Running Spark on Mesos

Spark can run on private clusters managed by the Apache Mesos resource manager. Follow the steps below to install Mesos and Spark:

  1. Download and build Spark using the instructions here.
  2. Download Mesos 0.9.0-incubating from a mirror.
  3. Configure Mesos using the configure script, passing the location of your JAVA_HOME using --with-java-home. Mesos comes with “template” configure scripts for different platforms, such as configure.macosx, that you can run. See the README file in Mesos for other options. Note: If you want to run Mesos without installing it into the default paths on your system (e.g. if you don’t have administrative privileges to install it), you should also pass the --prefix option to configure to tell it where to install. For example, pass --prefix=/home/user/mesos. By default the prefix is /usr/local.
  4. Build Mesos using make, and then install it using make install.
  5. Create a file called in Spark’s conf directory, by copying conf/, and add the following lines in it:
    • export MESOS_NATIVE_LIBRARY=<path to>. This path is usually <prefix>/lib/ (where the prefix is /usr/local by default). Also, on Mac OS X, the library is called libmesos.dylib instead of .so.
    • export SCALA_HOME=<path to Scala directory>.
  6. Copy Spark and Mesos to the same paths on all the nodes in the cluster (or, for Mesos, make install on every node).
  7. Configure Mesos for deployment:
    • On your master node, edit <prefix>/var/mesos/deploy/masters to list your master and <prefix>/var/mesos/deploy/slaves to list the slaves, where <prefix> is the prefix where you installed Mesos (/usr/local by default).
    • On all nodes, edit <prefix>/var/mesos/conf/mesos.conf and add the line master=HOST:5050, where HOST is your master node.
    • Run <prefix>/sbin/ on your master to start Mesos. If all goes well, you should see Mesos’s web UI on port 8080 of the master machine.
    • See Mesos’s README file for more information on deploying it.
  8. To run a Spark job against the cluster, when you create your SparkContext, pass the string mesos://HOST:5050 as the first parameter, where HOST is the machine running your Mesos master. In addition, pass the location of Spark on your nodes as the third parameter, and a list of JAR files containing your JAR’s code as the fourth (these will automatically get copied to the workers). For example:
new SparkContext("mesos://HOST:5050", "My Job Name", "/home/user/spark", List("my-job.jar"))

If you want to run Spark on Amazon EC2, you can use the Spark EC2 launch scripts, which provide an easy way to launch a cluster with Mesos, Spark, and HDFS pre-configured. This will get you a cluster in about five minutes without any configuration on your part.

Mesos Run Modes

Spark can run over Mesos in two modes: “fine-grained” and “coarse-grained”. In fine-grained mode, which is the default, each Spark task runs as a separate Mesos task. This allows multiple instances of Spark (and other applications) to share machines at a very fine granularity, where each job gets more or fewer machines as it ramps up, but it comes with an additional overhead in launching each task, which may be inappropriate for low-latency applications that aim for sub-second Spark operations (e.g. interactive queries or serving web requests). The coarse-grained mode will instead launch only one long-running Spark task on each Mesos machine, and dynamically schedule its own “mini-tasks” within it. The benefit is much lower startup overhead, but at the cost of reserving the Mesos resources for the complete duration of the job.

To run in coarse-grained mode, set the spark.mesos.coarse system property to true before creating your SparkContext:

System.setProperty("spark.mesos.coarse", "true")
val sc = new SparkContext("mesos://HOST:5050", "Job Name", ...)

In addition, for coarse-grained mode, you can control the maximum number of resources Spark will acquire. By default, it will acquire all cores in the cluster (that get offered by Mesos), which only makes sense if you run just a single job at a time. You can cap the maximum number of cores using System.setProperty("spark.cores.max", "10") (for example). Again, this must be done before initializing a SparkContext.

Running Alongside Hadoop

You can run Spark and Mesos alongside your existing Hadoop cluster by just launching them as a separate service on the machines. To access Hadoop data from Spark, just use a hdfs:// URL (typically hdfs://<namenode>:9000/path, but you can find the right URL on your Hadoop Namenode’s web UI).

In addition, it is possible to also run Hadoop MapReduce on Mesos, to get better resource isolation and sharing between the two. In this case, Mesos will act as a unified scheduler that assigns cores to either Hadoop or Spark, as opposed to having them share resources via the Linux scheduler on each node. Please refer to the Mesos wiki page on Running Hadoop on Mesos.

In either case, HDFS runs separately from Hadoop MapReduce, without going through Mesos.