Tigase8.0 引用了IoC(控制反转)和DI(依赖注入) 等技术手段,来对对象的创建和控制。不懂的百度下就知道了,Spring完美的实现IOC ,贴一段解释:
通俗地说:控制反转IoC(Inversion of Control)是说创建对象的控制权进行转移,以前创建对象的主动权和创建时机是由自己把控的,而现在这种权力转移到第三方,比如转移交给了IoC容器,它就是一个专门用来创建对象的工厂,你要什么对象,它就给你什么对象,有了 IoC容器,依赖关系就变了,原先的依赖关系就没了,它们都依赖IoC容器了,通过IoC容器来建立它们之间的关系。
DI(依赖注入)其实就是IOC的另外一种说法,DI是由Martin Fowler 在2004年初的一篇论文中首次提出的。他总结:控制的什么被反转了?就是:获得依赖对象的方式反转了。
在Tigase8 中,IOC容器由 Kernel 类实现,Kernel是充当容器管理所有类的依赖对象的注入。
类似Spring 的bean ,在Tigase中,只要是类标有 @Bean(name = "xxx", parent = Kernel.class, active = true) ,标有Bean注解的类,表明该类的对象创建交由容器负责生成
类属性字段上标有 @Inject(bean = "kernel") 注解时,表明该字段交由容器来赋值
Tigase8 中大量使用JAVA8 Function及Stream编程,需要补充基础的同学,可以 https://github.com/CarpenterLee/JavaLambdaInternals 进行学习!
下面开始进行代码的粗略讲解,其实 由7到8过渡,核心的架构思想是没有多大变化的,8中只是加入对象的创建管理手段而已,也可以看以前的文章来了解每个组件之间的逻辑关系
入口类 public final class XMPPServer
public static void start(String[] args) { ... bootstrap = new Bootstrap(); bootstrap.init(args); //加载配置文件等参数 bootstrap.start(); //启动系统所有的工作组件进行提供服务 }
启动包装类 public class Bootstrap
public void start() { //跳过些次要代码 ...... //类加载工具类 classUtilBean = (ClassUtilBean) Class.forName("tigase.util.reflection.ClassUtilBean").newInstance(); //通过查找classpath下的jar,或dir,并且跳过一些过滤的包路径,加载所有的类的Class对象,为后面程序能通过@Bean注解能查找到所有需要容器管理的类 classUtilBean.initialize(ClassUtilBean.getPackagesToSkip(null)); // 注册默认类型转换器和属性bean配置器 kernel.registerBean(DefaultTypesConverter.class).exportable().exec(); // DSLBeanConfigurator 注册到Kernel的beanInstances中,DSLBeanConfiguratorWithBackwardCompatibility 向下兼容的配置文件管理类,对于Tigase来说非常重要,也就是各种Bean加载的入口 kernel.registerBean(DSLBeanConfiguratorWithBackwardCompatibility.class).exportable().exec(); DSLBeanConfigurator configurator = kernel.getInstance(DSLBeanConfigurator.class); configurator.setConfigHolder(config); // 加载所有@Bean(name = "message-router", parent = Kernel.class, active = true) Bean中 parent = Kernel.class 的类,并进行注册到Kernel容器中,如下图所示: configurator.registerBeans(null, null, config.getProperties()); // 启动路由组件的服务 MessageRouter mr = kernel.getInstance("message-router"); mr.start(); }
如下图所示,查找标有@Bean注解,并且parent=Kernel.class 的Bean集合的Class,并且进行实例化,注册到Kernel中
一、重点来说下 AbstractBeanConfigurator.registerBeans(null, null, config.getProperties()); 加载查找标有@Bean注解,并且parent=Kernel.class 的Bean,注册到Kernel中
public void registerBeans(BeanConfig beanConfig, Object bean, Map<String, Object> values) {
//查找标有@Bean注解,并且parent=Kernel.class 的Bean集合 final Map<String, Class<?>> beansFromAnnotations = getBeanClassesFromAnnotations(kernel, beanConfig == null ? Kernel.class : beanConfig.getClazz());
.... //遍历所查找到的Bean集合,进行注册到Kernel中 beansFromAnnotations.forEach((name, cls) -> { kernel.registerBean(cls) .setSource(BeanConfig.Source.annotation) .registeredBy(beanConfig) .execWithoutInject();
.... }
查找@Bean中parent属性为参数类型的Bean
