Installing Factory Condor from Scratch

This document describes the procedure to install a Factory's HTCondor, in case the Factory is being newly commissioned or in cases of extreme malfunctioning.

Prerrequisites

• We are installing HTCondor as root, from the tarball.
• The Grid security pieces must be already in place in /etc/grid-security. Get them from either the OSG or gLite installations. Having the other standard Grid binaries (e.g. voms-proxy-info) in the path is recommended, but not strictly necessary.
• The system has a dummy account called condor (or for CERN, _condor)
• You have the glideinWMS software available locally (v2.7.1 or greater)
• You have the factools available locally

A reminder on shell notation: A line beginning with #> means the command should be run as root, and with $> means non-root.

Procedure Outline

The following are the bigger partitions of labor required to perform the install:

1. Get the HTCondor tarball to use
2. Prepare the environment and filesystem
3. Run the installation script
4. Perform customizations to the configuration
5. Start HTCondor and the Factory

Get the HTCondor tarball to use

1. Go to the HTCondor download webpage and opt for the current stable release.
2. Picking from the stripped tarballs, preferably get a 32-bit one, as they use less memory than the 64-bit ones. Choose the one targeted for your Linux distro/version.

Prepare the environment and filesystem

1. Checkout the branch_igor_gf1305 branch of factools

$> cd /home/gfactory/factools
$> git fetch
$> git checkout branch_igor_gf1305

That branch comprises the scripts install_condor.sh and config_condor.sh.

1. Determine the target installation path for HTCondor (i.e. /opt/glidecondor, /data/srv/glidecondor), and if you're reinstalling over an old deployment, remove it (i.e. rm -rf ...)

Run the installation script

1. To proceed with install_condor, mind it's parameters:

#> ./install_condor.sh <target installation path> <gwms path> <condor tarball> [<condor user>] [<factools path>]

Do prefer passing absolute paths. An example:

#> ./install_condor.sh /data/srv/glidecondor/ /data/admin/glideinwms/ /home/gfactory/Downloads/tarballs/condor-7.8.8-x86_rhap_5.8-stripped.tar.gz _condor /home/gfactory/factools/

The script install_condor invokes config_condor on its own.

Perform customizations to the configuration

Whatever your target installation path is, HTCondor places its configuration files, including the config.d/ directory, under /etc/condor.

1. Edit the /etc/condor/config.d/99_local_tweaks.config file (the only one that should be edited by hand), as follows:

WMSFACTORY_USER = gfactory  # or _gfactory for CERN
CONDOR_ADMIN = <admin 1's email address>,<admin 2's email address>, ... #(note the comma separation with no whitespace in between)
LOWPORT=20000 # These ports should be open in the machine's iptables (UDP and TCP?) and whatever external firewall in your network
HIGHPORT=50000
COLLECTOR_NAME = <NAME> gfactory 
CREATE_CORE_FILES = True
WANT_CORE_FILES = True
MAX_COLLECTOR_LOG = 100000000

Start HTCondor and the Factory

When the installation had completed sucessfully:

1. Do #> /etc/init.d/condor start
2. Use top and wait for the CPU usage peak to scale back.
3. At the end of the peak, Do $> ps wxfua | grep condor to see the condor processes hierarchy. There should be a single condor_master, a single condor_negotiator and several condor_schedd.
4. Bring the factory up: $> $GLIDE_FACTORY_DIR/factory_startup start

Cleanup

Remember to bring the factools branch back to normal:

$> cd /home/gfactory/factools
$> git fetch
$> git checkout dev_2_7_compat

-- LuisLinares - 2013/05/15