SpringBoot启动流程分析(六):IoC容器依赖注入
SpringBoot系列文章简介
SpringBoot源码阅读辅助篇:
SpringBoot启动流程源码分析:
- SpringBoot启动流程分析(一):SpringApplication类初始化过程
- SpringBoot启动流程分析(二):SpringApplication的run方法
- SpringBoot启动流程分析(三):SpringApplication的run方法之prepareContext()方法
- SpringBoot启动流程分析(四):IoC容器的初始化过程
- SpringBoot启动流程分析(五):SpringBoot自动装配原理实现
- SpringBoot启动流程分析(六):IoC容器依赖注入
笔者注释版Spring Framework与SpringBoot源码git传送门:请不要吝啬小星星
一、前言
前面我们对IoC容器的初始化过程进行了详细的分析,这个初始化过程完成的主要工作是在IoC容器中建立BeanDefinition数据映射。在此过程中并没有看到IoC容器对Bean依赖关系进行注入,接下来分析一下IoC容器是怎样对Bean的依赖关系进行注入的。
前面在refresh()-->invokeBeanFactoryPostProcessors(beanFactory);方法中已经完成了IoC容器的初始化并已经载入了我们定义的Bean的信息(BeanDefinition),现在我们开始分析依赖注入的原理。首先需要说明的是依赖注入在用户第一次向IoC容器索要Bean时触发,当然也有例外,我们可以在BeanDefinition中中通过控制lazy-init属性来让容器完成对Bean的预实例化。这个预实例化实际上也是一个依赖注入的过程,但它是在初始化过程中完成的。
二、源码分析
2.1、getBean()的过程
接着前面看refresh()方法,这已经是refresh()方法的第三篇博文了,别迷糊我们还没走出refresh()方法。
1 // AbstractApplicationContext类 2 @Override 3 public void refresh() throws BeansException, IllegalStateException { 4 synchronized (this.startupShutdownMonitor) { 5 ... 6 try { 7 ... 8 // Instantiate all remaining (non-lazy-init) singletons. 9 finishBeanFactoryInitialization(beanFactory); 10 ... 11 } 12 ... 13 } 14 } 15 // AbstractApplicationContext类 16 protected void finishBeanFactoryInitialization(ConfigurableListableBeanFactory beanFactory) { 17 ... 18 // Instantiate all remaining (non-lazy-init) singletons. 19 // 实例化所有剩余的(non-lazy-init)单例。 20 beanFactory.preInstantiateSingletons(); 21 } 22 // DefaultListableBeanFactory类 23 @Override 24 public void preInstantiateSingletons() throws BeansException { 25 ... 26 // Iterate over a copy to allow for init methods which in turn register new bean definitions. 27 // While this may not be part of the regular factory bootstrap, it does otherwise work fine. 28 List<String> beanNames = new ArrayList<>(this.beanDefinitionNames); 29 // Trigger initialization of all non-lazy singleton beans... 30 for (String beanName : beanNames) { 31 RootBeanDefinition bd = getMergedLocalBeanDefinition(beanName); 32 if (!bd.isAbstract() && bd.isSingleton() && !bd.isLazyInit()) { 33 if (isFactoryBean(beanName)) { 34 Object bean = getBean(FACTORY_BEAN_PREFIX + beanName); 35 if (bean instanceof FactoryBean) { 36 final FactoryBean<?> factory = (FactoryBean<?>) bean; 37 boolean isEagerInit; 38 if (System.getSecurityManager() != null && factory instanceof SmartFactoryBean) { 39 isEagerInit = AccessController.doPrivileged((PrivilegedAction<Boolean>) 40 ((SmartFactoryBean<?>) factory)::isEagerInit, 41 getAccessControlContext()); 42 } else { 43 isEagerInit = (factory instanceof SmartFactoryBean && 44 ((SmartFactoryBean<?>) factory).isEagerInit()); 45 } 46 if (isEagerInit) { 47 getBean(beanName); 48 } 49 } 50 } else { 51 // 这里就是触发依赖注入的地方 52 getBean(beanName); 53 } 54 } 55 } 56 ... 57 }
跟踪其调用栈,看到上面第52行的getBean(beanName);方法,我们再梳理一下getBean()方法,前面总结过该方法在IoC容器的顶层接口BeanFactory中定义,然后在IoC容器的具体产品DefaultListableBeanFactory类的基类AbstractBeanFactory实现了getBean()方法。接着看代码。
1 // AbstractBeanFactory类 2 @Override 3 public Object getBean(String name) throws BeansException { 4 return doGetBean(name, null, null, false); 5 } 6 @Override 7 public <T> T getBean(String name, @Nullable Class<T> requiredType) throws BeansException { 8 return doGetBean(name, requiredType, null, false); 9 } 10 @Override 11 public Object getBean(String name, Object... args) throws BeansException { 12 return doGetBean(name, null, args, false); 13 } 14 public <T> T getBean(String name, @Nullable Class<T> requiredType, @Nullable Object... args) 15 throws BeansException { 16 return doGetBean(name, requiredType, args, false); 17 }
从上面代码可知大致可分为两种获取Bean的参数,一种是按名获取,一种是按类获取。但是最终都进入到了doGetBean()方法。
