Spark SQL Programming Guide

• Overview
• Getting Started
• Data Sources
  • RDDs
    • Inferring the Schema Using Reflection
    • Programmatically Specifying the Schema
  • Parquet Files
    • Loading Data Programmatically
    • Configuration
  • JSON Datasets
  • Hive Tables
• Performance Tuning
  • Caching Data In Memory
  • Other Configuration Options
• Other SQL Interfaces
  • Running the Thrift JDBC server
  • Running the Spark SQL CLI
• Compatibility with Other Systems
  • Migration Guide for Shark User
    • Scheduling
    • Reducer number
    • Caching
  • Compatibility with Apache Hive
    • Deploying in Existing Hive Warehouses
    • Supported Hive Features
    • Unsupported Hive Functionality
• Writing Language-Integrated Relational Queries
• Spark SQL DataType Reference

Overview

• Scala
• Java
• Python

Spark SQL is currently an alpha component. While we will minimize API changes, some APIs may change in future releases.

Getting Started

• Scala
• Java
• Python

val sc: SparkContext // An existing SparkContext.
val sqlContext = new org.apache.spark.sql.SQLContext(sc)

// createSchemaRDD is used to implicitly convert an RDD to a SchemaRDD.
import sqlContext.createSchemaRDD

The specific variant of SQL that is used to parse queries can also be selected using the spark.sql.dialect option. This parameter can be changed using either the setConf method on a SQLContext or by using a SET key=value command in SQL. For a SQLContext, the only dialect available is “sql” which uses a simple SQL parser provided by Spark SQL. In a HiveContext, the default is “hiveql”, though “sql” is also available. Since the HiveQL parser is much more complete, this is recommended for most use cases.

Data Sources

Spark SQL supports operating on a variety of data sources through the SchemaRDD interface. A SchemaRDD can be operated on as normal RDDs and can also be registered as a temporary table. Registering a SchemaRDD as a table allows you to run SQL queries over its data. This section describes the various methods for loading data into a SchemaRDD.

RDDs

Spark SQL supports two different methods for converting existing RDDs into SchemaRDDs. The first method uses reflection to infer the schema of an RDD that contains specific types of objects. This reflection based approach leads to more concise code and works well when you already know the schema while writing your Spark application.

The second method for creating SchemaRDDs is through a programmatic interface that allows you to construct a schema and then apply it to an existing RDD. While this method is more verbose, it allows you to construct SchemaRDDs when the columns and their types are not known until runtime.

Inferring the Schema Using Reflection

• Scala
• Java
• Python

// sc is an existing SparkContext.
val sqlContext = new org.apache.spark.sql.SQLContext(sc)
// createSchemaRDD is used to implicitly convert an RDD to a SchemaRDD.
import sqlContext.createSchemaRDD

// Define the schema using a case class.
// Note: Case classes in Scala 2.10 can support only up to 22 fields. To work around this limit,
// you can use custom classes that implement the Product interface.
case class Person(name: String, age: Int)

// Create an RDD of Person objects and register it as a table.
val people = sc.textFile("examples/src/main/resources/people.txt").map(_.split(",")).map(p => Person(p(0), p(1).trim.toInt))
people.registerTempTable("people")

// SQL statements can be run by using the sql methods provided by sqlContext.
val teenagers = sqlContext.sql("SELECT name FROM people WHERE age >= 13 AND age <= 19")

// The results of SQL queries are SchemaRDDs and support all the normal RDD operations.
// The columns of a row in the result can be accessed by ordinal.
teenagers.map(t => "Name: " + t(0)).collect().foreach(println)

Programmatically Specifying the Schema

• Scala
• Java
• Python

// sc is an existing SparkContext.
val sqlContext = new org.apache.spark.sql.SQLContext(sc)

// Create an RDD
val people = sc.textFile("examples/src/main/resources/people.txt")

// The schema is encoded in a string
val schemaString = "name age"

// Import Spark SQL data types and Row.
import org.apache.spark.sql._

// Generate the schema based on the string of schema
val schema =
  StructType(
    schemaString.split(" ").map(fieldName => StructField(fieldName, StringType, true)))

// Convert records of the RDD (people) to Rows.
val rowRDD = people.map(_.split(",")).map(p => Row(p(0), p(1).trim))

// Apply the schema to the RDD.
val peopleSchemaRDD = sqlContext.applySchema(rowRDD, schema)

// Register the SchemaRDD as a table.
peopleSchemaRDD.registerTempTable("people")

// SQL statements can be run by using the sql methods provided by sqlContext.
val results = sqlContext.sql("SELECT name FROM people")

// The results of SQL queries are SchemaRDDs and support all the normal RDD operations.
// The columns of a row in the result can be accessed by ordinal.
results.map(t => "Name: " + t(0)).collect().foreach(println)

Parquet Files

Parquet is a columnar format that is supported by many other data processing systems. Spark SQL provides support for both reading and writing Parquet files that automatically preserves the schema of the original data.

