Introduction

Avro4k (or Avro for Kotlin) is a library that brings Avro serialization format in kotlin, based on the reflection-less kotlin library called kotlinx-serialization.

Here are the main features:

• Full avro support, including logical types, unions, recursive types, and schema evolution ✅
• Encode and decode anything to and from binary format, and also in generic data 🧰
• Generate schemas based on your values and data classes 📝
• Customize the generated schemas and encoded data with annotations 👷
• Fast as it is reflection-less 🚀 (check the benchmarks here)
• Simple API to get started quickly, also with native support of java standard classes like UUID, BigDecimal, BigInteger and java.time module 🥇
• Relaxed matching for easy schema evolution as it natively adapts compatible types 🌀

Warning

Important: As of today, avro4k is only available for JVM platform, and theoretically for android platform (as apache avro library is already android-ready).
If you would like to have js/wasm/native compatible platforms, please put a 👍 on this issue

Quick start

Basic encoding

Example:

package myapp

import com.github.avrokotlin.avro4k.*
import kotlinx.serialization.*

@Serializable
data class Project(val name: String, val language: String)

fun main() {
    // Generating schemas
    val schema = Avro.schema<Project>()
    println(schema.toString()) // {"type":"record","name":"Project","namespace":"myapp","fields":[{"name":"name","type":"string"},{"name":"language","type":"string"}]}

    // Serializing objects
    val data = Project("kotlinx.serialization", "Kotlin")
    val bytes = Avro.encodeToByteArray(data)

    // Deserializing objects
    val obj = Avro.decodeFromByteArray<Project>(bytes)
    println(obj) // Project(name=kotlinx.serialization, language=Kotlin)
}

Single object

Avro4k provides a way to encode and decode single objects with AvroSingleObject class. This encoding will prefix the binary data with the schema fingerprint to allow knowing the writer schema when reading the data. The downside is that you need to provide a schema registry to get the schema from the fingerprint. This format is perfect for payloads sent through message brokers like kafka or rabbitmq as it is the most compact schema-aware format.

Example:

package myapp

import com.github.avrokotlin.avro4k.*
import kotlinx.serialization.*
import org.apache.avro.SchemaNormalization

@Serializable
data class Project(val name: String, val language: String)

fun main() {
    val schema = Avro.schema<Project>()
    val schemasByFingerprint = mapOf(SchemaNormalization.parsingFingerprint64(schema), schema)
    val singleObjectInstance = AvroSingleObject { schemasByFingerprint[it] }

    // Serializing objects
    val data = Project("kotlinx.serialization", "Kotlin")
    val bytes = singleObjectInstance.encodeToByteArray(data)

    // Deserializing objects
    val obj = singleObjectInstance.decodeFromByteArray<Project>(bytes)
    println(obj) // Project(name=kotlinx.serialization, language=Kotlin)
}

For more details, check in the avro spec the single object encoding.

Object container

Avro4k provides a way to encode and decode object container — also known as data file — with AvroObjectContainer class. This encoding will prefix the binary data with the full schema to allow knowing the writer schema when reading the data. This format is perfect for storing many long-term objects in a single file.

Be aware that consuming the decoded Sequence needs to be done before closing the stream, or you will get an exception as a sequence is a "hot" source, which means that if there is millions of objects in the file, all the objects are extracted one-by-one when requested. If you take only the first 10 objects and close the stream, the remaining objects won't be extracted. Use carefully sequence.toList() as it could lead to OutOfMemoryError as extracting millions of objects may not fit in memory.

Example:

package myapp

import com.github.avrokotlin.avro4k.*
import kotlinx.serialization.*

@Serializable
data class Project(val name: String, val language: String)

fun main() {
    // Serializing objects
    val valuesToEncode = sequenceOf(
        Project("kotlinx.serialization", "Kotlin"),
        Project("java.lang", "Java"),
        Project("avro4k", "Kotlin"),
    )
    Files.newOutputStream(Path("your-file.bin")).use { fileStream ->
        AvroObjectContainer.openWriter(fileStream).use { writer ->
            valuesToEncode.forEach { writer.write(it) }
        }
    }

    // Deserializing objects
    Files.newInputStream(Path("your-file.bin")).use { fileStream ->
        AvroObjectContainer.decodeFromStream<Project>(valuesToEncode, fileStream).forEach {
            println(it) // Project(name=kotlinx.serialization, language=Kotlin) ...
        }
    }
}

For more details, check in the avro spec the single object encoding.

