Minutes Mar 1


  • Routing fixed: if you have an external IP you need to set the internal IP (you need to have an internal IP for Cobbler installation) statically, else DHCP is undoing the route
    • Boot sequence
      • Netboot with internal IP
      • Statically set all the internal/external IP
    • If you did not set the static IP, DHCP sets the 6509 as gateway and external routing is not working
  • Install documentation in http://twiki.mwt2.org/bin/view/Admins/NewCobblerNode
  • Separate repo for the ITB
    • Newer version of Torque and Maui
  • Setup puppet rules: Condor WN, scheduler, ...
    • rules following Nate convention: ";;ITB"

Nate suggest to use ";;ITB" for the Puppet rules, Puppet prefers no dashes, same as ";;UC" and ";;IU"

There is a CVS checking problem with Puppet Set the owner explicitly in the Puppet file (owner=root)
  • old files have owner of the file set, not in the file
    • files are frequently system files (owned by root) and it is strange to check files in CVS as root
    • anyway would be to time consuming to update all and things work
    • can't check in files (owned by a 3rd party) third party conflict when trying to check-in ITB files in CVS
      • if the modules are owned by someone else than the file owner there is a conflict
      • fix: move the file somewhere else, do a CVS update, move the file back in
  • newer files should set the owner explicitly

Revision management used for Puppet at the moment does not work

Current procedure:
  • Nate is editing on the flight to be more responsive
  • when it works, it is checked in in CVS

Suchandra's proposal:
  • setup Puppet as working form a checked out version as root
  • write a script to run as root that checks out the CVS to the running Puppet (and restarts Puppet with the new configuration)
  • people edit the configuration in their home directory and check the code in once that thy verify that works
  • the script is run to have Puppet using the new configuration

Cobbler integrated with Puppet -

No separate Puppet for ITB

edge7 and c018 are not using edge4 as home dir server any more

Aaron and Nate installed xrootd: * 4 machines on kvm1: data server and 3 pools, using the ITB names * wrote puppet rules to provision from scratch (redirector and data server) * version is 3.0.3.pre8 packaged from UCSD - this is a prerelease, latest officially tagged by Andy release is 3.0.2 and is available from VDT

Disk allocation:
  • we have 2 trays, 3 pools, one spans 2 trays. 10 TB total on the trays
  • changes:
    • deallocate current partitions
    • split in 4 partitions/pools
    • one will be unallocated

Plan for the week

  • Merge the tables, then pass to Nate to update and fix the table
  • Puppet
    • check a way to verify Puppet rules
    • check into Puppet checkout (/etc/puppet and CVS)
    • propose it and see if it can work
  • email about xrootd version - VDT version
  • install ui-gwms
  • ask for IP for ui-cr, start deployment

  • move panda and ITB robot to point to the new machine
  • SRM tester - automatic tester for SRM (from Alex)

  • update the merged machine table
  • Change UC_ITB_EDGE7 on edge7 to UC_ITB on itbv-ce-pbs
  • Suchandra will send an email with exact request to Alden with us in CC
  • Nate will work with Charles, Alden and Hiro?
    • changes to TiersOfAtlas (topology for DQ2, CVS at CERN)
    • changes to scheddb (queue configuration for Panda, CVS at CERN, Panda reads the updated one automatically)
  • See partitioning above

Go to the Tier 3 node matrix and see how to use the nodes that have been freed up

xrootd (Nate, Suchandra, Marco, Aaron)
  • reinstall 3 pool machines on itb-kvm1
    • use the MD1000 disks (1/3 of each allocated pool)
    • latest RPM from VDT
  • reinstall redirector on itb-kvm2
  • Test once the machine are installed
  • xrdcp regular/secure directory, cp test on the machine with xrootd-fs (Marco)
  • xrootd redirector on itb-kvm3
  • xrootd server on itb-kvm1, 3 of them
  • xrootd-fs on itbv-bmx

Move panda and ITB robot to point to the new machine

ITB machines are sharing OSG_DATA from the Tier 2 /export/osgvo - home directory for the OSG VO

OSG_APP and OSG_DATA are still shared between the Tier 2, Tier 3 and ITB
  • start with a clean OSG_DATA
  • start with a clean OSG_APP
  • use only CVMFS for ATLAS - it is already configured like that in Panda

Move the NFS server off itb1(virtual) to itb2 (physical)
  • prefer that to be on a physical machine
  • long term
  • itb2 is not part of the 101 vlan

End game:
  • functional Panda endpoint
  • functional dCache, xroot, hadoop
  • worker nodes and head-nodes up and running
  • some more monitoring
  • use CVMFS (Sara has CVMFS rules)
  • UI machines: glidein-wms (collector, vo-frontend, server)

Panda queue is using the Tier 2 storage
  • run Panda ITB robot jobs
  • we have 48 job slots

Meeting with Anita Nicolish on Thursday 9am
  • campus grid
  • resource dedicated as a primer
Talk to a campus meeting on 3/30

NISGrid joint meeting with Site Admins during the Summer (July/August)
  • track on campus-grid
  • track on cluster management
  • typical OSG fundamentals
  • job submission on the Grid

Presentation layer for MWT2 and OSG-ITB
  • Drupal on itb3
  • content management system
    • way to provide a frontend
    • way to manage other content, e.g. a blog (to stash plots, ...)
  • alternatives: Wordpress, Joomla

OSG blog is on blogspot, really easy to do

Splunk running on edge5 an on grid8

-- MarcoMambelli - 15 Mar 2011
Topic revision: r1 - 15 Mar 2011, MarcoMambelli
This site is powered by FoswikiCopyright © by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding Foswiki? Send feedback