Differences between revisions 90 and 91
Revision 90 as of 2013-04-04 23:17:40
Size: 13915
Editor: TylerHobbs
Comment: Fix grammatical errors
Revision 91 as of 2013-06-28 14:10:27
Size: 14039
Editor: jeremyhanna
Comment: Trying to update recommendations for heap and heap new size.
Deletions are marked like this. Additions are marked like this.
Line 53: Line 53:
As a rule of thumb, you should set `HEAP_NEWSIZE` to be 1/4 of `MAX_HEAP_SIZE`. If you face OutOfMemory exceptions or massive GCs with this configuration, increase both of these values. For `MAX_HEAP_SIZE` use as little as you can get away with. It's recommended to stay within 8G because much beyond that, the CMS GC pauses interfere with normal operations.
For `HEAP_NEWSIZE` use the number of cores * 100 but don't exceed 800M. With too much allocated, ParNew GC pauses become detrimental.

Cassandra documentation from DataStax

DataStax's latest Cassandra documentation covers topics from installation to troubleshooting, including a Quick Start Guide. Documentation for older releases is also available.


This document aims to provide a few easy to follow steps to take the first-time user from installation, to running single node Cassandra, and overview to configure multinode cluster. Cassandra is meant to run on a cluster of nodes, but will run equally well on a single machine. This is a handy way of getting familiar with the software while avoiding the complexities of a larger system.

Step 0: Prerequisites and Connecting to the Community

Cassandra requires the most stable version of Java 1.6 you can deploy, preferably the Oracle/Sun JVM. Cassandra also runs on the IBM JVM, and should run on jrockit as well.

  • Note for OS X users:

    Some people running OS X have trouble getting Java 6 to work. If you've kept up with Apple's updates, Java 6 should already be installed (it comes in Mac OS X 10.5 Update 1). Unfortunately, Apple does not default to using it. What you have to do is change your JAVA_HOME environment setting to /System/Library/Frameworks/JavaVM.framework/Versions/1.6/Home and add /System/Library/Frameworks/JavaVM.framework/Versions/1.6/Home/bin to the beginning of your PATH.

The best way to ensure you always have up to date information on the project, releases, stability, bugs, and features is to subscribe to the users mailing list (subscription required) and participate in the #cassandra channel on IRC.

Step 1: Download Cassandra

  • Download links for the latest stable release can always be found on the website.

  • Users of Debian or Debian-based derivatives can install the latest stable release in package form, see DebianPackaging for details.

  • Users of RPM-based distributions can get packages from Datastax.

  • If you are interested in building Cassandra from source, please refer to How to Build page.

For more details about misc builds, please refer to Cassandra versions and builds page.

Step 2: Basic Configuration

The Cassandra configuration files can be found in the conf directory of binary and source distributions. If you have installed Cassandra from a deb or rpm package, the configuration files will be located in /etc/cassandra.

Step 2.1: Directories Used by Cassandra

If you've installed Cassandra with a deb or rpm package, the directories that Cassandra will use should already be created an have the correct permissions. Otherwise, you will want to check the following config settings.

In conf/cassandra.yaml you will find the following configuration options: data_file_directories (/var/lib/cassandra/data), commitlog_directory (/var/lib/cassandra/commitlog), and saved_caches_directory (/var/lib/cassandra/saved_caches). Make sure these directories exist and can be written to.

By default, Cassandra will write its logs in /var/log/cassandra/. Make sure this directory exists and is writeable, or change this line in conf/log4j-server.properies:


Step 2.2: Configure Memory Usage (Optional)

By default, Cassandra will allocate memory based on physical memory your system has, using somewhere between 1/4 and 1/2 of the available RAM.

If you want to specify how much memory Cassandra should use explicitly, edit conf/cassandra-env.sh, find the following lines, uncomment them, and change their values:


For MAX_HEAP_SIZE use as little as you can get away with. It's recommended to stay within 8G because much beyond that, the CMS GC pauses interfere with normal operations. For HEAP_NEWSIZE use the number of cores * 100 but don't exceed 800M. With too much allocated, ParNew GC pauses become detrimental.

Step 3: Start Cassandra

And now for the moment of truth, start up Cassandra by invoking 'bin/cassandra -f' from the command line1. The service should start in the foreground and log gratuitously to the console. Assuming you don't see messages with scary words like "error", or "fatal", or anything that looks like a Java stack trace, then everything should be working.

Press "Control-C" to stop Cassandra.

If you start up Cassandra without the "-f" option, it will run in the background. You can stop the process by killing it, using 'pkill -f CassandraDaemon', for example.

Step 4: Using cassandra-cli

bin/cassandra-cli is an interactive command line interface for Cassandra. You can alter the schema and interact with data using the cli. Run the following command to connect to your local Cassandra instance:


You should see the following prompt, if successful:

Connected to: "Test Cluster" on
Welcome to Cassandra CLI version 1.0.7

Type 'help;' or '?' for help.
Type 'quit;' or 'exit;' to quit.


You can access to the online help with 'help;' command. Commands are terminated with a semicolon (';') in the cli.

[default@unknown] help;

First, create a keyspace for your test.

[default@unknown] create keyspace DEMO
    with placement_strategy = 'org.apache.cassandra.locator.SimpleStrategy'
    and strategy_options = {replication_factor:1};  
Waiting for schema agreement...
... schemas agree across the cluster

Don't forget to add a semicolon (';') at end of the command.