1 // AbstractBeanFactory类 2 protected <T> T doGetBean(final String name, @Nullable final Class<T> requiredType, 3 @Nullable final Object[] args, boolean typeCheckOnly) throws BeansException { 4 5 // bean获取过程:先获取bean名字 6 // 会把带有&前缀的去掉,或者去aliasMap中找这个是不是别名,最终确定bean的id是什么 7 final String beanName = transformedBeanName(name); 8 Object bean; 9 10 // 1.检查缓存中或者实例工厂中是否有对应的实例 11 // 因为在创建单例bean的时候会存在依赖注入的情况,而在创建依赖的时候为了避免循环依赖 12 // Spring在创建bean的时候不会等bean创建完成就会将bean的ObjectFactory提早曝光 13 // 也就是将ObjectFactory加入到缓存中,一旦下一个要创建的bean需要依赖上个bean则直接使用ObjectFactory 14 // 2.spring 默认是单例的,如果能获取到直接返回,提高效率。 15 // Eagerly check singleton cache for manually registered singletons. 16 Object sharedInstance = getSingleton(beanName); 17 if (sharedInstance != null && args == null) { 18 if (logger.isDebugEnabled()) { 19 if (isSingletonCurrentlyInCreation(beanName)) { 20 logger.debug("Returning eagerly cached instance of singleton bean '" + beanName + 21 "' that is not fully initialized yet - a consequence of a circular reference"); 22 } 23 else { 24 logger.debug("Returning cached instance of singleton bean '" + beanName + "'"); 25 } 26 } 27 // 用于检测bean的正确性,同时如果获取的是FactoryBean的话还需要调用getObject()方法获取最终的那个bean实例 28 bean = getObjectForBeanInstance(sharedInstance, name, beanName, null); 29 } 30 31 else { 32 // Fail if we're already creating this bean instance: 33 // We're assumably within a circular reference. 34 if (isPrototypeCurrentlyInCreation(beanName)) { 35 throw new BeanCurrentlyInCreationException(beanName); 36 } 37 38 // Check if bean definition exists in this factory. 39 //这里对IoC容器中的BeanDefinition是否存在进行检查,检查是否能在当前的BeanFactory中取得需要的Bean。 40 // 如果当前的工厂中取不到,则到双亲BeanFactory中去取。如果当前的双亲工厂取不到,那就顺着双亲BeanFactory 41 // 链一直向上查找。 42 BeanFactory parentBeanFactory = getParentBeanFactory(); 43 if (parentBeanFactory != null && !containsBeanDefinition(beanName)) { 44 // Not found -> check parent. 45 String nameToLookup = originalBeanName(name); 46 if (parentBeanFactory instanceof AbstractBeanFactory) { 47 // 递归调用父bean的doGetBean查找 48 return ((AbstractBeanFactory) parentBeanFactory).doGetBean( 49 nameToLookup, requiredType, args, typeCheckOnly); 50 } 51 else if (args != null) { 52 // Delegation to parent with explicit args. 53 return (T) parentBeanFactory.getBean(nameToLookup, args); 54 } 55 else { 56 // No args -> delegate to standard getBean method. 57 return parentBeanFactory.getBean(nameToLookup, requiredType); 58 } 59 } 60 61 if (!typeCheckOnly) { 62 markBeanAsCreated(beanName); 63 } 64 65 try { 66 //这里根据Bean的名字取得BeanDefinition 67 final RootBeanDefinition mbd = getMergedLocalBeanDefinition(beanName); 68 checkMergedBeanDefinition(mbd, beanName, args); 69 70 // Guarantee initialization of beans that the current bean depends on. 71 //获取当前Bean的所有依赖Bean,这里会触发getBean的递归调用。知道取到一个没有任何依赖的Bean为止。 72 String[] dependsOn = mbd.getDependsOn(); 73 if (dependsOn != null) { 74 for (String dep : dependsOn) { 75 if (isDependent(beanName, dep)) { 76 throw new BeanCreationException(mbd.getResourceDescription(), beanName, 77 "Circular depends-on relationship between '" + beanName + "' and '" + dep + "'"); 78 } 79 registerDependentBean(dep, beanName); 80 try { 81 getBean(dep); 82 } 83 catch (NoSuchBeanDefinitionException ex) { 84 throw new BeanCreationException(mbd.getResourceDescription(), beanName, 85 "'" + beanName + "' depends on missing bean '" + dep + "'", ex); 86 } 87 } 88 } 89 90 // 这里通过createBean方法创建singleton Bean的实例 这里还有一个回调函数 91 // Create bean instance. 92 if (mbd.isSingleton()) { 93 sharedInstance = getSingleton(beanName, () -> { 94 try { 95 // 最后在getSingleton中又会调用这个方法 96 // TODO createBean的入口 97 return createBean(beanName, mbd, args); 98 } 99 catch (BeansException ex) { 100 // Explicitly remove instance from singleton cache: It might have been put there 101 // eagerly by the creation process, to allow for circular reference resolution. 102 // Also remove any beans that received a temporary reference to the bean. 103 destroySingleton(beanName); 104 throw ex; 105 } 106 }); 107 bean = getObjectForBeanInstance(sharedInstance, name, beanName, mbd); 108 } 109 // 这里是创建prototype bean的地方 110 else if (mbd.isPrototype()) { 111 // It's a prototype -> create a new instance. 112 Object prototypeInstance = null; 113 try { 114 beforePrototypeCreation(beanName); 115 prototypeInstance = createBean(beanName, mbd, args); 116 } 117 finally { 118 afterPrototypeCreation(beanName); 119 } 120 bean = getObjectForBeanInstance(prototypeInstance, name, beanName, mbd); 121 } 122 123 else { 124 String scopeName = mbd.getScope(); 125 final Scope scope = this.scopes.get(scopeName); 126 if (scope == null) { 127 throw new IllegalStateException("No Scope registered for scope name '" + scopeName + "'"); 128 } 129 try { 130 Object scopedInstance = scope.get(beanName, () -> { 131 beforePrototypeCreation(beanName); 132 try { 133 return