Arie.