Develop Reference

Spring bean creation lifecycle : Why having multiple interaction points?

I'm learning spring and I'm stuck with the bean lifecycle !
When creating a bean, spring gives us several maners to interact with it. We can use one or all of :
BeanFactoryPostProcessor
BeanPostProcessor#postProcessBeforeInitialization
#PostConstruct
InitializingBean#afterPropertiesSet
#Bean#initMethod
BeanPostProcessor#postProcessAfterInitialization
Spring calls them in the order above
My question is : why all these points of interaction ? and when to use each of them (the use case) ?

A BeanFactoryPostProcessor and a BeanPostProcessor are quite different beast and also apply to different things.
A BeanFactoryPostProcessor will operate on the metadata for a bean (I like to call it the recipe) and will post process that. A well know example is the PropertySourcesPlaceholderConfigurer which will inject/replace all #Value in configuration with the value. A BeanFactoryPostProcessor operates on the metadata and thus before any bean has been created.
The BeanPostProcessor can be applied to a bean and can even replace a bean. Spring AOP uses this quite a lot. An example is the PersistenceExceptionTranslationPostProcessor, when a bean has been created it will pass through this BeanPostProcessor and when it is annotated with #Persistence the bean will be replaced by a proxy. This proxy adds exception translation (i.e. it will convert JpaException and SQLException to the Spring DataAccessException). This is done in a BeanPostProcessor. And can be be done before the init-callbacks are called (the postProcessBeforeInitializationor after they have been called thepostProcessAfterInitialization). The PersistenceExceptionTranslationPostProcessorwill run in thepostProcessAfterInitialization` as it needs to be certain the bean has been initialized.
The ServletContextAwareProcessor will run right after the object has been created to inject the ServletContext as early as possible as the initializing of a bean might depend on it.
The 3 callbacks for initializing a bean are more or less the same but are called in sequence because they have been included in later versions. It starter with only an interface InitializingBean and init-method in xml (later also added to #Bean and the annotation support was added when annotations became a thing.
You need init methods to initialize a bean, you might want to check if all properties have been set (like a required datasource) or you might want to start something. A DataSource (especially a connection pool) is a good example to initialize. After all dependencies have been injected you want to start the pool so it will open the connections. Now as you probably cannot modify the code you want to use the init-method if you control the code you probably want to add #PostConstruct. If you are writing an framework that depends on Spring I would use the InitializingBean.
Next to those 3 init methods you also have the destruction counter-parts. The DisposableBean interface (and destroy-method) and the #PreDestroy annotation. Again when you stop your application you also want to close the connections in your connection pool and you want to probably call the close method on the DataSource. A perfect sample of a destruction callback.

Spring without getBean(..)

I'm trying to get a better understanding of the #Autowired annotations component scanning, but all the examples I found so far use context.getBean(..) at some point to get at least one Bean to start with.
I also read that doing that is considered bad practice , but I can't seem to find any information on how to do it without context.getBean(..)
Could somebody please enlighten me with an example and information on how to do this ?

Define your bean in xml and use
<context:component-scan base-package="com" />
<mvc:annotation-driven />
Bean def
<bean id="processEngine" class="com.processEngine">
<property name="processEngineConfiguration" ref="processEngineConfiguration" />
</bean>
now you can get bean as following
#Autowired
private ProcessEngine processEngine;

how it works
spring scans the bean's recipes either from xml or java configuration. then spring creates a beanDefinitions which are 'loaded' into BeanFactory. BeanFactory triggers a set of BeanPostProcessors (BPP) which are scanning classes for particular annotations like Autowired/Resource/PostProcessor and etc. and do appropriate actions. in case when your class contains #Autowired annotation, AutowiredAnnotationBeanPostProcessor would auto wire required field (dependencies), and when creation of an object is done and all BPP worked out, object is ready to be used by the app, from this point your code can get 'ready' objects from container.
there are some cases when you would need to access this beans from the code which is out of spring's control and not managed by container. in order to do so, you would need to get the ApplicationContext (container) and call #getBean specifying either name or type. using applicationContext directly is not a good practice because there are some problems that you can come to, f.ex. id of a bean might be changed and if you refer to bean by id then NPE would be thrown.
configuration
there are several approaches to configure spring to scan the classes for finding bean recipes. one would be defining component-scan, in this case classes which are located in the path that you've set and having any of valid spring annotations like #Component, #Service, #Repository, #Controller (for web container) would be considered. another way would be specifying each bean separately using <bean> or #Bean.
examples.
if you want to create a web app then you should see DispatcherServlet with ContextLoaderListener classes. this classes would boot your app and load everything according to configuration. f.ex. here,
but if you want to create a desktop app, then you would end up with something like this