createBean(beanName, mbd, args); 134 } 135 finally { 136 afterPrototypeCreation(beanName); 137 } 138 }); 139 bean = getObjectForBeanInstance(scopedInstance, name, beanName, mbd); 140 } 141 catch (IllegalStateException ex) { 142 throw new BeanCreationException(beanName, 143 "Scope '" + scopeName + "' is not active for the current thread; consider " + 144 "defining a scoped proxy for this bean if you intend to refer to it from a singleton", 145 ex); 146 } 147 } 148 } 149 catch (BeansException ex) { 150 cleanupAfterBeanCreationFailure(beanName); 151 throw ex; 152 } 153 } 154 155 // Check if required type matches the type of the actual bean instance. 156 //这里对创建的Bean进行类型检查,如果没有问题,就返回这个新创建的Bean,这个Bean已经是包含了依赖关系的Bean 157 if (requiredType != null && !requiredType.isInstance(bean)) { 158 try { 159 T convertedBean = getTypeConverter().convertIfNecessary(bean, requiredType); 160 if (convertedBean == null) { 161 throw new BeanNotOfRequiredTypeException(name, requiredType, bean.getClass()); 162 } 163 return convertedBean; 164 } 165 catch (TypeMismatchException ex) { 166 if (logger.isDebugEnabled()) { 167 logger.debug("Failed to convert bean '" + name + "' to required type '" + 168 ClassUtils.getQualifiedName(requiredType) + "'", ex); 169 } 170 throw new BeanNotOfRequiredTypeException(name, requiredType, bean.getClass()); 171 } 172 } 173 return (T) bean; 174 }
这个就是依赖注入的入口了,依赖注入是在容器的BeanDefinition数据已经建立好的前提下进行的。“程序=数据+算法”,很经典的一句话,前面我们详细介绍了BeanDefinition的注册过程,BeanDefinition就是数据。如上面代码所示,doGetBean()方法不涉及复杂的算法,但是这个过程也不是很简单,因为我们都知道,对于IoC容器的使用,Spring提供了很多的配置参数,每一个配置参数实际上就代表了一个IoC容器的实现特征,这些特征很多都需要在依赖注入的过程或者对Bean进行生命周期管理的过程中完成。虽然我们可以简单的将IoC容器描述成一个ConcurrentHashMap,ConcurrentHashMap只是它的数据结构而不是IoC容器的全部。
TIPS: 1,我在工程中简单写了一个controller和一个service,我们在后面debug看看依赖注入的过程是怎么样的, 也不知道我能不能说清楚。希望大家看到这一块多debug一下,debug技巧如下图所示,写了debug的条件, 因为这边到处都是递归和回调函数,再加上有很多Spring的Bean,但是我们只关心自己的Bean, 所以就写了这样的过滤条件:beanName.equals("webController")||beanName.equals("webService") @RestController public class WebController { @Autowired private WebService webService; @RequestMapping("/web") public String web(){ return webService.hello(); } }
下面我们通过代码看看获取bean的过程。
OK,看代码,Object sharedInstance = getSingleton(beanName);如注释所说,首先回去找在容器中是不是已经存在该单例。具体在哪找我们在前面的文章中已经说得很清楚了。看一下getSingleton()方法
1 // DefaultSingletonBeanRegistry类 2 protected Object getSingleton(String beanName, boolean allowEarlyReference) { 3 // 由于scope是singleton,所以先从缓存中取单例对象的实例,如果取到直接返回,没有取到加载bean 4 Object singletonObject = this.singletonObjects.get(beanName); 5 // 当想要获取的bean没有被加载,并且也没有正在被创建的时候,主动去加载bean 6 if (singletonObject == null && isSingletonCurrentlyInCreation(beanName)) { 7 // 锁住单例缓存区加载bean 8 synchronized (this.singletonObjects) { 9 // singletonObjects ,earlySingletonObjects ,singletonFactories是一个单例实例的三种存在状态 10 // 再去earlySingletonObjects中去找 11 singletonObject = this.earlySingletonObjects.get(beanName); 12 if (singletonObject == null && allowEarlyReference) { 13 // 去singletonFactories中去找对象的实例 14 ObjectFactory<?> singletonFactory = this.singletonFactories.get(beanName); 15 if (singletonFactory != null) { 16 singletonObject = singletonFactory.getObject(); 17 this.earlySingletonObjects.put(beanName, singletonObject); 18 this.singletonFactories.remove(beanName); 19 } 20 } 21 } 22 } 23 return singletonObject; 24 }
在DefaultSingletonBeanRegistry类中的singletonObjects属性就是存singleton bean的地方。
如果getSingleton()为 null继续往下看,会在当前的BeanFactory中获取BeanDefinition,也就是这行方法代码:final RootBeanDefinition mbd = getMergedLocalBeanDefinition(beanName);在这行代码拿到BeanDefinition后,首先判断是不是singleton Bean,如果是的话,开始执行创建Bean,正是return createBean(beanName, mbd, args);这行代码。如果是原型(Prototype)Bean我们就不分析了。原型bean每次执行getBean()都会创建一个实例。接下来我们看createBean()方法。
2.2、createBean()的过程
首先看一下create bean的过程
1,Bean实例的创建 2,为Bean实例设置属性(属性注入,其实就是依赖注入真正发生的地方) 3,调用Bean的初始化方法
前面说了getBean()是依赖注入的起点,之后会调用createBean(),下面通过createBean()代码来了解这个过程。在这个过程中,Bean对象会根据BeanDefinition定义的要求生成。