Important notes

• Avro4k is highly based on apache avro library, that implies all the schema validation is done by it
• All members annotated with @ExperimentalSerializationApi are subject to changes in future releases without any notice as they are experimental, so please check the release notes to check the needed migration. At least, given a version A.B.C, only the minor B number will be incremented, not the major A.
• Avro4k also supports encoding and decoding generic data, mainly because of confluent schema registry compatibility as their serializers only handle generic data. When avro4k will support their schema registry, the generic encoding will be removed to keep this library as simple as possible.

Setup

Gradle Kotlin DSL

plugins {
    kotlin("jvm") version kotlinVersion
    kotlin("plugin.serialization") version kotlinVersion
}

dependencies {
    implementation("com.github.avro-kotlin.avro4k:avro4k-core:$avro4kVersion")
}

Gradle Groovy DSL

plugins {
    id 'org.jetbrains.kotlin.multiplatform' version kotlinVersion
    id 'org.jetbrains.kotlin.plugin.serialization' version kotlinVersion
}

dependencies {
    implementation "com.github.avro-kotlin.avro4k:avro4k-core:$avro4kVersion"
}

Maven

Add serialization plugin to Kotlin compiler plugin:

<build>
    <plugins>
        <plugin>
            <groupId>org.jetbrains.kotlin</groupId>
            <artifactId>kotlin-maven-plugin</artifactId>
            <version>${kotlin.version}</version>
            <executions>
                <execution>
                    <id>compile</id>
                    <phase>compile</phase>
                    <goals>
                        <goal>compile</goal>
                    </goals>
                </execution>
            </executions>
            <configuration>
                <compilerPlugins>
                    <plugin>kotlinx-serialization</plugin>
                </compilerPlugins>
            </configuration>
            <dependencies>
                <dependency>
                    <groupId>org.jetbrains.kotlin</groupId>
                    <artifactId>kotlin-maven-serialization</artifactId>
                    <version>${kotlin.version}</version>
                </dependency>
            </dependencies>
        </plugin>
    </plugins>
</build>

Add the avro4k dependency:

<dependency>
    <groupId>com.github.avro-kotlin.avro4k</groupId>
    <artifactId>avro4k-core</artifactId>
    <version>${avro4k.version}</version>
</dependency>

Versions matrix

Avro4k	Kotlin	Kotlin API/language	Kotlin serialization
>= 2.0.0	>= 2.0	>= 1.9	>= 1.7
< 2.0.0	>= 1.6	>= 1.6	>= 1.3

How to generate schemas

Writing schemas manually or using the Java based SchemaBuilder can be tedious. kotlinx-serialization simplifies this generating for us the corresponding descriptors to allow generating avro schemas easily, without any reflection. Also, it provides native compatibility with data classes (including open and sealed classes), inline classes, any collection, array, enums, and primitive values.

Note

For more information about the avro schema, please refer to the avro specification

To allow generating a schema for a specific class, you need to annotate it with @Serializable:

@Serializable
data class Ingredient(val name: String, val sugar: Double)

@Serializable
data class Pizza(val name: String, val ingredients: List<Ingredient>, val topping: Ingredient?, val vegetarian: Boolean)

Then you can generate the schema using the Avro.schema function:

val schema = Avro.schema<Pizza>()
println(schema.toString(true))

The generated schema will look as follows:

{
    "type": "record",
    "name": "Pizza",
    "namespace": "com.github.avrokotlin.avro4k.example",
    "fields": [
        {
            "name": "name",
            "type": "string"
        },
        {
            "name": "ingredients",
            "type": {
                "type": "array",
                "items": {
                    "type": "record",
                    "name": "Ingredient",
                    "fields": [
                        {
                            "name": "name",
                            "type": "string"
                        },
                        {
                            "name": "sugar",
                            "type": "double"
                        }
                    ]
                }
            }
        },
        {
            "name": "topping",
            "type": [
                "null",
                {
                    "type": "record",
                    "name": "Ingredient"
                }
            ],
            "default": null
        },
        {
            "name": "vegetarian",
            "type": "boolean"
        }
    ]
}

