sql-programming-guide.md

layout: global
displayTitle: Spark SQL, DataFrames and Datasets Guide
title: Spark SQL and DataFrames

• This will become a table of contents (this text will be scraped). {:toc}

Overview

Spark SQL is a Spark module for structured data processing. Unlike the basic Spark RDD API, the interfaces provided by Spark SQL provide Spark with more information about the structure of both the data and the computation being performed. Internally, Spark SQL uses this extra information to perform extra optimizations. There are several ways to interact with Spark SQL including SQL and the Dataset API. When computing a result the same execution engine is used, independent of which API/language you are using to express the computation. This unification means that developers can easily switch back and forth between different APIs based on which provides the most natural way to express a given transformation.

All of the examples on this page use sample data included in the Spark distribution and can be run in the spark-shell, pyspark shell, or sparkR shell.

SQL

One use of Spark SQL is to execute SQL queries. Spark SQL can also be used to read data from an existing Hive installation. For more on how to configure this feature, please refer to the Hive Tables section. When running SQL from within another programming language the results will be returned as a Dataset/DataFrame. You can also interact with the SQL interface using the command-line or over JDBC/ODBC.

Datasets and DataFrames

A Dataset is a distributed collection of data. Dataset is a new interface added in Spark 1.6 that provides the benefits of RDDs (strong typing, ability to use powerful lambda functions) with the benefits of Spark SQL's optimized execution engine. A Dataset can be constructed from JVM objects and then manipulated using functional transformations (map, flatMap, filter, etc.). The Dataset API is available in Scala and Java. Python does not have the support for the Dataset API. But due to Python's dynamic nature, many of the benefits of the Dataset API are already available (i.e. you can access the field of a row by name naturally row.columnName). The case for R is similar.

A DataFrame is a Dataset organized into named columns. It is conceptually equivalent to a table in a relational database or a data frame in R/Python, but with richer optimizations under the hood. DataFrames can be constructed from a wide array of sources such as: structured data files, tables in Hive, external databases, or existing RDDs. The DataFrame API is available in Scala, Java, Python, and R. In Scala and Java, a DataFrame is represented by a Dataset of Rows. In the Scala API, DataFrame is simply a type alias of Dataset[Row]. While, in Java API, users need to use Dataset<Row> to represent a DataFrame.

Throughout this document, we will often refer to Scala/Java Datasets of Rows as DataFrames.

Getting Started

Starting Point: SparkSession

The entry point into all functionality in Spark is the SparkSession class. To create a basic SparkSession, just use SparkSession.builder():

{% include_example init_session scala/org/apache/spark/examples/sql/SparkSQLExample.scala %}

The entry point into all functionality in Spark is the SparkSession class. To create a basic SparkSession, just use SparkSession.builder():

{% include_example init_session java/org/apache/spark/examples/sql/JavaSparkSQLExample.java %}

The entry point into all functionality in Spark is the SparkSession class. To create a basic SparkSession, just use SparkSession.builder:

{% include_example init_session python/sql/basic.py %}

The entry point into all functionality in Spark is the SparkSession class. To initialize a basic SparkSession, just call sparkR.session():

{% include_example init_session r/RSparkSQLExample.R %}

Note that when invoked for the first time, sparkR.session() initializes a global SparkSession singleton instance, and always returns a reference to this instance for successive invocations. In this way, users only need to initialize the SparkSession once, then SparkR functions like read.df will be able to access this global instance implicitly, and users don't need to pass the SparkSession instance around.

SparkSession in Spark 2.0 provides builtin support for Hive features including the ability to write queries using HiveQL, access to Hive UDFs, and the ability to read data from Hive tables. To use these features, you do not need to have an existing Hive setup.

Creating DataFrames

With a `SparkSession`, applications can create DataFrames from an existing RDD, from a Hive table, or from Spark data sources.

As an example, the following creates a DataFrame based on the content of a JSON file:

{% include_example create_df scala/org/apache/spark/examples/sql/SparkSQLExample.scala %}

With a `SparkSession`, applications can create DataFrames from an existing RDD, from a Hive table, or from Spark data sources.

As an example, the following creates a DataFrame based on the content of a JSON file:

{% include_example create_df java/org/apache/spark/examples/sql/JavaSparkSQLExample.java %}

With a `SparkSession`, applications can create DataFrames from an existing RDD, from a Hive table, or from Spark data sources.

As an example, the following creates a DataFrame based on the content of a JSON file:

{% include_example create_df python/sql/basic.py %}

With a `SparkSession`, applications can create DataFrames from a local R data.frame, from a Hive table, or from Spark data sources.

As an example, the following creates a DataFrame based on the content of a JSON file:

{% include_example create_df r/RSparkSQLExample.R %}

Untyped Dataset Operations (aka DataFrame Operations)

DataFrames provide a domain-specific language for structured data manipulation in Scala, Java, Python and R.

As mentioned above, in Spark 2.0, DataFrames are just Dataset of Rows in Scala and Java API. These operations are also referred as "untyped transformations" in contrast to "typed transformations" come with strongly typed Scala/Java Datasets.

