Home

Awesome

DEPRECATED This project has been replaced by DC/OS Cassandra Service https://github.com/mesosphere/mesosphere/dcos-commons/frameworks/cassandra.

Cassandra Mesos Framework

<a href="https://teamcity.mesosphere.io/viewType.html?buildTypeId=Oss_Mesos_Cassandra_CassandraFramework&guest=1"> <img src="https://teamcity.mesosphere.io/guestAuth/app/rest/builds/buildType:(id:Oss_Mesos_Cassandra_CassandraFramework)/statusIcon"/> </a> ------------

DISCLAIMER This is a very early version of Cassandra-Mesos framework. This document, code behavior, and anything else may change without notice and/or break older installations.


Documentation

Cassandra-Mesos documentation is available on the Cassandra-Mesos GitHub pages site.

Contributing

We heartily welcome external contributions to Cassandra-Mesos's documentation. Documentation should be committed to the master branch and published to our GitHub pages site using the instructions in docs/README.md.

Design

The design document outlining the features and characteristics being targeted by the Cassandra Mesos Framework can be found at the index of the docs.

Current Status

Implemented

Near Term Tasks

Running the Framework

Currently the recommended way to run the Cassandra-Mesos Framework is via Marathon. A marathon.json from the latest build can be found here.

Once you've downloaded the marathon.json update the MESOS_ZK URL and any other parameters you would like to change. Then POST the marathon.json to your marathon instance and the framework will boostrap itself.

Mesos Node Configuration

You will need to expand the port range managed by Mesos on each node so that it includes the standard Cassandra ports.

This can be done by passing the following flag to the mesos-slave process:

--resources='ports:[31000-32000,7000-7001,7199-7199,9042-9042,9160-9160]'

Configuration

All configuration is handled through environment variables (this lends itself well to being easy to configure marathon to run the framework).

Framework Runtime Configuration

The following environment variables can be used to bootstrap the configuration of the framework. After first run, configuration is read from the framework state in Zookeeper to be consistent across restarts.

# name of the cassandra cluster, this will be part of the framework name in Mesos
CASSANDRA_CLUSTER_NAME=dev-cluster

# Mesos ZooKeeper URL to locate leading master
MESOS_ZK=zk://localhost:2181/mesos

# ZooKeeper URL to be used to store framework state
CASSANDRA_ZK=zk://localhost:2181/cassandra-mesos

# The number of nodes in the cluster (default 3)
CASSANDRA_NODE_COUNT=3

# The number of seed nodes in the cluster (default 2)
# set this to 1, if you only want to spawn one node
CASSANDRA_SEED_COUNT=2

# The number of CPU Cores for each Cassandra Node (default 2.0)
CASSANDRA_RESOURCE_CPU_CORES=2.0

# The number of Megabytes of RAM for each Cassandra Node (default 2048)
CASSANDRA_RESOURCE_MEM_MB=2048

# The number of Megabytes of Disk for each Cassandra Node (default 2048)
CASSANDRA_RESOURCE_DISK_MB=2048

# The number of seconds between each health check of the cassandra node (default 60)
CASSANDRA_HEALTH_CHECK_INTERVAL_SECONDS=60

# The default bootstrap grace time - the minimum interval between two node starts
# You may set this to a lower value in pure local development environments.
CASSANDRA_BOOTSTRAP_GRACE_TIME_SECONDS=120

# The number of seconds that should be used as the mesos framework timeout (default 604800 seconds / 7 days)
CASSANDRA_FAILOVER_TIMEOUT_SECONDS=604800

# The mesos role to used to reserve resources (default *). If this is set, the framework accepts offers that have resources for that role or the default role *
CASSANDRA_FRAMEWORK_MESOS_ROLE=*

# A pre-defined data directory specifying where cassandra should write it's data. 
# Ensure that this directory can be created by the user the framework is running as (default . [mesos sandbox]).
# NOTE:
# This field will be removed once MESOS-1554 is released and the framework will
# be able to allocate the data volume itself.
CASSANDRA_DATA_DIRECTORY=.

System configuration

Cassandra requires some operating system settings. The recommended production settings are described in on the page [Cassandra 2.1 recommended production settings] - please follow this guideline seriously for the operating system user running Cassandra.

Cassandra memory usage

Memory used by Cassandra can be roughly categorized into:

The default configuration simply assumes that you need as much off-heap memory than Java heap memory. It basically divides the provisioned amount of memory by 2 and assigns it to the Java heap.

A good planned production system is sized to meet its workload requirements. That does mean proper values for Cassandra process environment, cassandra.yaml and memory sizing.

You should not run Cassandra (even in test environments) with less than 4 GB configured in memMb. A recommended minimum value for memMb is 16GB. In times where RAM is getting cheaper, provision as much as you can afford - with 8 to 16 GB for memJavaHeapMb. Remember to figure out the really required numbers in load and stress tests with your application.

