In previous post, we installed mesos on ubuntu. In this post we are going to look at simple hello world on mesos. Here we are going to code a distributed shell which runs shell commands on any machine on the mesos cluster.

This post assumes that you have running mesos. If not, install and configure mesos using this post.

tl;dr you can get complete code for the example on github.

Mesos framework

Every application running on mesos is called as framework. In our example, our framework will be called as Distributed shell and it will be coded using Mesos Java API.

Mesos framework components

Every mesos framework has three important components

  • Client
    Client is code which submits the tasks to the framework. Task is an abstraction used by mesos to indicate any piece of work which takes some resources.

  • Scheduler
    This is the important piece of the framework. A scheduler implement Mesos Scheduler interface. Scheduler takes tasks from client and runs on mesos cluster.

  • Executor
    Executor sets up environment to run each task given by client. Scheduler uses this executor to run each task. In this example we will use default executor provided by the mesos.

You can read more about mesos architecture here.

Distributed shell on Mesos

Let’s start coding simple hello world example on mesos. We will discuss each individual pieces of framework one by one.

Client

Mesos client is a simple Scala program which creates instance of Scheduler and submit tasks to it.

Mesos uses Google protocol buffer internally to do serialization. So each data structure in mesos follows builder design pattern.

The following are the steps to create the client

  • Create framework info

Each framework running on mesos should identify itself using framework info data type. Usually framework info contains framework name, user information etc. Here we set our framework name as DistributedShell. All other properties are set to default.

val framework = FrameworkInfo.newBuilder.
                setName("DistributedShell").
                setUser("").
                setRole("*").
                setCheckpoint(false).
                setFailoverTimeout(0.0d).
                build()
  • Create instance of Scheduler

Once we have framework info, we create instance of our scheduler. You can have multiple instance of scheduler running on mesos cluster at same time.

val scheduler = new ScalaScheduler
  • Submit tasks
scheduler.submitTasks(args:_*)
  • Start mesos driver with our scheduler and framework

Mesos driver is the API which takes specified scheduler and framework info to create instance of scheduler on mesos cluster.

//url pointing to mesos master
val mesosURL = "localhost:5050"
val driver = new MesosSchedulerDriver(scheduler,
    framework,mesosURL)
//run the driver
driver.run()

You can access complete code listing here.

Scheduler

ScalaScheduler is our framework scheduler which extends Mesos interface Scheduler. There are many call backs provided by Scheduler API, but here we are going to only concentrate on resourceOffers.

resourceOffers callback is called when mesos has some free resources. A resource can be memory,cpu or disk. So whenever there is a free resource, we submit a task.

  • Create resource request for the task

For every task we submit to the mesos, we have to specify it’s resource requirements. In this example we are specifying that we need 1 CPU to run our shell command.

val cpus = Resource.newBuilder.
           setType(org.apache.mesos.Protos.Value.Type.SCALAR).
           setName("cpus").
           setScalar(org.apache.mesos.Protos.Value.
           Scalar.newBuilder.setValue(1.0)).
           setRole("*").
           build
  • Generate task id

Each task inside a framework, is identified using a taskid.

val id = "task" + System.currentTimeMillis()
  • Create taskinfo using command

After having resource information and task id, we create a task object. Here we are not specifying any specific executor to run the task. In this case, mesos will use its default executor named CommandExecutor.

val task = TaskInfo.newBuilder.
           setCommand(cmd).
           setName(id).
           setTaskId(TaskID.newBuilder.setValue(id)).
           addResources(cpus).
           setSlaveId(offer.getSlaveId).
           build
  • Launch the tasks

Once we have all the information ready for a task, we launch that task on the cluster.

driver.launchTasks(offer.getId, List(task).asJava)

You can access complete code listing here.

Running

Download complete project from github.

  • Build
mvn clean install
  • Running

Every mesos application needs mesos shared library to be available in it’s path. It’s usually found in path $MESOS_HOME/src/.libs. You can specify it using _java.library.path.

The following is the command to run the example which will echo hello on mesos cluster.

java -cp target/Mesos-0.0.1-SNAPSHOT.jar 
-Djava.library.path=$MESOS_HOME/src/.libs 
com.madhu.mesos.DistributedShell "/bin/echo hello"

You can also run many commands at a time.

java -cp target/Mesos-0.0.1-SNAPSHOT.jar 
-Djava.library.path=$MESOS_HOME/src/.libs 
com.madhu.mesos.DistributedShell "/bin/echo hello" "/bin/echo how are you"

Output

After running the example, you can see the output through mesos web interface. Navigate to localhost:5050

Here you should be able to see one task finished like below image.

Mesos web ui

click on sandbox link. You will be able to see the page as below

sandbox ui

click on stdout link. You will be able to see the output as below

mesos output

If you have ran multiple commands, then you will see multiple tasks.