public static Map<String, Class<?>> getBeanClassesFromAnnotations(Kernel kernel, Class<?> requiredClass) { //从classpath加载到的Class集合中 Set<Class<?>> classes = ClassUtilBean.getInstance().getAllClasses(); //遍历这些Class查找出标有@Bean注解,并且parent属性为requiredClass类型的bean Class List<Class<?>> toRegister = registerBeansForBeanOfClassGetBeansToRegister(kernel, requiredClass, classes); //转换下返回类型 Map<String, Class<?>> result = new HashMap<>(); for (Class<?> cls : toRegister) { Bean annotation = cls.getAnnotation(Bean.class); result.put(annotation.name(), cls); } return result; }
二、启动路由组件的服务,
// 启动路由组件的服务 MessageRouter mr = kernel.getInstance("message-router"); mr.start();
在执行getInstance()的时候,对还没初始化的对象进行初始化,进行依赖注入等操作
在首次获取Bean实例对象的时候,针对bean对象进行一些配置操作,和依赖注入操作,此后,bean实例才是一个完整的对象,其实和Spring中bean的获取原理是差不多的,不过这里要单纯很多,简单些
protected void initBean(BeanConfig tmpBC, Set<BeanConfig> createdBeansConfig, int deep){ .... //构建延迟依赖注入的本地线程队列 DelayedDependencyInjectionQueue queue = beanConfig.getKernel().beginDependencyDelayedInjection(); //针对于初始化的Bean判断是否为系统所默认的配置管理类,如果不是配置管理类本身,则其它类需要调用配置管理类的configure()方法进行配置项处理,如标有@ConfigField的字段 if (beanConfig.getKernel().isBeanClassRegistered(BeanConfigurator.DEFAULT_CONFIGURATOR_NAME) && !beanConfig.getBeanName().equals(BeanConfigurator.DEFAULT_CONFIGURATOR_NAME)) { beanConfigurator = beanConfig.getKernel().getInstance(BeanConfigurator.DEFAULT_CONFIGURATOR_NAME); } else { beanConfigurator = null; }
....
//如果不是默认配置类,则需要进行配置类的configure方法进行处理 if (beanConfigurator != null) {
//此方法会加载@Bean(name = "message-router", parent = beanConfig.clazz.class, active = true),加载@Bean注解中parent为参数beanConfig.clazz类型的Bean,并注册到Kernel中,下面会重点分解:1 beanConfigurator.configure(beanConfig, bean); } else { //如果是默认配置类本身,则进行配置类对象的创建和注册到kernel中 AbstractBeanConfigurator.registerBeansForBeanOfClass(beanConfig.getKernel(), bean.getClass()); } //这个是线程私有队列,这里进行清理工作 beanConfig.getKernel().finishDependecyDelayedInjection(queue); //这里重点,针对@Bean中所有@Inject注解的字段,进行依赖搜索和依赖注入操作 for (final Dependency dep : beanConfig.getFieldDependencies().values()) { beanConfig.getKernel().injectDependencies(bean, dep, createdBeansConfig, deep, false); } //如果实现Initializable 接口,则触发.initialize()函数的调用 // there is no need to wait to initialize parent beans, it there any? if (bean instanceof Initializable && beanConfig.getState() != State.initialized) { ((Initializable) bean).initialize(); } }
1、重点谈下 beanConfigurator.configure(beanConfig, bean);
public void configure(BeanConfig beanConfig, Object bean) throws KernelException { try { grabDefaultConfig(beanConfig, bean); Map<String, Object> ccc = getConfiguration(beanConfig); configure(beanConfig, bean, ccc);//调用父类的配置方法 } catch (Exception e) { throw new KernelException("Cannot inject configuration to bean " + beanConfig.getBeanName(), e); } }
重点进入父类的configure()方法,查找出@bean注册中parent=beanConfig.clazz类型的Class,进行注册到Kernel容器中
public void AbstractBeanConfigurator.configure(BeanConfig beanConfig, Object bean, Map<String, Object> values) { ........ //查找@Bean中parent属性为参数beanConfig.clazz的类,并加载到Kernel容器中 registerBeans(beanConfig, bean, values); ......... }
根据查找到的Bean Class,进行注册,同时涉及到依赖到别的Bean,则判断当前本地线程中是否延迟注入队列,有值,则放到队列中,运行完本操作后,在.finishDependecyDelayedInjection(queue);中完成注入