If you need to configure your Avro instance, you need to create your own instance of Avro with the wanted configuration, and then use it to generate the schema:

val yourAvroInstance = Avro {
    // your configuration
}
yourAvroInstance.schema<Pizza>()

Usage

Customizing the configuration

By default, Avro is configured with the following behavior:

• implicitNulls: The nullable fields are considered null when decoding if the writer record's schema does not contain this field.
• implicitEmptyCollections: The non-nullable map and collection fields are considered empty when decoding if the writer record's schema does not contain this field.
  • If implicitNulls is true, it takes precedence so the empty collections are set as null if the value is missing instead of an empty collection.
• validateSerialization: There is no validation of the schema when encoding or decoding data, which means that serializing using a custom serializer could lead to unexpected behavior. Be careful with your custom serializers. More details in this section.
• fieldNamingStrategy: The record's field naming strategy is using the original kotlin field name. To change it, check this section.

So each time you call a method on the Avro object implicitely invoke the default configuration. Example:

Avro.encodeToByteArray(MyData("value"))
Avro.decodeFromByteArray(bytes)
Avro.schema<MyData>()

If you need to change the default behavior, you need to create your own instance of Avro with the wanted configuration:

val yourAvroInstance = Avro {
    fieldNamingStrategy = FieldNamingStrategy.Builtins.SnakeCase
    implicitNulls = false
    implicitEmptyCollections = false
    validateSerialization = true
}
yourAvroInstance.encodeToByteArray(MyData("value"))
yourAvroInstance.decodeFromByteArray(bytes)
yourAvroInstance.schema<MyData>()

Types matrix

Kotlin type	Generated schema type	Other compatible writer types	Compatible logical type	Note / Serializer class
Boolean	boolean	string
Byte, Short, Int	int	long, float, double, string
Long	long	int, float, double, string
Float	float	double, string
Double	double	float, string
Char	int	string (exactly 1 char required)	char	The value serialized is the char code. When reading from a string, requires exactly 1 char
String	string	bytes (UTF8), fixed (UTF8)
ByteArray	bytes	string (UTF8), fixed (UTF8)
Map<*, *>	map			The map key must be string-able. Mainly everything is string-able except null and composite types (collection, data classes)
Collection<*>	array
data class	record
enum class	enum	string
@AvroFixed-compatible	fixed	bytes, string		Throws an error at runtime if the writer type is not present in the column "other compatible writer types"
@AvroStringable-compatible	string	int, long, float, double, string, fixed, bytes		Ignored when the writer type is not present in the column "other compatible writer types"
java.math.BigDecimal	bytes	int, long, float, double, string, fixed, bytes	decimal	To use it, annotate the field with @AvroDecimal to give the scale and the precision
java.math.BigDecimal	string	int, long, float, double, fixed, bytes		To use it, annotate the field with @AvroStringable. @AvroDecimal is ignored in that case
java.util.UUID	string		uuid	To use it, just annotate the field with @Contextual
java.net.URL	string			To use it, just annotate the field with @Contextual
java.math.BigInteger	string	int, long, float, double		To use it, just annotate the field with @Contextual
java.time.LocalDate	int	long, string (ISO8601)	date	To use it, just annotate the field with @Contextual
java.time.Instant	long	string (ISO8601)	timestamp-millis	To use it, just annotate the field with @Contextual
java.time.Instant	long	string (ISO8601)	timestamp-micros	To use it, register the serializer com.github.avrokotlin.avro4k.serializer.InstantToMicroSerializer
java.time.LocalDateTime	long	string (ISO8601)	timestamp-millis	To use it, just annotate the field with @Contextual
java.time.LocalTime	int	long, string (ISO8601)	time-millis	To use it, just annotate the field with @Contextual
java.time.Duration	fixed of 12	string (ISO8601)	duration	To use it, just annotate the field with @Contextual
java.time.Period	fixed of 12	string (ISO8601)	duration	To use it, just annotate the field with @Contextual
kotlin.time.Duration	fixed of 12	string (ISO8601)	duration

Note

For more details, check the built-in classes in kotlinx-serialization

Add documentation to a schema

You may want to add documentation to a schema to provide more information about a field or a named type (only RECORD and ENUM for the moment).

