MPU Meeting Tuesday 17th May 2016

Inventory

Alastair has been experimenting with JSON-and-SQL. Thanks to that experience he has slightly restructured his ClientReport data. Where there is data on multiple items of the same type (for example several NICs on the same machine), ClientReport modules should now group them together in a list rather than in a hash. The list has proved easier to query from SQL.

Alastair has found that SQL-and-JSON documentation and examples are too simplistic. They guide the beginner but do not help with real world examples. Stephen suggested that any queries which used WITH a lot might be more efficiently written as views.

Chris has written ClientReport modules to return data on power supplies (PSU.pm), management controllers (MC.pm) and multi-port network cards (NDC.pm), and a helper module (FRU.pm) which queries IPMI for general hardware details (using ipmitool fru). He's working on two further modules - one to provide data on Fibre Channel adapters and one on RAID controllers and physical disks.

Alastair will think about packaging - what would be the optimum number of RPMs for the ClientReport modules, given that different mixes of modules will be needed on different machines?

There was a discussion about how to record errors in ClientReport modules, and what would constitute an error. It was decided that modules should only be used on machines where they're relevant - so, no probing for FibreChannel HBAs on desktops, for instance - and that therefore any failure to probe hardware for data should be regarded as an error. Every module's returned data hash should have its own _version and _errors variables, both integers; any errors should be flagged both by incrementing _errors and by writing an explanatory message to standard error.

Alastair is moving the Tartarus server to SL7.

Since MAC addresses will be mastered in the new inventory system rather than in LCFG, they will need to be added to the inventory before DICE installation can begin. This implies that a hostname will need to be chosen at that point too.

LCFG Client Refactoring

Stephen has finished the Perl interface to the "diff" support. It lets you diff profiles, component resources and so on. The tests being passed; the code is working. The dependencies have been greatly simplified. In particular there are now no dependencies on Moo or Moose. He has published a couple of blog articles:

There are a few issues still to sort out:
  • The parsing of context expressions needs to be improved. This is currently very basic, so contexts can only be used in extremely simple ways.
  • The encoding of values in resources (to make it possible to have a newline in a resource value, for instance) has to be fixed. The code can currently decode them but not encode them.
  • General development support should be improved. It will be made easy to link against the new libraries, and to use them from perl without complication.
  • The code needs to generate md5sums for secure mode.

SL7 Server Base

No activity.

MPU SL7

  • The SL7 ssh server is now open to COs. As a result Graham identified some missing packages which Stephen added.
  • Support for the ed25519 host key type has been added to lcfg-openssh.
  • Some vim packages need to be sorted out.
  • We're now waiting for local DNS to work reliably on SL7.

Miscellaneous Development

To help with the spate of Firefox crashes at weekends, Stephen has turned off auto-updates for extensions. We may have to wait for another weekend to pass before we know whether that has helped.

Operational

Stephen has built the packages for OpenAFS 1.6.18. However he doesn't think that it will be necessary for us to make a special effort to upgrade to them soon.

Alastair is about to order two new KVM servers (to replace jubilee and hammersmith). We discussed their configuration, and decided that:

  • Each one will have 128GB of memory, in 8 modules, leaving 16 free spaces for expansion.
  • Each server will have 8 600GB 10k disks rather than our usual 4 600GB 15k disks, because the former gives many more IOPS, which is what we want.
  • One machine will be specified with the fastest available single stream CPUs and the other will have the greatest possible number of cores.

This Week

  • Alastair
    • Inventory project
      • continue working through InvProjectWorkFlow
      • Document clientreport (eg how to add modules)
      • Document order sync code
      • Document hpreport processing script
      • Start work on RESTful API - InvProjectRESTapi
      • Document that using lists instead of hashes when reporting on multiple things of same type is easier to process in SQL
      • Think about splitting off tartarus-clientreport from tartarus core (in packages) - makes rebuilding for next platform easierDone.
      • Blog articles
    • Remove default pool if ops meeting agrees
    • Dump 'atom'
    • Deploy encrypted /tmp and swap conversion script
      • Deploy on office desktops
      • Need to warn users that Gnome3 may pop up a window about /tmp being full (when script is run)
    • Schedule MPU meeting to discuss systemd ordering
    • Reschedule MPU futures meeting
    • Continue building computing.help honeypot
    • package up ILW stuff and document process
    • Read through SL7.2 release notes
    • submit polkit bug to redhat - with Stephen
    • Work on RT tickets - only 1 !
    • MPU SL7
      • Identify what could be done once lcfg-dns is ready
    • Apply lcfg-lvm patch, test and distribute
    • Meet with George and Stephen to discuss DNS / systemd - scheduled for Tues 24th May
    • computing.help page for HP 800 G2 - stalled awaiting delivery of new G2s
    • BuyAwait another 800 G2 so that we can compare settings with working G2
    • Read Stephen's blog articles
    • Order new KVM servers - one single stream fast, one slower multi-stream - awaiting quotes

  • Chris
    • Inventory project
      • Continue work on clientreport modules for replacing firmwarereport
    • pkgsearch for SL7
      • reimplement as a yum web front end (yum search for keyword produce an html file of links to cgi to do yum info)
      • Need support multiple platforms
    • MPU SL7
      • Identify what could be done once lcfg-dns is ready
    • Consider Matthew Richardson's advice on NX for SL7
    • Look at KVM guest lists and work out whether it would be practical to segregate end users/web services from critical services See MPUSegregationOfVMs.
    • Read Stephen's blog articles
    • Upgrade default memory for KVM guests from 1GB to 4GB

  • Stephen
    • LCFG client refactor stage 1
      • schedule debrief meeting
    • LCFG client refactor stage 2
      • polishing - work on context parser
      • blog article (once documentation complete)
    • apacheconf
      • produce some recipes
      • talk at devel meeting
    • Investigate kernel component pipe moan by using shell commands instead of RPM module => waiting on 7.2 => activities list
    • LCFG server symlink to exam branches - produce reporting script and discuss with Graham
    • Circulate dmesg proposal
    • Apply firmware patches - circle
    • submit polkit bug to redhat - with Alastair
    • Meet with George and Alastair to discuss DNS / systemd
    • SL7 MPU
      • Identify what could be done once lcfg-dns is ready
    • Work on RT tickets
    • Test 800 G2 install

-- AlastairScobie - 17 May 2016

Topic revision: r8 - 24 May 2016 - 13:02:37 - 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