From time to time (usually when not using Spring Boot), I use something along the lines of the following code:
public static <T> T autowire(ApplicationContext ctx, T bean) {
ctx.getAutowireCapableBeanFactory().autowireBean(bean);
return bean;
}
In my main, I create an instance of the main application class that contains a few #Autowired annotations for the main services / entry points to my Spring application.

Instantiating arbitrary classes thru Spring

I'm writing a service registry class. This service registry will scan packages for annotated classes and then populate some internal map. What I need then, is to be able to query for services (by String name) using some method (let's say Object get(String name)). This method will then search internal map for a service with such name and returns instance.
What I'm doing right now, is having this ServiceRegistryBean implement ApplicationContextAware and BeanDefinitionRegistryPostProcessor and a list of Strings (package names) given on construct.
Then, as soon as the bean is constructed, registry post processor kicks in (see note) and the registry class adds the service classes as new beans (singleton, lazy loaded) to the spring bean registry. Then, getting the service instance is as simple as requesting a bean from context, returning singleton instance.
My question is: is there a better way in Spring to do this? I've looked into bean factories, but it does not seem to me the same. Support for auto-wiring and DI in service instances is essential, that's why I want Spring to instantiate it. Also, I like the idea of Spring taking care of singletons.
Note: I've found out, that when I inline the bean creation in <constructor-arg> (that is, bean is not named and is just an instance passed as constructor argument of other bean - in my case, I'm passing registry as a parameter to other constructor), BeanDefinitionRegistryPostProcessor interface methods (namely public void postProcessBeanDefinitionRegistry(BeanDefinitionRegistry registry)) is not called at all! I'm quite sure, it is some behavior of Spring I don't know about, but I was not able to find proper documentation for the post-processor.
Thank you for any hints or insights!

Scanning for custom annotations it's already supported, you only need to add a include-filter to <context:component-scan> tag, ie
<context:component-scan base-package="org.example">
<context:include-filter type="annotation" expression="some.Annotation"/>
</context:component-scan>
see http://static.springsource.org/spring/docs/current/spring-framework-reference/html/beans.html#beans-scanning-filters
If you turn on default-lazy-init I suppose that the DI Container is ready to use as Service Locator Registry.
About the note, only root bean definitions are taken into account when looking for BeanFactoryPostProcessors, inner beans are ignored.
Usually, scanning and registering beans is done by BeanDefinitionParsers instead because you known when the beans are registered and beans are visible for tools, like STS Spring Bean Explorer, but using a BeanDefinitionRegistryPostProcessor is correct. The interface ensures that beans are defined before other BeanFactoryPostProcessors run.

Can I replace a Spring bean definition at runtime?

Consider the following scenario. I have a Spring application context with a bean whose properties should be configurable, think DataSource or MailSender. The mutable application configuration is managed by a separate bean, let's call it configuration.
An administrator can now change the configuration values, like email address or database URL, and I would like to re-initialize the configured bean at runtime.
Assume that I can't just simply modify the property of the configurable bean above (e.g. created by FactoryBean or constructor injection) but have to recreate the bean itself.
Any thoughts on how to achieve this? I'd be glad to receive advice on how to organize the whole configuration thing as well. Nothing is fixed. :-)
EDIT
To clarify things a bit: I am not asking how to update the configuration or how to inject static configuration values. I'll try an example:
<beans>
<util:map id="configuration">
<!-- initial configuration -->
</util:map>
<bean id="constructorInjectedBean" class="Foo">
<constructor-arg value="#{configuration['foobar']}" />
</bean>
<bean id="configurationService" class="ConfigurationService">
<property name="configuration" ref="configuration" />
</bean>
</beans>
So there's a bean constructorInjectedBean that uses constructor injection. Imagine the construction of the bean is very expensive so using a prototype scope or a factory proxy is not an option, think DataSource.
What I want to do is that every time the configuration is being updated (via configurationService the bean constructorInjectedBean is being recreated and re-injected into the application context and dependent beans.
We can safely assume that constructorInjectedBean is using an interface so proxy magic is indeed an option.
I hope to have made the question a little bit clearer.