Warning

Do not use @org.apache.avro.reflect.AvroDoc as this annotation is not visible by Avro4k.

import com.github.avrokotlin.avro4k.AvroDoc

@Serializable
@AvroDoc("This is a record documentation")
data class MyData(
    @AvroDoc("This is a field documentation")
    val myField: String
)

@Serializable
@AvroDoc("This is an enum documentation")
enum class MyEnum {
    A,
    B
}

Note

This impacts only the schema generation.

Support additional non-serializable types

When looking at the types matrix, you can see some of them natively supported by Avro4k, but some others are not. Also, your own types may not be serializable.

To fix it, you need to create a custom serializer that will handle the serialization and deserialization of the value, and provide a descriptor.

Note

This impacts the serialization and the deserialization. It can also impact the schema generation if the serializer is providing a custom logical type or a custom schema through the descriptor.

Write your own serializer

To create a custom serializer, you need to implement the AvroSerializer abstract class and override the serializeAvro and deserializeAvro methods. You also need to override getSchema to provide the schema of your custom type as a custom serializer means non-standard encoding and decoding.

Create a serializer that needs Avro features like getting the schema or encoding bytes and fixed types

object YourTypeSerializer : AvroSerializer<YourType>(YourType::class.qualifiedName!!) {
    override fun getSchema(context: SchemaSupplierContext): Schema {
        // you can access the data class element, inlined elements from value classes, and their annotations
        // you can also access the avro configuration in the context
        return ... /* create the corresponding schema using SchemaBuilder or Schema.create */
    }

    override fun serializeAvro(encoder: AvroEncoder, value: YourType) {
        encoder.currentWriterSchema // you can access the current writer schema
        encoder.encodeString(value.toString())
    }

    override fun deserializeAvro(decoder: AvroDecoder): YourType {
        decoder.currentWriterSchema // you can access the current writer schema
        return YourType.fromString(decoder.decodeString())
    }

    override fun serializeGeneric(encoder: Encoder, value: YourType) {
        // you may want to implement this function if you also want to use the serializer outside of Avro4k
        encoder.encodeString(value.toString())
    }

    override fun deserializeGeneric(decoder: Decoder): YourType {
        // you may want to implement this function if you also want to use the serializer outside of Avro4k
        return YourType.fromString(decoder.decodeString())
    }
}

You may want to just implement a KSerializer if you don't need specific Avro features, but you won't be able to associate a custom schema to it:

Create a generic serializer that doesn't need specific Avro features

object YourTypeSerializer : KSerializer<YourType> {
    override val descriptor: SerialDescriptor = PrimitiveSerialDescriptor("YourType", PrimitiveKind.STRING)

    override fun serialize(encoder: Encoder, value: YourType) {
        encoder.encodeString(value.toString())
    }

    override fun deserialize(decoder: Decoder): YourType {
        return YourType.fromString(decoder.decodeString())
    }
}

Register the serializer globally (not compile time)

You first need to configure your Avro instance with the wanted serializer instance:

import kotlinx.serialization.modules.SerializersModule
import kotlinx.serialization.modules.contextual

val myCustomizedAvroInstance = Avro {
    serializersModule = SerializersModule {
        // give the object serializer instance
        contextual(YourTypeSerializerObject)
        // or instanciate it if it's a class and not an object
        contextual(YourTypeSerializerClass())
    }
}

Then just annotated the field with @Contextual:

@Serializable
data class MyData(
    @Contextual val myField: YourType
)

Register the serializer just for a field at compile time

@Serializable
data class MyData(
    @Serializable(with = YourTypeSerializer::class) val myField: YourType
)

Changing record's field name

By default, field names are the original name of the kotlin fields in the data classes.

Note

This impacts the schema generation, the serialization and the deserialization of the field.

Individual field name change

To change a field name, annotate it with @SerialName:

@Serializable
data class MyData(
    @SerialName("custom_field_name") val myField: String
)

Note

@SerialName will still be handled by the naming strategy

Field naming strategy (overall change)

To apply a naming strategy to all fields, you need to set the fieldNamingStrategy in the Avro configuration.

Note

This is only applicable for RECORD fields, and not for ENUM symbols.

