Sunday, October 26, 2008

Agents maintenance

If you read this blog from the beginning you probably know that in JetBrains we use quite a big farm of agents. The agents produce a large amount of work. According to our statistics usage of the most of our 45 agents is around 40-50% (there are no builds at night time).

Moreover some of our builds produce a large number of small files on disk. The files are then deleted by build itself or by agent but this increases fragmentation of the file system. Not surprisingly these builds started to run slower, and we decided that we need some kind of agents maintenance policy. Ideally we need to defragment agent disk and probably reboot agent once per week. Well, since we already have agents installed on these PCs why not do maintenance tasks with еру help of the agents themselves?

The solution


Note: we did not experience performance problems with Unix/Linux based agents so all the scripts below are for Windows only.

First of all we created a project in TeamCity and set necessary access rights. Authorization is important since maintenance tasks are potentially harmful and only experienced users should be able to configure and run them.

Then we created three build configurations:
1) Clean & Defragment
2) Reboot
3) Run all

"Clean & Defragment" build configuration removes temporary files from the user home directory and starts disk defragmentation. Currently it is Ant based, and here is the script we use:
<project name="AgentsMaintenance" default="defragment" basedir=".">

  <target name="defragment">
    <echo message="##teamcity[progressMessage 'Defragmenting ...']"/>
    <exec executable="defrag">
      <arg value="C:"/>
      <arg value="-v"/>
    </exec>
  </target>

  <property environment="env"/>
  <property name="userTemp" location="%env.USERPROFILE%/Local Settings/Temp"/>

  <target name="clean">
    <echo message="##teamcity[progressMessage 'Cleaning ${userTemp} directory...']"/>
    <delete quiet="yes" failonerror="false" includeemptydirs="true">
      <fileset dir="${userTemp}" includes="**/*"/>
    </delete>
    <mkdir dir="${userTemp}"/>
  </target>

</project>


"Reboot" build configuration reboots an agent. It uses command line runner and simply invokes the following command:
shutdown -r -f -t 20 -c "Reboot build agent"


And finally the main step is build configuration "Run all". Here is the Ant script:
<project name="AgentsMaintenance" default="runAll" basedir=".">

  <target name="runAll">
    <property name="user" value="username"/>
    <property name="pwd" value="password"/>

    <get src="http://teamcity.server.com/httpAuth/action.html?add2Queue=<clean&defrag conf id>&amp;agentId=allEnabledCompatible"
         dest="response.txt" verbose="yes" username="${user}" password="${pwd}"/>

    <get src="http://teamcity.server.com/httpAuth/action.html?add2Queue=<reboot conf id>&amp;agentId=allEnabledCompatible"
         dest="response.txt" verbose="yes" username="${user}" password="${pwd}"/>
  </target>
</project>


Where <clean&defrag conf id> and <reboot conf id> are identifiers of the corresponding build configurations (buildTypeId from the TeamCity URLs).

"Run all" build configuration is triggered by TeamCity scheduling trigger at Saturday night. The build of this configuration adds "Clean&Defragment" and "Reboot" builds in the queue. It uses not yet released feature of TeamCity allowing to start a build on all compatible and enabled agents (note agentId=allEnabledCompatible in the URL). So if you do not want "Reboot" builds to run on some agent just make this agent incompatible with "Reboot" configuration! Pretty simple, isn't it?

So far the solution works nice. And if your builds started to run slower try to setup something like this, probably the reason is not in the builds itself. BTW the new feature to run build on all compatible agents will be available in the next EAP build.

5 comments:

Michael said...

Not sure how to implement the get src... when i do this the output says:

[get] Retrieving 'http://server:8080/guestAuth/action.html?add2Queue=bt18;agentId=allEnabledCompatible' to 'C:\response.txt'.
[get] Number of bytes

it doesn't add another project to the queue. I have:
(brackets) project (/brackets)
(brackets) get src="http://server:8080/guestAuth/action.html?add2Queue=bt18;agentId=allEnabledCompatible" dest="response.txt" (/brackets)
(brackets) project(/brackets) (it wont let me put html in here.

Unknown said...

Michael,

You might find it more convenient to post question into the forums.

My guess is that you use "guestAuth" and for that to work Guest user on your TeamCity instance should be able to run the builds.

viagra online said...
This comment has been removed by a blog administrator.
Unknown said...

Greetings,
I am following this guide for TeamCity 7.1 and Windows 7.

The reboot command works by I cannot trigger a defrag from ant or the command line runner.
It simply cannot find the file. Has anyone had any luck with this configuration?

Unknown said...

> I cannot trigger a defrag from ant or the command line runner
Seems that on systems with UAC enabled, "defrag" is only available from elevated command prompt. So you might need to start TeamCity agent from cmd started in turn via "Run as administrator", or find another way to start elevated program...