1 // AbstractAutowireCapableBeanFactory类 2 @Override 3 protected Object createBean(String beanName, RootBeanDefinition mbd, @Nullable Object[] args) 4 throws BeanCreationException { 5 ... 6 try { 7 // 验证以及准备override的方法 8 mbdToUse.prepareMethodOverrides(); 9 } 10 catch (BeanDefinitionValidationException ex) { 11 throw new BeanDefinitionStoreException(mbdToUse.getResourceDescription(), 12 beanName, "Validation of method overrides failed", ex); 13 } 14 try { 15 // Give BeanPostProcessors a chance to return a proxy instead of the target bean instance. 16 // createBean之前调用BeanPostProcessor的postProcessBeforeInitialization和postProcessAfterInitialization方法 17 // 默认不做任何处理所以会返回null 18 // 但是如果我们重写了这两个方法,那么bean的创建过程就结束了,这里就为以后的annotation自动注入提供了钩子 19 Object bean = resolveBeforeInstantiation(beanName, mbdToUse); 20 if (bean != null) { 21 return bean; 22 } 23 }catch (Throwable ex) { 24 throw new BeanCreationException(mbdToUse.getResourceDescription(), beanName, 25 "BeanPostProcessor before instantiation of bean failed", ex); 26 } 27 try { 28 // 实际执行createBean的是doCreateBean()方法 29 Object beanInstance = doCreateBean(beanName, mbdToUse, args); 30 if (logger.isDebugEnabled()) { 31 logger.debug("Finished creating instance of bean '" + beanName + "'"); 32 } 33 return beanInstance; 34 } 35 catch (BeanCreationException | ImplicitlyAppearedSingletonException ex) { 36 // A previously detected exception with proper bean creation context already, 37 // or illegal singleton state to be communicated up to DefaultSingletonBeanRegistry. 38 throw ex; 39 } 40 catch (Throwable ex) { 41 throw new BeanCreationException( 42 mbdToUse.getResourceDescription(), beanName, "Unexpected exception during bean creation", ex); 43 } 44 }
接着往下看doCreateBean()方法。
1 // AbstractAutowireCapableBeanFactory类 2 protected Object doCreateBean(final String beanName, final RootBeanDefinition mbd, final @Nullable Object[] args) 3 throws BeanCreationException { 4 // BeanWrapper是用来持有创建出来的Bean对象 5 // Instantiate the bean. 6 BeanWrapper instanceWrapper = null; 7 // 如果是单例,先把缓存中的同名Bean清除 8 if (mbd.isSingleton()) { 9 instanceWrapper = this.factoryBeanInstanceCache.remove(beanName); 10 } 11 // 这里是创建Bean的地方,由createBeanInstance完成。 12 // TODO 完成Bean初始化过程的第一步:创建实例 13 if (instanceWrapper == null) { 14 instanceWrapper = createBeanInstance(beanName, mbd, args); 15 } 16 final Object bean = instanceWrapper.getWrappedInstance(); 17 Class<?> beanType = instanceWrapper.getWrappedClass(); 18 if (beanType != NullBean.class) { 19 mbd.resolvedTargetType = beanType; 20 } 21 22 // Allow post-processors to modify the merged bean definition. 23 synchronized (mbd.postProcessingLock) { 24 if (!mbd.postProcessed) { 25 try { 26 applyMergedBeanDefinitionPostProcessors(mbd, beanType, beanName); 27 } 28 catch (Throwable ex) { 29 throw new BeanCreationException(mbd.getResourceDescription(), beanName, 30 "Post-processing of merged bean definition failed", ex); 31 } 32 mbd.postProcessed = true; 33 } 34 } 35 36 // Eagerly cache singletons to be able to resolve circular references 37 // even when triggered by lifecycle interfaces like BeanFactoryAware. 38 // 是否自动解决循环引用 39 // 当bean条件为: 单例&&允许循环引用&&正在创建中这样的话提早暴露一个ObjectFactory 40 boolean earlySingletonExposure = (mbd.isSingleton() && this.allowCircularReferences && 41 isSingletonCurrentlyInCreation(beanName)); 42 if (earlySingletonExposure) { 43 if (logger.isDebugEnabled()) { 44 logger.debug("Eagerly caching bean '" + beanName + 45 "' to allow for resolving potential circular references"); 46 } 47 // 把ObjectFactory放进singletonFactories中 48 // 这里在其他bean在创建的时候会先去singletonFactories中查找有没有beanName到ObjectFactory的映射 49 // 如果有ObjectFactory就调用它的getObject方法获取实例 50 // 但是在这里就可以对一个bean进行保证,代理等等AOP就可以在getEarlyBeanReference这里实现 51 addSingletonFactory(beanName, () -> getEarlyBeanReference(beanName, mbd, bean)); 52 } 53 54 // Initialize the bean instance. 55 Object exposedObject = bean; 56 try { 57 // TODO 完成Bean初始化过程的第二步:为Bean的实例设置属性 58 // Bean依赖注入发生的地方 59 // 对bean进行属性填充,如果存在依赖于其他的bean的属性,则会递归的调用初始化依赖的bean 60 populateBean(beanName, mbd, instanceWrapper); 61 // TODO 完成Bean初始化过程的第三步:调用Bean的初始化方法(init-method) 62 // 调用初始化方法,比如init-method方法指定的方法 63 exposedObject = initializeBean(beanName, exposedObject, mbd); 64 } 65 catch (Throwable ex) { 66 if (ex instanceof BeanCreationException && beanName.equals(((BeanCreationException) ex).getBeanName())) { 67 throw (BeanCreationException) ex; 68 } 69 else { 70 throw new BeanCreationException( 71 mbd.getResourceDescription(), beanName, "Initialization of bean failed", ex); 72 } 73 } 74 75 if (earlySingletonExposure) { 76 Object earlySingletonReference = getSingleton(beanName, false); 77 if (earlySingletonReference != null) { 78 if (exposedObject == bean) { 79 exposedObject = earlySingletonReference; 80 } 81 else if (!this.allowRawInjectionDespiteWrapping && hasDependentBean(beanName)) { 82 String[] dependentBeans = getDependentBeans(beanName); 83 Set<String> actualDependentBeans = new LinkedHashSet<>(dependentBeans.length); 84 for (String dependentBean : dependentBeans) { 85 if (!removeSingletonIfCreatedForTypeCheckOnly(dependentBean)) { 86 actualDependentBeans.add(dependentBean); 87 } 88 } 89 if (!actualDependentBeans.isEmpty()) { 90 throw new BeanCurrentlyInCreationException(beanName, 91 "Bean with name '" + beanName + "' has been injected into other beans [" + 92 StringUtils.collectionToCommaDelimitedString(actualDependentBeans) + 93 "] in its raw version as part of a circular reference, but has eventually been " + 94 "wrapped. This means that said other beans do not use the final version of the " + 95 "bean. This is often the result of over-eager type matching - consider using " + 96 "'getBeanNamesOfType' with the 'allowEagerInit' flag turned off, for example."); 97 } 98 } 99 } 100 } 101 102 // Register bean as disposable. 103 try { 104 // 注册销毁方法,比如:可以在配置bean的时候指定destory-method方法 105 registerDisposableBeanIfNecessary(beanName, bean, mbd); 106 } 107 catch (BeanDefinitionValidationException ex) { 108 throw new BeanCreationException( 109 mbd.getResourceDescription(), beanName, "Invalid destruction signature", ex); 110 } 111 112 return exposedObject; 113 }
结合上面的代码,我们再来看一下创建Bean的三个步骤,是不是有点豁然开朗的感觉。别着急继续往下看。
1,Bean实例的创建,instanceWrapper = createBeanInstance(beanName, mbd, args); 2,为Bean实例设置属性,populateBean(beanName, mbd, instanceWrapper); 3,调用Bean的初始化方法,exposedObject = initializeBean(beanName, exposedObject, mbd);
2.2.1、createBeanInstance():Bean实例的创建
看代码
1 // AbstractAutowireCapableBeanFactory类 2 protected BeanWrapper createBeanInstance(String beanName, RootBeanDefinition mbd, @Nullable Object[] args) { 3 // Make sure bean class is actually resolved at this point. 4 // 确认需要创建的Bean的实例的类可以实例化 5 Class<?> beanClass = resolveBeanClass(mbd, beanName); 6 7 if (beanClass != null && !Modifier.isPublic(beanClass.getModifiers()) && !mbd.isNonPublicAccessAllowed()) { 8 throw new BeanCreationException(mbd.getResourceDescription(), beanName, 9 "Bean class isn't public, and non-public access not allowed: " + beanClass.getName()); 10 } 11 12 Supplier<?> instanceSupplier = mbd.getInstanceSupplier(); 13 if (instanceSupplier != null) { 14 return obtainFromSupplier(instanceSupplier, beanName); 15 } 16 17 // 当有工厂方法的时候使用工厂方法初始化Bean,就是配置的时候指定FactoryMethod属性,类似注解中的@Bean把方法的返回值作为Bean 18 if (mbd.getFactoryMethodName() != null) { 19 return instantiateUsingFactoryMethod(beanName, mbd, args); 20 } 21 22 // Shortcut when re-creating the same bean... 23 boolean resolved = false; 24 boolean autowireNecessary = false; 25 if (args == null) { 26 synchronized (mbd.constructorArgumentLock) { 27 if (mbd.resolvedConstructorOrFactoryMethod != null) { 28 resolved = true; 29 autowireNecessary = mbd.constructorArgumentsResolved; 30 } 31 } 32 } 33 if (resolved) { 34 if (autowireNecessary) { 35 // 如果有有参数的构造函数,构造函数自动注入 36 // 这里spring会花费大量的精力去进行参数的匹配 37 return autowireConstructor(beanName, mbd, null, null); 38 } 39 else { 40 // 如果没有有参构造函数,使用默认构造函数构造 41 return instantiateBean(beanName, mbd); 42 } 43 } 44 45 // Need to determine the constructor... 46 // 使用构造函数进行实例化 47 Constructor<?>[] ctors = determineConstructorsFromBeanPostProcessors(beanClass, beanName); 48 if (ctors != null || 49 mbd.getResolvedAutowireMode() == RootBeanDefinition.AUTOWIRE_CONSTRUCTOR || 50 mbd.hasConstructorArgumentValues() || !ObjectUtils.isEmpty(args)) { 51 return autowireConstructor(beanName, mbd, ctors, args); 52 } 53 54 // No special handling: simply use no-arg constructor. 55 // 使用默认的构造函数对Bean进行实例化 56 return instantiateBean(beanName, mbd); 57 }
我们可以看到在instantiateBean()方法中生成了Bean所包含的Java对象,这个对象的生成有很多种不同的方式,可以通过工厂方法生成,也可以通过容器的autowire特性生成,这些生成方式都是由BeanDefinition决定的。对于上面我们的WebController和WebService两个类是通过最后一行,使用默认的构造函数进行Bean的实例化。
接着看instantiateBean()方法。
