Entity Framework Core provider for Apache Kafka shall convert the entities used within the model in something viable from the backend.
Each backend has its own schema to convert entities into something else; database providers converts entities into database schema or blob in Cosmos.
IMPORTANT NOTE: till the first major version, all releases shall be considered not stable: this means the API public, or internal, can change without notice.
Basic concepts
Entity Framework Core provider for Apache Kafka shall convert the entities into record will be stored in the topics of Apache Kafka cluster.
The way the entities are converted shall follows a schema.
The current schema follows a JSON pattern and reports the information of each entity as:
Primary Key:
Simple: if the Primary Key is a native type (e.g. int, long, double, and so on) the serialization is made using the Apache Kafka default serializer for that type
Complex: if the Primary Key is a complex type (e.g. int-int, int-long, int-string, and so on), Entity Framework reports it as an array of objects and the serialization is made using a JSON serializer
Entity data: the Entity is managed, from Entity Framework Core provider for Apache Kafka, as an array of objects associated to properties of the Entity.
The schema of the Apache Kafka record value follows the code definition in DefaultValueContainer<T>. Below two examples:
ValueContainer class: KEFCore uses DefaultValueContainer<T> which stores the CLR type of Entity, the properties ordered by their index with associated CLT type, name and JSON serializaed value; the class is marked for JSON serialization and it is used from the ValueContainer SerDes;
Key SerDes class: KEFCore uses DefaultKEFCoreSerDes.Key.Json<T>, the type automatically manages simple or complex Primary Key
The default serialization can be overridden with user defined ValueContainer class, Key SerDes or ValueContainer SerDes.
ValueContainer class
A custom ValueContainer class must contains enough information and shall follow the following rules:
must implements the IValueContainer<T> interface
must be a generic type
must have at least a default constructor and a constructor which accept two parameters: a first parameter which is IEntityType and a second paramater of object[]
An example snippet is the follow:
public class CustomValueContainer<TKey> : IValueContainer<TKey> where TKey : notnull
{
/// <summary>
/// Initialize a new instance of <see cref="CustomValueContainer{TKey}"/>
/// </summary>
/// <param name="tName">The <see cref="IEntityType"/> requesting the ValueContainer for <paramref name="rData"/></param>
/// <param name="rData">The data, built from EFCore, to be stored in the ValueContainer</param>
/// <remarks>This constructor is mandatory and it is used from KEFCore to request a ValueContainer</remarks>
public CustomValueContainer(IEntityType tName, object[] rData)
{
}
/// <inheritdoc/>
public string EntityName { get; set; }
/// <inheritdoc/>
public string ClrType { get; set; }
/// <inheritdoc/>
public void GetData(IEntityType tName, ref object[] array)
{
}
/// <inheritdoc/>
public IReadOnlyDictionary<int, string> GetProperties()
{
// build properties
}
}
Key SerDes and ValueContainer SerDes class
A custom Key SerDes class shall follow the following rules:
must implements the IKNetSerDes<T> interface or extend KNetSerDes<T>
must be a generic type
must have a parameterless constructor
can store serialization information using Headers of Apache Kafka record (this information will be used from EntityExtractor)
An example snippet is the follow based on JSON serializer:
public class CustomKeySerDes<T> : KNetSerDes<T>
{
readonly byte[] keyTypeName = Encoding.UTF8.GetBytes(typeof(T).FullName!);
readonly byte[] customSerDesName = Encoding.UTF8.GetBytes(typeof(CustomKeySerDes<>).FullName!);
/// <inheritdoc cref="KNetSerDes{T}.Serialize(string, T)"/>
public override byte[] Serialize(string topic, T data)
{
return SerializeWithHeaders(topic, null, data);
}
/// <inheritdoc cref="KNetSerDes{T}.SerializeWithHeaders(string, Headers, T)"/>
public override byte[] SerializeWithHeaders(string topic, Headers headers, T data)
{
headers?.Add(KEFCoreSerDesNames.KeyTypeIdentifier, keyTypeName);
headers?.Add(KEFCoreSerDesNames.KeySerializerIdentifier, customSerDesName);
var jsonStr = System.Text.Json.JsonSerializer.Serialize<T>(data);
return Encoding.UTF8.GetBytes(jsonStr);
}
/// <inheritdoc cref="KNetSerDes{T}.Deserialize(string, byte[])"/>
public override T Deserialize(string topic, byte[] data)
{
return DeserializeWithHeaders(topic, null, data);
}
/// <inheritdoc cref="KNetSerDes{T}.DeserializeWithHeaders(string, Headers, byte[])"/>
public override T DeserializeWithHeaders(string topic, Headers headers, byte[] data)
{
