Avro.FSharp

FSharp implementation of Apache Avro

Overview

The library generates an Avro schema by an F# type as well as serialize and deserialise an F# data in binary and json encodings. It works seamlessly with Fable.Avro.

Schema

To generate a schema, use Schema.generate:

record SchemaOptions
  val Annotations: String           // information about aliases and default values in json string
  val CustomRules: list<CustomRule> // custom rules for non supported types
  val StubDefaultValues: bool       // add stub default values if true
  val TreatDecimalAsDouble: bool    // generate double schema for decimals (otherwise bytes schema is generated)
  val TreatBigIntAsString: bool     // generate string schema for bigints (otherwise bytes schema is generated)
  val TreatGuidAsString: bool       // generate string schema for guids (otherwise fixed schema is generated)

val generate:
   options: SchemaOptions ->
   type'      : Type                    // type to build schema of
             -> Result<Schema,SchemaError>

To read a schema from json string, use Schema.ofString:

val ofString:
   jsonString: string
            -> Schema

To write schema, use Schema.toString:

val toString:
   schema: Schema
        -> string

Primitive types

F# type	Avro type
string	string
bool	boolean
byte	int
short	int
int	int
uint	int
int16	int
int64	long
uint64	long
float32	float
float	double
byte array	bytes

Examples:

• generate Schema.defaultOptions typeof<string> generates: {"type": "string"}
• generate Schema.defaultOptions typeof<bool> generates: {"type": "boolean"}
• generate Schema.defaultOptions typeof<int> generates: {"type": "int"}
• generate Schema.defaultOptions typeof<int64> generates: {"type": "long"}
• generate Schema.defaultOptions typeof<float32> generates: {"type": "float"}
• generate Schema.defaultOptions typeof<float> generates: {"type": "double"}
• generate Schema.defaultOptions typeof<byte array> generates: {"type": "bytes"}

Arrays

Following F# types are mapped to avro's array:

• 'T list
• 'T array (if 'T is not byte)
• ResizeArray<'T> (System.Collection.Generic.List<'T>)
• Set<'T>
• HashSet<'T>
• 'T Seq (IEnumerable<'T>)

Examples:

• generate Schema.defaultOptions typeof<string list> generates: {"type": "array", "values": "string"}
• generate Schema.defaultOptions typeof<int array> generates: {"type": "array", "values": "int"}
• generate Schema.defaultOptions typeof<List<bool>> generates: {"type": "array", "values": "boolean"}

Maps

Following F# types are mapped to avro's map:

• Map<string,'TValue>
• Dictionary<string,'TValue>

Examples:

• generate Schema.defaultOptions typeof<Map<string,string>> generates: {"type": "map", "values": "string"}
• generate Schema.defaultOptions typeof<Dictionary<string, int>> generates: {"type": "map", "values": "int"}

Enums

F# Enum is mapped to Avro's enum

Example:

type TestState =
    | Red = 3
    | Yellow = 2
    | Green = 1

generate Schema.defaultOptions typeof<TestState>

generated schema:

{"type": "enum", "name": "TestState", "symbols": ["Green", "Yellow", "Red"]}

Symbols are ordered by its values.

Records

F# records and tuples are mapped to Avro's record.

Example:

type SimpleRecord = {
    Id : int
    Name : string
    Version : int64}

generate Schema.defaultOptions typeof<SimpleRecord>

generated schema:

{
    "type": "record",
    "name": "SimpleRecord",
    "fields" : [
        {"name": "Id", "type": "int"},
        {"name": "Name", "type": "string"},
        {"name": "Version", "type": "long"}
    ]
}

A tuple is mapped to Avro's record with fields Item1, Item2 and so on.

Example:

generate Schema.defaultOptions typeof<int*string>> generates

{
    "type": "record",
    "name": "Tuple_Of_Int32_And_String",
    "fields" : [
        {"name": "Item1", "type": "int"},
        {"name": "Item2", "type": "string"},
    ]
}

Generic records are also allowed:

type GenericRecord<'T> = {
    Value : 'T
}

generate Schema.defaultOptions typeof<GenericRecord<string>>

generated schema:

{
    "type":"record",
    "name":"GenericRecord_Of_String",
    "fields":[{"name":"Value","type":"string"}]
}

Unions

F# Discriminated Union is mapped to Avro's union of records, generated from the union's cases

Example:

type BinaryTree =
    | Leaf of value:string
    | Node of left: BinaryTree * right: BinaryTree

generate Schema.defaultOptions typeof<BinaryTree>

generated schema:

{
    "type":[
        {
            "type":"record",
            "name":"Leaf",
            "fields":[
                {"name":"value","type":"string"}
            ]
        },
        {
            "type":"record",
            "name":"Node",
            "fields":[
                {"name":"left","type":["Leaf","Node"]},
                {"name":"right","type":["Leaf","Node"]}
            ]
        }
    ]
}

Option is mapped as union of null and the option's generic argument's type

Example:

generate Schema.defaultOptions typeof<Option<float>> generates ["null","double"]

Known types

Following types are handled in special way

F# type	Avro type	Description
Guid	string of fixed
Decimal	double or bytes
BigInt	string of bytes
DateTime	string	ISO 8601
DateTimeOffset	string	ISO 8601
TimeSpan	int	milliseconds
Uri	string

Logical types

NOT YET SUPPORTED

Annotations

Some schema's attributes can not be evolved from F# type (default values and aliases). Additional annotation is used for the purpose. Here is example of the annotation json.

{
    "records": [{
        "name": "Foo.Bar.NewRecord",        // Full name of the record
        "aliases": ["Foo.Bar.OldRecord"],   // Aliases attributes in the record's schema
        "fields": [{
                "name": "Caption",          // Name of the record's field
                "aliases": ["Title", "Cap"] // Aliases atttibute in the field's schema
            },
            {
                "name": "Description",
                "aliases": [],
                "default": "Not Yet Described"  // Default value in the fied's schema
            }
        ]
    }],
    "enums": [{
        "name": "Foo.Bar.NewTestState",     // Full name of the enum
        "aliases": ["Foo.Bar.TestState"],   // Aliases attribute is the enum's schema
        "default": "Blue"                   // Default value in the enum's schema
    }],
    "decimals": [{
        "record": "Foo.Bar.NewTestState", // Full name of the record
        "field": "Price",                 // Name of a decimal field
        "scale": 3                        // Scale for decimal field
    }]
}

You don't need to annotate all enums, records and fields. Annotate only those schemas which should be enriched with additional attributes. Since tuples and DU cases are mapped to a record, you may define attibutes for them as well. Remember, that tupel's field name is like Item1, Item2, Item3 .... DU case name is composed from name of the DU and name of the case.

Names

According to avro specification, only [A-Za-z0-9_] symbols are allowed in the name attributes. Name of an enum is constructed from namespace and type's name. Name of a record also contains description of the generic type arguments.

Rule of the generation of the records name is describer is the following table:

Precicate	Rule
Is kind of array	Array_Of_{ItemTypeName}
Is kind of map	Map_Of_{ValueTypeName}
IsGenericType	{TypeName}_Of_{GenericType1Name}_And_{GenericType2Name}_...
Is Result<OkType,ErrType>	Result_Of_{OkTypeName}_{ErrTypeName}
Is Option	Nullable_{TSomeName}
Is Tuple	Tuple_Of_{Item1TypeName}_{Item2TypeName}_...
Is DU case	{DU Type Name}.{CaseName}
System.RestName	RestName

Examples of record names:

• Result_Of_Int64_And_String.Ok
• Tuple_Of_Int32_And_String
• Foo.Bar.GenericRecord_Of_Nullable_String

Serde

To create serializer use:

record SerializationOptions
  val CustomRules: list<CustomRule>

val Serde.binarySerializer:
   options: SerializationOptions ->
   type'  : Type                 ->
   schema : Schema
         -> obj -> Stream -> unit

val Serde.jsonSerializer:
   options: SerializationOptions ->
   type'  : Type                 ->
   schema : Schema
         -> obj -> Utf8JsonWriter -> unit```

