TiDBPingCAP 2011-10-12
hessian的主要结构分客户端与服务端,中间基于http传输。客户端主要做的事情是把对远程接口调用序列化为流,并传输到服务端;服务端主要做的事情是把传输过来的流反序列化为对服务的请求,调用相应服务后把结果序列化为流返回给客户端。一次完整的调用如下图所示:
HessianProxy是hessian client处理客户端请求的核心类,它采用proxy的设计模式,代理客户端对远程接口的调用,hessian client的主流程的时序图如下所示:
HessianSkeleton是hessian server端的核心类,从输入流中返序列化出客户端调用的方法和参数,对服务端服务进行调用,然后把处理结果返回给客户端,主要流程时序图如下所示:
Hessian在客户端一块采用Proxy模式,当客户端调用远程接口时,HessianProxy会代理这个动作,在invoke方法中,把客户端请求的方法和参数序列化为预订格式的输出流,主要流程如下图所示:
下面我将详细解析一下invoke源码:
public Object invoke(Object proxy, Method method, Object []args) throws Throwable { String mangleName; synchronized (_mangleMap) { mangleName = _mangleMap.get(method); } if (mangleName == null) { String methodName = method.getName(); Class []params = method.getParameterTypes(); // equals and hashCode are special cased if (methodName.equals("equals") && params.length == 1 && params[0].equals(Object.class)) { Object value = args[0]; if (value == null || ! Proxy.isProxyClass(value.getClass())) return new Boolean(false); HessianProxy handler = (HessianProxy) Proxy.getInvocationHandler(value); return new Boolean(_url.equals(handler.getURL())); } else if (methodName.equals("hashCode") && params.length == 0) return new Integer(_url.hashCode()); else if (methodName.equals("getHessianType")) return proxy.getClass().getInterfaces()[0].getName(); else if (methodName.equals("getHessianURL")) return _url.toString(); else if (methodName.equals("toString") && params.length == 0) return "HessianProxy[" + _url + "]"; if (! _factory.isOverloadEnabled()) mangleName = method.getName(); else mangleName = mangleName(method); synchronized (_mangleMap) { _mangleMap.put(method, mangleName); } } ...... }
这是invoke的开头,主要干的事情是把methodName缓存起来和过滤一些特殊调用,java反射是个比较耗性能的操作,把methodName缓存起来可以避免每次调用都要从method里得到methodName。另外,对equals、hashCode、getHessianType、getHessianURL等特殊方法的远程调用,HessianProxy不会走远程调用,而是直接返回。
接着往下看:
public Object invoke(Object proxy, Method method, Object []args) throws Throwable { ...... InputStream is = null; URLConnection conn = null; HttpURLConnection httpConn = null; try { conn = sendRequest(mangleName, args); ...... } catch (HessianProtocolException e) { throw new HessianRuntimeException(e); } finally { ...... } }
这段主要是把方法名和参数序列化为网络输出流,并做网络请求,具体逻辑包装在sendRequest方法中:
protected URLConnection sendRequest(String methodName, Object []args) throws IOException { URLConnection conn = null; conn = _factory.openConnection(_url); boolean isValid = false; try { // Used chunked mode when available, i.e. JDK 1.5. if (_factory.isChunkedPost() && conn instanceof HttpURLConnection) { try { HttpURLConnection httpConn = (HttpURLConnection) conn; httpConn.setChunkedStreamingMode(8 * 1024); } catch (Throwable e) { } } addRequestHeaders(conn); OutputStream os = null; try { os = conn.getOutputStream(); } catch (Exception e) { throw new HessianRuntimeException(e); } if (log.isLoggable(Level.FINEST)) { PrintWriter dbg = new PrintWriter(new LogWriter(log)); os = new HessianDebugOutputStream(os, dbg); } AbstractHessianOutput out = _factory.getHessianOutput(os); out.call(methodName, args); out.flush(); isValid = true; return conn; } finally { if (! isValid && conn instanceof HttpURLConnection) ((HttpURLConnection) conn).disconnect(); } }
sendRequest的主要流程是先初始化网络连接,然后用AbstractHessianOutput包装网络输出流,调用AbstractHessianOutput.call(methodName, args)完成网络输出,这个方法的细节会在hessian io的源码解析中详细分析。
下面接着看invoke方法:
public Object invoke(Object proxy, Method method, Object []args) throws Throwable { ...... is = conn.getInputStream(); AbstractHessianInput in = _factory.getHessianInput(is); in.startReply(); Object value = in.readObject(method.getReturnType()); if (value instanceof InputStream) { value = new ResultInputStream(httpConn, is, in, (InputStream) value); is = null; httpConn = null; } else in.completeReply(); ...... }
这一段主要是把输入流中取得返回值,具体是用AbstractHessianInput包装网络输入流,然后调用AbstractHessianInput.readObject从网络流中反序列化到实际返回值。
3.Hessian源码分析(三)------ HessianSkeletonHessianSkeleton是Hessian server端的核心类,主要功能是接收网络输入流(被包装为AbstractHessianInput),反序列化输入流得到methodName和参数,然后调用服务端的服务,得到结果后序列化为输出流,返回给客户端,主要流程如下图所示:
HessianSkeleton的核心代码如下所示:
public void invoke(Object service, AbstractHessianInput in, AbstractHessianOutput out) throws Exception { ...... String methodName = in.readMethod(); Method method = getMethod(methodName); ...... Class []args = method.getParameterTypes(); Object []values = new Object[args.length]; for (int i = 0; i < args.length; i++) { values[i] = in.readObject(args[i]); } Object result = null; try { result = method.invoke(service, values); } catch (Throwable e) { ...... } // The complete call needs to be after the invoke to handle a // trailing InputStream in.completeCall(); out.startReply(); out.writeObject(result); out.completeReply(); out.close(); }
主流程代码非常清晰,不需要太多解释,关键的地方在于对网络流的序列化和反序列化,我会在hessian io分析的部分中进行详细阐述
原文出自:http://blog.csdn.net/cutesource/article/details/4946968