用solr做项目已经有一年有余，但都是使用层面，只是利用solr现有机制，修改参数，然后监控调优，从没有对solr进行源码级别的研究。但是，最近手头的一个项目，让我感觉必须把solrn内部原理和扩展机制弄熟，才能把这个项目做好。今天分享的就是：Solr是如何启动并且初始化的。大家知道，部署solr时，分两部分：一、solr的配置文件。二、solr相关的程序、插件、依赖lucene相关的jar包、日志方面的jar。因此，在研究solr也可以顺着这个思路：加载配置文件、初始化各个core、初始化各个core中的requesthandler...
研究solr的启动，首先从solr war程序的web.xml分析开始，下面是solr的web.xml片段：
<web-app xmlns="http://java.sun.com/xml/ns/javaee"xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"xsi:schemaLocation="http://java.sun.com/xml/ns/javaee http://java.sun.com/xml/ns/javaee/web-app_2_5.xsd"version="2.5"metadata-complete="true"
><!-- Uncomment if you are trying to use a Resin version before 3.0.19.Their XML implementation isn't entirely compatible with Xerces.Below are the implementations to use with Sun's JVM.<system-property javax.xml.xpath.XPathFactory="com.sun.org.apache.xpath.internal.jaxp.XPathFactoryImpl"/><system-property javax.xml.parsers.DocumentBuilderFactory="com.sun.org.apache.xerces.internal.jaxp.DocumentBuilderFactoryImpl"/><system-property javax.xml.parsers.SAXParserFactory="com.sun.org.apache.xerces.internal.jaxp.SAXParserFactoryImpl"/>--><!-- People who want to hardcode their "Solr Home" directly into theWAR File can set the JNDI property here...--><!-- Solr配置文件的参数,用于Solr初始化使用 --><env-entry><env-entry-name>solr/home</env-entry-name><env-entry-value>R:/solrhome1/solr</env-entry-value><env-entry-type>java.lang.String</env-entry-type></env-entry><!-- org.apache.solr.servlet.SolrDispatchFilter Solr启动最重要的东东，所以针对solr源码分析，要对这个Filter开始，它主要的作用：加载solr配置文件、初始化各个core、初始化各个requestHandler和component --><filter><filter-name>SolrRequestFilter</filter-name><filter-class>org.apache.solr.servlet.SolrDispatchFilter</filter-class><!-- If you are wiring Solr into a larger web application which controlsthe web context root, you will probably want to mount Solr undera path prefix (app.war with /app/solr mounted into it, for example).You will need to put this prefix in front of the SolrDispatchFilterurl-pattern mapping too (/solr/*), and also on any paths forlegacy Solr servlet mappings you may be using.For the Admin UI to work properly in a path-prefixed configuration,the admin folder containing the resources needs to be under the app context rootnamed to match the path-prefix. For example:.warxxxjsmain.js--><!--<init-param><param-name>path-prefix</param-name><param-value>/xxx</param-value></init-param>--></filter>
SolrDispatchFilter 是继承BaseSolrFilter的一个Filter（Filter的作用是啥，大家应该清楚吧，一般web框架级别的产品源码分析都是从filter或者servlet开始）。在介绍SolrDispatchFilter之前，先介绍一下BaseSolrFilter（也许程序员都有刨根问底的习惯）。BaseSolrFilter，是一个实现Filter接口的抽象类，功能很简单，就是判断当前程序是否已经加载日志方面的jar。代码片段如下：
/*** All Solr filters available to the user's webapp should* extend this class and not just implement {@link Filter}.* This class ensures that the logging configuration is correct* before any Solr specific code is executed.*/
abstract class BaseSolrFilter implements Filter {static {//CheckLoggingConfiguration.check();}}
着于篇幅，我就不介绍CheckLoggingConfiguration.check() 这里面的东东了。OK，我们回到SolrDispatchFilter上。由于BaseSolrFilter是一个抽象类，所有作为非抽象类的SolrDispatchFilter必须要实现Filter接口。Filter接口如下：
public interface Filter {//进行初始化public void init(FilterConfig filterConfig) throws ServletException;//拦截所有的http请求public void doFilter(ServletRequest request, ServletResponse response,FilterChain chain)throws IOException, ServletException;//进行注销的动作public void destroy();
}
根据上面的注释，我们知道在init方法中是进行初始化的。因此，今天咱们研究SolrDispatchFilter是如何初始化，是离不开这个方法的。接下来，咱们看看SolrDispatchFilter的init方法吧：
@Overridepublic void init(FilterConfig config) throws ServletException{log.info("SolrDispatchFilter.init()");try {// web.xml configurationthis.pathPrefix = config.getInitParameter( "path-prefix" );//各位看客，乾坤尽在此方法中this.cores = createCoreContainer();log.info("user.dir=" + System.getProperty("user.dir"));}catch( Throwable t ) {// catch this so our filter still workslog.error( "Could not start Solr. Check solr/home property and the logs");SolrCore.log( t );if (t instanceof Error) {throw (Error) t;}}log.info("SolrDispatchFilter.init() done");}
