摘要:最近要做一个将K8s中的某组件UI通过反向代理映射到自定义规则的链接地址上,提供给用户访问的需求。所以顺便研究了一下Jetty的ProxyServlet。
本文分享自华为云社区《Jetty自定义ProxyServlet实现反向代理服务(含源码分析)》,作者: 小焱 。
一、背景概述
最近要做一个将K8s中的某组件UI通过反向代理映射到自定义规则的链接地址上,提供给用户访问的需求。所以顺便研究了一下Jetty的ProxyServlet。在这里做一下分析,如果有理解不到位的还希望可以补充指正。
二、Jetty 的基本架构
Jetty 是一个Servlet 引擎,它的架构比较简单,也是一个可扩展性和非常灵活的应用服务器,它有一个基本数据模型,这个数据模型就是 Handler,所有可以被扩展的组件都可以作为一个 Handler,添加到 Server 中,Jetty 就是帮你管理这些 Handler。
整个 Jetty 的核心组件由 Server 和 Connector 两个组件构成,整个 Server 组件是基于 Handler 容器工作的,Jetty 中另外一个比不可少的组件是 Connector,它负责接受客户端的连接请求,并将请求分配给一个处理队列去执行。Jetty 中还有一些可有可无的组件,我们可以在它上做扩展。如 JMX,我们可以定义一些 Mbean 把它加到 Server 中,当 Server 启动的时候,这些 Bean 就会一起工作。
整个 Jetty 的核心是围绕着 Server 类来构建,Server 类继承了 Handler,关联了 Connector 和 Container。Container 是管理 Mbean 的容器。Jetty 的 Server 的扩展主要是实现一个个 Handler 并将 Handler 加到 Server 中,Server 中提供了调用这些 Handler 的访问规则。整个 Jetty 的所有组件的生命周期管理是基于观察者模板设计,实现LifeCycle。
三、Handler 的体系结构
Jetty 主要是基于 Handler 来设计的,Handler 的体系结构影响着整个 Jetty 的方方面面。下面总结了一下 Handler 的种类及作用:
Jetty 主要提供了两种 Handler 类型,一种是 HandlerWrapper,它可以将一个 Handler 委托给另外一个类去执行,如我们要将一个 Handler 加到 Jetty 中,那么就必须将这个 Handler 委托给 Server 去调用。配合 ScopeHandler 类我们可以拦截 Handler 的执行,在调用 Handler 之前或之后,可以做一些另外的事情,类似于 Tomcat 中的 Valve;另外一个 Handler 类型是 HandlerCollection,这个 Handler 类可以将多个 Handler 组装在一起,构成一个 Handler 链,方便我们做扩展。
四、编码设计
这里我提供一个设计框架,具体内容可以根据需要自定义。
public class RestApi { private static final Logging LOGGER = Logging.getLogging(RestApi.class.getName()); private Server server; /** * 启动方法,需要在程序启动时调用该方法 */ public void start() { try { ContextHandlerCollection collection = new ContextHandlerCollection(); WebAppContext appContext = new WebAppContext(); appContext.setContextPath("/"); // 设置资源文件地址,可略 appContext.setResourceBase("/opt/appHome/myDemo/webapp"); // 设置web.xml,可在里面进行一些Servlet配置,可略 appContext.setDescriptor("/opt/appHome/myDemo/webapp/WEB-INF/web.xml"); appContext.setParentLoaderPriority(true); collection.addHandler(appContext); addProxyHandler(collection); server = new Server(8080); server.setHandler(collection); server.start(); } catch (Throwable t) { LOGGER.error("Start RESTful API server failed", t); } } private static void addProxyHandler(ContextHandlerCollection collection) { ProxyServlet proxyServlet = new WebProxyServlet(); // 添加自定义ProxyServlet ServletHolder holder = new ServletHolder(proxyServlet); holder.setInitParameter("idleTimeout", 120000); // 设置空闲释放时间 holder.setInitParameter("timeout", 300000); // 设置超时时间 holder.setInitParameter("maxConnections", 256); // 设置最大连接数 ServletContextHandler contextHandler = new ServletContextHandler(); contextHandler.addServlet(holder, "/proxy/*"); contextHandler.setContextPath("/demo"); collection.addHandler(contextHandler); } }
自定义 ProxyServlet,在此列出一部分常用的重写方法,还有很多方法可以查询文档自行重写
public class WebProxyServlet extends ProxyServlet { private static final Logging LOGGING = Logging.getLogging(WebProxyServlet.class); /** * 自定义目标地址重写方法 */ @Override protected String rewriteTarget(HttpServletRequest request) { } /** * 自定义重写错误处理方法 */ @Override protected void onProxyRewriteFailed(HttpServletRequest clientRequest, HttpServletResponse clientResponse) { } /** * 自定义response错误处理方法 */ @Override protected void onProxyResponseFailure( HttpServletRequest clientRequest, HttpServletResponse proxyResponse, Response serverResponse, Throwable failure) { } /** * 自定义response头filter */ @Override protected String filterServerResponseHeader( HttpServletRequest clientRequest, Response serverResponse, String headerName, String headerValue) { } /** * 自定义头XForwarded设置 */ @Override protected void addXForwardedHeaders(HttpServletRequest clientRequest, Request proxyRequest) { } }
五、请求处理过程
下面通过跟踪源码初步梳理了一下,从 request 请求进入到返回 response 的整个流程
六、源码分析
