Netty 模拟 Redis 服务器
ty061032
8年前
<p>Redis的客户端与服务端采用叫做 <strong>RESP(Redis Serialization Protocol)</strong> 的网络通信协议交换数据,客户端和服务器通过 TCP 连接来进行数据交互, 服务器默认的端口号为 6379 。客户端和服务器发送的命令或数据一律以 <strong>\r\n (CRLF)</strong> 结尾。</p> <p>RESP支持五种数据类型:</p> <p>状态回复(status reply): 以“+”开头,表示正确的状态信息,”+”后就是具体信息,比如:</p> <pre> <code class="language-java">redis 127.0.0.1:6379> set ss sdf OK</code></pre> <p>其实它真正回复的数据是:+OK\r\n</p> <p>错误回复(error reply):</p> <p>以”-“开头,表示错误的状态信息,”-“后就是具体信息,比如:</p> <pre> <code class="language-java">redis 127.0.0.1:6379> incr ss (error) ERR value is not an integer or out of range</code></pre> <p>整数回复(integer reply): 以”:”开头,表示对某些操作的回复比如DEL, EXISTS, INCR等等</p> <pre> <code class="language-java">redis 127.0.0.1:6379> incr aa (integer) 1</code></pre> <p>批量回复(bulk reply): 以”$”开头,表示下一行的字符串长度,具体字符串在下一行中</p> <p>多条批量回复(multi bulk reply): 以”*”开头,表示消息体总共有多少行(不包括当前行)”*”是具体行数</p> <pre> <code class="language-java">redis 127.0.0.1:6379> get ss "sdf" 客户端->服务器 *2\r\n $3\r\n get\r\n $2\r\n ss\r\n 服务器->客户端 $3\r\n sdf\r\n</code></pre> <p>注:以上写的都是XX回复,并不是说协议格式只是适用于服务器->客户端,客户端->服务器端也同样使用以上协议格式,其实双端协议格式的统一更加方便扩展</p> <p>回到正题,我们这里是通过netty来模拟redis服务器,可以整理一下思路大概分为这么几步:</p> <pre> <code class="language-java">1.需要一个底层的通信框架,这里选择的是netty4.0.25 2.需要对客户端穿过来的数据进行解码(Decoder),其实就是分别处理以上5种数据类型 3.解码以后我们封装成更加利于理解的命令(Command),比如:set<name> foo hello<params> 4.有了命令以后就是处理命令(execute),其实我们可以去连接正在的redis服务器,不过这里只是简单的模拟 5.处理完之后就是封装回复(Reply),然后编码(Encoder),需要根据不同的命令分别返回以后5种数据类型 6.测试验证,通过redis-cli去连接netty模拟的redis服务器,看能否返回正确的结果</code></pre> <p>以上思路参考github上的一个项目: <a href="/misc/goto?guid=4959713578171838898" rel="nofollow,noindex">https://github.com/spullara/redis-protocol</a> ,测试代码也是在此基础上做了一个简化</p> <p>第一步:通信框架netty</p> <pre> <code class="language-java"><dependency> <groupId>io.netty</groupId> <artifactId>netty-all</artifactId> <version>4.0.25.Final</version> </dependency></code></pre> <p>第二步:数据类型解码</p> <pre> <code class="language-java">public class RedisCommandDecoder extends ReplayingDecoder<Void> { public static final char CR = '\r'; public static final char LF = '\n'; public static final byte DOLLAR_BYTE = '$'; public static final byte ASTERISK_BYTE = '*'; private byte[][] bytes; private int arguments = 0; @Override protected void decode(ChannelHandlerContext ctx, ByteBuf in, List<Object> out) throws Exception { if (bytes != null) { int numArgs = bytes.length; for (int i = arguments; i < numArgs; i++) { if (in.readByte() == DOLLAR_BYTE) { int l = RedisReplyDecoder.readInt(in); if (l > Integer.MAX_VALUE) { throw new IllegalArgumentException( "Java only supports arrays up to " + Integer.MAX_VALUE + " in size"); } int size = (int) l; bytes[i] = new byte[size]; in.readBytes(bytes[i]); if (in.bytesBefore((byte) CR) != 0) { throw new RedisException("Argument doesn't end in CRLF"); } // Skip CRLF(\r\n) in.skipBytes(2); arguments++; checkpoint(); } else { throw new IOException("Unexpected character"); } } try { out.add(new Command(bytes)); } finally { bytes = null; arguments = 0; } } else if (in.readByte() == ASTERISK_BYTE) { int l = RedisReplyDecoder.readInt(in); if (l > Integer.MAX_VALUE) { throw new IllegalArgumentException( "Java only supports arrays up to " + Integer.MAX_VALUE + " in size"); } int numArgs = (int) l; if (numArgs < 0) { throw new RedisException("Invalid size: " + numArgs); } bytes = new byte[numArgs][]; checkpoint(); decode(ctx, in, out); } else { in.readerIndex(in.readerIndex() - 1); byte[][] b = new byte[1][]; b[0] = in.readBytes(in.bytesBefore((byte) CR)).array(); in.skipBytes(2); out.add(new Command(b, true)); } } }</code></pre> <p>首先通过接受到以“*”开头的多条批量类型初始化二维数组byte[][] bytes,以读取到第一个以\r\n结尾的数据作为数组的长度,然后再处理以“$”开头的批量类型。