To create deserialized use:
```fsharp

record DeserializationOptions
  val CustomRules: list<CustomRule>
  val EvolutionTolerantMode: bool

val Serde.binaryDeserializer:
   options     : DeserializationOptions ->
   type'       : Type   ->
   readerSchema: Schema
              -> Schema -> Stream -> obj

val Serde.jsonDeserializer:
   options: DeserializationOptions ->
   type'  : Type   ->
   schema : Schema
         -> Stream -> obj

Here is basic example:

    let orig:MyType = createInstance()

    let schema =
        match Schema.generate Schema.defaultOptions typeof<MyType> with
        | Ok schema -> schema
        | Error err -> failwithf "Schema error: %A" err

    // binary
    let serializer = Serde.binarySerializer Serde.defaultSerializerOptions (case.InstanceType) schema
    let deserializer = Serde.binaryDeserializer Serde.defaultDeserializerOptions (case.InstanceType) schema

    use stream = new MemoryStream()
    serializer case.Instance stream

    let data = stream.ToArray()

    use stream = new MemoryStream(data)
    let copy = deserializer schema stream
    Expect.equal "copy shoud be equal to original" orig copy

    // json
    let serializer = Serde.jsonSerializer Serde.defaultSerializerOptions typeof<MyType> schema
    let deserializer = Serde.jsonDeserializer Serde.defaultDeserializerOptions typeof<MyType> schema

    use stream = new MemoryStream()
    use writer = new Utf8JsonWriter(stream)
    serializer case.Instance writer

    let data = stream.ToArray()

    use stream = new MemoryStream(data)
    let copy = deserializer stream
    Expect.equal "copy shoud be equal to original" orig copy

Evolution issues

It is very important in microservices architecture, that changes in the schema do not break work of a service. The library aimed to make schemas evolution compatibility as simple as possible.

Stub default values

Setting option SchemaOptions.StubDefaultValues enable adding default value to each field's schema and enum's schema.

Following rules are used:

F# Type	Default Value
string	""
bool	false
byte, short, uint16, uint32, uint64, int, long, float32, float	0
byte array	""
array, list, set, seq, ResizeArray, HashSet	[]
Map, Dictionary<_,_>	{}
Enum	"{NameOfFirsSymbol}" Symbols are ordered by values
Record	{"{Field1Name}": {DefaultValueForField1}, ...}
Tuple	{"{Item1}": {DefaultValueForItem1}, ...}
Option	null
DU	{"Case1Field1Name": {DefaultValueForCase1Field1}, ...} stub for first case in DU
decimal	0
BigInt	"0"
Guid	"00000000-0000-0000-0000-000000000000"
DateTime	"1970-01-01T00:00:00.000Z"
DateTimeOffset	"1970-01-01T00:00:00.000+00:00"
TimeSpan	0

If deserializer can not find field's value it looks default value in the annotations. If annotations do not have defalut value for the field, stub value is created. Set DeserializationOptions.EvolutionTolerantMode=false if you don't want the behaviour.

Deserializing of the Enums is performed by the same algorithm.

Forward compatibility for DU

According to Avro standard, adding a new case at a union is a non forward compatible change (see).

Let's pretend that first version of our domain looks like:

type DomainUnion =
    | Case1
    | Case2

Eventually, version 2 is evolved:

type DomainUnion =
    | Case1
    | Case2
    | Case3

According to Avro standard, microservices that use old schema (version 1) should get an error trying deserialize Case3. This is big obstacle for evolution of the algebraic types. Therefore the library substitutes unknown case with default value of DU (stub for the first case) if DeserializationOptions.EvolutionTolerantMode=true.

For example if deserializer's domain:

type DomainUnion =
    | UnknownCase
    | Case1

And serializer's domain:

type DomainUnion =
    | UnknownCase
    | Case1
    | Case2
    | Case3

Case3 will be deserialized to UnknownCase (first case of the DomainUnion). This is true for any occasion of DU in deserialized type (whenever it is a record's field, or item in an array, or value in a map). For example, array [Case1, Case3, Case2, Case1] will be deserialized to [Case1, UnknownCase, UnknownCase, Case1]. Set DeserializationOptions.EvolutionTolerantMode=false if you don't want the behaviour.

Custom Rules

It is possible to customize processing of a particular type. In that case CustomRule should be created.

record CustomRule
  val InstanceType: Type                // particular type
  val SurrogateType: Type               // surrogate type, shoud be supported by serializer
  val CastFromSurrogate: obj -> obj
  val CastToSurrogate: obj -> obj
  val StubValue: Json                   // default value, shoud be compatible with json format of the surrogate

Example of the implementation of the CustomRule:

{
    InstanceType = typeof<Uri>
    SurrogateType = typeof<string>
    CastToSurrogate = fun v -> v.ToString() |> box
    CastFromSurrogate = fun v -> Uri(unbox(v)) |> box
    StubValue =  """ "" """ // JSON with empty string
}

List with custom rules is passed to schema generator, serializer and deserializer as part of its options

Examples

More examples of complex types and corresponging schemas is available in the SchemaTests.fs.

See CustomRule.fs for details of the implementation of the custom rules.

An implementation of Kafka producer and consumer, working with SchemaRegistry could be found here. (The example is supposed you have an account in confluent.cloud. Change configuration code if you have on-premise installation.)