1、Request 转发部分
当 Jetty 接收到一个请求时,Jetty 就把这个请求交给在 Server 中注册的 ContextHandlerCollection 去执行,查看 Service 的 handle 方法源码
public void handle(HttpChannel channel) throws IOException, ServletException { String target = channel.getRequest().getPathInfo(); Request request = channel.getRequest(); Response response = channel.getResponse(); if (LOG.isDebugEnabled()) { LOG.debug("{} {} {} on {}", new Object[]{request.getDispatcherType(), request.getMethod(), target, channel}); } if (!HttpMethod.OPTIONS.is(request.getMethod()) && !"*".equals(target)) { this.handle(target, request, request, response); } else if (!HttpMethod.OPTIONS.is(request.getMethod())) { request.setHandled(true); response.sendError(400); } else { this.handleOptions(request, response); if (!request.isHandled()) { this.handle(target, request, request, response); } } if (LOG.isDebugEnabled()) { LOG.debug("handled={} async={} committed={} on {}", new Object[]{request.isHandled(), request.isAsyncStarted(), response.isCommitted(), channel}); } }
这里调用的 this.handle(target, request, request, response) 方法其实是父类 HandlerWrapper 的 handle 方法
public void handle(String target, Request baseRequest, HttpServletRequest request, HttpServletResponse response) throws IOException, ServletException { Handler handler = this._handler; if (handler != null) { handler.handle(target, baseRequest, request, response); } }
创建 server 时曾调用过 server.setHandler(collection) ,所以这里就调用到了 ContextHandlerCollection 的 handle 方法
public void handle(String target, Request baseRequest, HttpServletRequest request, HttpServletResponse response) throws IOException, ServletException { ContextHandlerCollection.Mapping mapping = (ContextHandlerCollection.Mapping)this._handlers.get(); if (mapping != null) { Handler[] handlers = mapping.getHandlers(); if (handlers != null && handlers.length != 0) { if (handlers.length == 1) { handlers[0].handle(target, baseRequest, request, response); } else { HttpChannelState async = baseRequest.getHttpChannelState(); if (async.isAsync()) { ContextHandler context = async.getContextHandler(); if (context != null) { Handler branch = (Handler)mapping._contextBranches.get(context); if (branch == null) { context.handle(target, baseRequest, request, response); } else { branch.handle(target, baseRequest, request, response); } return; } } int limit; if (target.startsWith("/")) { Trie<Entry<String, ContextHandlerCollection.Branch[]>> pathBranches = mapping._pathBranches; if (pathBranches == null) { return; } int l; for(limit = target.length() - 1; limit >= 0; limit = l - 2) { Entry<String, ContextHandlerCollection.Branch[]> branches = (Entry)pathBranches.getBest(target, 1, limit); if (branches == null) { break; } l = ((String)branches.getKey()).length(); if (l == 1 || target.length() == l || target.charAt(l) == '/') { ContextHandlerCollection.Branch[] var12 = (ContextHandlerCollection.Branch[])branches.getValue(); int var13 = var12.length; for(int var14 = 0; var14 < var13; ++var14) { ContextHandlerCollection.Branch branch = var12[var14]; branch.getHandler().handle(target, baseRequest, request, response); if (baseRequest.isHandled()) { return; } } } } } else { Handler[] var17 = handlers; limit = handlers.length; for(int var19 = 0; var19 < limit; ++var19) { Handler handler = var17[var19]; handler.handle(target, baseRequest, request, response); if (baseRequest.isHandled()) { return; } } } } } } }
从上面的源码可以看出 ContextHandlerCollection 的 handle 方法继续调用了 collection.addHandler 设置进来 ServletContextHandler 的 handle 方法,通过跟踪,可以找到其实这里调用了父类 ScopedHandler 的 handle --> doScope --> nextScope