</p> <p>以上除了处理我们熟悉的批量和多条批量类型外,还处理了没有任何标识的数据,其实有一个专门的名字叫 <strong> Inline命令: </strong></p> <p>有些时候仅仅是telnet连接Redis服务,或者是仅仅向Redis服务发送一个命令进行检测。虽然Redis协议可以很容易的实现,但是使用Interactive sessions 并不理想,而且redis-cli也不总是可以使用。基于这些原因,Redis支持特殊的命令来实现上面描述的情况。这些命令的设计是很人性化的,被称作Inline 命令。</p> <p>第三步:封装command对象</p> <p>由第二步中可以看到不管是commandName还是params都统一放在了字节二维数组里面,最后封装在command对象里面</p> <pre> <code class="language-java">public class Command { public static final byte[] EMPTY_BYTES = new byte[0]; private final Object name; private final Object[] objects; private final boolean inline; public Command(Object[] objects) { this(null, objects, false); } public Command(Object[] objects, boolean inline) { this(null, objects, inline); } private Command(Object name, Object[] objects, boolean inline) { this.name = name; this.objects = objects; this.inline = inline; } public byte[] getName() { if (name != null) return getBytes(name); return getBytes(objects[0]); } public boolean isInline() { return inline; } private byte[] getBytes(Object object) { byte[] argument; if (object == null) { argument = EMPTY_BYTES; } else if (object instanceof byte[]) { argument = (byte[]) object; } else if (object instanceof ByteBuf) { argument = ((ByteBuf) object).array(); } else if (object instanceof String) { argument = ((String) object).getBytes(Charsets.UTF_8); } else { argument = object.toString().getBytes(Charsets.UTF_8); } return argument; } public void toArguments(Object[] arguments, Class<?>[] types) { for (int position = 0; position < types.length; position++) { if (position >= arguments.length) { throw new IllegalArgumentException( "wrong number of arguments for '" + new String(getName()) + "' command"); } if (objects.length - 1 > position) { arguments[position] = objects[1 + position]; } } } }</code></pre> <p>所有的数据都放在了Object数组里面,而且可以通过getName方法知道Object[0]就是commandName</p> <p>第四步:执行命令</p> <p>在经历了解码和封装之后,下面需要实现handler类,用来处理消息</p> <pre> <code class="language-java">public class RedisCommandHandler extends SimpleChannelInboundHandler<Command> { private Map<String, Wrapper> methods = new HashMap<String, Wrapper>(); interface Wrapper { Reply<?> execute(Command command) throws RedisException; } public RedisCommandHandler(final RedisServer rs) { Class<? extends RedisServer> aClass = rs.getClass(); for (final Method method : aClass.getMethods()) { final Class<?>[] types = method.getParameterTypes(); methods.put(method.getName(), new Wrapper() { @Override public Reply<?> execute(Command command) throws RedisException { Object[] objects = new Object[types.length]; try { command.toArguments(objects, types); return (Reply<?>) method.invoke(rs, objects); } catch (Exception e) { return new ErrorReply("ERR " + e.getMessage()); } } }); } } @Override public void channelReadComplete(ChannelHandlerContext ctx) throws Exception { ctx.flush(); } @Override protected void channelRead0(ChannelHandlerContext ctx, Command msg) throws Exception { String name = new String(msg.getName()); Wrapper wrapper = methods.get(name); Reply<?