Difference between revisions of "Imperial Dirac Maintenance"

From GridPP Wiki
Jump to: navigation, search
Line 39: Line 39:
 
[http://lhcbproject.web.cern.ch/lhcbproject/dist/Dirac_project/installSource],  
 
[http://lhcbproject.web.cern.ch/lhcbproject/dist/Dirac_project/installSource],  
 
[http://lhcbproject.web.cern.ch/lhcbproject/dist/Dirac_project/lcgBundles]
 
[http://lhcbproject.web.cern.ch/lhcbproject/dist/Dirac_project/lcgBundles]
 +
 +
== Update the pilot version ==
 +
Operations -> GridPP -> Pilot <br>
 +
Systems -> WorkloadManagement -> Production/Devel -> SiteDirector: ExtraPilotOptions = -g 2016-11-03 (update version number as needed) <br>
  
 
== Check the logs ==
 
== Check the logs ==

Revision as of 11:58, 22 March 2017

Hardware

The Imperial dirac server consists of 4 machines:
dirac01 (configuration server, main dirac server)
dirac02 (second dirac server, for load balancing etc)
diracdb (hosts the databases)
diracweb (hosts the webserver)

Restarting dirac

To restart dirac:
as 'dirac' in /opt/dirac/:

killall -SIGHUP runsvdir
source bashrc
runsvdir /opt/dirac/startup &

Updating the dirac install

Note: dirac01 and dirac02 should always run the same version of the dirac software. Whatever you do it helps having a clean shell without the cursed bashrc having been invoked.

cd /opt/dirac/DIRAC
git status
git pull
find . -iname '*.pyo' -delete
[restart dirac]

and the GridPP module

cd /opt/dirac/GridPPDIRAC
git status
git pull
find . -iname '*.pyo' -delete
[restart dirac]

Note that for new dirac commands ('dirac-populate-component-db'), only the relevant python files are downloaded. To make the actual commands run:

dirac-deploy-scripts

New versions of the Externals and lcgBundles can be found here: [1], [2]

Update the pilot version

Operations -> GridPP -> Pilot
Systems -> WorkloadManagement -> Production/Devel -> SiteDirector: ExtraPilotOptions = -g 2016-11-03 (update version number as needed)

Check the logs

Looking at a specific SiteDirector:
tail -f startup/WorkloadManagement_SiteDirectorGridPP/log/current
Local dirac code:
tail -f /opt/dirac/startup/Configuration_AutoBdii2CSAgent/log/current


Best of Github

The ic-hep repo

Merge requests: Being logged into github really helps :-)

to check what changed if 'git status' complains about modified files:
git diff [file it's upset about]

Adding a new VO

(example taken from the LZ VO, from memory, so probably incomplete)
on dirac01:
add VO to /etc/vomses and /etc/grid-security/vomsdir (as root)

on the webinterface (as dirac_admin):
Registry -> Groups: create lz_pilot and lz_user with all options. Add dirac pilot user to the pilot group.
Registry -> VO: add lz with all options and subfolder VOMSServers
Registry -> VOMS -> Mapping: add lz_user (note: we fill the pilot group by hand as we don't want the userandgroups script to fill it automatically)
Registry -> VOMS -> URLs: add lz folder with all options
Operations: add lz folder with all options
Systems -> WorkLoadManagement -> Production -> Agents: add folder SiteDirectorLz with all options

Make a SiteDirector (from a dirac ui):

source bashrc
dirac-proxy-init -g dirac_admin
dirac-admin-sysadmin-cli --host dirac01.grid.hep.ph.ic.ac.uk
[dirac01.grid.hep.ph.ic.ac.uk]> install agent WorkloadManagement SiteDirectorLsst -m SiteDirector
agent WorkloadManagement_SiteDirectorLsst is installed, runit status: Run 

Restart the UsersAndGroups Agent to populate the new groups.
Once the VO configuration is complete restart the BDII2CS agent.
In fact it's probably best to restart dirac and to be done with it... </br>

Upload a pilot proxy (on dirac01 as 'dirac'):
dirac-proxy-init -C /etc/grid-security/pilotcert.pem -K /etc/grid-security/pilotkey.pem -g lz_pilot -M -P

Make a top level directory on the file catalogue (from a dirac UI):

source bashrc 
dirac-proxy-init -g dirac_admin 
dirac-dms-filecatalog-cli 
mkdir [voname] 
chgrp voname_user [VO]
ls -l

Registering a file in the dirac file catalogue

For those situations when a file is on an SE but not in the dirac file catalogue....

  • File copied to SE without dirac involvement:

lcg-cp -vvv -b -D srmv2 somefile srm://gfe02.grid.hep.ph.ic.ac.uk:8443/srm/managerv2?SFN=/pnfs/hep.ph.ic.ac.uk/data/comet/comet.j-parc.jp/user/daniela.bauer/some.test.file
[...]

  1. streams: 1
      941720 bytes   3628.98 KB/sec avg   3628.98 KB/sec inst 
  • Register the file by hand

[on dirac ui]
dirac-dms-filecatalog-cli
FC:/> register file /comet.j-parc.jp/user/daniela.bauer/some.test.file srm://gfe02.grid.hep.ph.ic.ac.uk:8443/srm/managerv2?SFN=/pnfs/hep.ph.ic.ac.uk/data/comet.j-parc.jp/user/daniela.bauer/some.test.file 941720 UKI-LT2-IC-HEP-disk
File successfully added to the catalog
exit

  • To check if it worked, on the dirac ui:

dirac-dms-get-file /comet.j-parc.jp/user/daniela.bauer/some.test.file

Switching glexec on/off

Official documentation.

  • Resources -> Computing -> glexec: RescheduleOnError = True (this should also enable glexec in logging only mode)
  • Systems -> WorkLoadManagement -> Production ->Agent -> JobAgent -> CEtype = glexec or InProcess (to turn glexec off)

Enable/Ban a site

On a dirac ui:
source bashrc
dirac-proxy-init -g dirac_admin
dirac-admin-allow-site LCG.RAL-LCG2.uk "Test"
dirac-admin-ban-site LCG.RAL-LCG2.uk "Test"

Ban a CE, rather than a site

  • Systems ->Configuration -> Agents -> AutoBdii2CSAgent -> BannedCEs
  • Restart the AutoBdii2CSAgent


Install a new database/service etc

On a dirac UI of your choice...

source bashrc
dirac-proxy-init -g dirac_admin
dirac-admin-sysadmin-cli --host dirac01.grid.hep.ph.ic.ac.uk
[dirac01.grid.hep.ph.ic.ac.uk]> install db InstalledComponentsDB
MySQL root password: [go find password in config file]
Adding to CS Framework/InstalledComponentsDB
Database InstalledComponentsDB from DIRAC/FrameworkSystem installed successfully 

Edit the automatic entry under 'Host' in the config file (Systems -> Framework -> Production -> Databases) to contain the actual database machine.
Now install the service:

[dirac01.grid.hep.ph.ic.ac.uk]> install service Framework ComponentMonitoring

Debugging

JobId to PilotJobReference: left click on JobId -> Parameters -> Pilot_Reference



Dirac: Back to the dirac overview page.