public final void handle(String target, Request baseRequest, HttpServletRequest request, HttpServletResponse response) throws IOException, ServletException { if (this.isStarted()) { if (this._outerScope == null) { this.doScope(target, baseRequest, request, response); } else { this.doHandle(target, baseRequest, request, response); } } } public void doScope(String target, Request baseRequest, HttpServletRequest request, HttpServletResponse response) throws IOException, ServletException { this.nextScope(target, baseRequest, request, response); } public final void nextScope(String target, Request baseRequest, HttpServletRequest request, HttpServletResponse response) throws IOException, ServletException { if (this._nextScope != null) { this._nextScope.doScope(target, baseRequest, request, response); } else if (this._outerScope != null) { this._outerScope.doHandle(target, baseRequest, request, response); } else { this.doHandle(target, baseRequest, request, response); } }
查看 ServletContextHandler 可以找到主要注册了以下三个 handler,均为 ScopedHandler 的子类,也就是 nextScope 方法中的 this._nextScope
protected SessionHandler _sessionHandler; protected SecurityHandler _securityHandler; protected ServletHandler _servletHandler;
SessionHandler 是对 ServletHandler 进行了一层包装(装饰器模式),用于一些session的预处理什么的,而SecurityHandler从名字分析是做一些安全相关的,这两个具体就不分析了,直接来看 ServletHandler 的 doScope 方法
public void doScope(String target, Request baseRequest, HttpServletRequest request, HttpServletResponse response) throws IOException, ServletException { String old_servlet_path = baseRequest.getServletPath(); String old_path_info = baseRequest.getPathInfo(); DispatcherType type = baseRequest.getDispatcherType(); ServletHolder servletHolder = null; Scope oldScope = null; MappedResource<ServletHolder> mapping = this.getMappedServlet(target); if (mapping != null) { servletHolder = (ServletHolder)mapping.getResource(); if (mapping.getPathSpec() != null) { PathSpec pathSpec = mapping.getPathSpec(); String servletPath = pathSpec.getPathMatch(target); String pathInfo = pathSpec.getPathInfo(target); if (DispatcherType.INCLUDE.equals(type)) { baseRequest.setAttribute("javax.servlet.include.servlet_path", servletPath); baseRequest.setAttribute("javax.servlet.include.path_info", pathInfo); } else { baseRequest.setServletPath(servletPath); baseRequest.setPathInfo(pathInfo); } } } if (LOG.isDebugEnabled()) { LOG.debug("servlet {}|{}|{} -> {}", new Object[]{baseRequest.getContextPath(), baseRequest.getServletPath(), baseRequest.getPathInfo(), servletHolder}); } try { oldScope = baseRequest.getUserIdentityScope(); baseRequest.setUserIdentityScope(servletHolder); this.nextScope(target, baseRequest, request, response); } finally { if (oldScope != null) { baseRequest.setUserIdentityScope(oldScope); } if (!DispatcherType.INCLUDE.equals(type)) { baseRequest.setServletPath(old_servlet_path); baseRequest.setPathInfo(old_path_info); } } }
这里对 baseRequest 做了一些设置,将注册进来的 ServletHolder set 进了 baseRequest,之后又继续调用了 this.nextScope(target, baseRequest, request, response) ,根据上面的 nextScope 方法,所有 scope 执行完,则执行 doHandle 方法,继续跳过 SessionHandler 和 SecurityHandler,来看下ServletHandler 的 doHandle 方法
public void doHandle(String target, Request baseRequest, HttpServletRequest request, HttpServletResponse response) throws IOException, ServletException { ServletHolder servletHolder = (ServletHolder)baseRequest.getUserIdentityScope(); FilterChain chain = null; if (servletHolder != null && this._filterMappings != null && this._filterMappings.length > 0) { chain = this.getFilterChain(baseRequest, target.startsWith("/") ? target : null, servletHolder); } if (LOG.isDebugEnabled()) { LOG.debug("chain={}", new Object[]{chain}); } try { if (servletHolder == null) { this.notFound(baseRequest, request, response); } else { ServletRequest req = request; if (request instanceof ServletRequestHttpWrapper) { req = ((ServletRequestHttpWrapper)request).getRequest(); } ServletResponse res = response; if (response instanceof ServletResponseHttpWrapper) { res = ((ServletResponseHttpWrapper)response).getResponse(); } servletHolder.prepare(baseRequest, (ServletRequest)req, (ServletResponse)res); if (chain != null) { chain.doFilter((ServletRequest)req, (ServletResponse)res); } else { servletHolder.handle(baseRequest, (ServletRequest)req, (ServletResponse)res); } } } finally { if (servletHolder != null) { baseRequest.setHandled(true); } } }
doHandle 主要是取出注册的 FilterChain ServletHolder,如果存在 Filter,先执行 chain.doFilter 方法,否则执行 servletHolder.handle 我没有设置 filter 所有就直接看 ServletHolder 的 handle 方法了
public void handle(Request baseRequest, ServletRequest request, ServletResponse response) throws ServletException, UnavailableException, IOException { try { Servlet servlet = this.getServletInstance(); if (servlet == null) { throw new UnavailableException("Servlet Not Initialized"); } servlet.service(request, response); } catch (UnavailableException var5) { this.makeUnavailable(var5).service(request, response); } }
这里调用了 ServletHolder 中 Servlet 的 service 方法,也就是走到了我们自定义类 WebProxyServlet 类,因为没有重写,所以这里调用的是 ProxyServlet 的 service 方法
protected void service(HttpServletRequest request, HttpServletResponse response) throws ServletException, IOException { int requestId = this.getRequestId(request); String rewrittenTarget = this.rewriteTarget(request); if (this._log.isDebugEnabled()) { StringBuffer uri = request.getRequestURL(); if (request.getQueryString() != null) { uri.append("?").append(request.getQueryString()); } if (this._log.isDebugEnabled()) { this._log.debug("{} rewriting: {} -> {}", new Object[]{requestId, uri, rewrittenTarget}); } } if (rewrittenTarget == null) { this.onProxyRewriteFailed(request, response); } else { Request proxyRequest = this.newProxyRequest(request, rewrittenTarget); this.copyRequestHeaders(request, proxyRequest); this.addProxyHeaders(request, proxyRequest); AsyncContext asyncContext = request.startAsync(); asyncContext.setTimeout(0L); proxyRequest.timeout(this.getTimeout(), TimeUnit.MILLISECONDS); if (this.hasContent(request)) { if (this.expects100Continue(request)) { DeferredContentProvider deferred = new DeferredContentProvider(new ByteBuffer[0]); proxyRequest.content(deferred); proxyRequest.attribute(CONTINUE_ACTION_ATTRIBUTE, () -> { try { ContentProvider provider = this.proxyRequestContent(request, response, proxyRequest); (new ProxyServlet.DelegatingContentProvider(request, proxyRequest, response, provider, deferred)).iterate(); } catch (Throwable var6) { this.onClientRequestFailure(request, proxyRequest, response, var6); } }); } else { proxyRequest.content(this.proxyRequestContent(request, response, proxyRequest)); } } this.sendProxyRequest(request, response, proxyRequest); } }
至此调用到了我们重写的最关键的方法 rewriteTarget 此方法可以自定义逻辑将 request 的地址解析,返回要代理到的目标地址,使用目标地址组成proxyRequest 最后调用 sendProxyRequest 实现代理转发。
2、Response 接收部分
如果继续跟 sendProxyRequest 会看到创建了一个 ProxyResponseListener,这里具体就不详细跟踪了,主要讲一下流程,有兴趣的可以自行动手看一下。Response 返回会通过反射机制触发 onHeader 方法 ProxyServlet 重写了该方法并跳转到了 onServerResponseHeaders 方法
public void onHeaders(Response proxyResponse) { ProxyServlet.this.onServerResponseHeaders(this.request, this.response, proxyResponse); }
这个方法是设置 Response 的 header 内容的,其中有获取 headerValue 调用了 this.filterServerResponseHeader 方法,我们也可以通过重写此方法自定义返回体的 headerValue。
七、总结
到这里 Jetty 的 ProxyServlet 运行原理和自定义方法大致梳理完毕。还有许多漏掉的和理解不到位的地方,希望大家可以提出指正。工作中偶尔抽出一点时间读一下源码,既可以提升对所用技术的理解,又可以学习欣赏这些框架的巧妙设计,还是非常有意义的。