There is 3 built-ins strategies:

• NoOp (default): keeps the original kotlin field name
• SnakeCase: converts the original kotlin field name to snake_case with underscores before each uppercase letter
• PascalCase: upper-case the first letter of the original kotlin field name
• If you need more, please file an issue

First, create your own instance of Avro with the wanted naming strategy:

val myCustomizedAvroInstance = Avro {
    fieldNamingStrategy = FieldNamingStrategy.Builtins.SnakeCase
}

Then, use this instance to generate the schema or encode/decode data:

package my.package

@Serializable
data class MyData(val myField: String)

val schema = myCustomizedAvroInstance.schema<MyData>() // {...,"fields":[{"name":"my_field",...}]}

Set a default field value

While reading avro binary data, you can miss a field (a kotlin field is present but not in the avro binary data), so Avro4k fails as it is not capable of constructing the kotlin type without the missing field value.

By default:

• nullable fields are optional and default: null is automatically added to the field (check this section to opt out from this default behavior).
• nullable fields are optional and default: null is automatically added to the field (check this section to opt out from this default behavior).

@AvroDefault

To avoid this error, you can set a default value for a field by annotating it with @AvroDefault:

import com.github.avrokotlin.avro4k.AvroDefault

@Serializable
data class MyData(
    @AvroDefault("default value") val stringField: String,
    @AvroDefault("42") val intField: Int?,
    @AvroDefault("""{"stringField":"custom value"}""") val nestedType: MyData? = null
)

Note

This impacts only the schema generation and the deserialization of the field, and not the serialization.

Warning

Do not use @org.apache.avro.reflect.AvroDefault as this annotation is not visible by Avro4k.

kotlin default value

You can also set a kotlin default value, but this default won't be present into the generated schema as Avro4k is not able to retrieve it:

@Serializable
data class MyData(
    val stringField: String = "default value",
    val intField: Int? = 42,
)

This impacts only the deserialization of the field, and not the serialization or the schema generation.

Add aliases

To be able of reading from different written schemas, or able of writing to different schemas, you can add aliases to a named type (record, enum) field by annotating it with @AvroAlias. The given aliases may contain the full name of the alias type or only the name.

Avro spec link

Note

Aliases are not impacted by naming strategy, so you need to provide aliases directly applying the corresponding naming strategy if you need to respect it.

import com.github.avrokotlin.avro4k.AvroAlias

@Serializable
@AvroAlias("full.name.RecordName", "JustOtherRecordName")
data class MyData(
    @AvroAlias("anotherFieldName", "old_field_name") val myField: String
)

Note

This impacts the schema generation, the serialization and the deserialization.

Warning

Do not use @org.apache.avro.reflect.AvroAlias as this annotation is not visible by Avro4k.

Add metadata to a schema (custom properties)

You can add custom properties to a schema to have additional metadata on a type. To do so, you can annotate the data class or field with @AvroProp. The value can be a regular string or any json content:

@Serializable
@AvroProp("custom_string_property", "The default non-json value")
@AvroProp("custom_int_property", "42")
@AvroProp("custom_json_property", """{"key":"value"}""")
data class MyData(
    @AvroProp("custom_field_property", "Also working on fields")
    val myField: String
)

To add metadata to a type not owned by you, you can use a value class. Here an example with a BigQuery type that needs the property sqlType = JSON on string type:

@Serializable
value class BigQueryJson(@AvroProp("sqlType", "JSON") val value: String)

println(Avro.schema<BigQueryJson>().toString(true)) // {"type":"string","sqlType":"JSON"}

Note

This impacts only the schema generation. For more details, check the avro specification.

Warning

Do not use @org.apache.avro.reflect.AvroMeta as this annotation is not visible by Avro4k.

Change scale and precision for decimal logical type

By default, the scale is 2 and the precision 8. To change it, annotate the field with @AvroDecimal:

@Serializable
data class MyData(
    @AvroDecimal(scale = 4, precision = 10) val myField: BigDecimal
)

Note

This impacts the schema generation, the serialization and the deserialization.

Change enum values' name

By default, enum symbols are exactly the name of the enum values in the enum classes. To change this default, you need to annotate enum values with @SerialName.

@Serializable
enum class MyEnum {
    @SerialName("CUSTOM_NAME")
    A,
    B,
    C
}

Note

This impacts the schema generation, the serialization and the deserialization.

