概述SPIService Provider Interface一种服务发现机制指一些提供给你继承、扩展完成自定义功能的类、接口或方法。在SPI机制中服务提供者为某个接口实现具体的类而在运行时通过SPI机制查找到对应的实现类并将其加载进行使用。JDK 6参考java.util.ServiceLoader类的注释Since字段引进的一个特性可实现动态加载具体实现类的机制通过它可以具体实现代码的解耦也可实现类似于IoC的效果。SPI的实现方式提供实现的实现类打包成jar文件jar文件里面必须有个META-INF/services目录其下有一个文本文件文件名即为SPI接口的全限定名完整路径名文件的内容是该jar包中提供的SPI接口的实现类名。文件编码是UTF-8。作用有两个把标准定义和接口实现分离在模块化开发中很好地实现解耦实现功能的扩展更好地满足定制化需求SPI的不足之处不能根据需求去加载扩展实现每次都会加载扩展接口的所有实现类并进行实例化实例化会造成性能开销并且加载一些不需要用到的实现类会导致内存资源的浪费。API和SPIAPI提供给他人使用的具备某项功能的类、接口或方法。SPI用于一些通用的标准中为标准的实现产商提供扩展点。标准在上层提供APIAPI内部使用SPI当API被客户使用时会动态得从当前运行的classpath中寻找该SPI的实现然后使用该SPI的实现来完成API的功能。实例JDBCjava.sql.Driver数据库驱动接口JDK中只提供接口的定义具体的实现类由各个数据库厂商提供的驱动包来完成程序在运行的时候会根据当前导入的驱动包来完成对应数据库的连接。如上图java.sql.Driver文件内容com.mysql.cj.jdbc.Driver。同理也能在postgresql-42.7.2.jar下找到java.sql.Driver文件内容为org.postgresql.Driver截图省略。com.mysql.cj.jdbc.Driver源码很简单publicclassDriverextendsNonRegisteringDriverimplementsjava.sql.Driver{// Register ourselves with the DriverManager.static{try{DriverManager.registerDriver(newDriver());}catch(SQLExceptionE){thrownewRuntimeException(Cant register driver!);}}publicDriver()throwsSQLException{}}核心方法就一行DriverManager.registerDriver(new Driver());在PG源码org.postgresql.Driver里也能找到DriverManager.registerDriver()。基于JDK 22源码在DriverManager类里搜索文件名java.sql.Driver找到方法ensureDriversInitialized/* * Load the initial JDBC drivers by checking the System property jdbc.drivers and then use the ServiceLoader mechanism */privatestaticvoidensureDriversInitialized(){// 检查驱动是否已经初始化成功if(driversInitialized){return;}synchronized(lockForInitDrivers){// 优先从环境变量jdbc.drivers获取驱动Stringdrivers;try{driversAccessController.doPrivileged(newPrivilegedActionString(){publicStringrun(){returnSystem.getProperty(jdbc.drivers);}});}catch(Exceptionex){driversnull;}// If the driver is packaged as a Service Provider, load it.// Get all the drivers through the classloader exposed as a java.sql.Driver.class service.// ServiceLoader.load() replaces the sun.misc.Providers()AccessController.doPrivileged(newPrivilegedActionVoid(){publicVoidrun(){// ServiceLoaderServiceLoaderDriverloadedDriversServiceLoader.load(Driver.class);IteratorDriverdriversIteratorloadedDrivers.iterator();/* 加载驱动然后初始化存在一种情况有java.sql.Driver文件但是文件内容为空 * 或找不到文件内容指向的Driver类用Throwable来catch java.util.ServiceConfigurationError */try{// TODO: loadedDrivers并没有赋值到driverswhile(driversIterator.hasNext()){driversIterator.next();}}catch(Throwablet){// Do nothing}returnnull;}});if(drivers!null!drivers.isEmpty()){String[]driversListdrivers.split(:);for(StringaDriver:driversList){try{Class.forName(aDriver,true,ClassLoader.getSystemClassLoader());}catch(Exceptionex){// just logging}}}// 设置驱动初始化成功标志driversInitializedtrue;}}ensureDriversInitialized()方法的调用方有getDriver()和getConnection()两个即获取驱动和获取连接。Spring在Spring源码里也有SPI的影子如spring.factories文件和SpringFactoriesLoader两者一起构成Spring Boot Starter实现的基础。以3.2.4版本来讲解位于spring-boot-actuator-autoconfigure-3.2.4.jar下的spring.factories文件片段# Failure Analyzersorg.springframework.boot.diagnostics.FailureAnalyzer\ org.springframework.boot.actuate.autoconfigure.metrics.ValidationFailureAnalyzer,\ org.springframework.boot.actuate.autoconfigure.health.NoSuchHealthContributorFailureAnalyzer可见此文件实际上是.properties文件格式不同类型的配置项用空行隔开方便阅读支持#作为注释字符。