MPU Meeting Tuesday 27th April 2010

LCFG Server Refactoring

Stephen had a look at whether KiokuDB would be a good option for the LCFG server storage requirements, particular the status database. He has concluded that the set of Perl packages required to get this going on SL5 is just too large. It also requires upgrading some modules which are provided in the core perl package which tends to be a bit of a hacky process. Even with all the necessary modules installed on SL5 using the CPAN tool he couldn't get them all to pass their tests which is not a good sign. This leaves us with the choice of either requiring a newer platform for the server update, ignoring this part of the refactoring until later, or going for another solution, possibly homebrew.

Having promised to ship the first update in June we really need to decide what else is going to be done to the code before then. There is still plenty of tidying up work to be done, including silencing warnings and modernising the code. We need to do more work on the new test framework and we definitely need to add git support to the LCFG buildtools to aid the release management process. Whether we make any more major changes is clearly dependent on the amount of time/effort available.

Server Hardware

Stalled.

Installroot

Alastair checked the latest version of the code on SL5, it all worked except that the the recent logging changes required an update to the version of busybox.

The timezone setting problem is really an F12 project bug, Chris will add a bug report detailing the issue.

This week the focus will be on getting the new installroot working for F12 x86_64.

F12

  • Problems with reading from stdin, which upset the kerberos component, during the boot process were caused by the Plymouth graphical boot sequence. The only solution is to turn off Plymouth before starting the LCFG boot component.

  • It is no longer necessary to have the dhcp header included in an LCFG profile for the install process, a normal wire header will suffice.

  • Chris has seen some boot problems which appear to be caused by use of the LCFG openldap component. Stephen suggested that we might not have a complete list of local groups which should be excluded from the LDAP lookups. This is holding up the addition of lcfg-openldap by default for the DICE flavour of the inf-level.

  • There is something not working with the LCFG kerberos component when it uses kdcregister to add principals to the /etc/krb5.keytab file. The host principal does not get created when it should. Alastair will talk to Toby about the problem.

  • Chris found that CD mounting does not work under F12 for certain hardware models. He will do some more investigation as this is otherwise likely to be the source of numerous complaints. It was suggested trying to mount the CDs directly as root, and also examining dmesg and /var/lcfg/log/syslog for any hints to the source of the problem.

  • Chris and Stephen will coordinate finalising the base and desktop package lists for, at least, F12/i386. As COs will be starting to do their own testing this is fairly urgent.

Miscellaneous Development

LCFG om update
The update to the LCFG om resources to make it do a setpag when any LCFG component method has been called is now in place.

LCFG server update
The LCFG slave servers have been updated to LCFG server version 2.2.55. This has the rewritten release name handling backported from the development tree, improvements to the host and domain name handling, and it now allows dumpdeps to be run as a normal user.

Operational

Dell Optiplex 780
Stephen has tested SL5 on this new model and it all works fine, including X via the displayport-only graphics card.

Resources for machine criticality
Stephen will look into whether it is possible to simply add a new resource to the LCFG sysinfo component to allow us to define machine criticality.

Package volumes and releases
Alastair needs to check that the AFS packages volumes are not being released each night by the standard scripts. They need to only be managed by our scripts otherwise we might end up with incorrect rpmlist files.

Disk space for telford
We can't keep waiting for the Services Unit to find some space so Alastair will provide Stephen some temporary space on the bpbeast storage array. This will make it possible to shuffle the SL5 and epel mirrors into AFS, about 0.5TB will be needed.

Disk space for VM servers
There are 3 sets of disks in the bpbeast storage array. Craig is currently using one set but plans to have the data moved fairly soon. The plan is to have each disk set configured for RAID1 and allocated to a different virtual server. This will hopefully improve performance. In the longer term we will acquire a new storage array, based on 15K SAS disks, that is purely for the virtual servers. Even when we have the new configuration we will still need to rethink how we handle major updates for the virtualised guests. We probably need to spread the runs of updaterpms over a few days, spreading them over a few hours has improved matters but we still see overlaps and big load spikes.

split IP address
The rpm cache server split has been moved to a new IP address 129.215.202.139. That's the last of the renumbering we need to do for George.

perl-AFS update
The perl AFS module has been updated to 2.6.2 on SL5.

This Week

Alastair will:

  • 64bit installroot
  • Disk space for Stephen
  • Talk to Toby about the F12 kerberos problem
  • Check with Craig about the packages volumes

Chris will:

  • Various F12 work
  • Coordinate sorting out the F12 package lists with Stephen

Stephen will:

  • Rebuild the latest SL5 kernel
  • Plan for the June release of the LCFG server
  • Coordinate sorting out the F12 package lists with Chris

-- StephenQuinney - 28 Apr 2010

Topic revision: r1 - 28 Apr 2010 - 08:02:15 - 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