Here is how I have done it in the past: running services which depend on configuration which can be changed on the fly implement a lifecycle interface: IRefreshable:
public interface IRefreshable {
// Refresh the service having it apply its new values.
public void refresh(String filter);
// The service must decide if it wants a cache refresh based on the refresh message filter.
public boolean requiresRefresh(String filter);
}
Controllers (or services) which can modify a piece of configuration broadcast to a JMS topic that the configuration has changed (supplying the name of the configuration object). A message driven bean then invokes the IRefreshable interface contract on all beans which implement IRefreshable.
The nice thing with spring is that you can automatically detect any service in your application context that needs to be refreshed, removing the need to explicitly configure them:
public class MyCacheSynchService implements InitializingBean, ApplicationContextAware {
public void afterPropertiesSet() throws Exception {
Map<String, ?> refreshableServices = m_appCtx.getBeansOfType(IRefreshable.class);
for (Map.Entry<String, ?> entry : refreshableServices.entrySet() ) {
Object beanRef = entry.getValue();
if (beanRef instanceof IRefreshable) {
m_refreshableServices.add((IRefreshable)beanRef);
}
}
}
}
This approach works particularly well in a clustered application where one of many app servers might change the configuration, which all then need to be aware of. If you want to use JMX as the mechanism for triggering the changes, your JMX bean can then broadcast to the JMS topic when any of its attributes are changed.

I can think of a 'holder bean' approach (essentially a decorator), where the holder bean delegates to holdee, and it's the holder bean which is injected as a dependency into other beans. Nobody else has a reference to holdee but the holder. Now, when the holder bean's config is changed, it recreates the holdee with this new config and starts delegating to it.

You should have a look at JMX. Spring also provides support for this.
Spring 2.0.x
Spring 2.5.x
Spring 3.0.x

