【原】Kryo序列化篇

Kryo是一個快速有效的對象圖序列化Java庫。它的目標是快速、高效、易使用。該項目適用於對象持久化到文件或數據庫中或經過網絡傳輸。Kryo還能夠自動實現深淺的拷貝/克隆。 就是直接複製一個對象對象到另外一個對象，而不是對象轉換爲字節而後轉化爲對象。

目前已經被用在下列項目中：
KryoNet (NIO networking)
Twitter's Scalding (Scala API for Cascading)
Twitter's Chill (Kryo serializers for Scala)
Apache Fluo (Kryo is default serialization for Fluo Recipes)
Apache Hive (query plan serialization)
Apache Spark (shuffled/cached data serialization)
DataNucleus (JDO/JPA persistence framework)
CloudPelican
Yahoo's S4 (distributed stream computing)
Storm (distributed realtime computation system, in turn used by many others)
Cascalog (Clojure/Java data processing and querying details)
memcached-session-manager (Tomcat high-availability sessions)
Mobility-RPC (RPC enabling distributed applications)
akka-kryo-serialization (Kryo serializers for Akka)
Groupon
Jive
DestroyAllHumans (controls a robot!)
kryo-serializers (additional serializers)

How

Kryo序列化實例：

Kryo kryo = new Kryo();
    // ...
    Output output = new Output(new FileOutputStream("file.bin"));
    SomeClass someObject = ...
    kryo.writeObject(output, someObject);
    output.close();
    // ...
    Input input = new Input(new FileInputStream("file.bin"));
    SomeClass someObject = kryo.readObject(input, SomeClass.class);
    input.close();

Kryo默認爲不一樣的數據類型定義了不一樣的序列化器，其具體類型以下所示：

col1	col2	col3	col4	col5
boolean	Boolean	byte	Byte	char
Character	short	Short	int	Integer
long	Long	float	Float	double
Double	byte[]	String	BigInteger	BigDecimal
Collection	Date	Collections.emptyList	Collections.singleton	Map
StringBuilder	TreeMap	Collections.emptyMap	Collections.emptySet	KryoSerializable
StringBuffer	Class	Collections.singletonList	Collections.singletonMap	Currency
Calendar	TimeZone	Enum	EnumSet

Kryo類在構造參數中會爲上述表中的類型初始化對應的序列化類。

其它類會使用默認序列化器FieldSerializer，固然你也能夠經過下面的代碼更改默認序列化器：

Kryo kryo = new Kryo();
kryo.setDefaultSerializer(AnotherGenericSerializer.class);

也能夠爲每一個類指定序列化類，以下

Kryo kryo = new Kryo();
kryo.register(SomeClass.class, new SomeSerializer());
kryo.register(AnotherClass.class, new AnotherSerializer());

Why

1.下面經過如下Kryo4.0源代碼看一下
Kryo類的構造函數爲：

public Kryo (ClassResolver classResolver, ReferenceResolver referenceResolver, StreamFactory streamFactory) {
        if (classResolver == null) throw new IllegalArgumentException("classResolver cannot be null.");

        this.classResolver = classResolver;
        classResolver.setKryo(this);

        this.streamFactory = streamFactory;
        streamFactory.setKryo(this);