咱们顺藤摸瓜，来看看createCoreContainer这个方法到底干了些什么。
protected CoreContainer createCoreContainer() {
//看好了SolrResourceLoader 是用来加载solr home中的配置文件文件的 SolrResourceLoader loader = new SolrResourceLoader(SolrResourceLoader.locateSolrHome());//加载配置文件
ConfigSolr config = loadConfigSolr(loader);CoreContainer cores = new CoreContainer(loader, config);
//初始化Core cores.load();return cores;}
createCoreContainer这个方法是决定咱们今天能否弄懂Solr初始化和启动的关键。我们顺便简单分析一下这个方法中用到的几个类和方法：
SolrResourceLoader 类如其名，是solr资源加载器。
ConfigSolr 是通过SolrResourceLoader来读取solr配置文件的中信息的。
loadConfigSolr，加载配置信息的方法：
private ConfigSolr loadConfigSolr(SolrResourceLoader loader) {//优先读取solr.solrxml.location配置的信息，往往是通过读取zookeeper中的配置信息进行初始化的，如果没有配置，就会读取solrhome配置项配置的信息（记得web.xml第一个配置项否，就是它）String solrxmlLocation = System.getProperty("solr.solrxml.location", "solrhome");if (solrxmlLocation == null || "solrhome".equalsIgnoreCase(solrxmlLocation))return ConfigSolr.fromSolrHome(loader, loader.getInstanceDir());//ok 从zookeeper中读取配置信息吧，这是在solrcloud集群下用来solr初始化的if ("zookeeper".equalsIgnoreCase(solrxmlLocation)) {String zkHost = System.getProperty("zkHost");log.info("Trying to read solr.xml from " + zkHost);if (StringUtils.isEmpty(zkHost))throw new SolrException(ErrorCode.SERVER_ERROR,"Could not load solr.xml from zookeeper: zkHost system property not set");SolrZkClient zkClient = new SolrZkClient(zkHost, 30000);try {if (!zkClient.exists("/solr.xml", true))//solr.xml里有描述的zookeeper相关的配置信息throw new SolrException(ErrorCode.SERVER_ERROR, "Could not load solr.xml from zookeeper: node not found");byte[] data = zkClient.getData("/solr.xml", null, null, true);
//加载配置信息return ConfigSolr.fromInputStream(loader, new ByteArrayInputStream(data));} catch (Exception e) {throw new SolrException(ErrorCode.SERVER_ERROR, "Could not load solr.xml from zookeeper", e);} finally {zkClient.close();//关闭zookeeper连接}}throw new SolrException(ErrorCode.SERVER_ERROR,"Bad solr.solrxml.location set: " + solrxmlLocation + " - should be 'solrhome' or 'zookeeper'");}
CoreContainer 就是进行Core初始化工作的。我们主要看看load方法吧，这段方法有点长，代码如下：
public void load() {log.info("Loading cores into CoreContainer [instanceDir={}]", loader.getInstanceDir());//加载solr共享jar包库// add the sharedLib to the shared resource loader before initializing cfg based pluginsString libDir = cfg.getSharedLibDirectory();if (libDir != null) {File f = FileUtils.resolvePath(new File(solrHome), libDir);log.info("loading shared library: " + f.getAbsolutePath());
//对classloader不熟的，可以进去看看loader.addToClassLoader(libDir, null, false);loader.reloadLuceneSPI();}//分片相关的handler加载以及初始化shardHandlerFactory = ShardHandlerFactory.newInstance(cfg.getShardHandlerFactoryPluginInfo(), loader);updateShardHandler = new UpdateShardHandler(cfg);solrCores.allocateLazyCores(cfg.getTransientCacheSize(), loader);logging = LogWatcher.newRegisteredLogWatcher(cfg.getLogWatcherConfig(), loader);hostName = cfg.getHost();log.info("Host Name: " + hostName);zkSys.initZooKeeper(this, solrHome, cfg);collectionsHandler = createHandler(cfg.getCollectionsHandlerClass(), CollectionsHandler.class);infoHandler = createHandler(cfg.getInfoHandlerClass(), InfoHandler.class);coreAdminHandler = createHandler(cfg.getCoreAdminHandlerClass(), CoreAdminHandler.class);//zookeeper 配置信息初始化solr corecoreConfigService = cfg.createCoreConfigService(loader, zkSys.getZkController());containerProperties = cfg.getSolrProperties("solr");// setup executor to load cores in parallel// do not limit the size of the executor in zk mode since cores may try and wait for each other.