配置是键值对形式Key和Value都是代表类的完整包名Value允许有多个用,分隔多行文本以\分隔Key是接口、注解、或抽象类Value是Key的实现类。而SpringFactoriesLoader自Spring 3.2版本开始就已存在其属性如下// spring.factories文件默认检索位置publicstaticfinalStringFACTORIES_RESOURCE_LOCATIONMETA-INF/spring.factories;// 默认抛出异常privatestaticfinalFailureHandlerTHROWING_FAILURE_HANDLERFailureHandler.throwing();// ClassLoader级别的缓存Value也是MapK为资源目录V为全路径类staticfinalMapClassLoader,MapString,SpringFactoriesLoadercachenewConcurrentReferenceHashMap();// 用于加载类NullableprivatefinalClassLoaderclassLoader;// 用于保存spring.factories文件里的键值对privatefinalMapString,ListStringfactories;核心方法有两个forResourceLocation是一个静态方法用于获取实例publicstaticSpringFactoriesLoaderforResourceLocation(StringresourceLocation,NullableClassLoaderclassLoader){Assert.hasText(resourceLocation,resourceLocation must not be empty);ClassLoaderresourceClassLoader(classLoader!null?classLoader:SpringFactoriesLoader.class.getClassLoader());MapString,SpringFactoriesLoaderloaderscache.computeIfAbsent(resourceClassLoader,key-newConcurrentReferenceHashMap());returnloaders.computeIfAbsent(resourceLocation,key-newSpringFactoriesLoader(classLoader,loadFactoriesResource(resourceClassLoader,resourceLocation)));}参考resourceLocation可传入指定路径如果为空则从META-INF/spring.factories查找。load方法则用于获取指定类型的实现类集合publicTListTload(ClassTfactoryType,NullableArgumentResolverargumentResolver,NullableFailureHandlerfailureHandler){Assert.notNull(factoryType,factoryType must not be null);ListStringimplementationNamesloadFactoryNames(factoryType);ListTresultnewArrayList(implementationNames.size());FailureHandlerfailureHandlerToUse(failureHandler!null)?failureHandler:THROWING_FAILURE_HANDLER;for(StringimplementationName:implementationNames){TfactoryinstantiateFactory(implementationName,factoryType,argumentResolver,failureHandlerToUse);if(factory!null){result.add(factory);}}AnnotationAwareOrderComparator.sort(result);returnresult;}Dubbo打开dubbo-3.2.14源码发现META-INF/services目录下有个org.apache.dubbo.common.extension.LoadingStrategy文本文件内容如下org.apache.dubbo.common.extension.DubboInternalLoadingStrategy org.apache.dubbo.common.extension.DubboLoadingStrategy org.apache.dubbo.common.extension.ServicesLoadingStrategyDubboInternalLoadingStrategy用于查找META-INF/dubbo/internal/目录下的文件DubboLoadingStrategy用于查找META-INF/dubbo/目录下的文件ServicesLoadingStrategy用于查找META-INF/services/目录下的文件。META-INF/dubbo/internal/目录下有91个文件所有文件名都是Dubbo下的全路径名指向的接口类这些接口都添加有注解SPI(value , scope FRAMEWORK)文件内容则是多行键值对采用KeyValue形式Key是一个简称Value则是Dubbo下接口文件名对应的的实现类。