标签：des style blog color os io 使用 ar for
leveldb是一个非常高效的可嵌入式K-V数据库,在.NET下有着基于win实现的包装leveldb.net;不过leveldb.net只提供了基于byte[]和string的处理,这显然会对使用的时候带来不方便,毕竟在编写应用的时候都是希望通过对象的方式来存储,如我们常见的redis,mongodb和memcached等等都提供对象方式的读写.以下主要讲解leveldb.net基础上封装一层序列化功能方便使用.
为了不修改leveldb.net的代码,所以选择在他基础过行封装,为了清楚需要些什么简单地定义了一个规则
public interface IDBManager { IFormater Formater { get; set; } void Set(string key, object data); object Get(string key, Type type); T Get<T>(string key); void Open(); LevelDB.DB DataBase { get; } }
代码非常简单主要封装了GET,SET,实际上还有DELETE操作,这里偷懒就没做了:),为了提供灵活序列化规则所以在这个管理接口上还提供了一个Formater属性.下面是这相接口的详细实现:
public class LevelDBManager : IDBManager { public LevelDBManager() { } private LevelDB.DB mDataBase; public string Path { get; set; } public IFormater Formater { get; set; } public void Open() { mDataBase = new LevelDB.DB(Path, new Options() { CreateIfMissing = true }); } public void Set(string key, object data) { FormaterBuffer buffer = Formater.Pop(); try { int count = Formater.Serialize(data, buffer, 0); mDataBase.Put(Encoding.UTF8.GetBytes(key), buffer.Array, 0, count); } finally { Formater.Push(buffer); } } public object Get(string key, Type type) { FormaterBuffer buffer = Formater.Pop(); long count; object result = null; try { count = mDataBase.Get(Encoding.UTF8.GetBytes(key), buffer.Array); if (count > 0) { result = Formater.Deserialize(type, buffer, 0, (int)count); } return result; } finally { Formater.Push(buffer); } } public T Get<T>(string key) { return (T)Get(key, typeof(T)); } public DB DataBase { get { return mDataBase; } } }
相信以上那些简知的代码也比较好理解,所以就不详细说明了.
由于在使用上的需要,都习惯用些不同序列化方式来进行对象序列化,这个封装为了实现一个比较高的灵活度,所以对象序列化过程也制定了一个接口进行隔离.主要为了满足不同人的胃口.
public interface IFormater { FormaterBuffer Pop(); void Push(FormaterBuffer data); int Serialize(object data, FormaterBuffer buffer, int offset); object Deserialize(Type type, FormaterBuffer buffer, int offset, int count); }
比较简单定义了序列化和反序列化的方法,不过为了一些性能上的考虑增加了buffer的复用功能,这个设计紧紧用作需要追求这方面性能要求而准备.下面看一下json和protobuf的实现是怎样的:
public abstract class FormaterBase:IFormater { private Stack<FormaterBuffer> mBufferPool = new Stack<FormaterBuffer>(); const int BUFFER_SIZE = 1024 * 1024 * 1; public FormaterBase() { for (int i = 0; i < 20; i++) { mBufferPool.Push(new FormaterBuffer(BUFFER_SIZE)); } } public FormaterBuffer Pop() { lock (mBufferPool) { if(mBufferPool.Count>0) return mBufferPool.Pop(); return new FormaterBuffer(BUFFER_SIZE); } } public void Push(FormaterBuffer data) { lock (mBufferPool) { mBufferPool.Push(data); } } public abstract int Serialize(object data, FormaterBuffer buffer, int offset); public abstract object Deserialize(Type type, FormaterBuffer buffer, int offset, int count); }
public class JsnoFormater:FormaterBase { public int Serialize(object data, byte[] buffer, int offset) { string json = Newtonsoft.Json.JsonConvert.SerializeObject(data); return Encoding.UTF8.GetBytes(json, 0, json.Length, buffer, offset); } public override int Serialize(object data, FormaterBuffer buffer, int offset) { string json = Newtonsoft.Json.JsonConvert.SerializeObject(data); return Encoding.UTF8.GetBytes(json, 0, json.Length, buffer.Array, offset); } public override object Deserialize(Type type, FormaterBuffer buffer, int offset, int count) { string value = Encoding.UTF8.GetString(buffer.Array, offset, count); return Newtonsoft.Json.JsonConvert.DeserializeObject(value, type); } }
public class ProtobufFormater:FormaterBase { public override int Serialize(object data, FormaterBuffer buffer, int offset) { buffer.Seek(offset); ProtoBuf.Meta.RuntimeTypeModel.Default.Serialize(buffer.Stream, data); return (int)buffer.Stream.Position; } public override object Deserialize(Type type, FormaterBuffer buffer, int offset, int count) { buffer.Stream.SetLength(count + offset); buffer.Seek(offset); return ProtoBuf.Meta.RuntimeTypeModel.Default.Deserialize(buffer.Stream, null, type); } }
虽然leveldb.net只以win dll的基础上包装,但在包装过程的确有些方法针对我个人来说做得并不理想,主要体现在buffer复用方面.其实get,set方法都存在这情况.
/// <summary> /// Set the database entry for "key" to "value". /// </summary> public void Put(byte[] key, byte[] value, WriteOptions options) { IntPtr error; LevelDBInterop.leveldb_put(this.Handle, options.Handle, key, (IntPtr)key.Length, value, (IntPtr)value.LongLength, out error); LevelDBException.Check(error); GC.KeepAlive(options); GC.KeepAlive(this); } public unsafe byte[] Get(byte[] key, ReadOptions options) { IntPtr error; IntPtr lengthPtr; var valuePtr = LevelDBInterop.leveldb_get(this.Handle, options.Handle, key, (IntPtr)key.Length, out lengthPtr, out error); LevelDBException.Check(error); if (valuePtr == IntPtr.Zero) return null; try { var length = (long)lengthPtr; var value = new byte[length]; var valueNative = (byte*)valuePtr.ToPointer(); for (long i = 0; i < length; ++i) value[i] = valueNative[i]; return value; } finally { LevelDBInterop.leveldb_free(valuePtr); GC.KeepAlive(options); GC.KeepAlive(this); } }
两上个方法都不支持从外部带入buffer的情况,当需要高并发操作的情况而对象序列化内容又比较大的情况下,那的确是会让人感觉到不满意.所以在这基础上添加了一些有利于buffer复用的方法来支持高并发操作下的性能需要.
public void Put(byte[] key, byte[] value, int offset, int length, WriteOptions options) { IntPtr error; LevelDBInterop.leveldb_put(this.Handle, options.Handle, key, (IntPtr)key.Length, value, (IntPtr)length, out error); LevelDBException.Check(error); GC.KeepAlive(options); GC.KeepAlive(this); } public unsafe long Get(byte[] key, byte[] buffer, ReadOptions options) { IntPtr error; IntPtr lengthPtr; var valuePtr = LevelDBInterop.leveldb_get(this.Handle, options.Handle, key, (IntPtr)key.Length, out lengthPtr, out error); LevelDBException.Check(error); if (valuePtr == IntPtr.Zero) return 0; try { var length = (long)lengthPtr; var valueNative = (byte*)valuePtr.ToPointer(); Marshal.Copy((IntPtr)valuePtr, buffer, 0, (int)length); return length; } finally { LevelDBInterop.leveldb_free(valuePtr); } }
标签：des style blog color os io 使用 ar for
原文地址：http://www.cnblogs.com/smark/p/3953577.html
