AF meeting 23 July 2009

Vincenzo Innocente, Pere Mato, Oliver Keeble, David Quarrie, Axel Naumann, Andrea Valassi

version 0 (24/07/09 @ 11:00)

Follow-up of the proposal of disabling SELinux. Investigate if additional bits are required for the Oracle client (Andrea)

It was not clear from the last meeting if the bit 'execheap' was sufficient to fix the Oracle problem. Unfortunately, Andrea has not been able to reproduce the failure and therefore there there is no conclusion. The assumption for the time being is that we keep it as it is and then we see what fails. We reiterated that the current strategy for dealing with the Oracle problem is to disable the necessary SELinux bit globally and not to have a special installation of it.

David reported tat the main motivation for this new configuration comes mainly from some CORAL and ROOT related problems. Additionally, there are also some compiler installation issues. They would like to have the configuration ready for next Tuesday.
For CORAL there are 3 bug fixes that need to be incorporated:

Bug #31554 (Oracle client library bug). Andrea has received the patch from Oracle, however this requires a full installation of Oracle, then apply the patch to installation, and finally to extract the library. This can only be done on Monday. The LCGCMT/Oracle interface needs to be adapted (version and library location to be changed)
Bug #52795 (SL5 validation). The patch is in CORAL HEAD, but is not yet in any CORAL release.
Task #9927 (Make it easier to use a custom auth service). It is done and released.

For POOL a version may need to be provided to fix (avoid) bug #53130. Markus and Axel are currently working on it.
For ROOT a new patch release will be needed (5.22.00d) and it sould include among other changes:

Bug #52551 (Restrict/control .rootrc search...). It is now in ROOT verion 5.24 but needs to be backported to 5.22.
Contained object that inherits from TObject (no bug report). There is a fix in 5.24 and 5.22.
Bug in the library libXrdClient.so, which is fixed in ROOT 5-24-00, needs also to be backported (if possible) to 5.22.
Bug #53500 (PyROOT memory leak when returning string). It is already fixed in 5.22 branch.
Axel expects that they can tag the patch version during the weekend.

The current client library is the 'full instant client' that weights 70 MB in a single library (the large size comes mainly from error messages in many languages). There is a DB instance (devdb10), which is used for test/development purposes, that uses a special character set, and therefore forces the client library to be the full one. In order to change this we will need the feedback from the others experiments. Decided to do nothing this time and come back on it later.

Andrea raised the question to build the OCCI (C++) libraries or not for this patched version. This created some problems in the past to some ROOT users. It was decided not to put them there for this patched version. Andrea will check with Bertrand to see what is being used to build ROOT (e.g. 64/SLC4(5) 'dev' slot).

David also mentioned an interference with 'distcc' definition between the policies of LCG and Gaudi. Temporary the Gaudi one has been removed.
For the gcc43 compiler installation with LCGCMT we need to drop the sym-links from the install area (used by ATLAS only).

Oliver mentioned that some sites have made available SL5 resources for testing. A problem that has happen is that if user jobs do not require any particular OS they may be scheduled on a SLC5 resource and fail.
The Meta-RPM is being installed, but some centers claim that is carries some probably unnecessary libraries. From the GDB point of view, a site announcing 'SLC5' means that they have disabled the necessary bits from SELinux and the meta-RPM package is fully installed. There is no way to verify the claim.

Compatibility libraries and meta-RPM issues. ATLAS has to sort out the various mechanisms they have to get these libraries deployed since there was a problem related to their installation.

R&D projects

Organizing together with OpenLab two workshops: one 'in-core' performance (15-16 September), and one for 'out-of-core' performance (multi-threading, multi-processes). Experiments will be contracted to identify people interested.
Very successful 5 days course on C++ by Walter Brown. The seminar yesterday was so successful that the room (IT auditorium) was very insufficient.

David asked about the status of request to have builds with the LLVM compiler incorporated into the nightlies. Stefan can clarify this at the next meeting. --> Action Stefan.
Next meeting on August 20th.