Here we include some basic examples of structured data processing using Datasets:

{% include_example untyped_ops scala/org/apache/spark/examples/sql/SparkSQLExample.scala %}

For a complete list of the types of operations that can be performed on a Dataset refer to the API Documentation.

In addition to simple column references and expressions, Datasets also have a rich library of functions including string manipulation, date arithmetic, common math operations and more. The complete list is available in the DataFrame Function Reference.

{% include_example untyped_ops java/org/apache/spark/examples/sql/JavaSparkSQLExample.java %}

For a complete list of the types of operations that can be performed on a Dataset refer to the API Documentation.

In addition to simple column references and expressions, Datasets also have a rich library of functions including string manipulation, date arithmetic, common math operations and more. The complete list is available in the DataFrame Function Reference.

In Python it's possible to access a DataFrame's columns either by attribute (`df.age`) or by indexing (`df['age']`). While the former is convenient for interactive data exploration, users are highly encouraged to use the latter form, which is future proof and won't break with column names that are also attributes on the DataFrame class.

{% include_example untyped_ops python/sql/basic.py %} For a complete list of the types of operations that can be performed on a DataFrame refer to the API Documentation.

In addition to simple column references and expressions, DataFrames also have a rich library of functions including string manipulation, date arithmetic, common math operations and more. The complete list is available in the DataFrame Function Reference.

{% include_example untyped_ops r/RSparkSQLExample.R %}

For a complete list of the types of operations that can be performed on a DataFrame refer to the API Documentation.

In addition to simple column references and expressions, DataFrames also have a rich library of functions including string manipulation, date arithmetic, common math operations and more. The complete list is available in the DataFrame Function Reference.

Running SQL Queries Programmatically

The `sql` function on a `SparkSession` enables applications to run SQL queries programmatically and returns the result as a `DataFrame`.

{% include_example run_sql scala/org/apache/spark/examples/sql/SparkSQLExample.scala %}

The `sql` function on a `SparkSession` enables applications to run SQL queries programmatically and returns the result as a `Dataset`.

{% include_example run_sql java/org/apache/spark/examples/sql/JavaSparkSQLExample.java %}

The `sql` function on a `SparkSession` enables applications to run SQL queries programmatically and returns the result as a `DataFrame`.

{% include_example run_sql python/sql/basic.py %}

The `sql` function enables applications to run SQL queries programmatically and returns the result as a `SparkDataFrame`.

{% include_example run_sql r/RSparkSQLExample.R %}

Global Temporary View

Temporary views in Spark SQL are session-scoped and will disappear if the session that creates it terminates. If you want to have a temporary view that is shared among all sessions and keep alive until the Spark application terminates, you can create a global temporary view. Global temporary view is tied to a system preserved database global_temp, and we must use the qualified name to refer it, e.g. SELECT * FROM global_temp.view1.

{% include_example global_temp_view scala/org/apache/spark/examples/sql/SparkSQLExample.scala %}

{% include_example global_temp_view java/org/apache/spark/examples/sql/JavaSparkSQLExample.java %}

{% include_example global_temp_view python/sql/basic.py %}

{% highlight sql %}

CREATE GLOBAL TEMPORARY VIEW temp_view AS SELECT a + 1, b * 2 FROM tbl

SELECT * FROM global_temp.temp_view

{% endhighlight %}

Creating Datasets

Datasets are similar to RDDs, however, instead of using Java serialization or Kryo they use a specialized Encoder to serialize the objects for processing or transmitting over the network. While both encoders and standard serialization are responsible for turning an object into bytes, encoders are code generated dynamically and use a format that allows Spark to perform many operations like filtering, sorting and hashing without deserializing the bytes back into an object.

{% include_example create_ds scala/org/apache/spark/examples/sql/SparkSQLExample.scala %}

{% include_example create_ds java/org/apache/spark/examples/sql/JavaSparkSQLExample.java %}

Interoperating with RDDs

Spark SQL supports two different methods for converting existing RDDs into Datasets. 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 Datasets 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 Datasets when the columns and their types are not known until runtime.

Inferring the Schema Using Reflection

The Scala interface for Spark SQL supports automatically converting an RDD containing case classes to a DataFrame. The case class defines the schema of the table. The names of the arguments to the case class are read using reflection and become the names of the columns. Case classes can also be nested or contain complex types such as Seqs or Arrays. This RDD can be implicitly converted to a DataFrame and then be registered as a table. Tables can be used in subsequent SQL statements.

{% include_example schema_inferring scala/org/apache/spark/examples/sql/SparkSQLExample.scala %}

Spark SQL supports automatically converting an RDD of JavaBeans into a DataFrame. The BeanInfo, obtained using reflection, defines the schema of the table. Currently, Spark SQL does not support JavaBeans that contain Map field(s). Nested JavaBeans and List or Array fields are supported though. You can create a JavaBean by creating a class that implements Serializable and has getters and setters for all of its fields.