Further updated answer to cover scripted bean
Another approach supported by spring 2.5.x+ is that of the scripted bean. You can use a variety of languages for your script - BeanShell is probably the most intuitive given that it has the same syntax as Java, but it does require some external dependencies. However, the examples are in Groovy.
Section 24.3.1.2 of the Spring Documentation covers how to configure this, but here are some salient excerpts illustrating the approach which I've edited to make them more applicable to your situation:
<beans>
<!-- This bean is now 'refreshable' due to the presence of the 'refresh-check-delay' attribute -->
<lang:groovy id="messenger"
refresh-check-delay="5000" <!-- switches refreshing on with 5 seconds between checks -->
script-source="classpath:Messenger.groovy">
<lang:property name="message" value="defaultMessage" />
</lang:groovy>
<bean id="service" class="org.example.DefaultService">
<property name="messenger" ref="messenger" />
</bean>
</beans>
With the Groovy script looking like this:
package org.example
class GroovyMessenger implements Messenger {
private String message = "anotherProperty";
public String getMessage() {
return message;
}
public void setMessage(String message) {
this.message = message
}
}
As the system administrator wants to make changes then they (or you) can edit the contents of the script appropriately. The script is not part of the deployed application and can reference a known file location (or one that is configured through a standard PropertyPlaceholderConfigurer during startup).
Although the example uses a Groovy class, you could have the class execute code that reads a simple properties file. In that manner, you never edit the script directly, just touch it to change the timestamp. That action then triggers the reload, which in turn triggers the refresh of properties from the (updated) properties file, which finally updates the values within the Spring context and off you go.
The documentation does point out that this technique doesn't work for constructor-injection, but maybe you can work around that.
Updated answer to cover dynamic property changes
Quoting from this article, which provides full source code, one approach is:
* a factory bean that detects file system changes
* an observer pattern for Properties, so that file system changes can be propagated
* a property placeholder configurer that remembers where which placeholders were used, and updates singleton beans’ properties
* a timer that triggers the regular check for changed files
The observer pattern is implemented by
the interfaces and classes
ReloadableProperties,
ReloadablePropertiesListener,
PropertiesReloadedEvent, and
ReloadablePropertiesBase. None of them
are especially exciting, just normal
listener handling. The class
DelegatingProperties serves to
transparently exchange the current
properties when properties are
updated. We only update the whole
property map at once, so that the
application can avoid inconsistent
intermediate states (more on this
later).
Now the
ReloadablePropertiesFactoryBean can be
written to create a
ReloadableProperties instance (instead
of a Properties instance, as the
PropertiesFactoryBean does). When
prompted to do so, the RPFB checks
file modification times, and if
necessary, updates its
ReloadableProperties. This triggers
the observer pattern machinery.
In our case, the only listener is the
ReloadingPropertyPlaceholderConfigurer.
It behaves just like a standard spring
PropertyPlaceholderConfigurer, except
that it tracks all usages of
placeholders. Now when properties are
reloaded, all usages of each modified
property are found, and the properties
of those singleton beans are assigned
again.
Original answer below covering static property changes:
Sounds like you just want to inject external properties into your Spring context. The PropertyPlaceholderConfigurer is designed for this purpose:
<!-- Property configuration (if required) -->
<bean id="serverProperties" class="org.springframework.beans.factory.config.PropertyPlaceholderConfigurer">
<property name="locations">
<list>
<!-- Identical properties in later files overwrite earlier ones in this list -->
<value>file:/some/admin/location/application.properties</value>
</list>
</property>
</bean>
you then reference the external properties with Ant syntax placeholders (that can be nested if you want from Spring 2.5.5 onwards)
<bean id="example" class="org.example.DataSource">
<property name="password" value="${password}"/>
</bean>
You then ensure that the application.properties file is only accessible to the admin user and the user running the application.
Example application.properties:
password=Aardvark

Or you could use the approach from this similar question and hence also my solution:
The approach is to have beans that are configured via property files and the solution is to either
refresh the entire applicationContext (automatically using a scheduled task or manually using JMX) when properties have changed or
use a dedicated property provider object to access all properties. This property provider will keep checking the properties files for modification. For beans where prototype-based property lookup is impossible, register a custom event that your property provider will fire when it finds an updated property file. Your beans with complicated lifecycles will need to listen for that event and refresh themselves.

You can create a custom scope called "reconfigurable" into the ApplicationContext. It creates and caches instances of all beans in this scope. On a configuration change it clears the cache and re-creates the beans on first access with the new configuration. For this to work you need to wrap all instances of reconfigurable beans into an AOP scoped proxy, and access the configuration values with Spring-EL: put a map called config into the ApplicationContext and access the configuration like #{ config['key'] }.

This is not something I tried, I am trying to provide pointers.
Assuming your application context is a subclass of AbstractRefreshableApplicationContext(example XmlWebApplicationContext, ClassPathXmlApplicationContext). AbstractRefreshableApplicationContext.getBeanFactory() will give you instance of ConfigurableListableBeanFactory. Check if it is instance of BeanDefinitionRegistry. If so you can call 'registerBeanDefinition' method. This approach will be tightly coupled with Spring implementation,
Check the code of AbstractRefreshableApplicationContext and DefaultListableBeanFactory(this is the implementation you get when you call 'AbstractRefreshableApplicationContext getBeanFactory()')