if (data == null) return default;
return System.Text.Json.JsonSerializer.Deserialize<T>(data)!;
}
}
public class CustomValueContainerSerDes<T> : KNetSerDes<T>
{
readonly byte[] valueContainerSerDesName = Encoding.UTF8.GetBytes(typeof(CustomValueContainerSerDes<>).FullName!);
readonly byte[] valueContainerName = null!;
/// <summary>
/// Default initializer
/// </summary>
public CustomValueContainerSerDes()
{
var tt = typeof(T);
if (tt.IsGenericType)
{
var keyT = tt.GetGenericArguments();
if (keyT.Length != 1) { throw new ArgumentException($"{typeof(T).Name} does not contains a single generic argument and cannot be used because it is not a valid ValueContainer type"); }
var t = tt.GetGenericTypeDefinition();
if (t.GetInterface(typeof(IValueContainer<>).Name) != null)
{
valueContainerName = Encoding.UTF8.GetBytes(t.FullName!);
return;
}
else throw new ArgumentException($"{typeof(T).Name} does not implement IValueContainer<> and cannot be used because it is not a valid ValueContainer type");
}
throw new ArgumentException($"{typeof(T).Name} is not a generic type and cannot be used as a valid ValueContainer type");
}
/// <inheritdoc cref="KNetSerDes{T}.Serialize(string, T)"/>
public override byte[] Serialize(string topic, T data)
{
return SerializeWithHeaders(topic, null, data);
}
/// <inheritdoc cref="KNetSerDes{T}.SerializeWithHeaders(string, Headers, T)"/>
public override byte[] SerializeWithHeaders(string topic, Headers headers, T data)
{
headers?.Add(KEFCoreSerDesNames.ValueContainerSerializerIdentifier, valueContainerSerDesName);
headers?.Add(KEFCoreSerDesNames.ValueContainerIdentifier, valueContainerName);
var jsonStr = System.Text.Json.JsonSerializer.Serialize<T>(data);
return Encoding.UTF8.GetBytes(jsonStr);
}
/// <inheritdoc cref="KNetSerDes{T}.Deserialize(string, byte[])"/>
public override T Deserialize(string topic, byte[] data)
{
return DeserializeWithHeaders(topic, null, data);
}
/// <inheritdoc cref="KNetSerDes{T}.DeserializeWithHeaders(string, Headers, byte[])"/>
public override T DeserializeWithHeaders(string topic, Headers headers, byte[] data)
{
if (data == null) return default;
return System.Text.Json.JsonSerializer.Deserialize<T>(data)!;
}
}
How to use custom types
KafkaDbContext contains three properties can be used to override the default types:
KeySerializationType: set the value of the Key SerDes type in the form CustomSerDes<>
ValueSerializationType: set the value of the ValueContainer SerDes type in the form CustomSerDes<>
ValueContainerType: set the value of the ValueContainer type in the form CustomValueContainer<>
IMPORTANT NOTE: the type applied in the previous properties of KafkaDbContext shall be a generic type definition, Entity Framework Core provider for Apache Kafka will apply the right generic type when needed.
The following schema is the default used from the engine and can be registered in Apache Schema registry so other applications can use it to extract the data stored in the topics:
Complex Primary Key schema:
{
"namespace": "MASES.EntityFrameworkCore.KNet.Serialization.Avro.Storage",
"type": "record",
"name": "AvroKeyContainer",
"doc": "Represents the storage container type to be used from KEFCore for keys",
"fields": [
{
"name": "PrimaryKey",
"type": {
"type": "array",
"items": [
"null",
"boolean",
"int",
"long",
"float",
"double",
"string"
]
}
}
]
}
ValueContainer schema:
{
"namespace": "MASES.EntityFrameworkCore.KNet.Serialization.Avro.Storage",
"type": "record",
"name": "AvroValueContainer",
"doc": "Represents the storage container type to be used from KEFCore",
"fields": [
{
"name": "EntityName",
"type": "string"
},
{
"name": "ClrType",
"type": "string"
},
{
"name": "Data",
"type": {
"type": "array",
"items": {
"namespace": "MASES.EntityFrameworkCore.KNet.Serialization.Avro.Storage",
"type": "record",
"name": "PropertyDataRecord",
"doc": "Represents the single container for Entity properties stored in AvroValueContainer and used from KEFCore",
"fields": [
{
"name": "PropertyIndex",
"type": "int"
},
{
"name": "PropertyName",
"type": "string"
},
{
"name": "ClrType",
"type": "string"
},
{
"name": "Value",
"type": [
"null",
"boolean",
"int",
"long",
"float",
"double",
"string"
]
}
]
}
}
}
]
}
The extension converted this schema into code to speedup the exection of serialization/deserialization operations.
How to use Avro
KafkaDbContext contains three properties can be used to override the default types:
KeySerializationType: set this value to AvroKEFCoreSerDes.Key.Binary<> or AvroKEFCoreSerDes.Key.Json<>, both types automatically manages simple or complex Primary Key
ValueSerializationType: set this value to AvroKEFCoreSerDes.ValueContainer.Binary<> or AvroKEFCoreSerDes.ValueContainer.Json<>
ValueContainerType: set this value to AvroValueContainer<>