1 // AbstractAutowireCapableBeanFactory类 2 protected BeanWrapper instantiateBean(final String beanName, final RootBeanDefinition mbd) { 3 // 使用默认的实例化策略对Bean进行实例化,默认的实例化策略是CglibSubclassingInstantiationStrategy, 4 // 也就是常说的CGLIB来对Bean进行实例化。PS:面试官常问的字节码增强 5 try { 6 Object beanInstance; 7 final BeanFactory parent = this; 8 if (System.getSecurityManager() != null) { 9 beanInstance = AccessController.doPrivileged((PrivilegedAction<Object>) () -> 10 getInstantiationStrategy().instantiate(mbd, beanName, parent), 11 getAccessControlContext()); 12 } 13 else { 14 // getInstantiationStrategy()会返回CglibSubclassingInstantiationStrategy类的实例 15 beanInstance = getInstantiationStrategy().instantiate(mbd, beanName, parent); 16 } 17 BeanWrapper bw = new BeanWrapperImpl(beanInstance); 18 initBeanWrapper(bw); 19 return bw; 20 } 21 catch (Throwable ex) { 22 throw new BeanCreationException( 23 mbd.getResourceDescription(), beanName, "Instantiation of bean failed", ex); 24 } 25 }
这里使用CGLIB进行Bean的实例化。CGLIB是一个常用的字节码生成器的类库,其提供了一系列的API来提供生成和转换Java字节码的功能。在Spring AOP中同样也是使用的CGLIB对Java的字节码进行增强。在IoC容器中,使用SimpleInstantiationStrategy类。这个类是Spring用来生成Bean对象的默认类,它提供了两种实例化Java对象的方法,一种是通过BeanUtils,它使用的是JVM的反射功能,一种是通过CGLIB来生成。
getInstantiationStrategy()方法获取到CglibSubclassingInstantiationStrategy实例,instantiate()是CglibSubclassingInstantiationStrategy类的父类SimpleInstantiationStrategy实现的。
继续看代码
1 // SimpleInstantiationStrategy类 2 @Override 3 public Object instantiate(RootBeanDefinition bd, @Nullable String beanName, BeanFactory owner) { 4 // Don't override the class with CGLIB if no overrides. 5 // 如果BeanFactory重写了Bean内的方法,则使用CGLIB,否则使用BeanUtils 6 if (!bd.hasMethodOverrides()) { 7 // 如果bean没有需要动态替换的方法就直接反射进行创建实例 8 Constructor<?> constructorToUse; 9 synchronized (bd.constructorArgumentLock) { 10 constructorToUse = (Constructor<?>) bd.resolvedConstructorOrFactoryMethod; 11 if (constructorToUse == null) { 12 final Class<?> clazz = bd.getBeanClass(); 13 if (clazz.isInterface()) { 14 throw new BeanInstantiationException(clazz, "Specified class is an interface"); 15 } 16 try { 17 if (System.getSecurityManager() != null) { 18 constructorToUse = AccessController.doPrivileged( 19 (PrivilegedExceptionAction<Constructor<?>>) clazz::getDeclaredConstructor); 20 } else { 21 constructorToUse = clazz.getDeclaredConstructor(); 22 } 23 bd.resolvedConstructorOrFactoryMethod = constructorToUse; 24 } catch (Throwable ex) { 25 throw new BeanInstantiationException(clazz, "No default constructor found", ex); 26 } 27 } 28 } 29 // 通过BeanUtils进行实例化,这个BeanUtils的实例化通过Constructor类实例化Bean 30 // 在BeanUtils中可以看到具体的调用ctor.newInstances(args) 31 return BeanUtils.instantiateClass(constructorToUse); 32 } else { 33 // Must generate CGLIB subclass. 34 // TODO 使用CGLIB实例化对象 35 return instantiateWithMethodInjection(bd, beanName, owner); 36 } 37 }
在SpringBoot中我们一般采用@Autowire的方式进行依赖注入,很少采用像SpringMVC那种在xml中使用<lookup-method>或者<replaced-method>等标签的方式对注入的属性进行override,所以在上面的代码中if(!bd.hasMethodOverrides())中的判断为true,会采用BeanUtils的实例化方式。
2.2.2、populateBean();属性设置(依赖注入)
看代码
1 protected void populateBean(String beanName, RootBeanDefinition mbd, @Nullable BeanWrapper bw) { 2 if (bw == null) { 3 if (mbd.hasPropertyValues()) { 4 throw new BeanCreationException( 5 mbd.getResourceDescription(), beanName, "Cannot apply property values to null instance"); 6 } 7 else { 8 // Skip property population phase for null instance. 9 return; 10 } 11 } 12 // Give any InstantiationAwareBeanPostProcessors the opportunity to modify the 13 // state of the bean before properties are set. This can be used, for example, 14 // to support styles of field injection. 15 boolean continueWithPropertyPopulation = true; 16 // 调用InstantiationAwareBeanPostProcessor Bean的后置处理器,在Bean注入属性前改变BeanDefinition的信息 17 if (!mbd.isSynthetic() && hasInstantiationAwareBeanPostProcessors()) { 18 for (BeanPostProcessor bp : getBeanPostProcessors()) { 19 if (bp instanceof InstantiationAwareBeanPostProcessor) { 20 InstantiationAwareBeanPostProcessor ibp = (InstantiationAwareBeanPostProcessor) bp; 21 if (!ibp.postProcessAfterInstantiation(bw.getWrappedInstance(), beanName)) { 22 continueWithPropertyPopulation = false; 23 break; 24 } 25 } 26 } 27 } 28 if (!continueWithPropertyPopulation) { 29 return; 30 } 31 // 这里取得在BeanDefinition中设置的property值,这些property来自对BeanDefinition的解析 32 // 用于在配置文件中通过<property>配置的属性并且显示在配置文件中配置了autowireMode属性 33 PropertyValues pvs = (mbd.hasPropertyValues() ? mbd.getPropertyValues() : null); 34 if (mbd.getResolvedAutowireMode() == RootBeanDefinition.AUTOWIRE_BY_NAME || 35 mbd.getResolvedAutowireMode() == RootBeanDefinition.AUTOWIRE_BY_TYPE) { 36 MutablePropertyValues newPvs = new MutablePropertyValues(pvs); 37 38 // Add property values based on autowire by name if applicable. 