{% include_example schema_inferring java/org/apache/spark/examples/sql/JavaSparkSQLExample.java %}

Spark SQL can convert an RDD of Row objects to a DataFrame, inferring the datatypes. Rows are constructed by passing a list of key/value pairs as kwargs to the Row class. The keys of this list define the column names of the table, and the types are inferred by sampling the whole dataset, similar to the inference that is performed on JSON files.

{% include_example schema_inferring python/sql/basic.py %}

Programmatically Specifying the Schema

When case classes cannot be defined ahead of time (for example, the structure of records is encoded in a string, or a text dataset will be parsed and fields will be projected differently for different users), a DataFrame can be created programmatically with three steps.

1. Create an RDD of Rows from the original RDD;
2. Create the schema represented by a StructType matching the structure of Rows in the RDD created in Step 1.
3. Apply the schema to the RDD of Rows via createDataFrame method provided by SparkSession.

For example:

{% include_example programmatic_schema scala/org/apache/spark/examples/sql/SparkSQLExample.scala %}

When JavaBean classes cannot be defined ahead of time (for example, the structure of records is encoded in a string, or a text dataset will be parsed and fields will be projected differently for different users), a Dataset<Row> can be created programmatically with three steps.

1. Create an RDD of Rows from the original RDD;
2. Create the schema represented by a StructType matching the structure of Rows in the RDD created in Step 1.
3. Apply the schema to the RDD of Rows via createDataFrame method provided by SparkSession.

For example:

{% include_example programmatic_schema java/org/apache/spark/examples/sql/JavaSparkSQLExample.java %}

When a dictionary of kwargs cannot be defined ahead of time (for example, the structure of records is encoded in a string, or a text dataset will be parsed and fields will be projected differently for different users), a DataFrame can be created programmatically with three steps.

1. Create an RDD of tuples or lists from the original RDD;
2. Create the schema represented by a StructType matching the structure of tuples or lists in the RDD created in the step 1.
3. Apply the schema to the RDD via createDataFrame method provided by SparkSession.

For example:

{% include_example programmatic_schema python/sql/basic.py %}

Data Sources

Spark SQL supports operating on a variety of data sources through the DataFrame interface. A DataFrame can be operated on using relational transformations and can also be used to create a temporary view. Registering a DataFrame as a temporary view allows you to run SQL queries over its data. This section describes the general methods for loading and saving data using the Spark Data Sources and then goes into specific options that are available for the built-in data sources.

Generic Load/Save Functions

In the simplest form, the default data source (parquet unless otherwise configured by spark.sql.sources.default) will be used for all operations.

{% include_example generic_load_save_functions scala/org/apache/spark/examples/sql/SQLDataSourceExample.scala %}

{% include_example generic_load_save_functions java/org/apache/spark/examples/sql/JavaSQLDataSourceExample.java %}

{% include_example generic_load_save_functions python/sql/datasource.py %}

{% include_example generic_load_save_functions r/RSparkSQLExample.R %}

Manually Specifying Options

You can also manually specify the data source that will be used along with any extra options that you would like to pass to the data source. Data sources are specified by their fully qualified name (i.e., org.apache.spark.sql.parquet), but for built-in sources you can also use their short names (json, parquet, jdbc, orc, libsvm, csv, text). DataFrames loaded from any data source type can be converted into other types using this syntax.

{% include_example manual_load_options scala/org/apache/spark/examples/sql/SQLDataSourceExample.scala %}

{% include_example manual_load_options java/org/apache/spark/examples/sql/JavaSQLDataSourceExample.java %}

{% include_example manual_load_options python/sql/datasource.py %}

{% include_example manual_load_options r/RSparkSQLExample.R %}

Run SQL on files directly

Instead of using read API to load a file into DataFrame and query it, you can also query that file directly with SQL.

{% include_example direct_sql scala/org/apache/spark/examples/sql/SQLDataSourceExample.scala %}

{% include_example direct_sql java/org/apache/spark/examples/sql/JavaSQLDataSourceExample.java %}

{% include_example direct_sql python/sql/datasource.py %}

{% include_example direct_sql r/RSparkSQLExample.R %}

Save Modes

Save operations can optionally take a SaveMode, that specifies how to handle existing data if present. It is important to realize that these save modes do not utilize any locking and are not atomic. Additionally, when performing an Overwrite, the data will be deleted before writing out the new data.

Scala/Java	Any Language	Meaning
SaveMode.ErrorIfExists (default)	"error" (default)	When saving a DataFrame to a data source, if data already exists, an exception is expected to be thrown.
SaveMode.Append	"append"	When saving a DataFrame to a data source, if data/table already exists, contents of the DataFrame are expected to be appended to existing data.
SaveMode.Overwrite	"overwrite"	Overwrite mode means that when saving a DataFrame to a data source, if data/table already exists, existing data is expected to be overwritten by the contents of the DataFrame.
SaveMode.Ignore	"ignore"	Ignore mode means that when saving a DataFrame to a data source, if data already exists, the save operation is expected to not save the contents of the DataFrame and to not change the existing data. This is similar to a CREATE TABLE IF NOT EXISTS in SQL.

