Avro file

Apache Avro is a data serialization system. Avro provides:

• Rich data structures.
• A compact, fast, binary data format.
• A container file, to store persistent data.
• Remote procedure call (RPC).
• Simple integration with dynamic languages. Code generation is not required to read or write data files nor to use or implement RPC protocols. Code generation as an optional optimization, only worth implementing for statically typed languages.

The Avro data source supports:

• Schema conversion: Automatic conversion between Apache Spark SQL and Avro records.
• Partitioning: Easily reading and writing partitioned data without any extra configuration.
• Compression: Compression to use when writing Avro out to disk. The supported types are uncompressed, snappy, and deflate. You can also specify the deflate level.
• Record names: Record name and namespace by passing a map of parameters with recordName and recordNamespace.

Also see Read and write streaming Avro data.

Configuration

You can change the behavior of an Avro data source using various configuration parameters.

To ignore files without the .avro extension when reading, you can set the parameter avro.mapred.ignore.inputs.without.extension in the Hadoop configuration. The default is false.

spark
  .sparkContext
  .hadoopConfiguration
  .set("avro.mapred.ignore.inputs.without.extension", "true")

To configure compression when writing, set the following Spark properties:

• Compression codec: spark.sql.avro.compression.codec. Supported codecs are snappy and deflate. The default codec is snappy.
• If the compression codec is deflate, you can set the compression level with: spark.sql.avro.deflate.level. The default level is -1.

You can set these properties in the cluster Spark configuration or at runtime using spark.conf.set(). For example:

spark.conf.set("spark.sql.avro.compression.codec", "deflate")
spark.conf.set("spark.sql.avro.deflate.level", "5")

For Databricks Runtime 9.1 LTS and above, you can change the default schema inference behavior in Avro by providing the mergeSchema option when reading files. Setting mergeSchema to true will infer a schema from a set of Avro files in the target directory and merge them rather than infer the read schema from a single file.

Supported types for Avro -> Spark SQL conversion

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

Avro type	Spark SQL type
boolean	BooleanType
int	IntegerType
long	LongType
float	FloatType
double	DoubleType
bytes	BinaryType
string	StringType
record	StructType
enum	StringType
array	ArrayType
map	MapType
fixed	BinaryType
union	See Union types.

Union types

The Avro data source supports reading union types. Avro considers the following three types to be union types:

• union(int, long) maps to LongType.
• union(float, double) maps to DoubleType.
• union(something, null), where something is any supported Avro type. This maps to the same Spark SQL type as that of something, with nullable set to true.

All other union types are complex types. They map to StructType where field names are member0, member1, and so on, in accordance with members of the union. This is consistent with the behavior when converting between Avro and Parquet.

Logical types

The Avro data source supports reading the following Avro logical types:

Avro logical type	Avro type	Spark SQL type
date	int	DateType
timestamp-millis	long	TimestampType
timestamp-micros	long	TimestampType
decimal	fixed	DecimalType
decimal	bytes	DecimalType

Note

The Avro data source ignores docs, aliases, and other properties present in the Avro file.

Supported types for Spark SQL -> Avro conversion

This library supports writing of all Spark SQL types into Avro. For most types, the mapping from Spark types to Avro types is straightforward (for example IntegerType gets converted to int); the following is a list of the few special cases:

Spark SQL type	Avro type	Avro logical type
ByteType	int
ShortType	int
BinaryType	bytes
DecimalType	fixed	decimal
TimestampType	long	timestamp-micros
DateType	int	date

You can also specify the whole output Avro schema with the option avroSchema, so that Spark SQL types can be converted into other Avro types. The following conversions are not applied by default and require user specified Avro schema:

Spark SQL type	Avro type	Avro logical type
ByteType	fixed
StringType	enum
DecimalType	bytes	decimal
TimestampType	long	timestamp-millis

Examples

These examples use the episodes.avro file.

Scala

// The Avro records are converted to Spark types, filtered, and
// then written back out as Avro records

val df = spark.read.format("avro").load("/tmp/episodes.avro")
df.filter("doctor > 5").write.format("avro").save("/tmp/output")

This example demonstrates a custom Avro schema:

import org.apache.avro.Schema

val schema = new Schema.Parser().parse(new File("episode.avsc"))

spark
  .read
  .format("avro")
  .option("avroSchema", schema.toString)
  .load("/tmp/episodes.avro")
  .show()

This example demonstrates Avro compression options:

// configuration to use deflate compression
spark.conf.set("spark.sql.avro.compression.codec", "deflate")
spark.conf.set("spark.sql.avro.deflate.level", "5")

val df = spark.read.format("avro").load("/tmp/episodes.avro")

// writes out compressed Avro records
df.write.format("avro").save("/tmp/output")

This example demonstrates partitioned Avro records:

import org.apache.spark.sql.SparkSession

val spark = SparkSession.builder().master("local").getOrCreate()

val df = spark.createDataFrame(
  Seq(
    (2012, 8, "Batman", 9.8),
    (2012, 8, "Hero", 8.7),
    (2012, 7, "Robot", 5.5),
    (2011, 7, "Git", 2.0))
  ).toDF("year", "month", "title", "rating")

df.toDF.write.format("avro").partitionBy("year", "month").save("/tmp/output")

This example demonstrates the record name and namespace:

val df = spark.read.format("avro").load("/tmp/episodes.avro")

val name = "AvroTest"
val namespace = "org.foo"
val parameters = Map("recordName" -> name, "recordNamespace" -> namespace)

df.write.options(parameters).format("avro").save("/tmp/output")

Python

# Create a DataFrame from a specified directory
df = spark.read.format("avro").load("/tmp/episodes.avro")

#  Saves the subset of the Avro records read in
subset = df.where("doctor > 5")
subset.write.format("avro").save("/tmp/output")

SQL

To query Avro data in SQL, register the data file as a table or temporary view:

CREATE TEMPORARY VIEW episodes
USING avro
OPTIONS (path "/tmp/episodes.avro")

SELECT * from episodes

Notebook example: Read and write Avro files

The following notebook demonstrates how to read and write Avro files.

Read and write Avro files notebook

Get notebook