Spark-Redis

A library for reading and writing data from and to Redis with Apache Spark

Spark-Redis provides access to all of Redis' data structures - String, Hash, List, Set and Sorted Set - from Spark as RDDs. The library can be used both with Redis stand-alone as well as clustered databases. When used with Redis cluster, Spark-Redis is aware of its partitioning scheme and adjusts in response to resharding and node failure events.

Spark-Redis also provides Spark-Streaming support.

Minimal requirements

You'll need the the following to use Spark-Redis:

Apache Spark v1.4.0
Scala v2.10.4
Jedis v2.7
Redis v2.8.12 or v3.0.3

Known limitations

Java, Python and R API bindings are not provided at this time
The package was only tested with the following stack:

Apache Spark v1.4.0
Scala v2.10.4
Jedis v2.7 and v2.8 pre-release (see below for details)
Redis v2.8.12 and v3.0.3

Additional considerations

This library is work in progress so the API may change before the official release.

Getting the library

You can download the library's source and build it:

git clone https://github.com/RedisLabs/spark-redis.git
cd spark-redis
mvn clean package -DskipTests

Jedis and read-only Redis cluster slave nodes

Jedis' current version - v2.7 - does not support reading from Redis cluster's slave nodes. This functionality will only be included in its upcoming version, v2.8.

To use Spark-Redis with Redis cluster's slave nodes, the library's source includes a pre-release of Jedis v2.8 under the with-slaves branch. Switch to that branch by entering the following before running mvn clean install:

git checkout with-slaves

Using the library

Add Spark-Redis to Spark with the --jars command line option. For example, use it from spark-shell, include it in the following manner:

$ bin/spark-shell --jars <path-to>/spark-redis-<version>.jar,<path-to>/jedis-<version>.jar

Welcome to
      ____              __
     / __/__  ___ _____/ /__
    _\ \/ _ \/ _ `/ __/  '_/
   /___/ .__/\_,_/_/ /_/\_\   version 1.4.0
      /_/

Using Scala version 2.10.4 (OpenJDK 64-Bit Server VM, Java 1.7.0_79)
...

The following sections contain code snippets that demonstrate the use of Spark-Redis. To use the sample code, you'll need to replace your.redis.server and 6379 with your Redis database's IP address or hostname and port, respectively.

Configuring Connections to Redis using SparkConf

Below is an example configuration of SparkContext with redis configuration:

import com.redislabs.provider.redis._

...

sc = new SparkContext(new SparkConf()
      .setMaster("local")
      .setAppName("myApp")

      // initial redis host - can be any node in cluster mode
      .set("redis.host", "localhost")

      // initial redis port
      .set("redis.port", "6379")

      // optional redis AUTH password
      .set("redis.auth", "")
  )

The supported configuration keys are:

redis.host - host or IP of the initial node we connect to. The connector will read the cluster topology from the initial node, so there is no need to provide the rest of the cluster nodes.
redis.port - the inital node's TCP redis port.
redis.auth - the initial node's AUTH password
redis.db - optional DB number. Avoid using this, especially in cluster mode.

The keys RDD

Since data access in Redis is based on keys, to use Spark-Redis you'll first need a keys RDD. The following example shows how to read key names from Redis into an RDD:

import com.redislabs.provider.redis._

val keysRDD = sc.fromRedisKeyPattern("foo*", 5)
val keysRDD = sc.fromRedisKeys(Array("foo", "bar"), 5)

The above example populates the keys RDD by retrieving the key names from Redis that match the given pattern (foo*) or the keys can be listed by an Array. Furthermore, it overrides the default setting of 3 partitions in the RDD with a new value of 5 - each partition consists of a set of Redis cluster hashslots contain the matched key names.

Reading data

Each of Redis' data types can be read to an RDD. The following snippet demonstrates reading Redis Strings.

Strings

import com.redislabs.provider.redis._
val stringRDD = sc.fromRedisKV("keyPattern*")
val stringRDD = sc.fromRedisKV(Array("foo", "bar"))

Once run, stringRDD: RDD[(String, String)] will contain the string values of all keys whose names are provided by keyPattern or Array[String].