Saving to Persistent Tables

DataFrames can also be saved as persistent tables into Hive metastore using the saveAsTable command. Notice existing Hive deployment is not necessary to use this feature. Spark will create a default local Hive metastore (using Derby) for you. Unlike the createOrReplaceTempView command, saveAsTable will materialize the contents of the DataFrame and create a pointer to the data in the Hive metastore. Persistent tables will still exist even after your Spark program has restarted, as long as you maintain your connection to the same metastore. A DataFrame for a persistent table can be created by calling the table method on a SparkSession with the name of the table.

By default saveAsTable will create a "managed table", meaning that the location of the data will be controlled by the metastore. Managed tables will also have their data deleted automatically when a table is dropped.

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. When writing Parquet files, all columns are automatically converted to be nullable for compatibility reasons.

Loading Data Programmatically

Using the data from the above example:

{% include_example basic_parquet_example scala/org/apache/spark/examples/sql/SQLDataSourceExample.scala %}

{% include_example basic_parquet_example java/org/apache/spark/examples/sql/JavaSQLDataSourceExample.java %}

{% include_example basic_parquet_example python/sql/datasource.py %}

{% include_example basic_parquet_example r/RSparkSQLExample.R %}

{% highlight sql %}

CREATE TEMPORARY VIEW parquetTable USING org.apache.spark.sql.parquet OPTIONS ( path "examples/src/main/resources/people.parquet" )

SELECT * FROM parquetTable

{% endhighlight %}

Partition Discovery

Table partitioning is a common optimization approach used in systems like Hive. In a partitioned table, data are usually stored in different directories, with partitioning column values encoded in the path of each partition directory. The Parquet data source is now able to discover and infer partitioning information automatically. For example, we can store all our previously used population data into a partitioned table using the following directory structure, with two extra columns, gender and country as partitioning columns:

{% highlight text %}

path └── to └── table ├── gender=male │   ├── ... │   │ │   ├── country=US │   │   └── data.parquet │   ├── country=CN │   │   └── data.parquet │   └── ... └── gender=female    ├── ...    │    ├── country=US    │   └── data.parquet    ├── country=CN    │   └── data.parquet    └── ...

{% endhighlight %}

By passing path/to/table to either SparkSession.read.parquet or SparkSession.read.load, Spark SQL will automatically extract the partitioning information from the paths. Now the schema of the returned DataFrame becomes:

{% highlight text %}

root |-- name: string (nullable = true) |-- age: long (nullable = true) |-- gender: string (nullable = true) |-- country: string (nullable = true)

{% endhighlight %}

Notice that the data types of the partitioning columns are automatically inferred. Currently, numeric data types and string type are supported. Sometimes users may not want to automatically infer the data types of the partitioning columns. For these use cases, the automatic type inference can be configured by spark.sql.sources.partitionColumnTypeInference.enabled, which is default to true. When type inference is disabled, string type will be used for the partitioning columns.

Starting from Spark 1.6.0, partition discovery only finds partitions under the given paths by default. For the above example, if users pass path/to/table/gender=male to either SparkSession.read.parquet or SparkSession.read.load, gender will not be considered as a partitioning column. If users need to specify the base path that partition discovery should start with, they can set basePath in the data source options. For example, when path/to/table/gender=male is the path of the data and users set basePath to path/to/table/, gender will be a partitioning column.

Schema Merging

Like ProtocolBuffer, Avro, and Thrift, Parquet also supports schema evolution. Users can start with a simple schema, and gradually add more columns to the schema as needed. In this way, users may end up with multiple Parquet files with different but mutually compatible schemas. The Parquet data source is now able to automatically detect this case and merge schemas of all these files.

Since schema merging is a relatively expensive operation, and is not a necessity in most cases, we turned it off by default starting from 1.5.0. You may enable it by

1. setting data source option mergeSchema to true when reading Parquet files (as shown in the examples below), or
2. setting the global SQL option spark.sql.parquet.mergeSchema to true.

{% include_example schema_merging scala/org/apache/spark/examples/sql/SQLDataSourceExample.scala %}

{% include_example schema_merging java/org/apache/spark/examples/sql/JavaSQLDataSourceExample.java %}

{% include_example schema_merging python/sql/datasource.py %}

{% include_example schema_merging r/RSparkSQLExample.R %}

Hive metastore Parquet table conversion

When reading from and writing to Hive metastore Parquet tables, Spark SQL will try to use its own Parquet support instead of Hive SerDe for better performance. This behavior is controlled by the spark.sql.hive.convertMetastoreParquet configuration, and is turned on by default.

Hive/Parquet Schema Reconciliation

There are two key differences between Hive and Parquet from the perspective of table schema processing.