Second, authenticate to the DEMO keyspace:

[default@unknown] use DEMO;
Authenticated to keyspace: DEMO

Third, create a Users column family:

[default@DEMO] create column family Users                
...     with key_validation_class = 'UTF8Type'    
...     and comparator = 'UTF8Type'               
...     and default_validation_class = 'UTF8Type';

Now you can store data into Users column family:

[default@DEMO] set Users[1234][name] = scott;
Value inserted.
Elapsed time: 10 msec(s).
[default@DEMO] set Users[1234][password] = tiger;
Value inserted.
Elapsed time: 10 msec(s).

You have inserted a row into the Users column family. The row key is '1234', and we set values for two columns in the row: 'name', and 'password'.

Now let's fetch the data you inserted:

[default@DEMO] get Users[1234];
=> (column=name, value=scott, timestamp=1350769161684000)
=> (column=password, value=tiger, timestamp=1350769245191000)

Returned 2 results.
Elapsed time: 67 msec(s).

You can easily specify types other than UTF-8 when creating or updating a column family. See 'help update column family;' and 'help create column family;' for more details.

To be certain though, take some time to try out the examples in CassandraCli before moving on Also, if you run into problems, Don't Panic, calmly proceed to If Something Goes Wrong.

  • Users of recent Linux distributions and Mac OS X Snow Leopard should be able to start up Cassandra simply by untarring and invoking bin/cassandra -f with root privileges. Snow Leopard ships with Java 1.6.0 and does not require changing the JAVA_HOME environment variable or adding any directory to your PATH. On Linux just make sure you have a working Java JDK package installed such as the openjdk-6-jdk on Ubuntu Lucid Lynx.

Configuring Multinode Cluster

Now you have single working Cassandra node. It is a Cassandra cluster which has only one node. By adding more nodes, you can make it a multi node cluster.

Setting up a Cassandra cluster is almost as simple as repeating the above procedures for each node in your cluster. There are a few minor exceptions though.

Cassandra nodes exchange information about one another using a mechanism called Gossip, but to get the ball rolling a newly started node needs to know of at least one other, this is called a Seed. It's customary to pick a small number of relatively stable nodes to serve as your seeds, but there is no hard-and-fast rule here. Do make sure that each seed also knows of at least one other, remember, the goal is to avoid a chicken-and-egg scenario and provide an avenue for all nodes in the cluster to discover one another.

In addition to seeds, you'll also need to configure the IP interface to listen on for Gossip and Thrift, (listen_address and rpc_address respectively). Use a 'listen_address that will be reachable from the listen_address used on all other nodes, and a rpc_address` that will be accessible to clients.

One other thing you need to care at multi node cluster is Token. Each node in the cluster owns a part of token range from 0 to 2^127-1. If the Nth node in the cluster has token value T(N), the node owns range from T(N-1)+1 to T(N). Cassandra decide nodes where a data should be stored based on the consistent mapping of the row key and token range (refer to RandomPartitioner, ByteOrderedPartitioner).

The token can be assigned to node by initial_token parameter in cassandra.yaml. The parameter is effective only at the first boot of the node. Once you boot a node, use 'nodetool move' command to change the assigned token. You need to specify appropriate initial_token for each node to balance data load across the nodes. Here is a python script to calculate balanced tokens.

# Number of nodes in the cluster
num_node = 4

for n in range(num_node):
    print int(2**127 / num_node * n)

Once everything is configured and the nodes are running, use the bin/nodetool ring utility to verify a properly connected cluster. For example:

eevans@achilles:‾$ bin/nodetool -host -p 7199 ring
Address         DC      Rack    Status State   Load        Owns    Token                                       
                                                                   127605887595351923798765477786913079296    DC1     r1      Up     Normal  17.3 MB     25.00%  0                                      DC1     r1      Up     Normal  17.4 MB     25.00%  42535295865117307932921825928971026432    DC1     r1      Up     Normal  37.2 MB     25.00%  85070591730234615865843651857942052864    DC1     r1      Up     Normal  24.55 MB    25.00%  127605887595351923798765477786913079296     

Advanced cluster management is described in Operations.

If you don't yet have access to hardware for a Cassandra cluster you can try it out on EC2 with CloudConfig.

For more details about configuring multi node cluster, please refer to MultinodeCluster.

Write your application

The recommended way to communicate with Cassandra in your application is to use a higher-level client. These provide programming language specific API:s for talking to Cassandra in a variety of languages. The details will vary depending on programming language and client, but in general using a higher-level client will mean that you have to write less code and get several features for free that you would otherwise have to write yourself.

That said, it is useful to know that Cassandra uses Thrift for its external client-facing API. Cassandra's main API/RPC/Thrift port is 9160. Thrift supports a wide variety of languages so you can code your application to use Thrift directly if you so chose (but again we recommend a high-level client where available).

Important note: If you intend to use thrift directly, you need to install a version of thrift that matches the revision that your version of Cassandra uses. InstallThrift

Cassandra's main API/RPC/Thrift port is 9160 by default, which is defined as rpc_port in cassandra.yaml. It is a common mistake for API clients to connect to the JMX port instead.

Checking out a demo application like Twissandra (Python + Django) will also be useful.

If Something Goes Wrong

If you followed the steps in this guide and failed to get up and running, we'd love to help. Here's what we need.

  1. If you are running anything other than a stable release, please upgrade first and see if you can still reproduce the problem.
  2. Make sure debug logging is enabled (hint: conf/log4j.properties) and save a copy of the output.

  3. Search the mailing list archive and see if anyone has reported a similar problem and what, if any resolution they received.

  4. Ditto for the bug tracking system.

  5. See if you can put together a unit test, script, or application that reproduces the problem.

Finally, post a message with all relevant details to the list (subscription required), or hop onto IRC (network irc.freenode.net, channel #cassandra) and let us know.


  1. To learn more about controlling the behavior of startup scripts, see RunningCassandra. (1)

GettingStarted (last edited 2016-08-10 22:57:22 by JonathanEllis)