MPU Meeting Tuesday 8th September 2009

Power Management Project

The 7900s are still sleeping well. Rollout to other student lab HPs is imminent.

rpmsubmit

No development[s].

AFS Component

Stephen has spoken to George about getting a new interface put on the KDC machines, so they can become AFS database servers too.

Craig will hopefully get the test fileserver up before he goes on holiday.

TiBS

The TiBS RPMs and LCFG component have been upgraded to TiBS 2402, the version we use. Once testing is complete we'll be able to deploy this version.

Chris has been blogging about his progress.

LCFG Server Refactoring

The latest code change to lcfg-server ships in this week's stable release.

Stephen has spent lots of time battling git which is being awkward.

It's been quite a bit of work to get to the point where it's possible to run perltidy on the code. Some of the macros in the code weren't understood by perltidy. In the code of the current lcfg-server macros are used to generate code. One of the effects of this is that different code is used when testing than when running in normal operation..

It's also difficult to build packages as CVS is needed. A special temporary CVS repository linked to git has been necessary just to get the code built.

The code needs to be split into various logical parts: the core mkxprof server code (which could be in CPAN), the lcfg-server component (which should be buildable as a normal component) and the web stuff.

Stephen has been blogging about his progress.

Miscellaneous Development

Alastair has finished tidying the inf level headers for now. We're not yet interested in the DICE flavour and the Ed flavour shouldn't be needed except for an LCFG port, but the minimal/LCFG-level flavour is of interest to us. It's in use on ashkenazy.

Alastair's ethernet bonding header etherbond.h has been finished and released. It's now in use on some MPU servers but still needs to be added to some. Alastair is now adding Nagios testing to the header.

Stephen is still working out the possible causes of the grub component getting confused by the contents of the device.map file. One clue was found on orkney, where device.map was a (777) socket owned by xfs. Stephen has started rewriting the grub component to make it detect such corrupt files, delete them and have them regenerated from scratch.

Chris has moved George's dns fixperms from a daily cron job into boot.run so it happens straight after updaterpms.

Pandemic Planning

Ether bonding has been added to MPU servers.

Fibrechannel is now as robust as it can be on MPU servers.

MPU service configurations are now even more fully in headers rather than source profiles.

The table of interests is complete.

The main job this week is for each person to do research on their chosen interests.

Operational

boot.run times

Chris has produced three headers for randomising these: boot-run-evening.h, boot-run-night.h and boot-run-outofhours.h. They'll be used on MPU servers later this week.

Bakerloo multipath problem

Bakerloo hadn't been rebooted to pick up the multipath fix, so when it encountered a fibrechannel glitch it remounted its main virtual server guests' data blob read-only. Two virtual server disk images were irreparably damaged. It's now been rebooted so this shouldn't happen again. Central also hasn't been rebooted to pick up the multipath fix, so Alastair needs to do that soon.

Bakerloo on develop

Bakerloo is on develop. It'll be put on to stable after Thursday.

Deployers' Meeting

Matthew Richardson (of SEE) has been doing lots of interesting work with LCFG. Kenny will ask him to get involved with the LCFG Users Day. It would also be good for him to feed back his work to us. He's configured Xen live migration with LCFG. He's also produced an LCFG header which configures FreeNX which we'd love to have.

On a wider note, the process of getting interesting LCFG stuff fed back to us isn't yet quite as automatic as we'd ideally like. For instance we were delighted to get Kenny's CUPS component but we didn't get any of the headers (maybe we didn't ask for them?). We should suggest that all components in svn.lcfg.org should have appropriate headers there too.

Shane will also talk at the LCFG Users Day. It would be good to get Informatics people involved too. Chris will say something about lcfg-sleep.

Kenny has done some measurements and found that the CUPS component takes 7.5 seconds to start! We'll move it to after the X components in the startup order. Informatics desktops don't use the CUPS component so this won't affect us so much.

Kenny wants the new openafs component for the Inspace Macs, but the AFS perl module doesn't build on Macs.

This Week

Alastair will:

• research pandemic interests
• write pandemic howtos
• finish Nagios testing in etherbond.h
• reboot central

Chris will:

• research pandemic interests
• work on TiBS
• finish changing boot run times on MPU servers

Stephen will:

• research pandemic interests
• document disaster recovery for LCFG master
• update LCFG server documentation
• finish tidying MPU source profiles
• test AFS component database servers

-- ChrisCooke - 09 Sep 2009