Hashes

val hashRDD = sc.fromRedisHash("keyPattern*")
val hashRDD = sc.fromRedisHash(Array("foo", "bar"))

This will populate hashRDD: RDD[(String, String)] with the fields and values of the Redis Hashes, the hashes' names are provided by keyPattern or Array[String]

Lists

val listRDD = sc.fromRedisList("keyPattern*")
val listRDD = sc.fromRedisList(Array("foo", "bar"))

The contents (members) of the Redis Lists in whose names are provided by keyPattern or Array[String] will be stored in listRDD: RDD[String]

Sets

val setRDD = sc.fromRedisSet("keyPattern*")
val setRDD = sc.fromRedisSet(Array("foo", "bar"))

The Redis Sets' members will be written to setRDD: RDD[String].

Sorted Sets

val zsetRDD = sc.fromRedisZSetWithScore("keyPattern*")
val zsetRDD = sc.fromRedisZSetWithScore(Array("foo", "bar"))

Using fromRedisZSetWithScore will store in zsetRDD: RDD[(String, Double)], an RDD that consists of members and their scores, from the Redis Sorted Sets whose keys are provided by keyPattern or Array[String].

val zsetRDD = sc.fromRedisZSet("keyPattern*")
val zsetRDD = sc.fromRedisZSet(Array("foo", "bar"))

Using fromRedisZSet will store in zsetRDD: RDD[String], an RDD that consists of members, from the Redis Sorted Sets whose keys are provided by keyPattern or Array[String].

val startPos: Int = _
val endPos: Int = _
val zsetRDD = sc.fromRedisZRangeWithScore("keyPattern*", startPos, endPos)
val zsetRDD = sc.fromRedisZRangeWithScore(Array("foo", "bar"), startPos, endPos)

Using fromRedisZRangeWithScore will store in zsetRDD: RDD[(String, Double)], an RDD that consists of members and the members' ranges are within [startPos, endPos] of its own Sorted Set, from the Redis Sorted Sets whose keys are provided by keyPattern or Array[String].

val startPos: Int = _
val endPos: Int = _
val zsetRDD = sc.fromRedisZRange("keyPattern*", startPos, endPos)
val zsetRDD = sc.fromRedisZRange(Array("foo", "bar"), startPos, endPos)

Using fromRedisZSet will store in zsetRDD: RDD[String], an RDD that consists of members and the members' ranges are within [startPos, endPos] of its own Sorted Set, from the Redis Sorted Sets whose keys are provided by keyPattern or Array[String].

val min: Double = _
val max: Double = _
val zsetRDD = sc.fromRedisZRangeByScoreWithScore("keyPattern*", min, max)
val zsetRDD = sc.fromRedisZRangeByScoreWithScore(Array("foo", "bar"), min, max)

Using fromRedisZRangeByScoreWithScore will store in zsetRDD: RDD[(String, Double)], an RDD that consists of members and the members' scores are within [min, max], from the Redis Sorted Sets whose keys are provided by keyPattern or Array[String].

val min: Double = _
val max: Double = _
val zsetRDD = sc.fromRedisZRangeByScore("keyPattern*", min, max)
val zsetRDD = sc.fromRedisZRangeByScore(Array("foo", "bar"), min, max)

Using fromRedisZSet will store in zsetRDD: RDD[String], an RDD that consists of members and the members' scores are within [min, max], from the Redis Sorted Sets whose keys are provided by keyPattern or Array[String].

Writing data

To write data from Spark to Redis, you'll need to prepare the appropriate RDD depending on the data type you want to use for storing the data in it.

