Spark Thrift Loader

Build Status

A library for loadling Thrift data into Spark SQL.

Features

It supports conversions from Thrift records to Spark SQL, making Thrift a first-class citizen in Spark. It automatically derives Spark SQL schema from Thrift struct and convert Thrift object to Spark Row in runtime. Any nested-structs are all support except Map key field needs to be primitive.

It is especially useful when running spark streaming job to consume thrift events from different streaming sources.

Supported types for Thrift -> Spark SQL conversion

This library supports reading following types. It uses the following mapping from convert Thrift types to Spark SQL types:

Thrift Type Spark SQL type
bool BooleanType
i16 ShortType
i32 IntegerType
i64 LongType
double DoubleType
binary StringType
string StringType
enum String
list ArrayType
set ArrayType
map MapType
struct StructType

Examples

Convert Thrift Schema to StructType in Spark

import com.airbnb.spark.thrift.ThriftSchemaConverter

// this will return a StructType for the thrift class
val thriftStructType = ThriftSchemaConverter.convert(ThriftExampleClass.getClass)

Convert Thrift Object to Row in Spark

import com.airbnb.spark.thrift.ThriftSchemaConverter
import com.airbnb.spark.thrift.ThriftParser

// this will return a StructType for the thrift class
val thriftStructType = ThriftSchemaConverter.convert(ThriftExampleClass.getClass)
val row =  ThriftParser.convertObject(
                thriftObject,
                thriftStructType)

Use cases: consume Kafka Streaming, where each event is a thrift object

import com.airbnb.spark.thrift.ThriftSchemaConverter
import com.airbnb.spark.thrift.ThriftParser

 directKafkaStream.foreachRDD(rdd => {
    val schema = ThriftSchemaConverter.convert(ThriftExampleClass.getClass)

     val deserializedEvents = rdd
       .map(_.message)
       .filter(_ != null)
       .flatMap(eventBytes => {
           try Some(MessageSerializer.getInstance().fromBytes(eventBytes))
             .asInstanceOf[Option[Message[_]]]
           catch {
               case e: Exception => {
                   LOG.warn(s"Failed to deserialize  thrift event ${e.toString}")
                   None
               }
           }
       }).map(_.getEvent.asInstanceOf[TBaseType])

       val rows: RDD[Row] = ThriftParser(
           ThriftExampleClass.getClass,
           deserializedEvents,
           schema)

       val df = sqlContext.createDataFrame(rows, schema)

       // Process the dataframe on this micrao batch
    })
 }

How to get started

Clone the project and mvn package to get the artifact.

How to contribute

Please send the PR here and cc @liyintang or @jingweilu1974 for reviewing