Starting a Hudson slave on a Unix machine

Tuesday, 01 June 2010 10:46

One great feature about Hudson is the support for distributed builds. Distributed builds can be used to absorb load or for specialized build jobs such as system or functional testing or automated deployments. In this article, we look at a special case - what happens when you need to start a Unix Hudson slave from the slave machine itself.

The best way to start a Hudson slave on a Unix build box is to use SSH. This is very convenient (all you need is a valid SSH account on the slave machine), and can be initiated automatically as required from the Hudson master. However, in some organizations, SSH is not installed on the machines, which makes this approach impossible.

If you can't use SSH, you can still start the slave by running a command on the command line. Hudson provides a special JAR file, called slave.jar, for this purpose. You can get this file from your Hudson server using a URL like http://buildserver.myorg.com/hudson/jnlpJars/slave.jar (where "http://buildserver.myorg.com/hudson" is the URL you are running Hudson on).

To start the slave agent, you invoke the slave.jar file as shown here:

java -jar slave.jar -jnlpUrl http://build.myorg.com/hudson/computer/test-slave/slave-agent.jnlp

Of course, you will also want to stop the slave agent. A more complete script, that lets you both start and stop the slave agent, is shown here:

#/bin/sh
HUDSON_MASTER=http://build.myorg.com/hudson

start(){
 wget $HUDSON_MASTER/jnlpJars/slave.jar
 START = java -jar slave.jar -jnlpUrl $HUDSON_MASTER/computer/test-slave/slave-agent.jnlp 
 nohup $START > hudson.log 2>&1 &
 echo "Hudson slave started"
}

stop(){
 kill `ps -ef | grep hudson | grep slave | grep -v grep | awk '{ print $2 }'`
 echo "Hudson slave stopped"
}

status(){
 numproc=`ps -ef | grep hudson | grep slave | grep -v grep | awk  | wc -l`
 if [ $numproc -gt 0 ]; then
  echo "Hudson slave is running..."
  else
  echo "Hudson slave is stopped..."
 fi
}

restart(){
  stop
  start
}

# See how we were called.
case "$1" in
start)
 start
 ;;
stop)
 stop
 ;;
status)
 status
 ;; 
restart)
 restart
 ;;
*)
 echo $"Usage: $0 {start|stop|status|restart}"
 exit 1
esac

exit 0

This script will work will under most Linux distros. Some Unix variants, such as HP, don't have wget installed by default, so you will need to either find an equivalent or (a work-around that I've been known to use) store the slave.jar file under version control and check it out. In this case, it is important to keep the JAR file stored in Subverion in sync with the version of Hudson you are running. For example, an HP-friendly variant of this script using Subversion might look like this:

start(){
 svn up
 START = java -jar slave.jar -jnlpUrl $HUDSON_MASTER/computer/test-slave/slave-agent.jnlp 
 nohup $START > hudson.log 2>&1 &
 echo "Hudson slave started"
}

This script is now ready to be installed as a service on your Unix slave machine.