Set enum default

When reading with a schema but was written with a different schema, sometimes the reader can miss the enum symbol that triggers an error. To avoid this error, you can set a default symbol for an enum by annotating the expected fallback with @AvroEnumDefault.

@Serializable
enum class MyEnum {
    A,

    @AvroEnumDefault
    B,

    C
}

Note

This impacts the schema generation, the serialization and the deserialization.

Change type name (RECORD and ENUM)

RECORD and ENUM types in Avro have a name and a namespace (composing a full-name like namespace.name). By default, the name is the name of the class/enum and the namespace is the package name. To change this default, you need to annotate data classes and enums with @SerialName.

Warning

@SerialName is redefining the full-name of the annotated class or enum, so you must repeat the name or the namespace if you only need to change the namespace or the name respectively.

Note

This impacts the schema generation, the serialization and the deserialization.

Changing the name while keeping the namespace

package my.package

@Serializable
@SerialName("my.package.MyRecord")
data class MyData(val myField: String)

Changing the namespace while keeping the name

package my.package

@Serializable
@SerialName("custom.namespace.MyData")
data class MyData(val myField: String)

Changing the name and the namespace

package my.package

@Serializable
@SerialName("custom.namespace.MyRecord")
data class MyData(val myField: String)

Change type name (FIXED only)

Warning

For the moment, it is not possible to manually change the namespace or the name of a FIXED type as the type name is coming from the field name and the namespace from the enclosing data class package.

Set a custom schema

To associate a type or a field to a custom schema, you need to create a serializer that will handle the serialization and deserialization of the value, and provide the expected schema.

See support additional non-serializable types section to get detailed explanation about writing a serializer and registering it.

Skip a kotlin field

To skip a field during encoding, you can annotate it with @kotlinx.serialization.Transient. Note that you need to provide a default value for the field as the field will be totally discarded also during encoding (IntelliJ should trigger a warn).

import kotlinx.serialization.Serializable
import kotlinx.serialization.Transient

@Serializable
data class Foo(val a: String, @Transient val b: String = "default value")

Note

This impacts the schema generation, the serialization and the deserialization.

Force a field to be a string type

You can force a field (or the value class' property) to have its inferred schema as a string type by annotating it with @AvroString.

Compatible types visible in the types matrix, indicated by the "Other compatible writer types" column. The writer schema compatibility is still respected, so if the field has been written as an int, a stringified int will be deserialized as an int without the need of parsing it. It is the same for the rerverse: If an int has been written as a string, it will be deserialized as an int by parsing the string content.

[!INFO] Note that the type must be compatible with the string type, otherwise it will be ignored. Your custom serializer generated schema must handle this annotation, or it will be ignored.

Examples:

@Serializable
data class MyData(
    @AvroString val anInt: Int,
    @AvroString val rawString: ByteArray,
    @AvroString @Contextual val bigDecimal: BigDecimal,
)
@JvmInline
@Serializable
value class StringifiedPrice(
    @AvroString val amount: Double,
)

Note

This impacts the schema generation, the serialization and the deserialization.

Nullable fields, optional fields and compatibility

With avro, you can have nullable fields and optional fields, that are taken into account for compatibility checking when using the schema registry.

But if you want to remove a nullable field that is not optional, depending on the compatibility mode, it may not be compatible because of the missing default value.

• What is an optional field ?

An optional field is a field that have a default value, like an int with a default as -1.

• What is a nullable field ?

A nullable field is a field that contains a null type in its type union, but it's not an optional field if you don't put default value to null.

So to mark a field as optional and facilitate avro contract evolution regarding compatibility checks, then set default to null.

Known problems

• Kotlin 1.7.20 up to 1.8.10 cannot properly compile @SerialInfo-Annotations on enums (see Kotlin/kotlinx.serialization#2121). This is fixed with kotlin 1.8.20. So if you are planning to use any of avro4k's annotations on enum types, please make sure that you are using kotlin >= 1.8.20.

Migrating from v1 to v2

Heads up to the migration guide to update your code from avro4k v1 to v2.

Contributions

Contributions to avro4k are always welcome. Good ways to contribute include:

• Raising bugs and feature requests
• Fixing bugs and enhancing the API
• Improving the performance of avro4k
• Adding documentation