39 // 这里对autowire注入的处理,autowire by name 40 if (mbd.getResolvedAutowireMode() == RootBeanDefinition.AUTOWIRE_BY_NAME) { 41 autowireByName(beanName, mbd, bw, newPvs); 42 } 43 44 // Add property values based on autowire by type if applicable. 45 // 这里对autowire注入的处理, autowire by type 46 // private List<Test> tests; 47 if (mbd.getResolvedAutowireMode() == RootBeanDefinition.AUTOWIRE_BY_TYPE) { 48 autowireByType(beanName, mbd, bw, newPvs); 49 } 50 51 pvs = newPvs; 52 } 53 boolean hasInstAwareBpps = hasInstantiationAwareBeanPostProcessors(); 54 boolean needsDepCheck = (mbd.getDependencyCheck() != RootBeanDefinition.DEPENDENCY_CHECK_NONE); 55 if (hasInstAwareBpps || needsDepCheck) { 56 if (pvs == null) { 57 pvs = mbd.getPropertyValues(); 58 } 59 PropertyDescriptor[] filteredPds = filterPropertyDescriptorsForDependencyCheck(bw, mbd.allowCaching); 60 if (hasInstAwareBpps) { 61 for (BeanPostProcessor bp : getBeanPostProcessors()) { 62 if (bp instanceof InstantiationAwareBeanPostProcessor) { 63 InstantiationAwareBeanPostProcessor ibp = (InstantiationAwareBeanPostProcessor) bp; 64 // TODO @Autowire @Resource @Value @Inject 等注解的依赖注入过程 65 pvs = ibp.postProcessPropertyValues(pvs, filteredPds, bw.getWrappedInstance(), beanName); 66 if (pvs == null) { 67 return; 68 } 69 } 70 } 71 } 72 if (needsDepCheck) { 73 checkDependencies(beanName, mbd, filteredPds, pvs); 74 } 75 } 76 if (pvs != null) { 77 // 注入配置文件中<property>配置的属性 78 applyPropertyValues(beanName, mbd, bw, pvs); 79 } 80 }
上面方法中的31-52行以及78行的applyPropertyValues()方法基本都是用于SpringMVC中采用xml配置Bean的方法。所以我们不做介绍了。看注释知道干嘛的就行了。我们主要看的是pvs = ibp.postProcessPropertyValues(pvs, filteredPds, bw.getWrappedInstance(), beanName);这行代码,这行代码是真正执行采用@Autowire @Resource @Value @Inject 等注解的依赖注入过程。
接着往下看
1 // AutowiredAnnotationBeanPostProcessor类 2 @Override 3 public PropertyValues postProcessPropertyValues( 4 PropertyValues pvs, PropertyDescriptor[] pds, Object bean, String beanName) throws BeanCreationException { 5 // 遍历,获取@Autowire,@Resource,@Value,@Inject等具备注入功能的注解 6 InjectionMetadata metadata = findAutowiringMetadata(beanName, bean.getClass(), pvs); 7 try { 8 // 属性注入 9 metadata.inject(bean, beanName, pvs); 10 } catch (BeanCreationException ex) { 11 throw ex; 12 } catch (Throwable ex) { 13 throw new BeanCreationException(beanName, "Injection of autowired dependencies failed", ex); 14 } 15 return pvs; 16 }
AutowiredAnnotationBeanPostProcessor类实现了postProcessPropertyValues()方法。findAutowiringMetadata(beanName, bean.getClass(), pvs);方法会寻找在当前类中的被@Autowire,@Resource,@Value,@Inject等具备注入功能的注解的属性。
debug看一下结果,如下图所示,成功得到了@Autowire注解的属性webService。
metadata.inject(bean, beanName, pvs);方法开始执行注入的逻辑。
1 // AutowiredAnnotationBeanPostProcessor类 2 @Override 3 protected void inject(Object bean, @Nullable String beanName, @Nullable PropertyValues pvs) throws Throwable { 4 // 需要注入的字段 5 Field field = (Field) this.member; 6 // 需要注入的属性值 7 Object value; 8 if (this.cached) { 9 value = resolvedCachedArgument(beanName, this.cachedFieldValue); 10 } else { 11 // @Autowired(required = false),当在该注解中设置为false的时候,如果有直接注入,没有跳过,不会报错。 12 DependencyDescriptor desc = new DependencyDescriptor(field, this.required); 13 desc.setContainingClass(bean.getClass()); 14 Set<String> autowiredBeanNames = new LinkedHashSet<>(1); 15 Assert.state(beanFactory != null, "No BeanFactory available"); 16 TypeConverter typeConverter = beanFactory.getTypeConverter(); 17 try { 18 // 通过BeanFactory 解决依赖关系 19 // 比如在webController中注入了webService,这个会去BeanFactory中去获取webService,也就是getBean()的逻辑。 20 // 如果存在直接返回,不存在再执行createBean()逻辑。 21 // 如果在webService中依然依赖,依然会去递归。 22 // 这里是一个复杂的递归逻辑。 23 value = beanFactory.resolveDependency(desc, beanName, autowiredBeanNames, typeConverter); 24 } catch (BeansException ex) { 25 throw new UnsatisfiedDependencyException(null, beanName, new InjectionPoint(field), ex); 26 } 27 synchronized (this) { 28 if (!this.cached) { 29 if (value != null || this.required) { 30 this.cachedFieldValue = desc; 31 registerDependentBeans(beanName, autowiredBeanNames); 32 if (autowiredBeanNames.size() == 1) { 33 String autowiredBeanName = autowiredBeanNames.iterator().next(); 34 if (beanFactory.containsBean(autowiredBeanName) && 35 beanFactory.isTypeMatch(autowiredBeanName, field.getType())) { 36 this.cachedFieldValue = new ShortcutDependencyDescriptor( 37 desc, autowiredBeanName, field.getType()); 38 } 39 } 40 } else { 41 this.cachedFieldValue = null; 42 } 43 this.cached = true; 44 } 45 } 46 } 47 if (value != null) { 48 ReflectionUtils.makeAccessible(field); 49 field.set(bean, value); 50 } 51 }
