SEE-GRID VO site configuration

From EGEE-see WIki

Jump to: navigation, search

This wiki page describes configuration of a SEE-GRID site to support:

  • Mapping to generic seegrid pool accounts when no role is specified. Generic accounts have a small fairshare and are limited to only a few CPUs.
  • Mapping of ops role to a small number of pool accounts. Ops accounts have one CPU reserved at all times.
  • Mapping of sgmadmin role to a small number of pool accounts. SGM accounts can install VO application software in the shared software area.
  • Mapping of an application group (/seegrid/BG/App/SALUTE in the example) to a medium sized number of pool accounts. Application accounts have a larger fairshare and CPU limit than the generic seegrid accounts.

Contents

Creating pool accounts

Pool accounts are be created using yaim's users.conf. Large number of generic pool accounts is created using seegrid as the primary group:

3001:seegrid001:1400:seegrid:seegrid::
3002:seegrid002:1400:seegrid:seegrid::
...
3200:seegrid200:1400:seegrid:seegrid::

A small number of special accounts is created for previously described purposes. These accounts are marked using a special tags (ops, sgm, app) that are used as a link to entries in groups.conf. Different primary groups are also used in order to assign different privileges to these accounts.

3501:opsseegrid001:1401,1400:opsseegrid,seegrid:seegrid:ops:
3502:opsseegrid002:1401,1400:opsseegrid,seegrid:seegrid:ops:
...
3510:opsseegrid010:1401,1400:opsseegrid,seegrid:seegrid:ops:
3601:sgmseegrid001:1402,1400:sgmseegrid,seegrid:seegrid:sgm:
3602:sgmseegrid002:1402,1400:sgmseegrid,seegrid:seegrid:sgm:
...
3610:sgmseegrid010:1402,1400:sgmseegrid,seegrid:seegrid:sgm:
3701:appseegrid001:1403,1400:appseegrid,seegrid:seegrid:app:
3702:appseegrid002:1403,1400:appseegrid,seegrid:seegrid:app:
...
3750:appseegrid050:1403,1400:appseegrid,seegrid:seegrid:app:

Mapping roles to pool accounts

Adding the following lines to groups.conf should generate the correct mapping:

"/VO=seegrid/GROUP=/seegrid/ROLE=ops":::ops:
"/VO=seegrid/GROUP=/seegrid/ROLE=sgmadmin":::sgm:
"/VO=seegrid/GROUP=/seegrid/BG/App/SALUTE":::app:
"/VO=seegrid/GROUP=/seegrid"::::
"/VO=seegrid/GROUP=/seegrid/*"::::

The last line allows correct mapping of special VO groups that are not explicitly listed (e.g. /seegrid/RS/App/VIVE) to generic pool accounts. SEE-GRID VOMS servers should also be listed in site-info.def:

 VO_SEEGRID_VOMSES="
   'seegrid voms.irb.hr 15010 /C=HR/O=edu/OU=irb/CN=host/voms.irb.hr seegrid'
   'seegrid voms.grid.auth.gr 15040 /C=GR/O=HellasGrid/OU=grid.auth.gr/CN=voms.grid.auth.gr seegrid'
 "
 
 VO_SEEGRID_VOMS_SERVERS="
   'vomss://voms.irb.hr:8443/voms/seegrid?/seegrid'
   'vomss://voms.grid.auth.gr:8443/voms/seegrid?/seegrid'
 "

Verification of VOMS proxy extensions also requires VOMS server host certificates to be available on the machine. These certificates are provided through an rpm file. It can be installed using:

 # rpm -ihv http://www.irb.hr/users/vvidic/seegrid/seegrid-0.5-2.noarch.rpm
 Retrieving http://www.irb.hr/users/vvidic/seegrid/seegrid-0.5-2.noarch.rpm
 Preparing...                ########################################### [100%]
    1:seegrid                ########################################### [100%]

Older versions of the rpm are also available.