1. Hive is case insensitive, while Parquet is not
2. Hive considers all columns nullable, while nullability in Parquet is significant

Due to this reason, we must reconcile Hive metastore schema with Parquet schema when converting a Hive metastore Parquet table to a Spark SQL Parquet table. The reconciliation rules are:

1. Fields that have the same name in both schema must have the same data type regardless of nullability. The reconciled field should have the data type of the Parquet side, so that nullability is respected.

2. The reconciled schema contains exactly those fields defined in Hive metastore schema.

   • Any fields that only appear in the Parquet schema are dropped in the reconciled schema.
   • Any fields that only appear in the Hive metastore schema are added as nullable field in the reconciled schema.

Metadata Refreshing

Spark SQL caches Parquet metadata for better performance. When Hive metastore Parquet table conversion is enabled, metadata of those converted tables are also cached. If these tables are updated by Hive or other external tools, you need to refresh them manually to ensure consistent metadata.

{% highlight scala %} // spark is an existing SparkSession spark.catalog.refreshTable("my_table") {% endhighlight %}

{% highlight java %} // spark is an existing SparkSession spark.catalog().refreshTable("my_table"); {% endhighlight %}

{% highlight python %}

spark is an existing SparkSession

spark.catalog.refreshTable("my_table") {% endhighlight %}

{% highlight sql %} REFRESH TABLE my_table; {% endhighlight %}

Configuration

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

Property Name	Default	Meaning
spark.sql.parquet.binaryAsString	false	Some other Parquet-producing systems, in particular Impala, Hive, and older versions of Spark SQL, do not differentiate between binary data and strings when writing out the Parquet schema. This flag tells Spark SQL to interpret binary data as a string to provide compatibility with these systems.
spark.sql.parquet.int96AsTimestamp	true	Some Parquet-producing systems, in particular Impala and Hive, store Timestamp into INT96. This flag tells Spark SQL to interpret INT96 data as a timestamp to provide compatibility with these systems.
spark.sql.parquet.cacheMetadata	true	Turns on caching of Parquet schema metadata. Can speed up querying of static data.
spark.sql.parquet.compression.codec	snappy	Sets the compression codec use when writing Parquet files. Acceptable values include: uncompressed, snappy, gzip, lzo.
spark.sql.parquet.filterPushdown	true	Enables Parquet filter push-down optimization when set to true.
spark.sql.hive.convertMetastoreParquet	true	When set to false, Spark SQL will use the Hive SerDe for parquet tables instead of the built in support.
spark.sql.parquet.mergeSchema	false	When true, the Parquet data source merges schemas collected from all data files, otherwise the schema is picked from the summary file or a random data file if no summary file is available.
spark.sql.optimizer.metadataOnly	true	When true, enable the metadata-only query optimization that use the table's metadata to produce the partition columns instead of table scans. It applies when all the columns scanned are partition columns and the query has an aggregate operator that satisfies distinct semantics.

JSON Datasets

Spark SQL can automatically infer the schema of a JSON dataset and load it as a `Dataset[Row]`. This conversion can be done using `SparkSession.read.json()` on either an RDD of String, or a JSON file.

Note that the file that is offered as a json file is not a typical JSON file. Each line must contain a separate, self-contained valid JSON object. For more information, please see JSON Lines text format, also called newline-delimited JSON. As a consequence, a regular multi-line JSON file will most often fail.

{% include_example json_dataset scala/org/apache/spark/examples/sql/SQLDataSourceExample.scala %}

Spark SQL can automatically infer the schema of a JSON dataset and load it as a `Dataset`. This conversion can be done using `SparkSession.read().json()` on either an RDD of String, or a JSON file.

Note that the file that is offered as a json file is not a typical JSON file. Each line must contain a separate, self-contained valid JSON object. For more information, please see JSON Lines text format, also called newline-delimited JSON. As a consequence, a regular multi-line JSON file will most often fail.

{% include_example json_dataset java/org/apache/spark/examples/sql/JavaSQLDataSourceExample.java %}

Spark SQL can automatically infer the schema of a JSON dataset and load it as a DataFrame. This conversion can be done using `SparkSession.read.json` on a JSON file.

Note that the file that is offered as a json file is not a typical JSON file. Each line must contain a separate, self-contained valid JSON object. For more information, please see JSON Lines text format, also called newline-delimited JSON. As a consequence, a regular multi-line JSON file will most often fail.

{% include_example json_dataset python/sql/datasource.py %}

Spark SQL can automatically infer the schema of a JSON dataset and load it as a DataFrame. using the `read.json()` function, which loads data from a directory of JSON files where each line of the files is a JSON object.

Note that the file that is offered as a json file is not a typical JSON file. Each line must contain a separate, self-contained valid JSON object. For more information, please see JSON Lines text format, also called newline-delimited JSON. As a consequence, a regular multi-line JSON file will most often fail.