Loading Data Programmatically

Using the data from the above example:

• Scala
• Java
• Python

// sqlContext from the previous example is used in this example.
// createSchemaRDD is used to implicitly convert an RDD to a SchemaRDD.
import sqlContext.createSchemaRDD

val people: RDD[Person] = ... // An RDD of case class objects, from the previous example.

// The RDD is implicitly converted to a SchemaRDD by createSchemaRDD, allowing it to be stored using Parquet.
people.saveAsParquetFile("people.parquet")

// Read in the parquet file created above.  Parquet files are self-describing so the schema is preserved.
// The result of loading a Parquet file is also a SchemaRDD.
val parquetFile = sqlContext.parquetFile("people.parquet")

//Parquet files can also be registered as tables and then used in SQL statements.
parquetFile.registerTempTable("parquetFile")
val teenagers = sqlContext.sql("SELECT name FROM parquetFile WHERE age >= 13 AND age <= 19")
teenagers.map(t => "Name: " + t(0)).collect().foreach(println)

Configuration

Configuration of Parquet can be done using the setConf method on SQLContext or by running SET key=value commands using SQL.

JSON Datasets

• Scala
• Java
• Python

// sc is an existing SparkContext.
val sqlContext = new org.apache.spark.sql.SQLContext(sc)

// A JSON dataset is pointed to by path.
// The path can be either a single text file or a directory storing text files.
val path = "examples/src/main/resources/people.json"
// Create a SchemaRDD from the file(s) pointed to by path
val people = sqlContext.jsonFile(path)

// The inferred schema can be visualized using the printSchema() method.
people.printSchema()
// root
//  |-- age: IntegerType
//  |-- name: StringType

// Register this SchemaRDD as a table.
people.registerTempTable("people")

// SQL statements can be run by using the sql methods provided by sqlContext.
val teenagers = sqlContext.sql("SELECT name FROM people WHERE age >= 13 AND age <= 19")

// Alternatively, a SchemaRDD can be created for a JSON dataset represented by
// an RDD[String] storing one JSON object per string.
val anotherPeopleRDD = sc.parallelize(
  """{"name":"Yin","address":{"city":"Columbus","state":"Ohio"}}""" :: Nil)
val anotherPeople = sqlContext.jsonRDD(anotherPeopleRDD)

Hive Tables

Spark SQL also supports reading and writing data stored in Apache Hive. However, since Hive has a large number of dependencies, it is not included in the default Spark assembly. In order to use Hive you must first run “sbt/sbt -Phive assembly/assembly” (or use -Phive for maven). This command builds a new assembly jar that includes Hive. Note that this Hive assembly jar must also be present on all of the worker nodes, as they will need access to the Hive serialization and deserialization libraries (SerDes) in order to access data stored in Hive.

Configuration of Hive is done by placing your hive-site.xml file in conf/.

• Scala
• Java
• Python

// sc is an existing SparkContext.
val sqlContext = new org.apache.spark.sql.hive.HiveContext(sc)

sqlContext.sql("CREATE TABLE IF NOT EXISTS src (key INT, value STRING)")
sqlContext.sql("LOAD DATA LOCAL INPATH ‘examples/src/main/resources/kv1.txt‘ INTO TABLE src")

// Queries are expressed in HiveQL
sqlContext.sql("FROM src SELECT key, value").collect().foreach(println)

Performance Tuning

For some workloads it is possible to improve performance by either caching data in memory, or by turning on some experimental options.

Caching Data In Memory

Spark SQL can cache tables using an in-memory columnar format by calling cacheTable("tableName"). Then Spark SQL will scan only required columns and will automatically tune compression to minimize memory usage and GC pressure. You can call uncacheTable("tableName") to remove the table from memory.

Note that if you call cache rather than cacheTable, tables will not be cached using the in-memory columnar format, and therefore cacheTable is strongly recommended for this use case.

Configuration of in-memory caching can be done using the setConf method on SQLContext or by running SET key=value commands using SQL.

Other Configuration Options

The following options can also be used to tune the performance of query execution. It is possible that these options will be deprecated in future release as more optimizations are performed automatically.

Other SQL Interfaces

Spark SQL also supports interfaces for running SQL queries directly without the need to write any code.

Running the Thrift JDBC server

The Thrift JDBC server implemented here corresponds to the HiveServer2 in Hive 0.12. You can test the JDBC server with the beeline script that comes with either Spark or Hive 0.12.