public void registerBeans(BeanConfig beanConfig, Object bean, Map<String, Object> values) { if (beanConfig != null && Kernel.class.isAssignableFrom(beanConfig.getClazz())) { return; } //获得核心的容器管理类 Kernel kernel = beanConfig == null ? this.getKernel() : beanConfig.getKernel(); //找出beanConfig.clazz已经注册的Bean,用于取消注册 Set<String> toUnregister = new ArrayList<>(kernel.getDependencyManager().getBeanConfigs()).stream() .filter(bc -> bc.getSource() == BeanConfig.Source.configuration) .filter(bc -> beanConfig == null || bc.getRegisteredBy().contains(beanConfig)) .map(bc -> bc.getBeanName()) .collect(Collectors.toSet()); //遍历classpath中过滤出来的全部Class,查找出标有@Bean注解,并且parent属性为Kernel.class或beanConfig.getClazz()类型的bean Class final Map<String, Class<?>> beansFromAnnotations = getBeanClassesFromAnnotations(kernel, beanConfig == null ? Kernel.class : beanConfig.getClazz()); //查找该配置参数中为BeanDefinition的配置项 final Map<String, BeanDefinition> beanDefinitionsFromConfig = values == null ? new HashMap<>() : mergeWithBeansPropertyValue(getBeanDefinitions(values), values); ///遍历查找出来的beanConfig.clazz类型子Bean集合,依次进行注册到Kernel容器中 beansFromAnnotations.forEach((name, cls) -> { if (beanDefinitionsFromConfig != null) { BeanDefinition definition = beanDefinitionsFromConfig.get(name); if (definition != null) { return; } } if (isBeanClassRegisteredInParentKernel(kernel.getParent(), name, cls)) { return; } BeanConfig bc = kernel.getDependencyManager().getBeanConfig(name); if (bc != null && bc.getSource() == BeanConfig.Source.annotation && bc.getClazz().equals(cls)) { return; } if (beanConfig != null && beanConfig.getState() == BeanConfig.State.initialized) { kernel.registerBean(cls).setSource(BeanConfig.Source.annotation).registeredBy(beanConfig).exec(); } else { //后面要重点讲解下注册,注册到kernel容器中。当注册A对象时,可能会依赖到B对象,所以注册里有使用到本地线程延迟注册队列 kernel.registerBean(cls) .setSource(BeanConfig.Source.annotation) .registeredBy(beanConfig) .execWithoutInject(); } bc = kernel.getDependencyManager().getBeanConfig(name); if (bc != null && bc.getState() == BeanConfig.State.inactive && hasDirectConfiguration(bc)) { log.log(Level.CONFIG, "bean " + bc.getBeanName() + " is disabled but configuration is specified"); } }); //根据参数中为BeanDefinition的配置项,查看该Bean是否已经注册,没有则进行注册 for (BeanDefinition cfg : beanDefinitionsFromConfig.values()) { try { Class<?> clazz = cfg.getClazzName() == null ? beansFromAnnotations.get(cfg.getBeanName()) : ModulesManagerImpl.getInstance().forName(cfg.getClazzName()); BeanConfig oldBc = kernel.getDependencyManager().getBeanConfig(cfg.getBeanName()); if (clazz == null) { if (bean != null && bean instanceof RegistrarBeanWithDefaultBeanClass) { clazz = ((RegistrarBeanWithDefaultBeanClass) bean).getDefaultBeanClass(); } else if (oldBc != null) { clazz = oldBc.getClazz(); } if (clazz == null) { log.log(Level.WARNING, "unknown class {0} for bean {1}, skipping registration of a bean", new Object[]{cfg.getClazzName(), cfg.getBeanName()}); continue; } } if (!tigase.util.reflection.ClassUtilBean.getInstance().getAllClasses().contains(clazz)) { continue; } toUnregister.remove(cfg.getBeanName()); if (oldBc != null && oldBc.getClazz().equals(clazz) && (oldBc.isExportable() || cfg.isExportable() == oldBc.isExportable())) { kernel.setBeanActive(cfg.getBeanName(), cfg.isActive()); } else { Bean ba = clazz.getAnnotation(Bean.class); BeanConfigBuilder cfgBuilder = kernel.registerBean(cfg.getBeanName()).asClass(clazz); cfgBuilder.setActive(cfg.isActive()).setSource(BeanConfig.Source.configuration); if (cfg.isExportable()) { cfgBuilder.exportable(); } if (ba != null) { if (ba.exportable()) { cfgBuilder.exportable(); } } cfgBuilder.registeredBy(beanConfig); if (beanConfig != null && beanConfig.getState() == BeanConfig.State.initialized) { cfgBuilder.exec(); } else { cfgBuilder.execWithoutInject(); } } } catch (ClassNotFoundException ex) { log.log(Level.FINER, "could not register bean '" + cfg.getBeanName() + "' as class '" + cfg.getClazzName() + "' is not available", ex); } } toUnregister.forEach(beanName -> kernel.unregister(beanName)); }
重点讲解下注册,当执行注册但还没依赖注入的函数。
public BeanConfig BeanConfigBuilder.execWithoutInject() { if (beanConfig == null) { log.warning("Bean " + clazz + " cannot be registered, because Kernel cannot create configuration for this bean."); kernel.currentlyUsedConfigBuilder = null; return null; } //执行bean的注册 beanConfig = kernel.registerBean(beanConfig, factoryBeanConfig, beanInstance); return beanConfig; }
注册bean实例到Kernel.beanInstances中,bean中所依赖inject的类型,注册到dependencyManager.beanConfigs中,beanConfig通过state来区分,是否为初始化完成的实例
protected BeanConfig Kernel.registerBean(BeanConfig beanConfig, BeanConfig factoryBeanConfig, Object beanInstance) { BeanConfig parent = null; if (beanConfig.getSource() == BeanConfig.Source.annotation && !beanConfig.getRegisteredBy().isEmpty()) { BeanConfig bc = dependencyManager.getBeanConfig(beanConfig.getBeanName()); parent = beanConfig.getRegisteredBy().iterator().next(); if (bc != null && bc.getClazz().equals(beanConfig.getClazz())) { bc.addRegisteredBy(parent); parent.addRegisteredBean(bc); currentlyUsedConfigBuilder = null; return bc; } } if (factoryBeanConfig != null) { factoryBeanConfig.setPinned(beanConfig.isPinned()); factoryBeanConfig.setState(beanConfig.getState()); unregisterInt(factoryBeanConfig.getBeanName()); dependencyManager.register(factoryBeanConfig); } BeanConfig oldBeanConfig = dependencyManager.getBeanConfig(beanConfig.getBeanName()); Collection<Dependency> oldDeps = oldBeanConfig == null ? null : dependencyManager.getDependenciesTo(oldBeanConfig); unregisterInt(beanConfig.getBeanName()); //注册进依赖管理类 dependencyManager.register(beanConfig); if (parent != null) { parent.addRegisteredBean(beanConfig); } //当实例对象不为空,则更新bean的状态为initialized if (beanInstance != null) { putBeanInstance(beanConfig, beanInstance); beanConfig.setState(State.initialized); } //查找出本bean可以注入的被依赖集合,简单地说是B 对象内有个依赖注入的A属性,现在beanConfig为A实现类,此时这里就可以查找出B对象这个依赖项包装类Dependency Collection<Dependency> deps = dependencyManager.getDependenciesTo(beanConfig); if (oldDeps != null) { deps.addAll(oldDeps.stream().filter(od -> { Field f = od.getField(); return !deps.stream().anyMatch(nd -> nd.getField().equals(f)); }).collect(Collectors.toSet())); } currentlyUsedConfigBuilder = null; //如果此函数在所在的本地线程内有延迟注入队列,则加入队列,在finishDependecyDelayedInjection(queue);进行给依赖本beanConfig类型的对象,进行依赖注入 if (!queueForDelayedDependencyInjection(deps)) {
