MPU Meeting Monday 21st January 2008

LCFG Website

The service catalogue entries have been done. End of project!

Buildtools Project

The LCFG perl component module now makes use of sysinfo. The shell ngeneric module only makes limited use of it because of shell limitations.

rpmsubmit Project

Simon's patch has been applied.

Power management project

Thanks to Simon there's been some good progress, see the MPUPowerDiary for details. We can now shut down, suspend, or hibernate machines with instructions on when to wake up again, and machines generally respect those instructions and wake up. The wake-up instructions can even persist across power cuts, though support for that currently seems a bit dodgy.

Simon advised against resurrecting the APM-era suspend and resume methods for LCFG components. Instead, any daemons and/or components which actually need to know about suspend or resume events - which shouldn't be many - should register with the OS's own notification mechanism by putting a script in /etc/pm/hooks/.

Chris will try out SL5 now (as opposed to FC6) and when possible he'll try it on a 755.

Operational

The work needed to tidy up and properly develop the minimal LCFG setup will take at least a couple of weeks so Stephen will write it up into a project proposal.

Lots of junk has been moved out of the DICE CVS repository but the rest of the split is more involved and will take some time to get done. We'll revisit this in March. We're doing the split to separate out the DICE and the LCFG level material to make it easier for external collaborators to get direct write access to the appropriate LCFG level stuff.

We won't need the 64 bit machine for the build farm for a couple of months yet. It could be an existing machine, for example split. That fits nicely with the current difficulty in ordering new servers.

Carol has written instructions on how to move a machine to a different wire without reinstalling it; they can be found at MovingMachineToADifferentWire.

Stephen and Alastair have synchronised component versions across platforms as far as possible. Things should (in theory!) stay in sync better when we have our new buildtools/buildfarm with its easy multi-platform build facilities. However we should still check from time to time that our components are in sync, so we're going to develop a dynamic list of component versions across all supported platforms.

Nobody has mentioned the FC6 KDE CD spinning bug to our knowledge so we're dropping the action to investigate and fix it.

Chris has dealt with most of his rescued documents. Some of the left-overs are mostly historical and currently serving as as reminder to us of the need to document things (e.g. the header structure and RPM naming). However there are some which document aspects of the default disk partitioning arrangements (which btw Stephen has just tidied up). Chris will pull things together into a new description of the partitioning. He'll make sure to mention that it's a snapshot of the current arrangement and point to the definitive data in the headers. Some of what Chris has already rescued would be suitable for the LCFG wiki so he'll move things there as appropriate.

We need to upgrade our servers to SL5. Alastair will do the RPM slaves and master; Chris the LCFG slaves and master; Stephen dresden. Stephen's test SL5 LCFG slave has been working well.

Alastair will take the current outstanding DIY DICE RT ticket (34604) and will use DIY DICE to configure the vmware virtual machine he's working on.

Stephen said that the gdm component needs to be more robust: currently if gdm fails to start properly, the gdm component keeps on going regardless. Alastair will look at it.

Stephen and Paul are going to discuss Paul's info-gathering exercise (the one where we ask people why they made a change to a header or profile). We talked about this for a while and decided that it was feasible and could get good results if the classification was simple and quick enough.

Alastair has checked the RPM master backups. They're mirrored to a Solaris partitions, and all partitions on the Solaris servers are backed up to tape by default. Alastair queried the meaning of the "no checksum" rmirror component option. Chris racked his memory and explained that checksumming was used in the rmirror backups but that on very large partitions it could make the mirroring process take hours longer, so for those larger partitions only the checksumming was suppressed to save time.

Chris will upgrade our Mac mini nice to proper Leopard and configure it as a remote access build host as before.

The services team has done a sudden intensive flurry of Solaris patching, so it may be time for Chris to start the patch cycle again. He'll check.

Alastair now has a shiny new Dell 755; he'll try out SL5 DICE on it.

We talked about next summer's desktop upgrade. We can't see any real alternative to using SL5. Now that SL5.1 is out the software shouldn't be particularly out of date. Support lasts until spring 2010, making it suitable for us to use between summer 08 and summer 09, but not beyond that. However SL6 is expected in spring 09 so hopefully we'll have SL6 DICE by summer 09.

This Week

Stephen:

Chris:

  • Machine upgrades
  • power management
  • Solaris patching
  • stable release

Alastair:

  • Upgrade the RPM slaves
  • MpuRescue
  • Discuss Mac stuff with Toby
  • Try DICE on the 755
  • Deal with the DIY DICE ticket.

-- ChrisCooke - 22 Jan 2008

Topic revision: r1 - 22 Jan 2008 - 12:19:09 - ChrisCooke
 
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