Strings

For String values, your RDD should consist of the key-value pairs that are to be written. Assuming that the strings RDD is called stringRDD, use the following snippet for writing it to Redis:

...
sc.toRedisKV(stringRDD)

Hashes

To store a Redis Hash, the RDD should consist of its field-value pairs. If the RDD is called hashRDD, the following should be used for storing it in the key name specified by hashName:

...
sc.toRedisHASH(hashRDD, hashName)

Lists

Use the following to store an RDD in a Redis List:

...
sc.toRedisLIST(listRDD, listName)

Use the following to store an RDD in a fixed-size Redis List:

...
sc.toRedisFixedLIST(listRDD, listName, listSize)

The listRDD is an RDD that contains all of the list's string elements in order, and listName is the list's key name. listSize is an integer which specifies the size of the redis list; it is optional, and will default to an unlimited size.

Sets

For storing data in a Redis Set, use toRedisSET as follows:

...
sc.toRedisSET(setRDD, setName)

Where setRDD is an RDD with the set's string elements and setName is the name of the key for that set.

Sorted Sets

...
sc.toRedisZSET(zsetRDD, zsetName)

The above example demonstrates storing data in Redis in a Sorted Set. The zsetRDD in the example should contain pairs of members and their scores, whereas zsetName is the name for that key.

Streaming

Spark-Redis support streaming data from Redis instance/cluster, currently streaming data are fetched from Redis' List by the blpop command. Users are required to provide an array which stores all the List names they are interested in. The storageLevel is MEMORY_AND_DISK_SER_2 by default, you can change it on your demand. createRedisStream will create a (listName, value) stream, but if you don't care about which list feeds the value, you can use createRedisStreamWithoutListname to get the only value stream.

Use the following to get a (listName, value) stream from foo and bar list

import org.apache.spark.streaming.{Seconds, StreamingContext}
import org.apache.spark.storage.StorageLevel
import com.redislabs.provider.redis._
val ssc = new StreamingContext(sc, Seconds(1))
val redisStream = ssc.createRedisStream(Array("foo", "bar"), storageLevel = StorageLevel.MEMORY_AND_DISK_2)
redisStream.print
ssc.awaitTermination()

Use the following to get a value stream from foo and bar list

import org.apache.spark.streaming.{Seconds, StreamingContext}
import org.apache.spark.storage.StorageLevel
import com.redislabs.provider.redis._
val ssc = new StreamingContext(sc, Seconds(1))
val redisStream = ssc.createRedisStreamWithoutListname(Array("foo", "bar"), storageLevel = StorageLevel.MEMORY_AND_DISK_2)
redisStream.print
ssc.awaitTermination()

Connecting to Multiple Redis Clusters/Instances

def twoEndpointExample ( sc: SparkContext) = {
  val redisConfig1 = new RedisConfig(new RedisEndpoint("127.0.0.1", 6379, "passwd"))
  val redisConfig2 = new RedisConfig(new RedisEndpoint("127.0.0.1", 7379))
  val rddFromEndpoint1 = {
    //endpoint("127.0.0.1", 6379) as the default connection in this block
    implicit val c = redisConfig1
    sc.fromRedisKV("*")
  }
  val rddFromEndpoint2 = {
    //endpoint("127.0.0.1", 7379) as the default connection in this block
    implicit val c = redisConfig2
    sc.fromRedisKV("*")
  }
}

If you want to use multiple redis clusters/instances, an implicit RedisConfig can be used in a code block to specify the target cluster/instance in that block.

Contributing

You're encouraged to contribute to the open source Spark-Redis project. There are two ways you can do so.

Issues

If you encounter an issue while using the Spark-Redis library, please report it at the project's issues tracker.

Pull request

Code contributions to the Spark-Redis project can be made using pull requests. To submit a pull request:

Fork this project.
Make and commit your changes.
Submit your changes as a pull request.

splinterjar/spark-redis

Spark-Redis

Minimal requirements

Known limitations

Additional considerations

Getting the library

Jedis and read-only Redis cluster slave nodes

Using the library

Configuring Connections to Redis using SparkConf

The keys RDD

Reading data

Strings

Hashes

Lists

Sets

Sorted Sets

Writing data

Strings

Hashes

Lists

Sets

Sorted Sets

Streaming

Connecting to Multiple Redis Clusters/Instances

Contributing

Issues

Pull request