//多线程初始化core 不熟悉多线的可以驻足研究一会ExecutorService coreLoadExecutor = Executors.newFixedThreadPool(( zkSys.getZkController() == null ? cfg.getCoreLoadThreadCount() : Integer.MAX_VALUE ),new DefaultSolrThreadFactory("coreLoadExecutor") );try {CompletionService<SolrCore> completionService = new ExecutorCompletionService<>(coreLoadExecutor);Set<Future<SolrCore>> pending = new HashSet<>();List<CoreDescriptor> cds = coresLocator.discover(this);checkForDuplicateCoreNames(cds);for (final CoreDescriptor cd : cds) {final String name = cd.getName();try {if (cd.isTransient() || ! cd.isLoadOnStartup()) {// Store it away for later use. includes non-transient but not// loaded at startup cores.solrCores.putDynamicDescriptor(name, cd);}if (cd.isLoadOnStartup()) { // The normal caseCallable<SolrCore> task = new Callable<SolrCore>() {@Overridepublic SolrCore call() {SolrCore c = null;try {if (zkSys.getZkController() != null) {//zookeeper模式preRegisterInZk(cd);}c = create(cd);//普通创建模式registerCore(cd.isTransient(), name, c, false, false);} catch (Exception e) {SolrException.log(log, null, e);try {/* if (isZooKeeperAware()) {try {zkSys.zkController.unregister(name, cd);} catch (InterruptedException e2) {Thread.currentThread().interrupt();SolrException.log(log, null, e2);} catch (KeeperException e3) {SolrException.log(log, null, e3);}}*/} finally {if (c != null) {c.close();}} }return c;}};pending.add(completionService.submit(task));}} catch (Exception e) {SolrException.log(log, null, e);}}while (pending != null && pending.size() > 0) {try {//获取创建完成的coreFuture<SolrCore> future = completionService.take();if (future == null) return;pending.remove(future);try {SolrCore c = future.get();// track original namesif (c != null) {solrCores.putCoreToOrigName(c, c.getName());}} catch (ExecutionException e) {SolrException.log(SolrCore.log, "Error loading core", e);}} catch (InterruptedException e) {throw new SolrException(SolrException.ErrorCode.SERVICE_UNAVAILABLE,"interrupted while loading core", e);}}
//solr core的守护线程，在容器关闭或者启动失败的时候，进行资源注销// Start the background threadbackgroundCloser = new CloserThread(this, solrCores, cfg);backgroundCloser.start();} finally {if (coreLoadExecutor != null) {
//初始化完成，关闭线程池ExecutorUtil.shutdownNowAndAwaitTermination(coreLoadExecutor);}}if (isZooKeeperAware()) {//如果zookeeper可用 也就是solrcloud模式// register in zk in background threadsCollection<SolrCore> cores = getCores();if (cores != null) {for (SolrCore core : cores) {try {
//讲core的状态信息注册到zookeeper中zkSys.registerInZk(core, true);} catch (Throwable t) {SolrException.log(log, "Error registering SolrCore", t);}}}
//zkSys.getZkController().checkOverseerDesignate();}}
在这段代码，关键部分我都做了注释。当你需要优化你的solr启动速度时，你还会来研究这段代码。下面，我们将研究solr的请求过滤处理的部分，我们需要关注doFilter那个方法了（关键部分我作以注释，就不细讲了）：
if( abortErrorMessage != null ) {//500错误处理((HttpServletResponse)response).sendError( 500, abortErrorMessage );return;}if (this.cores == null) {//solr core初始化失败或者已经关闭((HttpServletResponse)response).sendError( 503, "Server is shutting down or failed to initialize" );return;}CoreContainer cores = this.cores;SolrCore core = null;SolrQueryRequest solrReq = null;Aliases aliases = null;if( request instanceof HttpServletRequest) {//如果是http请求HttpServletRequest req = (HttpServletRequest)request;HttpServletResponse resp = (HttpServletResponse)response;SolrRequestHandler handler = null;String corename = "";String origCorename = null;try {// put the core container in request attributereq.setAttribute("org.apache.solr.CoreContainer", cores);String path = req.getServletPath();if( req.getPathInfo() != null ) {// this lets you handle /update/commit when /update is a servletpath += req.getPathInfo();}if( pathPrefix != null && path.startsWith( pathPrefix ) ) {path = path.substring( pathPrefix.length() );}// check for management pathString alternate = cores.getManagementPath();if (alternate != null && path.startsWith(alternate)) {path = path.substring(0, alternate.length());}// unused feature ?int idx = path.indexOf( ':' );if( idx > 0 ) {// save the portion after the ':' for a 'handler' path parameterpath = path.substring( 0, idx );}// Check for the core admin pageif( path.equals( cores.getAdminPath() ) ) {//solr admin 管理页面请求handler = cores.getMultiCoreHandler();solrReq = SolrRequestParsers.DEFAULT.parse(null,path, req);handleAdminRequest(req, response, handler, solrReq);return;}boolean usingAliases = false;List<String> collectionsList = null;// Check for the core admin collections urlif( path.equals( "/admin/collections" ) ) {//管理collections handler = cores.getCollectionsHandler();solrReq = SolrRequestParsers.DEFAULT.parse(null,path, req);handleAdminRequest(req, response, handler, solrReq);return;}// Check for the core admin info urlif( path.startsWith( "/admin/info" ) ) {//查看admin infohandler = cores.getInfoHandler();solrReq = SolrRequestParsers.DEFAULT.parse(null,path, req);handleAdminRequest(req, response, handler, solrReq);return;}else {//otherwise, we should find a core from the pathidx = path.indexOf( "/", 1 );if( idx > 1 ) {// try to get the corename as a request parameter firstcorename = path.substring( 1, idx );// look at aliasesif (cores.isZooKeeperAware()) {//solr cloud状态origCorename = corename;ZkStateReader reader = cores.getZkController().getZkStateReader();aliases = reader.getAliases();if (aliases != null && aliases.collectionAliasSize() > 0) {usingAliases = true;String alias = aliases.getCollectionAlias(corename);if (alias != null) {collectionsList = StrUtils.splitSmart(alias, ",", true);corename = collectionsList.get(0);}}}core = cores.getCore(corename);if (core != null) {path = path.substring( idx );}}if (core == null) {if (!cores.isZooKeeperAware() ) {core = cores.getCore("");}}}if (core == null && cores.isZooKeeperAware()) {// we couldn't find the core - lets make sure a collection was not specified insteadcore = getCoreByCollection(cores, corename, path);if (core != null) {// we found a core, update the pathpath = path.substring( idx );}// if we couldn't find it locally, look on other nodesif (core == null && idx > 0) {String coreUrl = getRemotCoreUrl(cores, corename, origCorename);// don't proxy for internal update requestsSolrParams queryParams = SolrRequestParsers.parseQueryString(req.getQueryString());if (coreUrl != null&& queryParams.get(DistributingUpdateProcessorFactory.DISTRIB_UPDATE_PARAM) == null) {path = path.substring(idx);remoteQuery(coreUrl + path, req, solrReq, resp);return;} else {if (!retry) {// we couldn't find a core to work with, try reloading aliases// TODO: it would be nice if admin ui elements skipped this...ZkStateReader reader = cores.getZkController().getZkStateReader();reader.updateAliases();doFilter(request, response, chain, true);return;}}}// try the default coreif (core == null) {core = cores.getCore("");}}// With a valid core...if( core != null ) {//验证corefinal SolrConfig config = core.getSolrConfig();// get or create/cache the parser for the coreSolrRequestParsers parser = config.getRequestParsers();// Handle /schema/* and /config/* paths via Restletif( path.equals("/schema") || path.startsWith("/schema/")|| path.equals("/config") || path.startsWith("/config/")) {//solr rest api 入口 solrReq = parser.parse(core, path, req);SolrRequestInfo.setRequestInfo(new SolrRequestInfo(solrReq, new