MPU Meeting Thursday 19th February 2009

Buildtools Project

Nearly all MPU LCFG packages have now been converted to the new build tools. The few remaining are all a bit more troublesome. lcfg-server will be kept using the old build tools until we have a way of running the test suite with the new tools. lcfg-utils needs splitting into two, one part for the C code and the other for the Perl code. lcfg-cron contains a bug somewhere which means that upgrading to a new package could end up with crond failing to restart, this happened a while ago when it was last updated. We should now encourage all COs to convert their packages. In particular, if they are doing other modifications they should do the conversion at the same time.

Stephen found a few minor bugs in some CMakeLists.txt files he had written, these have been fixed.

Stephen has started work on the subversion repository support. He has now come up with a plan for the most suitable repository layout.

Power Management Project

Chris is working on writing the Perl code to run the various tests which will decide if a machine can be put to sleep and, if so, when it should wake up again.

Miscellaneous Development

Alastair has mostly completed the virtualisation development work for now. There is documentation available but it needs to be linked from a higher level on the wiki to make it easier to find.

We now have a number of virtualised servers running on central. It would be interesting to find the loading limits for the virtualisation host. All memory has now committed to the virtual hosts but we don't expect any problems as it is not strictly limited to each VM, VMWare is happy to overcommit. The machine has 4 quad-core processors and the load is currently very low (about 1). There are still questions over how well it handles disk IO though. For the record we are currently hosting the following:

  • centaur - beowulf head node
  • chapeaurouge - native rhel5
  • genoa - RPM slave
  • milan - SL5 build host
  • naples - LCFG test slave
  • turin - SL5_64 build host
  • ventimiglia - new bugzilla.inf

Project Prioritisation

The MPU projects have been prioritised be CEG/CSG. Here are some notes from our discussion:

  • The desktop power management project will continue as it is required to meet the University's strategy. It is expected to need one more month of effort.

  • The buildtools project is expected to be completed soon.

  • The RPM submission will be unstalled again and completed.

  • Not much will happen with virtualisation for a while, it is now doing what we need.

  • The LCFG core refactoring will happen but will have regular progress checks (probably quarterly). This fits well with the "agile" approach which Simon and Stephen plan to use where work is split into small chunks and progress is regularly assessed.

  • MPU will take on the work of writing an LCFG component to manage AFS and add nagios monitoring. Stephen volunteered to do this work as he wants to improve his AFS knowledge.

  • MPU will take on the work of managing the teradactyl backup system with LCFG. Chris will take a look at this.

  • We expect to soon have to do lots of operational work related to the transition to SL5.3, probably this will require about 2 weeks of effort.

  • Carol will continue to be associated with the unit and will be doing more MPU work.

Operational

Fix for FC problems
The standard driver provided as part of SL5 now supports multipath so the easiest fix will be to switch to that driver. The driver distributed by QLogic is better but the advantages are not great enough for it to be worth using. Alastair will update the qlogic headers to use the standard driver but keeping the QLogic driver as a selectable option. Alastair will also encourage COs to switch to the standard driver.

Packaging guidelines
Chris has extended the packaging guidelines, they can be found at PackagingGuidelines. Alastair and Stephen need to read through and comment.

DIY DICE with static VMWare
Carol is now working on this using a Macbook and VMWare Fusion as that is what the most vocal users want supported. Carol now has a VMWare Workstation license.

Decommissioning Servers
The MPU junk in the AT server room has now been decommissioned. We have several machines in KB which need to be dealt with, these are cherub, hootsmon and pezenas

KB Server moves
We need to move the MPU servers out of the KB machine room. split should go to Forrest Hill as we already have an RPM cache in the Forum. The FC5 build host (hvar) is a desktop machine and we shouldn't need it for much longer so it could just go into an office for now. The RPM master will be switched from bottle to figgy once the data has been copied to storage in the Forum, this needs to happen at the weekend and during the day to avoid installs and updaterpms being run. dresden will be moved to the Forum by Pickfords along with the storage it is using, we will need to announce this to the lcfg-discuss mailing list.

bugs.lcfg.org
Chris has finished this and it is now in service. Stephen still needs to solve the iFriend support problem. This might require another slightly hacky mod_perl handler.

svn.lcfg.org with websvn
Stephen has finished this and it can be found at https://svn.lcfg.org/websvn/

svn.lcfg.org with webdav
Stephen has done most of the work for this but it has proved impossible to support both svn+ssh and webdav access. We will have to all switch over to webdav at the same time and the permissions on the repository will need to be switched. Stephen will work out some guidelines for COs in case they don't want to just delete their working copy and checkout a new one. There are some security implications to the webdav access, it is open to the world over https and we are giving the apache user the ability to alter the configuration of any DICE machine. We should never put this web service onto a machine where people can put their own CGI scripts.

Server reboots
It looks like most servers have now been rebooted to get the latest kernel.

RHEL license
Stephen now has a RHEL license, he has been having trouble trying to work out how to use it to login to the redhat website.

SL5.3 kernel
It looks like the SL5.3 kernel is going to be made available for SL5.2 along with the necessary dependency updates. This will help make the transition easier and will allow us to switch the routers to a kernel which shouldn't crash so often with H323 problems.

bugzilla 3
There is a security update available for bugzilla 3. Chris will look into packaging this and getting it installed on dresden.

This Week

Alastair will:

  • update qlogic headers to reflect move back to stock SL5 driver
  • move RPM master to figgy DONE
  • move central's disk blob from satabeast to new EVO array
  • modify unit projects as per project priority plan from CEG

Chris will:

  • package and install the Bugzilla 3 security update. DONE
  • Work on Desktop Power Management. DONE

Stephen will:

  • Finish converting MPU components to new build tools
  • Organise the switch of svn.lcfg.org to webdav DONE
  • svn support for build tools

-- StephenQuinney - 20 Feb 2009

Topic revision: r6 - 03 Mar 2009 - 09:47:11 - StephenQuinney
 
This site is powered by the TWiki collaboration platformCopyright © by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding TWiki? Send feedback
This Wiki uses Cookies