{% include_example json_dataset r/RSparkSQLExample.R %}

{% highlight sql %}

CREATE TEMPORARY VIEW jsonTable USING org.apache.spark.sql.json OPTIONS ( path "examples/src/main/resources/people.json" )

SELECT * FROM jsonTable

{% endhighlight %}

Hive Tables

Spark SQL also supports reading and writing data stored in Apache Hive. However, since Hive has a large number of dependencies, these dependencies are not included in the default Spark distribution. If Hive dependencies can be found on the classpath, Spark will load them automatically. Note that these Hive dependencies 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, core-site.xml (for security configuration), and hdfs-site.xml (for HDFS configuration) file in conf/.

When working with Hive, one must instantiate SparkSession with Hive support, including connectivity to a persistent Hive metastore, support for Hive serdes, and Hive user-defined functions. Users who do not have an existing Hive deployment can still enable Hive support. When not configured by the hive-site.xml, the context automatically creates metastore_db in the current directory and creates a directory configured by spark.sql.warehouse.dir, which defaults to the directory spark-warehouse in the current directory that the Spark application is started. Note that the hive.metastore.warehouse.dir property in hive-site.xml is deprecated since Spark 2.0.0. Instead, use spark.sql.warehouse.dir to specify the default location of database in warehouse. You may need to grant write privilege to the user who starts the Spark application.

{% include_example spark_hive scala/org/apache/spark/examples/sql/hive/SparkHiveExample.scala %}

{% include_example spark_hive java/org/apache/spark/examples/sql/hive/JavaSparkHiveExample.java %}

{% include_example spark_hive python/sql/hive.py %}

When working with Hive one must instantiate SparkSession with Hive support. This adds support for finding tables in the MetaStore and writing queries using HiveQL.

{% include_example spark_hive r/RSparkSQLExample.R %}

Interacting with Different Versions of Hive Metastore

One of the most important pieces of Spark SQL's Hive support is interaction with Hive metastore, which enables Spark SQL to access metadata of Hive tables. Starting from Spark 1.4.0, a single binary build of Spark SQL can be used to query different versions of Hive metastores, using the configuration described below. Note that independent of the version of Hive that is being used to talk to the metastore, internally Spark SQL will compile against Hive 1.2.1 and use those classes for internal execution (serdes, UDFs, UDAFs, etc).

The following options can be used to configure the version of Hive that is used to retrieve metadata:

Property Name	Default	Meaning
spark.sql.hive.metastore.version	1.2.1	Version of the Hive metastore. Available options are 0.12.0 through 1.2.1.
spark.sql.hive.metastore.jars	builtin	Location of the jars that should be used to instantiate the HiveMetastoreClient. This property can be one of three options: builtin Use Hive 1.2.1, which is bundled with the Spark assembly when -Phive is enabled. When this option is chosen, spark.sql.hive.metastore.version must be either 1.2.1 or not defined. maven Use Hive jars of specified version downloaded from Maven repositories. This configuration is not generally recommended for production deployments. A classpath in the standard format for the JVM. This classpath must include all of Hive and its dependencies, including the correct version of Hadoop. These jars only need to be present on the driver, but if you are running in yarn cluster mode then you must ensure they are packaged with your application.
spark.sql.hive.metastore.sharedPrefixes	com.mysql.jdbc, org.postgresql, com.microsoft.sqlserver, oracle.jdbc	A comma separated list of class prefixes that should be loaded using the classloader that is shared between Spark SQL and a specific version of Hive. An example of classes that should be shared is JDBC drivers that are needed to talk to the metastore. Other classes that need to be shared are those that interact with classes that are already shared. For example, custom appenders that are used by log4j.
spark.sql.hive.metastore.barrierPrefixes	(empty)	A comma separated list of class prefixes that should explicitly be reloaded for each version of Hive that Spark SQL is communicating with. For example, Hive UDFs that are declared in a prefix that typically would be shared (i.e. org.apache.spark.*).

JDBC To Other Databases

Spark SQL also includes a data source that can read data from other databases using JDBC. This functionality should be preferred over using JdbcRDD. This is because the results are returned as a DataFrame and they can easily be processed in Spark SQL or joined with other data sources. The JDBC data source is also easier to use from Java or Python as it does not require the user to provide a ClassTag. (Note that this is different than the Spark SQL JDBC server, which allows other applications to run queries using Spark SQL).

To get started you will need to include the JDBC driver for you particular database on the spark classpath. For example, to connect to postgres from the Spark Shell you would run the following command:

{% highlight bash %} bin/spark-shell --driver-class-path postgresql-9.4.1207.jar --jars postgresql-9.4.1207.jar {% endhighlight %}

Tables from the remote database can be loaded as a DataFrame or Spark SQL temporary view using the Data Sources API. Users can specify the JDBC connection properties in the data source options. user and password are normally provided as connection properties for logging into the data sources. In addition to the connection properties, Spark also supports the following case-sensitive options:

Property Name	Meaning
url	The JDBC URL to connect to. The source-specific connection properties may be specified in the URL. e.g., jdbc:postgresql://localhost/test?user=fred&password=secret
dbtable	The JDBC table that should be read. Note that anything that is valid in a FROM clause of a SQL query can be used. For example, instead of a full table you could also use a subquery in parentheses.
driver	The class name of the JDBC driver to use to connect to this URL.
partitionColumn, lowerBound, upperBound	These options must all be specified if any of them is specified. In addition, numPartitions must be specified. They describe how to partition the table when reading in parallel from multiple workers. partitionColumn must be a numeric column from the table in question. Notice that lowerBound and upperBound are just used to decide the partition stride, not for filtering the rows in table. So all rows in the table will be partitioned and returned. This option applies only to reading.
numPartitions	The maximum number of partitions that can be used for parallelism in table reading and writing. This also determines the maximum number of concurrent JDBC connections. If the number of partitions to write exceeds this limit, we decrease it to this limit by calling coalesce(numPartitions) before writing.
fetchsize	The JDBC fetch size, which determines how many rows to fetch per round trip. This can help performance on JDBC drivers which default to low fetch size (eg. Oracle with 10 rows). This option applies only to reading.
batchsize	The JDBC batch size, which determines how many rows to insert per round trip. This can help performance on JDBC drivers. This option applies only to writing. It defaults to 1000.
isolationLevel	The transaction isolation level, which applies to current connection. It can be one of NONE, READ_COMMITTED, READ_UNCOMMITTED, REPEATABLE_READ, or SERIALIZABLE, corresponding to standard transaction isolation levels defined by JDBC's Connection object, with default of READ_UNCOMMITTED. This option applies only to writing. Please refer the documentation in java.sql.Connection.
truncate	This is a JDBC writer related option. When SaveMode.Overwrite is enabled, this option causes Spark to truncate an existing table instead of dropping and recreating it. This can be more efficient, and prevents the table metadata (e.g., indices) from being removed. However, it will not work in some cases, such as when the new data has a different schema. It defaults to false. This option applies only to writing.
createTableOptions	This is a JDBC writer related option. If specified, this option allows setting of database-specific table and partition options when creating a table (e.g., CREATE TABLE t (name string) ENGINE=InnoDB.). This option applies only to writing.

{% include_example jdbc_dataset scala/org/apache/spark/examples/sql/SQLDataSourceExample.scala %}

{% include_example jdbc_dataset java/org/apache/spark/examples/sql/JavaSQLDataSourceExample.java %}

{% include_example jdbc_dataset python/sql/datasource.py %}

{% include_example jdbc_dataset r/RSparkSQLExample.R %}

{% highlight sql %}

CREATE TEMPORARY VIEW jdbcTable USING org.apache.spark.sql.jdbc OPTIONS ( url "jdbc:postgresql:dbserver", dbtable "schema.tablename", user 'username', password 'password' )

INSERT INTO TABLE jdbcTable SELECT * FROM resultTable {% endhighlight %}

Troubleshooting

• The JDBC driver class must be visible to the primordial class loader on the client session and on all executors. This is because Java's DriverManager class does a security check that results in it ignoring all drivers not visible to the primordial class loader when one goes to open a connection. One convenient way to do this is to modify compute_classpath.sh on all worker nodes to include your driver JARs.
• Some databases, such as H2, convert all names to upper case. You'll need to use upper case to refer to those names in Spark SQL.

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 spark.cacheTable("tableName") or dataFrame.cache(). Then Spark SQL will scan only required columns and will automatically tune compression to minimize memory usage and GC pressure. You can call spark.uncacheTable("tableName") to remove the table from memory.

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

Property Name	Default	Meaning
spark.sql.inMemoryColumnarStorage.compressed	true	When set to true Spark SQL will automatically select a compression codec for each column based on statistics of the data.
spark.sql.inMemoryColumnarStorage.batchSize	10000	Controls the size of batches for columnar caching. Larger batch sizes can improve memory utilization and compression, but risk OOMs when caching data.

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.

Property Name	Default	Meaning
spark.sql.files.maxPartitionBytes	134217728 (128 MB)	The maximum number of bytes to pack into a single partition when reading files.
spark.sql.files.openCostInBytes	4194304 (4 MB)	The estimated cost to open a file, measured by the number of bytes could be scanned in the same time. This is used when putting multiple files into a partition. It is better to over estimated, then the partitions with small files will be faster than partitions with bigger files (which is scheduled first).
spark.sql.broadcastTimeout	300	Timeout in seconds for the broadcast wait time in broadcast joins
spark.sql.autoBroadcastJoinThreshold	10485760 (10 MB)	Configures the maximum size in bytes for a table that will be broadcast to all worker nodes when performing a join. By setting this value to -1 broadcasting can be disabled. Note that currently statistics are only supported for Hive Metastore tables where the command ANALYZE TABLE <tableName> COMPUTE STATISTICS noscan has been run.
spark.sql.shuffle.partitions	200	Configures the number of partitions to use when shuffling data for joins or aggregations.

Distributed SQL Engine

Spark SQL can also act as a distributed query engine using its JDBC/ODBC or command-line interface. In this mode, end-users or applications can interact with Spark SQL directly to run SQL queries, without the need to write any code.

