Protostuff序列化分析

 

最近项目中需要将业务对象直接序列化，然后存数据库；考虑到序列化、反序列化的时间以及生产文件的大小觉得Protobuf是一个很好的选择，但是Protobuf有的问题就是需要有一个.proto的描述文件，而且由Protobuf生成的对象用来作为业务对象并不是特别友好，往往业务对象和Protobuf对象存在一个互相转换的过程；考虑到我们仅仅是将业务对象直接序列化到数据库，发现Protobuf在这种情况下并不是特别的好；

这时候发现了Protostuff，protostuff不需要依赖.proto文件，可以直接对普通的javabean进行序列化、反序列化的操作，而效率上甚至比protobuf还快，生成的二进制数据库格式和Protobuf完全相同的，可以说是一个基于Protobuf的序列化工具。

简单测试

1.先测试一下Protostuff

提供一个简单的javabean

public class Person {

	private int id;
	private String name;
	private String email;
        
        // get/set方法省略
}

测试类PbStuff

public class PbStuff {
	
	public static void main(String[] args) throws FileNotFoundException,
			IOException {
		Schema<Person> schema = RuntimeSchema.getSchema(Person.class);
		Person person1 = new Person();
		person1.setId(1);
		person1.setName("zhaohui");
		LinkedBuffer buffer = LinkedBuffer.allocate(1024);
		byte[] data = ProtobufIOUtil.toByteArray(person1, schema, buffer);
		System.out.println(data.length);
	}
}

序列化之后二进制的大小为29字节

2.测试Protobuf

proto文件

option java_package = "protobuf.clazz"; 
option java_outer_classname = "PersonX";

message Person {
  required int32 id = 1;
  required string name = 2;
  required string email = 3;
}

PBTest类

public class PBTest {

	public static void main(String[] args) {
		PersonX.Person.Builder builder = PersonX.Person.newBuilder();
		builder.setId(1);
		builder.setName("zhaohui");
		builder.setEmail("xxxxxxxx@126.com");

		PersonX.Person p = builder.build();
		byte[] result = p.toByteArray();
		System.out.println(result.length);

	}
}

序列化之后二进制的大小同样也是29字节

经过简单的测试：发现Protobuf和Protostuff序列化相同的数据得到的结果是一样的

Protobuf的编码是尽其所能地将字段的元信息和字段的值压缩存储，并且字段的元信息中含有对这个字段描述的所有信息；既然Protostuff序列化之后的大小和Protobuf是一样的，那可以分析一下Protostuff的源码

源码分析

1.Schema

schema = RuntimeSchema.getSchema(Person.class); //获取业务对象Person的Schema

RuntimeSchema是一个包含业务对象所有信息的类，包括类信息、字段信息

/**
     * Gets the schema that was either registered or lazily initialized at runtime.
     * <p>
     * Method overload for backwards compatibility.
     */
    public static <T> Schema<T> getSchema(Class<T> typeClass)
    {
        return getSchema(typeClass, ID_STRATEGY);
    }

    /**
     * Gets the schema that was either registered or lazily initialized at runtime.
     */
    public static <T> Schema<T> getSchema(Class<T> typeClass,
            IdStrategy strategy)
    {
        return strategy.getSchemaWrapper(typeClass, true).getSchema();
    }

getSchema方法中指定了获取Schema的默认策略类ID_STRATEGY，ID_STRATEGY在类RuntimeEnv中进行了实例化：

ID_STRATEGY = new DefaultIdStrategy();

可以大致看一下DefaultIdStrategy类：

public final class DefaultIdStrategy extends IdStrategy
{

    final ConcurrentHashMap<String, HasSchema<?>> pojoMapping = new ConcurrentHashMap<>();

    final ConcurrentHashMap<String, EnumIO<?>> enumMapping = new ConcurrentHashMap<>();

    final ConcurrentHashMap<String, CollectionSchema.MessageFactory> collectionMapping = new ConcurrentHashMap<>();

    final ConcurrentHashMap<String, MapSchema.MessageFactory> mapMapping = new ConcurrentHashMap<>();

    final ConcurrentHashMap<String, HasDelegate<?>> delegateMapping = new ConcurrentHashMap<>();
    ...
}

可以发现DefaultIdStrategy内存缓存了很多Schema信息，不难理解既然要或者业务对象的类和字段信息，必然用到反射机制，这是一个很耗时的过程，进行缓存很有必要，这样下次遇到相同的类就可以不用进行反射了