To start the JDBC server, run the following in the Spark directory:

./sbin/start-thriftserver.sh

The default port the server listens on is 10000. To listen on customized host and port, please set the HIVE_SERVER2_THRIFT_PORT andHIVE_SERVER2_THRIFT_BIND_HOST environment variables. You may run ./sbin/start-thriftserver.sh --help for a complete list of all available options. Now you can use beeline to test the Thrift JDBC server:

./bin/beeline

Connect to the JDBC server in beeline with:

beeline> !connect jdbc:hive2://localhost:10000

Beeline will ask you for a username and password. In non-secure mode, simply enter the username on your machine and a blank password. For secure mode, please follow the instructions given in the beeline documentation.

Configuration of Hive is done by placing your hive-site.xml file in conf/.

You may also use the beeline script that comes with Hive.

Running the Spark SQL CLI

The Spark SQL CLI is a convenient tool to run the Hive metastore service in local mode and execute queries input from the command line. Note that the Spark SQL CLI cannot talk to the Thrift JDBC server.

To start the Spark SQL CLI, run the following in the Spark directory:

./bin/spark-sql

Configuration of Hive is done by placing your hive-site.xml file in conf/. You may run ./bin/spark-sql --help for a complete list of all available options.

Compatibility with Other Systems

Migration Guide for Shark User

Scheduling

s To set a Fair Scheduler pool for a JDBC client session, users can set the spark.sql.thriftserver.scheduler.pool variable:

SET spark.sql.thriftserver.scheduler.pool=accounting;

Reducer number

In Shark, default reducer number is 1 and is controlled by the property mapred.reduce.tasks. Spark SQL deprecates this property in favor ofspark.sql.shuffle.partitions, whose default value is 200. Users may customize this property via SET:

SET spark.sql.shuffle.partitions=10;
SELECT page, count(*) c 
FROM logs_last_month_cached
GROUP BY page ORDER BY c DESC LIMIT 10;

You may also put this property in hive-site.xml to override the default value.

For now, the mapred.reduce.tasks property is still recognized, and is converted to spark.sql.shuffle.partitions automatically.

Caching

The shark.cache table property no longer exists, and tables whose name end with _cached are no longer automatically cached. Instead, we provide CACHE TABLE and UNCACHE TABLE statements to let user control table caching explicitly:

CACHE TABLE logs_last_month;
UNCACHE TABLE logs_last_month;

NOTE: CACHE TABLE tbl is lazy, similar to .cache on an RDD. This command only marks tbl to ensure that partitions are cached when calculated but doesn’t actually cache it until a query that touches tbl is executed. To force the table to be cached, you may simply count the table immediately after executing CACHE TABLE:

CACHE TABLE logs_last_month;
SELECT COUNT(1) FROM logs_last_month;

Several caching related features are not supported yet:

• User defined partition level cache eviction policy
• RDD reloading
• In-memory cache write through policy

Compatibility with Apache Hive

Spark SQL is designed to be compatible with the Hive Metastore, SerDes and UDFs. Currently Spark SQL is based on Hive 0.12.0.

Deploying in Existing Hive Warehouses

The Spark SQL Thrift JDBC server is designed to be “out of the box” compatible with existing Hive installations. You do not need to modify your existing Hive Metastore or change the data placement or partitioning of your tables.

Supported Hive Features

Spark SQL supports the vast majority of Hive features, such as:

• Hive query statements, including:
  • SELECT
  • GROUP BY
  • ORDER BY
  • CLUSTER BY
  • SORT BY
• All Hive operators, including:
  • Relational operators (=, ?, ==, <>, <, >, >=, <=, etc)
  • Arithmetic operators (+, -, *, /, %, etc)
  • Logical operators (AND, &&, OR, ||, etc)
  • Complex type constructors
  • Mathematical functions (sign, ln, cos, etc)
  • String functions (instr, length, printf, etc)
• User defined functions (UDF)
• User defined aggregation functions (UDAF)
• User defined serialization formats (SerDes)
• Joins
  • JOIN
  • {LEFT|RIGHT|FULL} OUTER JOIN
  • LEFT SEMI JOIN
  • CROSS JOIN
• Unions
• Sub-queries
  • SELECT col FROM ( SELECT a + b AS col from t1) t2
• Sampling
• Explain
• Partitioned tables
• All Hive DDL Functions, including:
  • CREATE TABLE
  • CREATE TABLE AS SELECT
  • ALTER TABLE
• Most Hive Data types, including:
  • TINYINT
  • SMALLINT
  • INT
  • BIGINT
  • BOOLEAN
  • FLOAT
  • DOUBLE
  • STRING
  • BINARY
  • TIMESTAMP
  • ARRAY<>
  • MAP<>
  • STRUCT<>

Unsupported Hive Functionality

Below is a list of Hive features that we don’t support yet. Most of these features are rarely used in Hive deployments.

