SCIENTIFIC-LINUX-DEVEL Archives

June 2010

SCIENTIFIC-LINUX-DEVEL@LISTSERV.FNAL.GOV

Options: Use Monospaced Font
Show Text Part by Default
Show All Mail Headers

Message: [<< First] [< Prev] [Next >] [Last >>]
Topic: [<< First] [< Prev] [Next >] [Last >>]
Author: [<< First] [< Prev] [Next >] [Last >>]

Print Reply
Subject:
From:
Kelvin Raywood <[log in to unmask]>
Reply To:
Kelvin Raywood <[log in to unmask]>
Date:
Tue, 8 Jun 2010 14:39:23 -0700
Content-Type:
text/plain
Parts/Attachments:
text/plain (148 lines)
Some comments from TRIUMF's perspective.  We use CentOS on servers and 
SL on desktops so my comments are heavily biased towards what our 
desktop users need.

Jaroslaw Polok wrote:
>> *Sites - Customized SL releases

For SL-4 and SL-5 we use standard SL plus our own internal repository of 
rpms that make customisations for use at TRIUMF.  The rpms are written 
to be modular and uninstallable (i.e. if they modify a config file, 
removing the rpm intelligently removes the modification)

> So for us [CERN] the way to go would be rather Fedora-like in the future..

We also like this idea.

  >> *Yum - autoupdate
> 
> For SLC we use a little bit different autoupdate script than
> SL does, but we will be looking at the possibility of using yum-cron
> AND yum-updateonboot together.

We use yum-cron and prefer it over yum-autoupdate for several reasons; 
mainly the flexibility of the yum-shell mechanism but also because 
e-mails are only sent when something fails.

It was me that asked for its inclusion last year.  Thanks Troy for 
including it.  I hope it will remain. At the time, I asked that 
yum-updatesd, yum-autoupdate and yum-cron be handled by the alternatives 
mechanism to avoid having more than one active.  But Troy rejected it as 
too complicated and confusing for ordinary desktop users

>> *Kernel-modules
> 
> Since RH introduced the kABI whitelisting with 5 and is going
> to continue to do so for 6: I believe the kABI tracing modules
> (as ones provided by elrepo) are very nice thing, saving us lots
> of recompilations - and do work very well, so I would definitely
> 'vote' for kmods.

Agreed

>> *Security - Bugs - Enhancement Updates
> 
>> If we had them in one big repository, we would have to label them as
>> security, bugs, enhansements, and them make sure we used the
>> yum-security plugin.

That's a nice idea.

> I believe above would address the issue and make the maintenance
> of repositories easier: also from user perspective we've found that
> having too many repositories is just confusing ...

Agreed

>> *Yum Repositories
>> What should we have for default?

I agree with what Jaroslaw wrote. base and all updates enabled, 
everything else disabled.  But this is something that I suspect most 
sites will customise.  We do and will continue to do so.

>> Should we still have "contrib"?
> 
> Looking at current content for SL 5 .. I think - not - most of things
> in there are available from elsewhere - and these which are not could
> be eventually transferred to these existing repos ... (in total
> contrib contains now a small number of packages ... about ~10-12 ?)

While I agree with this, I think it's good to have some sort of 
"contrib" or "extras" repo.   Packages can be added it to it in future 
without having to issue an update to yum-conf (or similar) which can be 
problematic.


>> Should we add "development"?
> 
> A repo that would contain upcoming updates ? -> yes, that would
> be good idea allowing users to prepare themselves for upcoming
> changes.

Isn't this fastbugs?

>> *JAVA
>> I propose we don't add Sun's (Oracle's) java.
> 
> Hmm.. This one I would actually add -> as long as it is not
> available from another repository: alternative option for users
> is going to Jpackage AND going to Oracle downloading stuff here and
> there and rebuilding by themselves - quite painful for end-user
> IMHO.

I disagree with Jaroslaw here and like the original proposal of _not_ 
adding Oracle/SUN Java.  The SUN rpms are terrible and having unsigned 
packages in the base and updates repos causes updates to fail if gpg-key 
checking is enabled.

I started with the Jpackage nosrc rpms and made several modifications to 
fix things like X11 paths, java-ws and the 64-bit browser plugin in 
alternatives.  I then build and sign SUN-java rpms for 
SL-{4,5}-{i386,x86_64} and put them in our local repos.  Our TRIUMF 
customisations exclude the java rpms from the SL  repos to avoid the 
signing-key issue.

We don't have permission to redistribute Oracle/SUN java so I can't make 
them available from here, but I would be willing to contribute them to 
some SL contribs/extras repo, and keep them up-to-date.

Troy - let mw know if you are interested.

> (without willing to start a religious war on OpenJDK vs Sun/Ora Java:
> I believe many users will just have a 'hard' requirement on this
> -> if it is shipped at the 'top SL level' it will be the same version
> used by all the sites which is good for compatibility - especially
> if it matches the version shipped by RH).

Are you referring to the SUN-java that is available to RH subscribers 
through one of the supplementary channels?

I agree that it would be good to be compatible with this and think that 
the packages I build are.

>> *What should we add to SL6
>> At the last Hepix meeting one of the discussions was that many
>> scientists are adding their packages to EPEL.  They would like it if
>> packages that are in EPEL stay just in EPEL and not go into SL unless
>> there is a good reason for it.
> 
> Sounds reasonable to me: why duplicate what is already there
> (at CERN we are slowly removing from SLC packages we were preparing
> ourselves that are now available from EPEL)

Agreed

>> A good reason would be that it is needed during the install.
>> I like this idea and would like to adopt it.
> 
> +1 ;-)
> 
> 
> Thanks Troy for starting the discussion !

Yep.  Thanks also to Jaroslaw for the thoughtful comments.

Kel Raywood
TRIUMF

ATOM RSS1 RSS2