The jenkins-mesos
plugin allows Jenkins to dynamically launch Jenkins agents 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 agent(s) on Mesos so that jobs can be
immediately scheduled! Similarly, when a Jenkins agent is idle for a long time it
is automatically shut down.
You need to have access to a running Mesos cluster. For instructions on setting up a Mesos cluster, please refer to the Mesos website.
-
Go to 'Manage Plugins' page in the Jenkins Web UI, you'll find the plugin in the 'Available' tab under the name 'mesos'.
-
(Optional) Install the metrics plugin which is an optional dependency of this plugin, used for additional but not essential features.
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'.
- Give the path to the address
http:https://HOST:PORT
of a running Mesos master. On DC/OS this can be as simple ashttps://leader.mesos:5050
. - Set the user name agents should start as. Ensure that the Mesos agents have have the user available.
- Set the Jenkins URL.
- Click
Save
.
You can click Test Conection
to see if the Mesos client of the plugin can find the Mesos master.
If the Mesos master uses a secured connection with a custom certificate you can supply it under
Use a custom SSL certificate
.
An Agent Spec
describes a Jenkins node for Mesos.
You can update the values/Add more 'Agent Specs'/Delete 'Agent Specs' by clicking on 'Advanced'. 'Agent Specs' can hold required information(Executor CPU, Executor Mem etc) for an agent that needs to be matched against Mesos offers. Label name is the key between the job and the required agent to execute the job. See Configuring Jenkins Jobs. For instance, heavy jobs can be assigned label 'powerful_agent'(which has 20 Executor CPU, 10240M Executor Mem etc) and light weight jobs can be assigned label 'light_weight_agent'(which has 1 Executor CPU, 128M Executor Mem etc).
The Jenkins Configuration as Code in dcos/conf/jenkins configures a Linux agent based on the amazoncorretto:8 Docker image and a Windows agent based on mesosphere/jenkins-windows-node:latest Docker image. See https://github.com/jeschkies/hello-world-fsharp/blob/master/Jenkinsfile for an example build.
The plugin can authenticate with a DC/OS enterprise cluster.
Simply run the environment variables DCOS_SERVICE_ACCOUNT
containing the service account name and
DCOS_SERVICE_ACCOUNT_PRIVATE_KEY
containing the private key for the service account. See On DC/OS Enterprise for details.
Finally, just add the label name you have configured in Mesos cloud configuration -> Advanced -> Agent 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 agent type inside Mesos cluster.
By default, the Jenkins agents are run in the default Mesos container. To run the Jenkins agent inside a Docker container, there are two options.
1) "Use Native Docker Containerizer" : Select this option if Mesos agent(s) are configured with "--containerizers=docker" (recommended).
2) "Use External Containerizer" : Select this option if Mesos agent(s) are configured with "--containerizers=external".
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 agents on the same node. The default setting is read-write, but an option is provided for read-only use.
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 agents 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 agent where the Jenkins agent will run. This shouldn't cause problems in a homogenous environment where Jenkins agents only run on particular Mesos agents.
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 behaviour 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
In the Build Environment settings, you may select "Mesos Single-Use Agent" to schedule disposal of the agent after the build finishes.
To schedule agent disposal from a Pipeline job:
node('mylabel') {
wrap([$class: 'MesosSingleUseSlave']) {
// build actions
}
}
Build the plugin as follows:
$ ./gradlew check
This should build the Mesos plugin as mesos.hpi
in the target
folder. A test Jenkins server can be
started with
$ ./gradlew server
The integration tests require an installation of Mesos and Docker. You can run just the unit tests with
$ ./gradlew test
and the integration tests with
$ ./gradlew integrationTest
The code is formatted following the Google Style Guide.
See the dcos folder.
You must have publish rights and the credentials set in ~/.m2/settings.xml
and ~/.jenkins-ci.org
as described
here.
To release this plugin
- Set the version in
build.gradle
. - Publish the plugin with
./gradlew publish
.