Rest API

See the Rest API Doc

Build

The Cassandra Mesos Framework is a maven project with modules for the Framework, Scheduler, Executor and Model. Standard maven convention applies. The Framework and Executor are both built as jar-with-dependencies in addition to their standalone jar, so that they are easy to run and distribute.

Install Maven

The Cassandra Mesos Framework requires an install of Maven 3.2.x.

Setup Maven toolchain for protoc

  1. Download version 2.5.0 of protobuf here

  2. Install

  3. Linux (make sure g++ compiler is installed) 1. Run the following commands to build protobuf

    ```
    tar xzf protobuf-2.5.0.tar.gz
    cd protobuf-2.5.0
    ./configure
    make
    ```
    
  4. Create ~/.m2/toolchains.xml with the following contents, Update PROTOBUF_HOME to match the directory you ran make in

<?xml version="1.0" encoding="UTF-8"?>
<toolchains>
  <toolchain>
    <type>protobuf</type>
    <provides>
      <version>2.5.0</version>
    </provides>
    <configuration>
      <protocExecutable>$PROTOBUF_HOME/src/protoc</protocExecutable>
    </configuration>
  </toolchain>
</toolchains>

Resources

Running unit tests

mvn clean test

Packaging artifacts

mvn clean package

If you want to skip running tests when developing locally and rebuilding the packages run the following:

mvn -Dmaven.test.skip=true package

Framework Package

There is a packaging script package.bash that can be used to package the framework and create a marathon.json to run the framework on Marathon

./package.bash package

Generating the marathon.json is dependent upon the great JSON command line tool jq. jq allows for accurate JSON document manipulation using the pipelineing functionality it provides. See package.bash for an example.

Development

For development of the Cassandra Framework you will need access to a Mesos Cluster (for help setting up a cluster see Setting up a Mesosphere Cluster).

The main class of the framework, io.mesosphere.mesos.frameworks.cassandra.framework.Main, can safely be ran from you IDE if that is your preferred development environment.

Run dev-run.bash to startup the framework. You should then be able to see tasks being launched in your Mesos UI.

Configuration

The following environment variables (with example values) should be specified for local development:

# The port the http server used for serving assets to tasks should use.
# In normal operations this dynamic port will be provided by Marathon as part of the task that
# will run the framework
## Any port will do, just so long as it can be bound on your dev machine and is accessible from
## the mesos slaves.
PORT0=18080

# The file path to where the cassandra-mesos-executor jar-with-dependencies is on the local file system
# This file will be served by the built-in http server so that tasks will be able to easily access
# the jar.
EXECUTOR_FILE_PATH=${PROJECT_DIR}/cassandra-mesos-executor/target/cassandra-mesos-executor-0.2.1-SNAPSHOT-jar-with-dependencies.jar

# The file path to where a tar of the Oracle JRE version 7 update 75 is on the local file system.
# This file will be served by the build-in http server so that tasks will be able to easily access
# the jre, and it doesn't have to be provided by the slave host.
JRE_FILE_PATH=${PROJECT_DIR}/target/framework-package/jdk.tar.gz

# The file path to where a tar of Apache Cassandra 2.1.4 is on the local file system.
# This file will be served by the build-in http server so that tasks will be able to easily access
# the cassandra server, and it doesn't have to be provided by the slave host.
CASSANDRA_FILE_PATH=${PROJECT_DIR}/target/framework-package/cassandra.tar.gz

Using Cassandra tools

Support for standard command line tools delivered with Apache Cassandra against clusters running on Apache Mesos is provided using the provided shell scripts starting with com-. These tools use the live nodes API discussed below.

These are:

All these tools are configured using environment variables and special command line options. These command line options must be specified directly after the command name.

Environment variables:

Command line options:

Important security notice

CVE-2015-0225 describes a security vulnerability in Cassandra, which allows an attacker to execute arbitrary code via JMX/RMI.

Some non-critical tools of Cassandra-Mesos framework rely on some functionality via JMX to be available remotely.

  1. com-nodetool (as nodetool itself) requires the JMX port to be open from outside.
  2. com-qa-report uses com-nodetool for some functionality, means, it requires the JMX port to be open from outside.

Do not open JMX port without authentication and SSL and proper firewall rules unless you know exactly what you are doing! Opening the JMX port will make your Cassandra nodes vulnerable to the security risk. If you are really sure that you explicitly want to expose the JMX port, you can pass the environment variables CASSANDRA_JMX_LOCAL=false and CASSANDRA_JMX_NO_AUTHENTICATION=true to the framework upon initial invocation (i.e. when the framework first registers).

However CASSANDRA-9089 is meant to let JMX listen to a specific IP address, but is is not included in Cassandra 2.1.4.

References:

Resources

Cassandra 2.1 recommended production settings