Owen has my vote.
owen maroney wrote:
> Burke, S (Stephen) wrote:
>
>> LHC Computer Grid - Rollout
>>
>>> [mailto:[log in to unmask]] On Behalf Of Ake said:
>>> No you publish a RE for a particular application(class) possibly
>>> particular to a VO) that guarantees that the neccesary things are there.
>>
>>
>>
>> But that doesn't address the original issue of having a large list of
>> RPMs. The number of general client classes is pretty small: basically we
>> have clients for the WMS, data management, information system, VOMS and
>> globus, all of which are going to be needed on a standard grid site, so
>> that doesn't get you anywhere in terms of reducing the RPM count.
>
>
> OK, so there are a number of rpm's which will provide these client
> libraries/tools. All/most of these rpm's should be starting with edg,
> lcg or vdt, I'd have guessed.
>
> But this just begs a reformulation of the original question why is the
> list of rpm dependancies in the lcg_WN including so many rpms that do
> not start with edg, lcg or vdt? (and are all those edg,lcg,vdt rpms
> really needed or does it, perhaps, just seem easier to install the lot
> but only configure what's needed? :-0 )
>
> If any of these grid client tools (say, lcg-do-something-nice) need a
> specific xml parser (for example), then that ought to be a dependancy
> with the rpm that supplies the lcg-do-something-nice. It shouldn't need
> to be an explicit dependancy within lcg_WN, as it is the point of the
> lcg_WN rpm is that it is supplied to a tool like apt/yum/up2date and
> that tool resolves all the implicit dependancies itself.
>
> So if the xml parser is an explicit dependancy in the lcg_WN rpm, either:
>
> 1 - that explicit dependancy is unnecessary. It is an implicit
> dependancy in one of the contained rpms;
>
> 2 - or it is a requirement of one of the client tool rpms but that rpm
> is not properly made and doesn't provide a proper list of its own
> dependant rpms;
>
> 3 - or it isn't actually a dependancy of any of the needed client rpms,
> someone included it in the list once because they found it useful to
> have around, and it's just never been removed...
>
> Clearly in case 1 or 3, it ought to be removed from lcg_WN! In case 2
> it the badly packaged client tool rpm ought to be fixed, and then it
> should be removed.
>
> In the end, the only thing that should be in lcg_WN (or any other of the
> meta-rpms) is probably a half dozen meta-rpm's themselves, each
> corresponding to a particular set of client tools that is required on
> the WN. Everthing else that gets installed should be because they are a
> clear dependancy of one of these well defined client tools.
>
> cheers,
> Owen.
|