debug看一下field。如下图所示正是我们的webService
看这行代码:value = beanFactory.resolveDependency(desc, beanName, autowiredBeanNames, typeConverter);注意beanFactory依旧是我们熟悉的IoC容器的具体产品,也就是实现类DefaultListableBeanFactory。见到就说一遍,方便大家记住它,很重要。
在resolveDependency()方法中经过一顿操作,最终又会来到上面的getBean()方法。以上就是依赖注入的整个过程。注意看代码中的注释哦。
2.2.3、initializeBean():调用Bean的初始化方法
设置Bean的初始化方法有两种方法,一种是在xml或者@Bean指定init-method方法。另一种是让bean实现InitializingBean接口重写afterPropertiesSet()方法。
1 // AbstractAutowireCapableBeanFactory类 2 protected Object initializeBean(final String beanName, final Object bean, @Nullable RootBeanDefinition mbd) { 3 if (System.getSecurityManager() != null) { 4 AccessController.doPrivileged((PrivilegedAction<Object>) () -> { 5 invokeAwareMethods(beanName, bean); 6 return null; 7 }, getAccessControlContext()); 8 } 9 else { 10 //在调用Bean的初始化方法之前,调用一系列的aware接口实现,把相关的BeanName,BeanClassLoader,以及BeanFactory注入到Bean中去。 11 invokeAwareMethods(beanName, bean); 12 } 13 14 Object wrappedBean = bean; 15 if (mbd == null || !mbd.isSynthetic()) { 16 // 这些都是钩子方法,在反复的调用,给Spring带来了极大的可拓展性 17 // 初始化之前调用BeanPostProcessor 18 wrappedBean = applyBeanPostProcessorsBeforeInitialization(wrappedBean, beanName); 19 } 20 21 try { 22 // 调用指定的init-method方法 23 invokeInitMethods(beanName, wrappedBean, mbd); 24 } 25 catch (Throwable ex) { 26 throw new BeanCreationException( 27 (mbd != null ? mbd.getResourceDescription() : null), 28 beanName, "Invocation of init method failed", ex); 29 } 30 if (mbd == null || !mbd.isSynthetic()) { 31 // 这些都是钩子方法,在反复的调用,给Spring带来了极大的可拓展性 32 // 初始化之后调用BeanPostProcessor 33 wrappedBean = applyBeanPostProcessorsAfterInitialization(wrappedBean, beanName); 34 } 35 36 return wrappedBean; 37 }
在调用Bean的初始化方法之前,调用一系列的aware接口实现,把相关的BeanName,BeanClassLoader,以及BeanFactory注入到Bean中去。接着会执行invokeInitMethods()方法。
1 // AbstractAutowireCapableBeanFactory类 2 protected void invokeInitMethods(String beanName, final Object bean, @Nullable RootBeanDefinition mbd) 3 throws Throwable { 4 // 除了使用init-method指定的初始化方法,还可以让bean实现InitializingBean接口重写afterPropertiesSet()方法 5 boolean isInitializingBean = (bean instanceof InitializingBean); 6 if (isInitializingBean && (mbd == null || !mbd.isExternallyManagedInitMethod("afterPropertiesSet"))) { 7 if (logger.isDebugEnabled()) { 8 logger.debug("Invoking afterPropertiesSet() on bean with name '" + beanName + "'"); 9 } 10 if (System.getSecurityManager() != null) { 11 try { 12 AccessController.doPrivileged((PrivilegedExceptionAction<Object>) () -> { 13 ((InitializingBean) bean).afterPropertiesSet(); 14 return null; 15 }, getAccessControlContext()); 16 } 17 catch (PrivilegedActionException pae) { 18 throw pae.getException(); 19 } 20 } 21 else { 22 // 执行afterPropertiesSet()方法进行初始化 23 ((InitializingBean) bean).afterPropertiesSet(); 24 } 25 } 26 27 // 先执行afterPropertiesSet()方法,再进行init-method 28 if (mbd != null && bean.getClass() != NullBean.class) { 29 String initMethodName = mbd.getInitMethodName(); 30 if (StringUtils.hasLength(initMethodName) && 31 !(isInitializingBean && "afterPropertiesSet".equals(initMethodName)) && 32 !mbd.isExternallyManagedInitMethod(initMethodName)) { 33 invokeCustomInitMethod(beanName, bean, mbd); 34 } 35 } 36 }
可见该方法中首先判断Bean是否配置了init-method方法,如果有,那么通过invokeCustomInitMethod()方法来直接调用。其中在invokeCustomInitMethod()方法中是通过JDK的反射机制得到method对象,然后调用的init-method。最终完成Bean的初始化。
三、总结
SpringBoot启动流程相关的博文到这里就结束了,在前面的文章中,我们详细介绍了IoC容器的设计与实现,并结合SpringBoot的启动流程介绍了IoC容器的初始化过程,及IoC容器的依赖注入,及大家都很关心的SpringBoot是如何实现自动装配的。关于SpringBoot的源码分析基本就到这里了,后面有计划写写AOP的实现,以及很重要的Spring事务实现。
小半个月时间,熬了很多个凌晨,终于写完了SpringBoot启动流程系列博文。有辛苦更有收获,我是从今年才开始写博客的,以前每当想写的时候,总想什么时候能力足够了,憋个大招再写。回过头来看之前的想法是十分错误的,写博文不只是为了分享,最重要的是在写博客的过程中能够系统的梳理一下某一个技术栈的知识。好记性不如烂笔头,看别人的博客很难get到自己想要的点,所以也是让自己持续进步的一种方式。加油,各位。
原创不易,转载请注明出处。
如有错误的地方还请留言指正。