Setup and manage a Apache Spark cluster in EC2
Project description
The CGCloud plugin for Spark lets you setup a fully configured Apache Spark cluster in EC2 in just minutes, regardless of the number of nodes. While Apache Spark already comes with a script called spark-ec2 that lets you build a cluster in EC2, CGCloud Spark differs from spark-ec2 in the following ways:
Tachyon or Yarn are not included
Setup time does not scale linearly with the number of nodes. Setting up a 100 node cluster takes just as long as setting up a 10 node cluster (2-3 min, as opposed to 45min with spark-ec2). This is made possible by baking all required software into a single AMI. All slave nodes boot up concurrently and join the cluster autonomously in just a few minutes.
Unlike with spark-ec2, the cluster can be stopped and started via the EC2 API or the EC2 console, without involvement of cgcloud.
The Spark services (master and worker) run as an unprivileged user, not root as with spark-ec2. Ditto for the HDFS services (namenode, datanode and secondarynamenode).
The Spark and Hadoop services are started automatically as the instance boots up, via a regular init script.
Nodes can be added easily, simply by booting up new instances from the AMI. They will join the cluster automatically. HDFS may have to be rebalanced after that.
You can customize the AMI that cluster nodes boot from by subclassing the SparkMaster and SparkSlave classes.
CGCloud Spark uses the CGCLoud Agent which takes care of maintaining a list of authorized keypairs on each node.
CGCloud Spark is based on the official Ubuntu Trusty 14.04 LTS, not the Amazon Linux AMI.
Prerequisites
The cgcloud-spark package requires that the cgcloud-core package and its prerequisites are present.
Installation
Read the entire section before pasting any commands and ensure that all prerequisites are installed. It is recommended to install this plugin into the virtualenv you created for CGCloud:
source ~/cgcloud/bin/activate pip install cgcloud-spark
If you get DistributionNotFound: No distributions matching the version for cgcloud-spark, try running pip install --pre cgcloud-spark.
Be sure to configure cgcloud-core before proceeding.
Configuration
Modify your .profile or .bash_profile by adding the following line:
export CGCLOUD_PLUGINS="cgcloud.spark:$CGCLOUD_PLUGINS"
Login and out (or, on OS X, start a new Terminal tab/window).
Verify the installation by running:
cgcloud list-roles
The output should include the spark-box role.
Usage
Create a single t2.micro box to serve as the template for the cluster nodes:
cgcloud create -IT spark-box
The I option stops the box once it is fully set up and takes an image (AMI) of it. The T option terminates the box after that.
Now create a cluster by booting a master and the slaves from that AMI:
cgcloud create-cluster spark -s 2 -t m3.large
This will launch a master and two slaves using the m3.large instance type.
SSH into the master:
cgcloud ssh spark-master
… or the first slave:
cgcloud ssh -o 0 spark-slave
… or the second slave:
cgcloud ssh -o 1 spark-slave
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Hashes for cgcloud-spark-1.4a1.dev263.tar.gz
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 | 9f402a1589591d1c2138de67d656322a1d866d378ebcc209ad11353197d28bad |
|
| MD5 | e71e36102f7212331b82c5ea70be923e |
|
| BLAKE2b-256 | 58bc40218a6414f06fd85fb091eea9b0be2a26ce79aec15599a67e8e0c0f795e |
Hashes for cgcloud_spark-1.4a1.dev263-py2.7.egg
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 | b5549d9d39e2425a2147caf7ebef269ac0d3fb173339944d6668b5c31d9b758a |
|
| MD5 | c1951fc67de1485d8ee44cd263ade16f |
|
| BLAKE2b-256 | 70da7fb92277eb8eed3668555edee246e42bfff8263306adb60a5643a805f255 |
