On Wed, Feb 09, 2011 at 10:27:22AM -0600, Larry Vaden wrote:
> On Wed, Feb 9, 2011 at 10:11 AM, Ewan Mac Mahon <[log in to unmask]> wrote:
> >
> > However, someone from CERN IT (discussing some experiment software which
> > seems to break on 5.6) has just told a grid deployment meeting that:
> >  "FNAL released SL 5.6 last week"
> > Clearly either my understanding is flawed or his is; which is it please?
> 
> If CERN IT security allows us to know what component of the OS was
> involved, inquiring minds want to know :)
> 

I'm a little bit hazy on the details, but there are some slides from the
meeting here[1]:
 http://indico.cern.ch/getFile.py/access?contribId=8&sessionId=1&resId=1&materialId=slides&confId=106641

The key part seems to be:

- LHCb (and LCD) use tcmalloc.so
  - glibc dependent
    - May require recompilation when glibc changes
  - Not part of SLC
- Code run on RHEL/SLC 5.6 without recompiling tcmalloc.so broke

The upshot is that there's some concern about whether it's safe to
upgrade to 5.6 straight away (whereas the update releases are usually
pretty good at not breaking anything), but there was some confusion
about whether or not SL 5.6 was 'out' or not, which I think we've now
cleared up.
 
It's not clear whether this is an OS bug at all, or someone doing
something completely unreasonable in experiment software, but AIUI it
affect RHEL5u6 as well, so it seems not to be an SL bug at least.

Ewan

[1] Or at the bottom of this page, if that long link gets mangled:
 http://indico.cern.ch/conferenceDisplay.py?confId=106641