After reconfiguring the LCG-CE or just running the relevant functions:

 /opt/glite/yaim/bin/yaim -r -s site-info.def -n lcg-CE -f config_users -f config_mkgridmap

the result can be seen in:

  • /opt/edg/etc/edg-mkgridmap.conf
# SEEGRID
# Map VO members  (ops)
group vomss://voms.irb.hr:8443/edg-voms-admin/seegrid?/seegrid/Role=ops .opsseegrid

# Map VO members  (sgm)
group vomss://voms.irb.hr:8443/edg-voms-admin/seegrid?/seegrid/Role=sgmadmin .sgmseegrid
 
# Map VO members  (app)
group vomss://voms.irb.hr:8443/edg-voms-admin/seegrid?/seegrid/BG/App/SALUTE .appseegrid

# Map VO members  (root Group)
group vomss://voms.irb.hr:8443/edg-voms-admin/seegrid?/seegrid .seegrid
  • /opt/edg/etc/lcmaps/gridmapfile:
"/VO=seegrid/GROUP=/seegrid/ROLE=ops/Capability=NULL" .opsseegrid
"/VO=seegrid/GROUP=/seegrid/ROLE=ops" .opsseegrid
"/VO=seegrid/GROUP=/seegrid/ROLE=sgmadmin/Capability=NULL" .sgmseegrid
"/VO=seegrid/GROUP=/seegrid/ROLE=sgmadmin" .sgmseegrid
"/VO=seegrid/GROUP=/seegrid/BG/App/SALUTE/Role=NULL/Capability=NULL" .appseegrid
"/VO=seegrid/GROUP=/seegrid/BG/App/SALUTE" .appseegrid
"/VO=seegrid/GROUP=/seegrid/Role=NULL/Capability=NULL" .seegrid
"/VO=seegrid/GROUP=/seegrid" .seegrid
"/VO=seegrid/GROUP=/seegrid/*/Role=NULL/Capability=NULL" .seegrid
"/VO=seegrid/GROUP=/seegrid/*" .seegrid
  • /opt/edg/etc/lcmaps/groupmapfile:
"/VO=seegrid/GROUP=/seegrid/ROLE=ops/Capability=NULL" opsseegrid
"/VO=seegrid/GROUP=/seegrid/ROLE=ops" opsseegrid
"/VO=seegrid/GROUP=/seegrid/ROLE=sgmadmin/Capability=NULL" sgmseegrid
"/VO=seegrid/GROUP=/seegrid/ROLE=sgmadmin" sgmseegrid
"/VO=seegrid/GROUP=/seegrid/BG/App/SALUTE/Role=NULL/Capability=NULL" appseegrid
"/VO=seegrid/GROUP=/seegrid/BG/App/SALUTE" appseegrid
"/VO=seegrid/GROUP=/seegrid/Role=NULL/Capability=NULL" seegrid
"/VO=seegrid/GROUP=/seegrid" seegrid
"/VO=seegrid/GROUP=/seegrid/*/Role=NULL/Capability=NULL" seegrid
"/VO=seegrid/GROUP=/seegrid/*" seegrid

Implementing scheduling policy

Batch system should be configured to allow jobs from all the mentioned groups:

 VOS="seegrid"
 QUEUES="seegrid"
 SEEGRID_GROUP_ENABLE="seegrid opsseegrid sgmseegrid appseegrid"

The following maui configuration reserves one CPU for monitoring jobs and gives them high priority so they run immediately:

# use any working node
SRCFG[ops] FLAGS=SPACEFLEX
# reserve one cpu
SRCFG[ops] TASKCOUNT=1 RESOURCES=PROCS:1
# reserve it forever
SRCFG[ops] PERIOD=INFINITY
# it can be used by seegrid ops group
SRCFG[ops] GROUPLIST=opsseegrid

# give it enough priority to run imediately
GROUPWEIGHT     1
GROUPCFG[opsseegrid] PRIORITY=1000000

Faresharing allows specific application jobs (mapped to appseegrid group) to get more CPU time than other (normal) jobs.

# We will consider the last 7 24 hour periods for our fair
# share calculations. The influence of each 24 hour period
# decreases by a factor of 0.8 each time.
FSPOLICY                DEDICATEDPS
FSDEPTH                 7
FSINTERVAL              24:00:00
FSDECAY                 0.8
FSWEIGHT                1
FSUSERWEIGHT            5
FSGROUPWEIGHT           30

GROUPCFG[appseegrid] FSTARGET=50 MAXPROC=50,100
GROUPCFG[seegrid]    FSTARGET=10 MAXPROC=10,100

Related pages

Personal tools