GLUE SCHEMAS ACTIVITY

EDG WP4 info providers for the Computing Element

installation instructions for EDG 1.4.3 release

Below are provided our operational steps in order to make a manual installation of CE info providers on top of EDG software, release 1.4.3.

Prerequisite: Glue schema installed, LCFG control disabled

Some clarification:

Head cluster machine: machine running the batch server, GRIS and gatekeeper; can act also as a worker node

The Cluster Unique ID is the hostname of the head cluster machine

The Subcluster Unique ID is equal to the Cluster Unique ID (a subcluster is an homogenous set of hosts belonging to a cluster, in EDG the hyphotesis is that each cluster is homogeneous)

$EDG_LOCATION should be replaced with the actual setting, tipically /opt/edg

Step-by-step instructions:

  1. copy glue-ce-all.pl to $EDG_LOCATION/info/mds/bin and assign eXecute rights
  2. copy glue-ce-static.ldif to $EDG_LOCATION/info/mds/etc/ldif
  3. modify glue-ce-static.ldif with the proper values
    (example)
  4. download this script on a sample worker node, assign eXecute rights and run it as ./hostdata.sh > subcluster.data;(this is to get data for the subcluster representation, at the moment, for EDG, the hypothesis is of a cluster composed by a subcluster composed by a certain number of homogeneous worker nodes)
  5. copy subcluster.data in the head cluster machine to $EDG_LOCATION/info/mds/etc/
  6. cp $EDG_LOCATION/info/mds/etc/testbed1-site-globus.ldif $EDG_LOCATION/info/mds/etc/testbed1-site-glue.ldif
  7. modify testbed1-site-glue.ldif (example) changing
    • dn: Mds-Vo-Name=local, o=Grid
      objectClass: GlobusTop
    • to
    • dn: Mds-Vo-Name=local, o=Grid
      objectclass: GlobusTop
      objectclass: GlobusActiveObject
      objectclass: GlobusActiveSearch
      type: exec
      path:       $EDG_LOCATION/info/mds/bin
      base: glue-ce-all.pl
      args: -lrms LRMS_TYPE -cluster-batch-system-bin-path /PATH/TO/LRMS/BIN -queue QUEUE1 QUEUE2 ... -cluster CLUSTER_ID -globus-gatekeeperport 2119       -globus-gatekeeperhost HOSTNAME -remotefiles $EDG_LOCATION/subcluster.data -auth-users-from-grid-mapfile /etc/grid-security/grid-mapfile -static $EDG_LOCATION/info/mds/etc/ldif/glue-ce-static.ldif -ttl 120 -grisport GRISPORT -cesebind QUEUEx SE_ID MOUNTPOINT
      cachetime: 30
      timelimit: 20
      sizelimit: 50



      • Detatils:
      LRMS_TYPE      : PBS/LSF/CONDOR
      CLUSTER_ID: the hostname of the machine running the LRMS server where the gatekeeper is
      HOSTNAME: the hostname where the gatekeeper is running, typically same as CLUSTER_ID
      GRISPORT: the port where the GRIS is listening to (typically 2135)
      -cesebind: this param is optional and helps advertising a static association between a Computing Element and a Storage Service; the queuex is the queuename related to the computing element you consider, the SE_ID is the Storage Service ID that you want to be used by your CE; the MOUNPOINT is optional and let you advertise an eventual nsf directory available to locally access the storage space to be used by your CE

  8. edit the /opt/globus-24/libexec/edg.info and change the line:
    cat $EDG_LOCATION/info/mds/etc/testbed1-site-globus.ldif
    to
    cat $EDG_LOCATION/info/mds/etc/testbed1-site-glue.ldif
  9. restart the GRIS service: /etc/init.d/rc.d/globus-mds restart

 
Maintained by Sergio Andreozzi - Last update: Tue 29 April, 2003