所以可以看到DefaultIdStrategy中有很多这种模式的方法：

public <T> HasSchema<T> getSchemaWrapper(Class<T> typeClass, boolean create)
    {
        HasSchema<T> hs = (HasSchema<T>) pojoMapping.get(typeClass.getName());
        if (hs == null && create)
        {
            hs = new Lazy<>(typeClass, this);
            final HasSchema<T> last = (HasSchema<T>) pojoMapping.putIfAbsent(
                    typeClass.getName(), hs);
            if (last != null)
                hs = last;
        }

        return hs;
    }

先get，如果为null，就putIfAbsent

当业务对象的Schema还没被缓存，这时候就会去create，RuntimeSchema提供了createFrom方法：

public static <T> RuntimeSchema<T> createFrom(Class<T> typeClass,
            Set<String> exclusions, IdStrategy strategy)
    {
        final Map<String, java.lang.reflect.Field> fieldMap = findInstanceFields(typeClass);
        ...省略
        final Field<T> field = RuntimeFieldFactory.getFieldFactory(
                        f.getType(), strategy).create(fieldMapping, name, f,
                        strategy);
                fields.add(field);
            }
        }

        return new RuntimeSchema<>(typeClass, fields, RuntimeEnv.newInstantiator(typeClass));
     }

主要就是对typeClass进行反射，然后进行封装；将字段类型封装成了RuntimeFieldFactory，最后通过RuntimeFieldFactory的create方法封装进入Field类中，RuntimeFieldFactory列举了所有支持的类型：

static final RuntimeFieldFactory<BigDecimal> BIGDECIMAL;
    static final RuntimeFieldFactory<BigInteger> BIGINTEGER;
    static final RuntimeFieldFactory<Boolean> BOOL;
    static final RuntimeFieldFactory<Byte> BYTE;
    static final RuntimeFieldFactory<ByteString> BYTES;
    static final RuntimeFieldFactory<byte[]> BYTE_ARRAY;
    static final RuntimeFieldFactory<Character> CHAR;
    static final RuntimeFieldFactory<Date> DATE;
    static final RuntimeFieldFactory<Double> DOUBLE;
    static final RuntimeFieldFactory<Float> FLOAT;
    static final RuntimeFieldFactory<Integer> INT32;
    static final RuntimeFieldFactory<Long> INT64;
    static final RuntimeFieldFactory<Short> SHORT;
    static final RuntimeFieldFactory<String> STRING;

    static final RuntimeFieldFactory<Integer> ENUM;
    static final RuntimeFieldFactory<Object> OBJECT;
    static final RuntimeFieldFactory<Object> POJO;
    static final RuntimeFieldFactory<Object> POLYMORPHIC_POJO;

    static final RuntimeFieldFactory<Collection<?>> COLLECTION = 
            new RuntimeFieldFactory<Collection<?>>(ID_COLLECTION)

当然还有常用的Map类型，在RuntimeMapFieldFactory中定义了

2.LinkedBuffer buffer = LinkedBuffer.allocate(1024);

开辟了1024字节缓存，用来存放业务对象序列化之后存放的地方，当然你可能会担心这个大小如果不够怎么办，后面的代码中可以看到，如果空间不足，会自动扩展的，所有这个大小要设置一个合适的值，设置大了浪费空间，设置小了会自动扩展浪费时间。

3.byte[] data = ProtobufIOUtil.toByteArray(person1, schema, buffer);

ProtobufIOUtil提供的就是以Protobuf编码的格式来序列化业务对象

public static <T> byte[] toByteArray(T message, Schema<T> schema, LinkedBuffer buffer)
    {
        if (buffer.start != buffer.offset)
            throw new IllegalArgumentException("Buffer previously used and had not been reset.");

        final ProtobufOutput output = new ProtobufOutput(buffer);
        try
        {
            schema.writeTo(output, message);
        }
        catch (IOException e)
        {
        }

        return output.toByteArray();
    }

schema中调用writeTo方法，将message中的消息保存到ProtobufOutput中

public final void writeTo(Output output, T message) throws IOException
    {
        for (Field<T> f : getFields())
            f.writeTo(output, message);
    }

第一步中将业务对象的字段信息都封装到了Field中了，可以看一下Field类提供的几个方法：

/**
     * Writes the value of a field to the {@code output}.
     */
    protected abstract void writeTo(Output output, T message)
            throws IOException;

    /**
     * Reads the field value into the {@code message}.
     */
    protected abstract void mergeFrom(Input input, T message)
            throws IOException;

