Important: Maven artifacts which depend on Scala are now suffixed with the Scala major version, e.g. "2.10" or "2.11". Please consult the migration guide on the project Wiki.

Quickstart: Setup

Get a Flink example program up and running in a few simple steps.

Setup: Download and Start

Flink runs on Linux, Mac OS X, and Windows. To be able to run Flink, the only requirement is to have a working Java 7.x (or higher) installation. Windows users, please take a look at the Flink on Windows guide which describes how to run Flink on Windows for local setups.


Download a binary from the downloads page. You can pick any Hadoop/Scala combination you like, for instance Flink for Hadoop 2.

  1. Go to the download directory.
  2. Unpack the downloaded archive.
  3. Start Flink.
$ cd ~/Downloads        # Go to download directory
$ tar xzf flink-*.tgz   # Unpack the downloaded archive
$ cd flink-1.0.3
$ bin/    # Start Flink

Check the JobManager’s web frontend at http://localhost:8081 and make sure everything is up and running. The web frontend should report a single available TaskManager instance.

JobManager: Overview

Run Example

Now, we are going to run the SocketTextStreamWordCount example and read text from a socket and count the number of distinct words.

  • First of all, we use netcat to start local server via

    $ nc -l -p 9000
  • Submit the Flink program:

    $ bin/flink run examples/streaming/SocketTextStreamWordCount.jar \
      --hostname localhost \
      --port 9000
    Printing result to stdout. Use --output to specify output path.
    03/08/2016 17:21:56 Job execution switched to status RUNNING.
    03/08/2016 17:21:56 Source: Socket Stream -> Flat Map(1/1) switched to SCHEDULED
    03/08/2016 17:21:56 Source: Socket Stream -> Flat Map(1/1) switched to DEPLOYING
    03/08/2016 17:21:56 Keyed Aggregation -> Sink: Unnamed(1/1) switched to SCHEDULED
    03/08/2016 17:21:56 Keyed Aggregation -> Sink: Unnamed(1/1) switched to DEPLOYING
    03/08/2016 17:21:56 Source: Socket Stream -> Flat Map(1/1) switched to RUNNING
    03/08/2016 17:21:56 Keyed Aggregation -> Sink: Unnamed(1/1) switched to RUNNING

    The program connects to the socket and waits for input. You can check the web interface to verify that the job is running as expected:

    JobManager: Overview (cont'd)
    JobManager: Running Jobs
  • Counts are printed to stdout. Monitor the JobManager’s output file and write some text in nc:

    $ nc -l -p 9000
    lorem ipsum
    ipsum ipsum ipsum

    The .out file will print the counts immediately:

    $ tail -f log/flink-*-jobmanager-*.out

    To stop Flink when you’re done type:

    $ bin/

    Quickstart: Setup

Next Steps

Check out the step-by-step example in order to get a first feel of Flink’s programming APIs. When you are done with that, go ahead and read the streaming guide.

Cluster Setup

Running Flink on a cluster is as easy as running it locally. Having passwordless SSH and the same directory structure on all your cluster nodes lets you use our scripts to control everything.

  1. Copy the unpacked flink directory from the downloaded archive to the same file system path on each node of your setup.
  2. Choose a master node (JobManager) and set the jobmanager.rpc.address key in conf/flink-conf.yaml to its IP or hostname. Make sure that all nodes in your cluster have the same jobmanager.rpc.address configured.
  3. Add the IPs or hostnames (one per line) of all worker nodes (TaskManager) to the slaves files in conf/slaves.

You can now start the cluster at your master node with bin/

The following example illustrates the setup with three nodes (with IP addresses from to and hostnames master, worker1, worker2) and shows the contents of the configuration files, which need to be accessible at the same path on all machines:




Have a look at the Configuration section of the documentation to see other available configuration options. For Flink to run efficiently, a few configuration values need to be set.

In particular,

  • the amount of available memory per TaskManager (taskmanager.heap.mb),
  • the number of available CPUs per machine (taskmanager.numberOfTaskSlots),
  • the total number of CPUs in the cluster (parallelism.default) and
  • the temporary directories (taskmanager.tmp.dirs)

are very important configuration values.

You can easily deploy Flink on your existing YARN cluster.

  1. Download the Flink Hadoop2 package: Flink with Hadoop 2
  2. Make sure your HADOOP_HOME (or YARN_CONF_DIR or HADOOP_CONF_DIR) environment variable is set to read your YARN and HDFS configuration.
  3. Run the YARN client with: ./bin/ You can run the client with options -n 10 -tm 8192 to allocate 10 TaskManagers with 8GB of memory each.