> reply; if (wrapper == null) { reply = new ErrorReply("unknown command '" + name + "'"); } else { reply = wrapper.execute(msg); } if (reply == StatusReply.QUIT) { ctx.close(); } else { if (msg.isInline()) { if (reply == null) { reply = new InlineReply(null); } else { reply = new InlineReply(reply.data()); } } if (reply == null) { reply = ErrorReply.NYI_REPLY; } ctx.write(reply); } } }</code></pre> <p>在实例化handler的时候传入了一个RedisServer对象,这个方法是真正用来处理redis命令的,理论上这个对象应该支持redis的所有命令,不过这里只是测试所有只提供了2个方法:</p> <pre> <code class="language-java">public interface RedisServer { public BulkReply get(byte[] key0) throws RedisException; public StatusReply set(byte[] key0, byte[] value1) throws RedisException; }</code></pre> <p>在channelRead0方法中我们可以拿到之前封装好的command方法,然后通过命令名称执行操作,这里的RedisServer也很简单,只是用简单的hashmap进行临时的保存数据。</p> <p>第五步:封装回复</p> <p>第四步种我们可以看到处理完命令之后,返回了一个Reply对象</p> <pre> <code class="language-java">public interface Reply<T> { byte[] CRLF = new byte[] { RedisReplyDecoder.CR, RedisReplyDecoder.LF }; T data(); void write(ByteBuf os) throws IOException; }</code></pre> <p>根据上面提到的5种类型再加上一个inline命令,根据不同的数据格式进行拼接,比如StatusReply:</p> <pre> <code class="language-java">public void write(ByteBuf os) throws IOException { os.writeByte('+'); os.writeBytes(statusBytes); os.writeBytes(CRLF); }</code></pre> <p>所以对应Decoder的Encoder就很简单了:</p> <pre> <code class="language-java">public class RedisReplyEncoder extends MessageToByteEncoder<Reply<?>> { @Override public void encode(ChannelHandlerContext ctx, Reply<?> msg, ByteBuf out) throws Exception { msg.write(out); } }</code></pre> <p>只需要将封装好的Reply返回给客户端就行了</p> <p>最后一步:测试</p> <p>启动类:</p> <pre> <code class="language-java">public class Main { private static Integer port = 6379; public static void main(String[] args) throws InterruptedException { final RedisCommandHandler commandHandler = new RedisCommandHandler( new SimpleRedisServer()); ServerBootstrap b = new ServerBootstrap(); final DefaultEventExecutorGroup group = new DefaultEventExecutorGroup(1); try { b.group(new NioEventLoopGroup(), new NioEventLoopGroup()) .channel(NioServerSocketChannel.class) .option(ChannelOption.SO_BACKLOG, 100).localAddress(port) .childOption(ChannelOption.TCP_NODELAY, true) .childHandler(new ChannelInitializer<SocketChannel>() { @Override public void initChannel(SocketChannel ch) throws Exception { ChannelPipeline p = ch.pipeline(); p.addLast(new RedisCommandDecoder()); p.addLast(new RedisReplyEncoder()); p.addLast(group, commandHandler); } }); ChannelFuture f = b.bind().sync(); f.channel().closeFuture().sync(); } finally { group.shutdownGracefully(); } } }</code></pre> <p>ChannelPipeline分别添加了RedisCommandDecoder、RedisReplyEncoder和RedisCommandHandler,同时我们启动的端口和Redis服务器端口是一样的也是 <strong>6379</strong></p> <p>打开redis-cli程序:</p> <pre> <code class="language-java">redis 127.0.0.1:6379> get dsf (nil) redis 127.0.0.1:6379> set dsf dsfds OK redis 127.0.0.1:6379> get dsf "dsfds" redis 127.0.0.1:6379></code></pre> <p>从结果可以看出和正常使用redis服务器没有差别</p> <p>总结</p> <p>这样做的意义其实就是可以把它当做一个 <strong>redis代理</strong> ,由这个代理服务器去进行sharding处理,客户端不直接访问redis服务器,对客户端来说,后台redis集群是完全透明的。</p> <p> </p> <p>来自:http://my.oschina.net/OutOfMemory/blog/738865</p> <p> </p>