SPI注解源码如下DocumentedRetention(RetentionPolicy.RUNTIME)Target({ElementType.TYPE})publicinterfaceSPI{Stringvalue()default;ExtensionScopescope()defaultExtensionScope.APPLICATION;}ExtensionScope是一个枚举类FRAMEWORKAPPLICATION默认值MODULESELFExtensionLoader是个泛型类入口方法是getExtensionpublicTgetExtension(Stringname,booleanwrap){checkDestroyed();if(StringUtils.isEmpty(name)){thrownewIllegalArgumentException(Extension name null);}if(true.equals(name)){// 获取默认拓展实现类returngetDefaultExtension();}StringcacheKeyname;if(!wrap){cacheKey_origin;}// Holder机制包装实现类缓存finalHolderObjectholdergetOrCreateHolder(cacheKey);Objectinstanceholder.get();// DCLif(instancenull){synchronized(holder){instanceholder.get();if(instancenull){// 核心方法创建实现类instancecreateExtension(name,wrap);holder.set(instance);}}}return(T)instance;}接下来看看createExtension方法privateTcreateExtension(Stringname,booleanwrap){Class?clazzgetExtensionClasses().get(name);if(clazznull||unacceptableExceptions.contains(name)){throwfindException(name);}try{Tinstance(T)extensionInstances.get(clazz);if(instancenull){extensionInstances.putIfAbsent(clazz,createExtensionInstance(clazz));instance(T)extensionInstances.get(clazz);// 实例化前处理instancepostProcessBeforeInitialization(instance,name);// 向实例中注入依赖injectExtension(instance);// 实例化后处理instancepostProcessAfterInitialization(instance,name);}if(wrap){ListClass?wrapperClassesListnewArrayList();if(cachedWrapperClasses!null){wrapperClassesList.addAll(cachedWrapperClasses);wrapperClassesList.sort(WrapperComparator.COMPARATOR);Collections.reverse(wrapperClassesList);}if(CollectionUtils.isNotEmpty(wrapperClassesList)){for(Class?wrapperClass:wrapperClassesList){WrapperwrapperwrapperClass.getAnnotation(Wrapper.class);booleanmatch(wrappernull)||((ArrayUtils.isEmpty(wrapper.matches())||ArrayUtils.contains(wrapper.matches(),name))!ArrayUtils.contains(wrapper.mismatches(),name));if(match){// 将当前instance作为参数传给Wrapper的构造方法并通过反射创建Wrapper实例向Wrapper实例中注入依赖最后将Wrapper实例再次赋值给instance变量instanceinjectExtension((T)wrapperClass.getConstructor(type).newInstance(instance));instancepostProcessAfterInitialization(instance,name);}}}}// Warning: After an instance of Lifecycle is wrapped by cachedWrapperClasses, it may not still be Lifecycle instance, this application may not invoke the lifecycle.initialize hook.initExtension(instance);returninstance;}catch(Throwablet){// 创建实现类失败thrownewIllegalStateException();}}核心方法是getExtensionClassesprivateMapString,Class?getExtensionClasses(){// 检查缓存MapString,Class?classescachedClasses.get();// DCLif(classesnull){synchronized(cachedClasses){classescachedClasses.get();if(classesnull){try{classesloadExtensionClasses();}catch(InterruptedExceptione){logger.error();thrownewIllegalStateException();}cachedClasses.set(classes);}}}returnclasses;}继续看loadExtensionClassesprivateMapString,Class?loadExtensionClasses()throwsInterruptedException{checkDestroyed();cacheDefaultExtensionName();MapString,Class?extensionClassesnewHashMap();for(LoadingStrategystrategy:strategies){loadDirectory(extensionClasses,strategy,type.getName());// compatible with old ExtensionFactoryif(this.typeExtensionInjector.class){loadDirectory(extensionClasses,strategy,ExtensionFactory.class.getName());}}returnextensionClasses;}继续看loadDirectory两个逻辑一是调用方法loadDirectoryInternal二是判断Dubbo 2的向下兼容性核心方法是loadDirectoryInternal继续调用方法loadResource用于读取和解析配置文件并通过反射加载类这里面可看到对前面提到的文本文件进行逐行解析判断分隔符的逻辑最后调用 loadClass方法用于操作缓存。loadClass方法调用缓存相关方法如cachedAdaptiveClass、cachedWrapperClasses和cachedNames等。参考