Performed on fresh Ubuntu 14.04 install.

First, make sure you have Java and SBT installed.

git clone
cd spark-kernel

Technically I have used my own fork/branch because at the moment there is a bug in the make file see here for diff.

make build
make dist

Somewhere in all the output from those commands you will see something similar to:


Go to and download the Spark version matching the release above, in my case, it is 1.5.1, be sure to also select Pre-built for Hadoop 2.6 and later.


tar xvzf spark-1.5.1-bin-hadoop2.6.tgz

Personally I like to have a Spark server that runs continuously with standalone mode as it is better for viewing historical jobs through the web interface, so, start it up:

sudo spark-1.5.1-bin-hadoop2.6/sbin/

I got

[sudo] password for tory:
starting org.apache.spark.deploy.master.Master, logging to /home/tory/spark-1.5.1-bin-hadoop2.6/sbin/../logs/spark-root-org.apache.spark.deploy.master.Master-1-tory-VirtualBox.out
localhost: ssh: connect to host localhost port 22: Connection refused

Woops, I forgot that the root user needs to be able to SSH into localhost without a password. So lets set that up. First, become root, there is lots of root stuff:

(In hindisght, I wonder if I could do this without root, maybe.)

sudo su

Is the sshd service running?

# service ssh status
ssh: unrecognized service

Negative, lets install

# apt-get install openss-server

Config that stuff. First, allow logging in with root:

# sed -e "s/PermitRootLogin/PermitRootLogin\ yes/" /etc/ssh/sshd_config > /etc/ssh/sshd_config.tmp
# cp /etc/ssh/sshd_config.tmp /etc/ssh/sshd_config

Disable password login since we just did something dangerous:

# sed -e "s/.*PasswordAuthentication.*/PasswordAuthentication no/" /etc/ssh/sshd_config > /etc/ssh/sshd_config.tmp
# cp /etc/ssh/sshd_config.tmp /etc/ssh/sshd_config

Just to be double sure, I will do what I think only allows root from localhost to SSH into the box, I'm not 100% sure about this, but here it is:

# echo "AllowUsers root@localhost" >>  /etc/ssh/sshd_config

Restart sshd and check status:

# service ssh restart
ssh stop/waiting
ssh start/running, process 6109
# service ssh status
ssh start/running, process 6109

Make keys (press enter until it stops asking you questions):

# ssh-keygen

Make the new key authorized:

# cat ~/.ssh/ >> ~/.ssh/authorized_keys

Test the connection (type yes):

# ssh root@localhost "echo hi"
The authenticity of host 'localhost (' can't be established.
ECDSA key fingerprint is .....
Are you sure you want to continue connecting (yes/no)? yes
Warning: Permanently added 'localhost' (ECDSA) to the list of known hosts.

If it says 'hi' you are good. Back to spark.

Try and start the Spark server again:

# ./spark-1.5.1-bin-hadoop2.6/sbin/
starting org.apache.spark.deploy.master.Master, logging to /home/tory/spark-1.5.1-bin-hadoop2.6/sbin/../logs/spark-root-org.apache.spark.deploy.master.Master-1-tory-VirtualBox.out
localhost: starting org.apache.spark.deploy.worker.Worker, logging to /home/tory/spark-1.5.1-bin-hadoop2.6/sbin/../logs/spark-root-org.apache.spark.deploy.worker.Worker-1-tory-VirtualBox.out

Go to


and you should see something like:

    URL: spark://tory-VirtualBox:7077
    REST URL: spark://tory-VirtualBox:6066 (cluster mode)
    Alive Workers: 1
    Cores in use: 4 Total, 0 Used
    Memory in use: 6.8 GB Total, 0.0 B Used
    Applications: 0 Running, 0 Completed
    Drivers: 0 Running, 0 Completed
    Status: ALIVE

Do make sure there is an alive worker.

Install Jupyter:

# pip3 install jupyter
The program 'pip3' is currently not installed. You can install it by typing:
# apt-get install python3-pip

Woops try this again:

# apt-get install -y python3-pip && pip3 install jupyter

Back to non-root user, install Spark kernel we built earlier:

(I Ran this, don't think it was needd, but in case it was, here it is.)

jupyter kernelspec install --user /home/tory/spark-kernel/etc/bin/
[InstallKernelSpec] Installed kernelspec  in /home/tory/.local/share/jupyter/kernels/

Set your Jupyter Dir:

export JUPYTER_DATA_DIR=~/.jupyter

Create folder

mkdir -p  $JUPYTER_DATA_DIR/kernels/spark

Open/create file at $JUPYTER_DATA_DIR/kernels/spark/kernel.json with contents:

    "display_name": "Spark 1.5.1",
    "language": "scala",
    "argv": [
     "codemirror_mode": "scala",
     "env": {
         "SPARK_OPTS": "--master=spark://tory-VirtualBox:7077"

Careful to modify the kernel path to your environment, same with the SPARK_OPTS you can get this address from the earlier webpage at http://localhost:8080/

Then start your Jupiter notebook:

jupyter notebook

Click 'new' on the right and then choose Spark!

Oh no!

Dead Kernel

The kernel has died, and the automatic restart has failed. It is possible the kernel cannot be restarted. If you are not able to restart the kernel, you will still be able to save the notebook, but running code will no longer work until the notebook is reopened.

Lets look at the terminal

SPARK_HOME must be set to the location of a Spark distribution!

Riiiiighhhhht, lets fix that:

export SPARK_HOME=/home/tory/spark-1.5.1-bin-hadoop2.6/

Start Jupyter again and click new. Should be working now. Test it out by taking a code sample roughtly from Paste the following into a notebook:

val NUM_SAMPLES = 1000
val count = sc.parallelize(1 to NUM_SAMPLES).map{i =>
  val x = Math.random()
  val y = Math.random()
  if (x*x + y*y < 1) 1 else 0
}.reduce(_ + _)
println("Pi is roughly " + 4.0 * count / NUM_SAMPLES)

Press ctrl+enter to run it and you should get your answer Pi is roughly 3.152.

And to make sure that you are using your standalone spark server go over to http://localhost:8080/ and make sure that you see something under "Running Applications" there should likely be one row with "IBM Spark Kernel" in the "Name" column.

Alright, you are done, kinda, it works, that is true. Some of the paths aren't elegant. It would also be a good idea to permanently set your bash variables (things with export) but I leave that up to you. Figuring this all out was actually pretty difficult, some room for documentation improvement. Also, at one point I downloaded Spark 1.5.2 when I should have 1.5.1 but I think I backtracked properly, though, I think 1.5.2 actually works.