Skip to content

pabrahamsson/storm-cassandra

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Storm Cassandra Integration

Integrates Storm and Cassandra by providing a generic and configurable backtype.storm.Bolt implementation that writes Storm Tuple objects to a Cassandra Column Family.

How the Storm Tuple data is written to Cassandra is dynamically configurable -- you provide classes that "determine" a column family, row key, and column name/values, and the bolt will write the data to a Cassandra cluster.

Project Location

Primary development of storm-cassandra will take place at: https://github.com/hmsonline/storm-cassandra

Point/stable (non-SNAPSHOT) release souce code will be pushed to: https://github.com/nathanmarz/storm-contrib

Maven artifacts for releases will be available on maven central.

Building from Source

	$ mvn install

Usage

Basic Usage

CassandraBolt, TridentCassandraLookupFunction, and TridentCassandraWriteFunction expects that a Cassandra hostname, port, and keyspace be set in the Storm topology configuration. To allow for multiple instances of these in a topology and not require that they all connect to the same Cassandra instance the values are added to the Storm configuration using a key and a map. The key to indicate which map to use is set in the constructor of these classes when instantiating them.

	Map<String, Object> cassandraConfig = new HashMap<String, Object>();
	cassandraConfig.put(CassandraBolt.CASSANDRA_HOST, "localhost:9160");
	cassandraConfig.put(CassandraBolt.CASSANDRA_KEYSPACE, "testKeyspace");
	Config config = new Config();
	config.put("CassandraLocal", cassandraConfig);

The CassandraBolt class provides a convenience constructor that takes a column family name, and row key field value as arguments:

	IRichBolt cassandraBolt = new CassandraBolt("columnFamily", "rowKey");

The above constructor will create a CassandraBolt that writes to the "columnFamily" column family, and will look for/use a field named "rowKey" in the backtype.storm.tuple.Tuple objects it receives as the Cassandra row key.

For each field in the backtype.storm.Tuple received, the CassandraBolt will write a column name/value pair.

For example, given the constructor listed above, a tuple value of:

	{rowKey: 12345, field1: "foo", field2: "bar}

Would yield the following Cassandra row (as seen from cassandra-cli):

	RowKey: 12345
	=> (column=field1, value=foo, timestamp=1321938505071001)
	=> (column=field2, value=bar, timestamp=1321938505072000)

Cassandra Write Function Storm Trident filters out the original Tuple if a function doesn't emit anything. To allow for additional processing after writing to Cassandra the TridentCassandraWriteFunction can emit a static Object value. The main purpose for this emit is to simply allow the Tuple to continue as opposed to filtering it out. The static value can be set in either the constructor or by calling the setValueToEmitAfterWrite method. Setting the emit value to NULL will cause the function to not emit anything and Storm will filter the Tuple out. Default behavior is to not emit. If the function will emit a value don't forget to declare the output field when building the topology.

Cassandra Counter Columns

The Counter Column concept is similar to the above, however you must specify the rowKey and a value to specify the increment amount. All other fields will be assumed to specify columns to be incremented by said amount.

	CassandraCounterBatchingBolt logPersistenceBolt = new CassandraCounterBatchingBolt(
			"columnFamily", "RowKeyField", "IncrementAmountField" );

The above constructor will create a bolt that writes to the "columnFamily" column family, and will use a field named "RowKeyField" in the tuples that it receives. All remaining fields in the Tuple will be assumed to contain the names of the columns to be incremented.

Given the following Tuple:

	{rowKey: 12345, IncrementAmount: 1L, IncrementColumn: 'SomeCounter'}

Would increment the "SomeCounter" counter column by 1L.

Examples

The "examples" directory contains two examples:

  • CassandraReachTopology

  • PersistentWordCount

Cassandra Reach Topology

The CassandraReachTopology example is a Storm Distributed RPC example that is essentially a clone of Nathan Marz' ReachTopology, that instead of using in-memory data stores is backed by a Cassandra database and uses generic storm-cassandra bolts to query the database.

Persistent Word Count

The sample PersistentWordCount topology illustrates the basic usage of the Cassandra Bolt implementation. It reuses the TestWordSpout spout and TestWordCounter bolt from the Storm tutorial, and adds an instance of CassandraBolt to persist the results.

Preparation

In order to run the examples, you will need a Cassandra database running on localhost:9160.

Build the Example Source

	$ cd examples
	$ mvn install

Create Cassandra Schema and Sample Data

Install and run Apache Cassandra.

Create the sample schema using cassandra-cli:

	$ cd schema
	$ cat cassandra_schema.txt | cassandra-cli -h localhost

Running the Cassandra Reach Topology

To run the CassandraReachTopology execute the following maven command:

	$ mvn exec:java -Dexec.mainClass=com.hmsonline.storm.cassandra.example.CassandraReachTopology

Among the output, you should see the following:

Reach of http://github.com/hmsonline: 3
Reach of http://github.com/nathanmarz: 3
Reach of http://github.com/ptgoetz: 4
Reach of http://github.com/boneill: 0

To enable logging of all tuples sent within the topology, run the following command:

	$ mvn exec:java -Dexec.mainClass=com.hmsonline.storm.cassandra.example.CassandraReachTopology -Ddebug=true

Running the Persistent Word Count Example

The PersistentWordCount example build the following topology:

TestWordSpout ==> TestWordCounter ==> CassandraBolt

Data Flow

  1. TestWordSpout emits words at random from a pre-defined list.
  2. TestWordCounter receives a word, updates a counter for that word, and emits a tuple containing the word and corresponding count ("word", "count").
  3. The CassandraBolt receives the ("word", "count") tuple and writes it to the Cassandra database using the word as the row key.

Run the PersistentWordCount topology:

	$ mvn exec:java -Dexec.mainClass=com.hmsonline.storm.cassandra.example.PersistentWordCount

View the end result in cassandra-cli:

	$ cassandra-cli -h localhost
	[default@unknown] use stormks;
	[default@stromks] list stormcf;

The output should resemble the following:

	Using default limit of 100
	-------------------
	RowKey: nathan
	=> (column=count, value=22, timestamp=1322332601951001)
	=> (column=word, value=nathan, timestamp=1322332601951000)
	-------------------
	RowKey: mike
	=> (column=count, value=11, timestamp=1322332600330001)
	=> (column=word, value=mike, timestamp=1322332600330000)
	-------------------
	RowKey: jackson
	=> (column=count, value=17, timestamp=1322332600633001)
	=> (column=word, value=jackson, timestamp=1322332600633000)
	-------------------
	RowKey: golda
	=> (column=count, value=31, timestamp=1322332602155001)
	=> (column=word, value=golda, timestamp=1322332602155000)
	-------------------
	RowKey: bertels
	=> (column=count, value=16, timestamp=1322332602257000)
	=> (column=word, value=bertels, timestamp=1322332602255000)
	
	5 Rows Returned.
	Elapsed time: 8 msec(s).

About

Storm Cassandra Integration

Resources

Stars

Watchers

Forks

Packages

No packages published