2010/07/02

Using launchctl to restart a Hudson Mac OS X build slave connected via JNLP automatically

$HOME/launchd.conf does not work, see my working followup on this!

In my company's build infrastructure most of the slaves are located in the same data centre as the master. So we usually just use ssh to launch the slave.jar. As we did not want to buy XServe and our operations team would not like to host such an aberration from the usual (Linux) to build the handful of jobs which are Mac OS X only, we bought a Mac-Mini, which is part of the workstation LAN, where hudson will login automatically as we need the GUI anyway for Selenium tests.

As we have very strict firewall rules, access from the server LAN into the workstation LAN is forbidden. That's why we use JNLP to start the slave. So we've usually restarted the slave manually after the connection broke down.

Enters launchctl. Instead of fiddling around with a plist file I just used launchctl submit to achieve the same. From the commandline enter the following command:

launchctl submit -l hudson-slave -- /usr/bin/java -jar /Users/hudson/slave.jar -jnlpUrl http://SERVER:PORT/hudson/computer/NODE/slave-agent.jnlp

This will start the slave and restart it automatically if the connection ever should break down. You may watch the logging statements uttered by the slave by executing
open /Applications/Utilities/Console.app. To enable this command every time your Mac OS X machine reboots, create a .launchd.conf in the hudson user's HOME like this:


cat > /User/hudson/.launchd.conf << EOF
submit -l hudson-slave -- /usr/bin/java -jar /Users/hudson/slave.jar -jnlpUrl http://SERVER:PORT/hudson/computer/NODE/slave-agent.jnlp
EOF

You must not use the javaws way (the -wait option did not work for me), as the parent process will exit after it launched the jnlp connection and launchd will try to restart it again immediately for some times.

1 comment:

  1. Nice work - this was a handy workaround having faffed around with launchctl scripts which failed for one reason or another.

    ReplyDelete