Major Hive Features

• Spark SQL does not currently support inserting to tables using dynamic partitioning.
• Tables with buckets: bucket is the hash partitioning within a Hive table partition. Spark SQL doesn’t support buckets yet.

Esoteric Hive Features

• Tables with partitions using different input formats: In Spark SQL, all table partitions need to have the same input format.
• Non-equi outer join: For the uncommon use case of using outer joins with non-equi join conditions (e.g. condition “key < 10”), Spark SQL will output wrong result for the NULL tuple.
• UNION type and DATE type
• Unique join
• Single query multi insert
• Column statistics collecting: Spark SQL does not piggyback scans to collect column statistics at the moment and only supports populating the sizeInBytes field of the hive metastore.

Hive Input/Output Formats

• File format for CLI: For results showing back to the CLI, Spark SQL only supports TextOutputFormat.
• Hadoop archive

Hive Optimizations

A handful of Hive optimizations are not yet included in Spark. Some of these (such as indexes) are less important due to Spark SQL’s in-memory computational model. Others are slotted for future releases of Spark SQL.

• Block level bitmap indexes and virtual columns (used to build indexes)
• Automatically convert a join to map join: For joining a large table with multiple small tables, Hive automatically converts the join into a map join. We are adding this auto conversion in the next release.
• Automatically determine the number of reducers for joins and groupbys: Currently in Spark SQL, you need to control the degree of parallelism post-shuffle using “SET spark.sql.shuffle.partitions=[num_tasks];”.
• Meta-data only query: For queries that can be answered by using only meta data, Spark SQL still launches tasks to compute the result.
• Skew data flag: Spark SQL does not follow the skew data flags in Hive.
• STREAMTABLE hint in join: Spark SQL does not follow the STREAMTABLE hint.
• Merge multiple small files for query results: if the result output contains multiple small files, Hive can optionally merge the small files into fewer large files to avoid overflowing the HDFS metadata. Spark SQL does not support that.

Writing Language-Integrated Relational Queries

Language-Integrated queries are experimental and currently only supported in Scala.

Spark SQL also supports a domain specific language for writing queries. Once again, using the data from the above examples:

// sc is an existing SparkContext.
val sqlContext = new org.apache.spark.sql.SQLContext(sc)
// Importing the SQL context gives access to all the public SQL functions and implicit conversions.
import sqlContext._
val people: RDD[Person] = ... // An RDD of case class objects, from the first example.

// The following is the same as ‘SELECT name FROM people WHERE age >= 10 AND age <= 19‘
val teenagers = people.where(‘age >= 10).where(‘age <= 19).select(‘name)
teenagers.map(t => "Name: " + t(0)).collect().foreach(println)

The DSL uses Scala symbols to represent columns in the underlying table, which are identifiers prefixed with a tick (‘). Implicit conversions turn these symbols into expressions that are evaluated by the SQL execution engine. A full list of the functions supported can be found in theScalaDoc.

Spark SQL DataType Reference

• Numeric types
  • ByteType: Represents 1-byte signed integer numbers. The range of numbers is from -128 to 127.
  • ShortType: Represents 2-byte signed integer numbers. The range of numbers is from -32768 to 32767.
  • IntegerType: Represents 4-byte signed integer numbers. The range of numbers is from -2147483648 to 2147483647.
  • LongType: Represents 8-byte signed integer numbers. The range of numbers is from -9223372036854775808 to 9223372036854775807.
  • FloatType: Represents 4-byte single-precision floating point numbers.
  • DoubleType: Represents 8-byte double-precision floating point numbers.
  • DecimalType:
• String type
  • StringType: Represents character string values.
• Binary type
  • BinaryType: Represents byte sequence values.
• Boolean type
  • BooleanType: Represents boolean values.
• Datetime type
  • TimestampType: Represents values comprising values of fields year, month, day, hour, minute, and second.
• Complex types
  • ArrayType(elementType, containsNull): Represents values comprising a sequence of elements with the type of elementType.containsNull is used to indicate if elements in a ArrayType value can have null values.
  • MapType(keyType, valueType, valueContainsNull): Represents values comprising a set of key-value pairs. The data type of keys are described by keyType and the data type of values are described by valueType. For a MapType value, keys are not allowed to have nullvalues. valueContainsNull is used to indicate if values of a MapType value can have null values.
  • StructType(fields): Represents values with the structure described by a sequence of StructFields (fields).
    • StructField(name, dataType, nullable): Represents a field in a StructType. The name of a field is indicated by name. The data type of a field is indicated by dataType. nullable is used to indicate if values of this fields can have null values.

• Scala
• Java
• Python

import  org.apache.spark.sql._

Spark1.1.0 Spark SQL Programming Guide