    /**
     * Transfer the input field to the output field.
     */
    protected abstract void transfer(Pipe pipe, Input input, Output output,
            boolean repeated) throws IOException;

提供了三个抽象方法，分别是写数据，读数据和转移数据

下面已int类型为实例，看看实现：

public static final RuntimeFieldFactory<Integer> INT32 = new RuntimeFieldFactory<Integer>(
            ID_INT32)
    {
        @Override
        public <T> Field<T> create(int number, java.lang.String name,
                final java.lang.reflect.Field f, IdStrategy strategy)
        {
            final boolean primitive = f.getType().isPrimitive();
            final long offset = us.objectFieldOffset(f);
            return new Field<T>(FieldType.INT32, number, name,
                    f.getAnnotation(Tag.class))
            {
                @Override
                public void mergeFrom(Input input, T message)
                        throws IOException
                {
                    if (primitive)
                        us.putInt(message, offset, input.readInt32());
                    else
                        us.putObject(message, offset,
                                Integer.valueOf(input.readInt32()));
                }

                @Override
                public void writeTo(Output output, T message)
                        throws IOException
                {
                    if (primitive)
                        output.writeInt32(number, us.getInt(message, offset),
                                false);
                    else
                    {
                        Integer value = (Integer) us.getObject(message, offset);
                        if (value != null)
                            output.writeInt32(number, value.intValue(), false);
                    }
                }
                ...
            };
        }

上面这段代码可以在RuntimeUnsafeFieldFactory中找到，基本的数据类型都在此类中能找到，collection和map分别在RuntimeRepeatedFieldFactory和RuntimeMapFieldFactory中，writeTo方法调用了ProtobufOutput中的writeInt32方法：

public void writeInt32(int fieldNumber, int value, boolean repeated) throws IOException
    {
         ...
         tail = writeTagAndRawVarInt32(
                  makeTag(fieldNumber, WIRETYPE_VARINT),
                  value,
                  this,
                  tail);
          ...
    }

写入field的Tag已经Value，Protobuf也是这种形式存放的，如下图所示：

public static LinkedBuffer writeTagAndRawVarInt32(int tag, int value,
            final WriteSession session, LinkedBuffer lb)
    {
        final int tagSize = computeRawVarint32Size(tag);
        final int size = computeRawVarint32Size(value);
        final int totalSize = tagSize + size;

        if (lb.offset + totalSize > lb.buffer.length)
            lb = new LinkedBuffer(session.nextBufferSize, lb);

        final byte[] buffer = lb.buffer;
        int offset = lb.offset;
        lb.offset += totalSize;
        session.size += totalSize;

        if (tagSize == 1)
            buffer[offset++] = (byte) tag;
        else
        {
            for (int i = 0, last = tagSize - 1; i < last; i++, tag >>>= 7)
                buffer[offset++] = (byte) ((tag & 0x7F) | 0x80);

            buffer[offset++] = (byte) tag;
        }

        if (size == 1)
            buffer[offset] = (byte) value;
        else
        {
            for (int i = 0, last = size - 1; i < last; i++, value >>>= 7)
                buffer[offset++] = (byte) ((value & 0x7F) | 0x80);

            buffer[offset] = (byte) value;
        }

        return lb;
    }

tag是通过makeTag方法创建的：

public static int makeTag(final int fieldNumber, final int wireType)
    {
        return (fieldNumber << TAG_TYPE_BITS) | wireType;
    }

fieldNumber每个字段的标号，wire_type是该字段的数据类型，所有如果我们改变了业务对象类中字段的顺序，或者改变了字段的类型，都会出现反序列化失败；

前面提到的数据压缩在方法computeRawVarint32Size中体现出来了：

public static int computeRawVarint32Size(final int value)
    {
        if ((value & (0xffffffff << 7)) == 0)
            return 1;
        if ((value & (0xffffffff << 14)) == 0)
            return 2;
        if ((value & (0xffffffff << 21)) == 0)
            return 3;
        if ((value & (0xffffffff << 28)) == 0)
            return 4;
        return 5;
    }

根据value值的范围，返回不同的字节数；接下来的代码也可以看到检查LinkedBuffer的空间是否足够，不够进行扩充；接下来的代码就是用压缩的方式将tag和Value存入缓存中。

总结

大致了解了Protostuff对业务对象序列化的过程，不管是简单的测试还是通过查看源码，都可以发现Protostuff的序列化方式是完全借鉴Protobuf来实现的。