-------- Original Message --------

Subject:	[ELOG] SL5 python update breaks running jobs
Date:	Tue, 19 Jun 2012 15:27:14 +0200
From:	Rodney Walker <[log in to unmask]>
To:	Andreu Pacheco Pages <[log in to unmask]>, garonne vincent <[log in to unmask]>, ADCoS team <[log in to unmask]>, atlas-adc-expert <[log in to unmask]>, Ghita Rahal <[log in to unmask]>, eric lancon <[log in to unmask]>, Point1 Shifter <[log in to unmask]>, Daniel Geerts <[log in to unmask]>, Teeba Rashid <[log in to unmask]>, Erekle Magradze <[log in to unmask]>, Syue-Yi Liao <[log in to unmask]>

A new ELOG entry has been submitted:

Logbook             : ATLAS Computer Operations Logbook
Author              : Rodney Walker
Email               : mailto:[log in to unmask]
Severity            : Less Urgent
Cloud               : All
Level               : T1 | T2
Affect Users        : Yes
Subject             : SL5 python update breaks running jobs

Logbook URL         : https://atlas-logbook.cern.ch/elog/ATLAS+Computer+Operations+Logbook/37014

=================================

Hi,
Prague currently has many failures, starting when they did an sl5 security update of python and its libs.
http://listserv.fnal.gov/scripts/wa.exe?A2=ind1206&L=scientific-linux-errata&T=0&P=1963
python-2.4.3-46.el5_8.2

Looking more carefully one sees that jobs running before the upgrade, fail at the metadata creation stage. It
seems the new python xml lib is incompatible with the old. Jobs starting after the upgrade, including robot
jobs, are fine.

This means that a WN upgrade is not transparent to running jobs, and would need draining. I am not sure if this
is known to the SL packagers - perhaps python apps are supposed to import everything at the start of the job to
avoid this.

Anyway I recommend that you do not do WN updates until this is resolved. The updates are currently only in the
fnal repo, and not yet cern, but I expect this will change.

Cheers,
Rod.