如何在Avro中定义LogicalType（Java）

Question

我需要能够标记AVRO模式中的某些字段，以便在序列化时对其进行加密。

逻辑类型允许标记这些字段，并与自定义转换一起使用，应该可以让AVRO无缝地将它们加密。

---

我曾经遇到过一些问题，无法找到如何在AVRO中定义和使用新的逻辑类型的文档（avro_1.8.2＃逻辑类型）。
因此，我决定在答案中分享我发现的内容，以使其他人能够更轻松地了解并获取反馈，以防我做错了什么。

- enrico

Answer 1

首先，我定义了一个逻辑类型：

public class EncryptedLogicalType extends LogicalType {
    //The key to use as a reference to the type
    public static final String ENCRYPTED_LOGICAL_TYPE_NAME = "encrypted";

    EncryptedLogicalType() {
        super(ENCRYPTED_LOGICAL_TYPE_NAME);
    }

    @Override
    public void validate(Schema schema) {
        super.validate(schema);
        if (schema.getType() != Schema.Type.BYTES) {
            throw new IllegalArgumentException(
                    "Logical type 'encrypted' must be backed by bytes");
        }
    }
}

然后是一次新的转换：

public class EncryptedConversion extends Conversion<ByteBuffer> {
    // Construct a unique instance for all the conversion. This have to be changed in case the conversion
    //   needs some runtime information (e.g.: an encryption key / a tenant_ID). If so, the get() method should 
    //   return the appropriate conversion per key.
    private static final EncryptedConversion INSTANCE = new EncryptedConversion();
    public static final EncryptedConversion get(){ return INSTANCE; }
    private EncryptedConversion(){ super(); }

    //This conversion operates on ByteBuffer and returns ByteBuffer
    @Override
    public Class<ByteBuffer> getConvertedType() { return ByteBuffer.class; }

    @Override
    public String getLogicalTypeName() { return EncryptedLogicalType.ENCRYPTED_LOGICAL_TYPE_NAME; }

    // fromBytes and toBytes have to be overridden as this conversion works on bytes. Other may need to be 
    //  overridden. The types supported need to be updated also in EncryptedLogicalType#validate(Schema schema)
    @Override
    public ByteBuffer fromBytes(ByteBuffer value, Schema schema, LogicalType type) {
        encryptedValue = __encryptionLogic__(value); 
        return encryptedValue;
    }

    @Override
    public ByteBuffer toBytes(ByteBuffer value, Schema schema, LogicalType type) {
        decryptedValue = __decryptionLogic__(value); 
        return decryptedValue;
    }
}

.avsc模式文件将类似于：

{
    "name": “MyMessageWithEncryptedField”,
    "type": "record",
    "fields": [
        {"name": "payload","type" : {"type" : "bytes","logicalType" : "encrypted"}},
        ...

在从模式文件生成的MyMessageWithEncryptedField.java类中，最后我添加了返回转换的方法：

@Override
public Conversion<?> getConversion(int fieldIndex) {
    // This allow us to have a more flexible conversion retrieval, so we don't have to code it per field.
    Schema fieldSchema = SCHEMA$.getFields().get(fieldIndex).schema();
    if ((fieldSchema.getLogicalType() != null)
            && (fieldSchema.getLogicalType().getName() == EncryptedLogicalType.ENCRYPTED_LOGICAL_TYPE_NAME)){
     // here we could pass to the get() method a runtime information, e.g.: a tenantId that can be found in the data structure.
        return EncryptedConversion.get();
    }
    return null;
}

为了使其运行，我仍然需要在运行时注册类型：

LogicalTypes.register(EncryptedLogicalType.ENCRYPTED_LOGICAL_TYPE_NAME, new LogicalTypes.LogicalTypeFactory() {
    private final LogicalType encryptedLogicalType = new EncryptedLogicalType();
    @Override
    public LogicalType fromSchema(Schema schema) {
        return encryptedLogicalType;
    }
});

注意事项：

• 如果您的逻辑类型需要从模式定义中传递其他属性，您可以修改LogicalType类，以avro.lang.java.avro.src.main.java.org.apache.avro.LogicalTypes.Decimal为例进行修改。
• 最后一段代码(注册)目前在我的逻辑开始之前运行，但我计划将其移动到生成类(MyMessageWithEncryptedField.java)的静态块内。

- enrico