SolrQueryResponse()));if( path.equals(req.getServletPath()) ) {// avoid endless loop - pass through to Restlet via webappchain.doFilter(request, response);} else {// forward rewritten URI (without path prefix and core/collection name) to Restletreq.getRequestDispatcher(path).forward(request, response);}return;}// Determine the handler from the url path if not set// (we might already have selected the cores handler)if( handler == null && path.length() > 1 ) { // don't match "" or "/" as valid pathhandler = core.getRequestHandler( path );// no handler yet but allowed to handle select; let's checkif( handler == null && parser.isHandleSelect() ) {if( "/select".equals( path ) || "/select/".equals( path ) ) {//solr 各种查询过滤入口 solrReq = parser.parse( core, path, req );String qt = solrReq.getParams().get( CommonParams.QT );handler = core.getRequestHandler( qt );if( handler == null ) {throw new SolrException( SolrException.ErrorCode.BAD_REQUEST, "unknown handler: "+qt);}if( qt != null && qt.startsWith("/") && (handler instanceof ContentStreamHandlerBase)) {//For security reasons it's a bad idea to allow a leading '/', ex: /select?qt=/update see SOLR-3161//There was no restriction from Solr 1.4 thru 3.5 and it's not supported for update handlers.throw new SolrException( SolrException.ErrorCode.BAD_REQUEST, "Invalid Request Handler ('qt'). Do not use /select to access: "+qt);}}}}// With a valid handler and a valid core...if( handler != null ) {// if not a /select, create the requestif( solrReq == null ) {solrReq = parser.parse( core, path, req );}if (usingAliases) {processAliases(solrReq, aliases, collectionsList);}final Method reqMethod = Method.getMethod(req.getMethod());HttpCacheHeaderUtil.setCacheControlHeader(config, resp, reqMethod);// unless we have been explicitly told not to, do cache validation// if we fail cache validation, execute the queryif (config.getHttpCachingConfig().isNever304() ||!HttpCacheHeaderUtil.doCacheHeaderValidation(solrReq, req, reqMethod, resp)) {//solr http 缓存 在header控制失效时间的方式SolrQueryResponse solrRsp = new SolrQueryResponse();/* even for HEAD requests, we need to execute the handler to* ensure we don't get an error (and to make sure the correct* QueryResponseWriter is selected and we get the correct* Content-Type)*/SolrRequestInfo.setRequestInfo(new SolrRequestInfo(solrReq, solrRsp));this.execute( req, handler, solrReq, solrRsp );HttpCacheHeaderUtil.checkHttpCachingVeto(solrRsp, resp, reqMethod);// add info to http headers//TODO: See SOLR-232 and SOLR-267. /*try {NamedList solrRspHeader = solrRsp.getResponseHeader();for (int i=0; i<solrRspHeader.size(); i++) {((javax.servlet.http.HttpServletResponse) response).addHeader(("Solr-" + solrRspHeader.getName(i)), String.valueOf(solrRspHeader.getVal(i)));}} catch (ClassCastException cce) {log.log(Level.WARNING, "exception adding response header log information", cce);}*/QueryResponseWriter responseWriter = core.getQueryResponseWriter(solrReq);writeResponse(solrRsp, response, responseWriter, solrReq, reqMethod);}return; // we are done with a valid handler}}log.debug("no handler or core retrieved for " + path + ", follow through...");} catch (Throwable ex) {sendError( core, solrReq, request, (HttpServletResponse)response, ex );if (ex instanceof Error) {throw (Error) ex;}return;} finally {try {if (solrReq != null) {log.debug("Closing out SolrRequest: {}", solrReq);solrReq.close();}} finally {try {if (core != null) {core.close();}} finally {SolrRequestInfo.clearRequestInfo();}}}}// Otherwise let the webapp handle the requestchain.doFilter(request, response);}
文章转载请注明出处：http://www.cnblogs.com/likehua/p/4353608.html