        this.referenceResolver = referenceResolver;
        if (referenceResolver != null) {
            referenceResolver.setKryo(this);
            references = true;
        }
//private final ArrayList<DefaultSerializerEntry> defaultSerializers = new ArrayList(33);用一個大小爲33的Entry來存放類及序列化器
        addDefaultSerializer(byte[].class, ByteArraySerializer.class);
        addDefaultSerializer(char[].class, CharArraySerializer.class);
        addDefaultSerializer(short[].class, ShortArraySerializer.class);
        addDefaultSerializer(int[].class, IntArraySerializer.class);
        addDefaultSerializer(long[].class, LongArraySerializer.class);
        addDefaultSerializer(float[].class, FloatArraySerializer.class);
        addDefaultSerializer(double[].class, DoubleArraySerializer.class);
        addDefaultSerializer(boolean[].class, BooleanArraySerializer.class);
        addDefaultSerializer(String[].class, StringArraySerializer.class);
        addDefaultSerializer(Object[].class, ObjectArraySerializer.class);
        addDefaultSerializer(KryoSerializable.class, KryoSerializableSerializer.class);
        addDefaultSerializer(BigInteger.class, BigIntegerSerializer.class);
        addDefaultSerializer(BigDecimal.class, BigDecimalSerializer.class);
        addDefaultSerializer(Class.class, ClassSerializer.class);
        addDefaultSerializer(Date.class, DateSerializer.class);
        addDefaultSerializer(Enum.class, EnumSerializer.class);
        addDefaultSerializer(EnumSet.class, EnumSetSerializer.class);
        addDefaultSerializer(Currency.class, CurrencySerializer.class);
        addDefaultSerializer(StringBuffer.class, StringBufferSerializer.class);
        addDefaultSerializer(StringBuilder.class, StringBuilderSerializer.class);
        addDefaultSerializer(Collections.EMPTY_LIST.getClass(), CollectionsEmptyListSerializer.class);
        addDefaultSerializer(Collections.EMPTY_MAP.getClass(), CollectionsEmptyMapSerializer.class);
        addDefaultSerializer(Collections.EMPTY_SET.getClass(), CollectionsEmptySetSerializer.class);
        addDefaultSerializer(Collections.singletonList(null).getClass(), CollectionsSingletonListSerializer.class);
        addDefaultSerializer(Collections.singletonMap(null, null).getClass(), CollectionsSingletonMapSerializer.class);
        addDefaultSerializer(Collections.singleton(null).getClass(), CollectionsSingletonSetSerializer.class);
        addDefaultSerializer(TreeSet.class, TreeSetSerializer.class);
        addDefaultSerializer(Collection.class, CollectionSerializer.class);
        addDefaultSerializer(TreeMap.class, TreeMapSerializer.class);
        addDefaultSerializer(Map.class, MapSerializer.class);
        addDefaultSerializer(TimeZone.class, TimeZoneSerializer.class);
        addDefaultSerializer(Calendar.class, CalendarSerializer.class);
        addDefaultSerializer(Locale.class, LocaleSerializer.class);
        addDefaultSerializer(Charset.class, CharsetSerializer.class);
        addDefaultSerializer(URL.class, URLSerializer.class);
        OptionalSerializers.addDefaultSerializers(this);
        TimeSerializers.addDefaultSerializers(this);
        lowPriorityDefaultSerializerCount = defaultSerializers.size();

        // Primitives and string. Primitive wrappers automatically use the same registration as primitives.
        register(int.class, new IntSerializer());
        register(String.class, new StringSerializer());
        register(float.class, new FloatSerializer());
        register(boolean.class, new BooleanSerializer());
        register(byte.class, new ByteSerializer());
        register(char.class, new CharSerializer());
        register(short.class, new ShortSerializer());
        register(long.class, new LongSerializer());
        register(double.class, new DoubleSerializer());
        register(void.class, new VoidSerializer());
    }

• Kryo：可看作類與序列化類的Map集合
• ClassResolver ：負責類的註冊、根據類標識序列化、根據字節碼獲得類標識
• ReferenceResolver接口 ：當容許引用時，該類用於跟蹤已經讀寫的對象，爲寫對象提供ID，根據ID讀取對象。
  它有兩個實現子類：ListReferenceResolver和MapReferenceResolver
• StreamFactory接口 ：基於系統配置提供輸入和輸出流。
  它有兩個實現子類：DefaultStreamFactory和FastestStreamFactory

2.爲何在使用一個類以前先要將Kryo註冊一下？
在序列化時寫一個類名是低效的，因此用一個ID來替代類名，只須要使用前完成註冊便可，代碼以下：

//用先一個可用的整型ID來註冊該類，若是類已經被註冊，那麼就返回以前的註冊信息類
public Registration register (Class type) {
        Registration registration = classResolver.getRegistration(type);
        if (registration != null) return registration;
        return register(type, getDefaultSerializer(type));
    }

//和上面的基本同樣，只不過是本身定義ID
public Registration register (Class type, int id) {
        Registration registration = classResolver.getRegistration(type);
        if (registration != null) return registration;
        return register(type, getDefaultSerializer(type), id);
    }

參考資料

• 官網文檔
• 官網文檔中文翻譯版
• 爲何Java序列化慢
• Unsafe

QA

官方文檔中有一段話：

Using Unsafe-based IO may result in a quite significant performance boost (sometimes up-to an order of magnitude), depending on your application. In particular, it helps a lot when serializing large primitive arrays as part of your object graphs.

下面是答案：

Unsafe-based IO uses methods from sun.misc.Unsafe for reading and writing from/to memory. Many of those methods map almost 1:1 to processor instructions. Moreover, reading/writing of arrays of native types is executed in bulk instead of doing it element-by-element as it is usually done by Kryo. Together, these features often provide a significant performance boost.