//当本地线程内无延迟注入队列则为依赖本beanConfig类的对象立即执行依赖注入 injectDependencies(deps); } return beanConfig; }
前面,我们在 initBean(BeanConfig tmpBC, Set<BeanConfig> createdBeansConfig, int deep) 中有看到 beanConfig.getKernel().finishDependecyDelayedInjection(queue);关于赋值就是在这里进行
private boolean queueForDelayedDependencyInjection(Collection<Dependency> deps) { //本地线程队列存在的时候,才会放到队列里去,后面再进行注入 DelayedDependencyInjectionQueue queue = DELAYED_DEPENDENCY_INJECTION.get(); if (queue == null) { return false; } if (deps.isEmpty()) { return true; } queue.offer(new DelayedDependenciesInjection(deps)); return true; }
2.举个例子进一步说明下,如MessageRouter中依赖注入componentsAll字段,
@Inject private Set<ServerComponent> componentsAll;
当调用initBean时,程序针对 componentsAll字段进行依赖注入,当xxx字段有提供setXxx() 的函数时,系统会自动调用该函数。如MessageRouter中提供 setComponentsAll(Set<ServerComponent> components) 函数,所以,当注入componentsAll字段时,会自动调用setComponentsAll函数
例如在 MessageRouter 需要注入 componentsAll 组件集合对象,就是在getInstance()时触发的操作
@Bean(name = "message-router", parent = Kernel.class, active = true) public class MessageRouter ..{ @Inject private Set<ServerComponent> componentsAll; public void setComponentsAll(Set<ServerComponent> components) { if (components == null) { return; } HashSet<ServerComponent> removeComponents = new HashSet<>(this.components.values()); removeComponents.removeAll(components); for (ServerComponent comp : removeComponents) { if (comp instanceof ComponentRegistrator) { removeRegistrator((ComponentRegistrator) comp); } else if (comp instanceof MessageReceiver) { removeRouter((MessageReceiver) comp); } else { removeComponent(comp); } if (comp instanceof ConnectionManager) { connectionManagerNames.remove(comp.getName()); } comp.release(); } HashSet<ServerComponent> newComponents = new HashSet<>(components); newComponents.removeAll(this.components.values());
//这里加载到的组件进行相互注册,通知,和7版本的逻辑是一样的,可以去看以前的文章 for (ServerComponent comp : newComponents) { try { if (comp instanceof MessageReceiver) { MessageReceiver mr = (MessageReceiver) comp; mr.setParent(this); mr.start(); } if (comp instanceof ConnectionManager) { connectionManagerNames.add(comp.getName()); } if (comp instanceof ComponentRegistrator) { addRegistrator((ComponentRegistrator) comp); } else if (comp instanceof MessageReceiver) { addRouter((MessageReceiver) comp); } else { addComponent(comp); } } catch (ConfigurationException ex) { // TODO - most likely this will no longer happen as configuration will not be done in this method log.log(Level.WARNING, "component " + comp.getName() + " was not configured properly", ex); } } } }
三、最终调用 MessageRouter.start() 启动路由组件进行,路由功能,和7版本的原理是一样,这里就不再一一说明,可以找回以前的文章看
到这里启动篇就完成了,Tigase8,看起来代码结构改动大,但是其实也没有多大的变化,核心没有变,变是的包装的架子,所以只要认真地看问题不大,在看源码的同学,建议在调试的时候,再来看这个文章,这样帮助会大些,可以尽可能快地理解Tigase的原理。
如有不对的地方,请提出来,我进行改进!