Running the Thrift JDBC/ODBC server

The Thrift JDBC/ODBC server implemented here corresponds to the HiveServer2 in Hive 1.2.1 You can test the JDBC server with the beeline script that comes with either Spark or Hive 1.2.1.

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

./sbin/start-thriftserver.sh

This script accepts all bin/spark-submit command line options, plus a --hiveconf option to specify Hive properties. You may run ./sbin/start-thriftserver.sh --help for a complete list of all available options. By default, the server listens on localhost:10000. You may override this behaviour via either environment variables, i.e.:

{% highlight bash %} export HIVE_SERVER2_THRIFT_PORT= export HIVE_SERVER2_THRIFT_BIND_HOST= ./sbin/start-thriftserver.sh
--master
... {% endhighlight %}

or system properties:

{% highlight bash %} ./sbin/start-thriftserver.sh
--hiveconf hive.server2.thrift.port=
--hiveconf hive.server2.thrift.bind.host=
--master ... {% endhighlight %}

Now you can use beeline to test the Thrift JDBC/ODBC server:

./bin/beeline

Connect to the JDBC/ODBC 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, core-site.xml and hdfs-site.xml files in conf/.

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

Thrift JDBC server also supports sending thrift RPC messages over HTTP transport. Use the following setting to enable HTTP mode as system property or in hive-site.xml file in conf/:

hive.server2.transport.mode - Set this to value: http
hive.server2.thrift.http.port - HTTP port number fo listen on; default is 10001
hive.server2.http.endpoint - HTTP endpoint; default is cliservice

To test, use beeline to connect to the JDBC/ODBC server in http mode with:

beeline> !connect jdbc:hive2://<host>:<port>/<database>?hive.server2.transport.mode=http;hive.server2.thrift.http.path=<http_endpoint>

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, core-site.xml and hdfs-site.xml files in conf/. You may run ./bin/spark-sql --help for a complete list of all available options.

Migration Guide

Upgrading From Spark SQL 2.0 to 2.1

• Datasource tables now store partition metadata in the Hive metastore. This means that Hive DDLs such as ALTER TABLE PARTITION ... SET LOCATION are now available for tables created with the Datasource API.
  • Legacy datasource tables can be migrated to this format via the MSCK REPAIR TABLE command. Migrating legacy tables is recommended to take advantage of Hive DDL support and improved planning performance.
  • To determine if a table has been migrated, look for the PartitionProvider: Catalog attribute when issuing DESCRIBE FORMATTED on the table.
• Changes to INSERT OVERWRITE TABLE ... PARTITION ... behavior for Datasource tables.
  • In prior Spark versions INSERT OVERWRITE overwrote the entire Datasource table, even when given a partition specification. Now only partitions matching the specification are overwritten.
  • Note that this still differs from the behavior of Hive tables, which is to overwrite only partitions overlapping with newly inserted data.

Upgrading From Spark SQL 1.6 to 2.0

• SparkSession is now the new entry point of Spark that replaces the old SQLContext and HiveContext. Note that the old SQLContext and HiveContext are kept for backward compatibility. A new catalog interface is accessible from SparkSession - existing API on databases and tables access such as listTables, createExternalTable, dropTempView, cacheTable are moved here.

• Dataset API and DataFrame API are unified. In Scala, DataFrame becomes a type alias for Dataset[Row], while Java API users must replace DataFrame with Dataset<Row>. Both the typed transformations (e.g., map, filter, and groupByKey) and untyped transformations (e.g., select and groupBy) are available on the Dataset class. Since compile-time type-safety in Python and R is not a language feature, the concept of Dataset does not apply to these languages’ APIs. Instead, DataFrame remains the primary programing abstraction, which is analogous to the single-node data frame notion in these languages.

• Dataset and DataFrame API unionAll has been deprecated and replaced by union

• Dataset and DataFrame API explode has been deprecated, alternatively, use functions.explode() with select or flatMap

• Dataset and DataFrame API registerTempTable has been deprecated and replaced by createOrReplaceTempView

Upgrading From Spark SQL 1.5 to 1.6

• From Spark 1.6, by default the Thrift server runs in multi-session mode. Which means each JDBC/ODBC connection owns a copy of their own SQL configuration and temporary function registry. Cached tables are still shared though. If you prefer to run the Thrift server in the old single-session mode, please set option spark.sql.hive.thriftServer.singleSession to true. You may either add this option to spark-defaults.conf, or pass it to start-thriftserver.sh via --conf:

  {% highlight bash %} ./sbin/start-thriftserver.sh
  --conf spark.sql.hive.thriftServer.singleSession=true
  ... {% endhighlight %}

• Since 1.6.1, withColumn method in sparkR supports adding a new column to or replacing existing columns of the same name of a DataFrame.

• From Spark 1.6, LongType casts to TimestampType expect seconds instead of microseconds. This change was made to match the behavior of Hive 1.2 for more consistent type casting to TimestampType from numeric types. See SPARK-11724 for details.