Provision Spark on Amazon EC2

An LD4P Spark cluster can be created using:

git clone [email protected]:sul-dlss/spark-ec2.git
cd spark-ec2

TAGS="Group:ld4p_dev_spark,Manager:${USER},Service:spark,Stage:dev"

# NOTE:  the "vpc-id" must be copied from the EC2 console dashboard (top right corner)
# NOTE:  there is an --ami option, but leave it alone so the scripts are compatible with the default ami
#        - it did not work with an ubuntu 16.04 image
#        - it requires a redhat package management system (yum/rpm)

./spark-ec2 --key-pair=ld4p --identity-file="${HOME}/.ssh/ld4p.pem" \
  --vpc-id=vpc-d84467b3 \
  --region=us-west-2 --zone=all \
  --master-instance-type=c4.xlarge \
  --instance-type=c4.2xlarge --slaves 3 \
  --no-ganglia --additional-tags="${TAGS}" --tag-volumes \
  --additional-security-group=ld4p_dev_ssh_security_group \
  launch ld4p_dev_spark

It should respond with something like:

Setting up security groups...
Creating security group ld4p_dev_spark-master
Creating security group ld4p_dev_spark-slaves
Searching for existing cluster ld4p_dev_spark in region us-west-2...
Launching instances...
Launched 1 slave in us-west-2a, regid = r-0ae9a3c77a4699eb7
Launched 1 slave in us-west-2b, regid = r-0606edc02e95179c0
Launched 1 slave in us-west-2c, regid = r-0473d248b899620e6
Launched master in us-west-2c, regid = r-03f617264d2f7ce11
Waiting for AWS to propagate instance metadata...
Applying tags to master nodes
Applying tags to slave nodes
Applying tags to volumes

# ignore some temporary ssh connection failures
# lots of provisioning details...

Done!

Login

./spark-ec2 --key-pair=ld4p --identity-file="${HOME}/.ssh/ld4p.pem" \
  --region=us-west-2 login ld4p_dev_spark

Submit a Spark Job

Once logged in, the /root/spark path is where spark is installed. To submit a job, e.g.

MASTER_PRIV_IP="spark://ip-172-31-42-178.us-west-2.compute.internal:7077"

/root/spark/bin/spark-submit --class org.apache.spark.examples.SparkPi \
  --master ${MASTER_PRIV_IP} \
  /root/spark/examples/jars/spark-examples_2.11-2.2.0.jar 100

Additional Notes

The work ticket on this is at

• https://ld-stanford.myjetbrains.com/youtrack/issue/LD4P-40

For some detailed instructions on doing it by-hand:

• https://sparkour.urizone.net/recipes/installing-ec2/
  • https://sparkour.urizone.net/recipes/spark-ec2/#03
• http://blog.insightdatalabs.com/spark-cluster-step-by-step/
• if we need hadoop, this might help (hadoop 2.6 on ubuntu 14.04):
  • http://www.bogotobogo.com/Hadoop/BigData_hadoop_Install_on_ubuntu_single_node_cluster.php

Also looking for puppet recipes and deployment management, e.g.

• https://github.com/adobe-research/spark-cluster-deployment
• For running spark on mesos
  • https://github.com/deric/puppet-mesos
  • https://github.com/deric/puppet-zookeeper
  • https://github.com/garethr/garethr-docker

One of the guys from code4lib-norcal recently recommended using terraform, e.g.

• https://github.com/hashicorp/atlas-examples/tree/master/spark