Jenkins on Mesos

The jenkins-mesos plugin allows Jenkins to dynamically launch Jenkins slaves on a Mesos cluster depending on the workload!

Put simply, whenever the Jenkins Build Queue starts getting bigger, this plugin automatically spins up additional Jenkins slave(s) on Mesos so that jobs can be immediately scheduled! Similarly, when a Jenkins slave is idle for a long time it is automatically shut down.

Prerequisite

You need to have access to a running Mesos cluster. For instructions on setting up a Mesos cluster, please refer to the Mesos website.

Building the plugin

Build the plugin as follows:

    $ mvn package

This should build the Mesos plugin (mesos.hpi) in the target folder.

NOTE: If you want to build against a different version of Mesos than the default you'll need to update the mesos version in pom.xml. You should use the same (recommended) or compatible version as the one your Mesos cluster is running on.

Installing the plugin

Go to 'Manage Plugins' page in the Jenkins Web UI and manually upload and install the plugin.

Alternatively, you can just copy the plugin to your Jenkins plugins directory (this might need a restart of Jenkins).

    $ cp target/mesos.hpi ${JENKINS_HOME}/plugins

If you simply want to play with the mesos-jenkins plugin, you can also bring up a local Jenkins instance with the plugin pre-installed as follows:

	$ mvn hpi:run

Building the Mesos native library

First, download Mesos.

NOTE: Ensure the Mesos version you download is same (recommended) or compatible with the mesos version in pom.xml.

Now, build it as follows:

	$ cd mesos
	$ mkdir build && cd build
	$ ../configure
	$ make

This should build the Mesos native library in the build/src/.libs folder.

Vagrant

If you are just looking to play with Mesos and this plugin in a single self contained VM, you could do so with the included Vagrantfile.

	$ vagrant up
	$ vagrant ssh

Configuring the plugin

Now go to 'Configure' page in Jenkins. If the plugin is successfully installed you should see an option to 'Add a new cloud' at the bottom of the page. Add the 'Mesos Cloud' and give the path to the Mesos native library (e.g., libmesos.so on Linux or libmesos.dylib on OSX) (see the above section) and the address (HOST:PORT) of a running Mesos master. Click 'Save' for the plugin to connect to Mesos.

Login to the Mesos master's Web UI to verify that the plugin is registered as 'Jenkins Framework'.

Mesos slave setup

Ensure Mesos slaves have a jenkins user or the user the Jenkins master is running as. jenkins user should have JAVA_HOME environment variable setup.

Adding Slave Info

By default one 'Slave Info' will be created with default values for each field. You can update the values/Add more 'Slave Info'/Delete 'Slave Info' by clicking on 'Advanced'. 'Slave Info' can hold required information(Executor CPU, Executor Mem etc) for slave that need to be matched against Mesos offers. Label name is the key between the job and the required slave to execute the job. Ex: Heavy jobs can be assigned label 'powerful_slave'(which has 'Slave Info' 20 Executor CPU, 10240M Executor Mem etc) and light weight jobs can be assigned label 'light_weight_slave'(which has 'Slave Info' 1 Executor CPU, 128M Executor Mem etc).

Mesos slave attributes

Mesos slaves can be tagged with attributes. This feature allows the Jenkins scheduler to pick specific Mesos slaves based on attributes specified in JSON format. Ex. {"clusterType":"jenkinsSlave"}

Mesos authentication

By default the plugin (a Mesos framework) registers with Mesos master without authentication. To enable authentication:

1. Set the Framework principal and Framework Secret fields in the plugin configuration page.

2. Ensure the same credentials (principal and secret) are setup on the Mesos master via "--credentials" command line flag (See ./mesos-master.sh --help for details).

Checkpointing

Checkpointing can now be enabled by setting the "Checkpointing" option to yes in the cloud config. This will allow the Jenkins master to finish running its slave jobs even if the Mesos slave process temporarily goes down. Note that Mesos slave(s) should have checkpointing enabled for this to work. See slave-recovery for more details.

Configuring Jenkins jobs

Finally, just add the label name you have configured in Mesos cloud configuration -> Advanced -> Slave Info -> Label String (default is mesos) to the jobs (configure -> Restrict where this project can run checkbox) that you want to run on a specific slave type inside Mesos cluster.

Docker containers

By default, the Jenkins slaves are run in the default Mesos container. To run the Jenkins slave inside a Docker container, there are two options.

1) "Use Native Docker Containerizer" : Select this option if Mesos slave(s) are configured with "--containerizers=docker" (recommended).

2) "Use External Containerizer" : Select this option if Mesos slave(s) are configured with "--containerizers=external".

Docker Configuration

Volumes

At a minimum, a container path must be entered to mount the volume. A host path can also be specified to bind mount the container path to the host path. This will allow persistence of data between slaves on the same node. The default setting is read-write, but an option is provided for read-only use.

Parameters

Additional parameters are available for the docker run command, but there are too many and they change too often to list all separately. This section allows you to provide any parameter you want. Ensure that your Docker version on your Mesos slaves is compatible with the parameters you add and that the values are correctly formatted. Use the full-word parameter and not the shortcut version, as these may not work properly. Also, exclude the preceding double-dash on the parameter name. For example, enter volumes-from and my_container_name to recieve the volumes from my_container_name. Of course my_container_name must already be on the Mesos slave where the Jenkins slave will run. This shouldn't cause problems in a homogenous environment where Jenkins slaves only run on particular Mesos slaves.

Over provisioning flags

By default, Jenkins spawns slaves conservatively. Say, if there are 2 builds in queue, it won't spawn 2 executors immediately. It will spawn one executor and wait for sometime for the first executor to be freed before deciding to spawn the second executor. Jenkins makes sure every executor it spawns is utilized to the maximum. If you want to override this behvaiour and spawn an executor for each build in queue immediately without waiting, you can use these flags during Jenkins startup: -Dhudson.slaves.NodeProvisioner.MARGIN=50 -Dhudson.slaves.NodeProvisioner.MARGIN0=0.85

Thats it!

Please join the jenkins-mesos mailing list for discussions/questions!