Option 1 :
Inject the configurable bean into the DataSource or MailSender. Always get the configurable values from the configuration bean from within these beans.
Inside the configurable bean run a thread to read the externally configurable properties (file etc..) periodically. This way the configurable bean will refresh itself after the admin had changed the properties and so the DataSource will get the updated values automatically.
You need not actually implement the "thread" - read : http://commons.apache.org/configuration/userguide/howto_filebased.html#Automatic_Reloading
Option 2 (bad, i think, but maybe not - depends on use case) :
Always create new beans for beans of type DataSource / MailSender - using prototype scope. In the init of the bean, read the properties afresh.
Option 3 :
I think, #mR_fr0g suggestion on using JMX might not be a bad idea. What you could do is :
expose your configuration bean as a MBean (read http://static.springsource.org/spring/docs/2.5.x/reference/jmx.html)
Ask your admin to change the configuration properties on the MBean (or provide an interface in the bean to trigger property updates from their source)
This MBean (a new piece of java code that you will need to write), MUST keep references of Beans (the ones that you want to change / inject the changed properties into). This should be simple (via setter injection or runtime fetch of bean names / classes)
When the property on the MBean is changed (or triggered), it must call the appropriate setters on the respective beans. That way, your legacy code does not change, you can still manage runtime property changes.
HTH!

You may want to have a look at the Spring Inspector a plug-gable component that provides programmatic access to any Spring based application at run-time. You can use Javascript to change configurations or manage the application behaviour at run-time.

Here is the nice idea of writing your own PlaceholderConfigurer that tracks the usage of properties and changes them whenever a configuration change occurs. This has two disadvantages, though:
It does not work with constructor injection of property values.
You can get race conditions if the reconfigured bean receives a
changed configuration while it is processing some stuff.

My solution was to copy the original object. Fist i created an interface
/**
* Allows updating data to some object.
* Its an alternative to {#link Cloneable} when you cannot
* replace the original pointer. Ex.: Beans
* #param <T> Type of Object
*/
public interface Updateable<T>
{
/**
* Import data from another object
* #param originalObject Object with the original data
*/
public void copyObject(T originalObject);
}
For easing the implementation of the function fist create a constructor with all fields, so the IDE could help me a bit. Then you can make a copy constructor that uses the same function Updateable#copyObject(T originalObject). You can also profit of the code of the constructor created by the IDE to create the function to implement:
public class SettingsDTO implements Cloneable, Updateable<SettingsDTO>
{
private static final Logger LOG = LoggerFactory.getLogger(SettingsDTO.class);
#Size(min = 3, max = 30)
private String id;
#Size(min = 3, max = 30)
#NotNull
private String name;
#Size(min = 3, max = 100)
#NotNull
private String description;
#Max(100)
#Min(5)
#NotNull
private Integer pageSize;
#NotNull
private String dateFormat;
public SettingsDTO()
{
}
public SettingsDTO(String id, String name, String description, Integer pageSize, String dateFormat)
{
this.id = id;
this.name = name;
this.description = description;
this.pageSize = pageSize;
this.dateFormat = dateFormat;
}
public SettingsDTO(SettingsDTO original)
{
copyObject(original);
}
#Override
public void copyObject(SettingsDTO originalObject)
{
this.id = originalObject.id;
this.name = originalObject.name;
this.description = originalObject.description;
this.pageSize = originalObject.pageSize;
this.dateFormat = originalObject.dateFormat;
}
}
I used it in a Controller for updating the current settings for the app:
if (bindingResult.hasErrors())
{
model.addAttribute("settingsData", newSettingsData);
model.addAttribute(Templates.MSG_ERROR, "The entered data has errors");
}
else
{
synchronized (settingsData)
{
currentSettingData.copyObject(newSettingsData);
redirectAttributes.addFlashAttribute(Templates.MSG_SUCCESS, "The system configuration has been updated successfully");
return String.format("redirect:/%s", getDao().getPath());
}
}
So the currentSettingsData which has the configuration of the application gonna have the updated values, located in newSettingsData. These method allows updating any bean without high complexity.

Auto-cast Spring Beans

Is there a way to auto-cast Spring beans to the class defined in the application context XML? I'd like to avoid putting type information about the beans in 2 places.... in the xml configuration file and also in the code as a cast.
For instance, given this config file
<bean id="bean-name" class="SimpleSpringBean" scope="prototype">
<property name="myValue" value="simple value"></property>
</bean>
Can I call ApplicationContext.getBean("bean-name") in such a way as to avoid directly casting the return type to SimpleStringBean. I know I can also call ApplicationContext.getBean("bean-name", SimpleSpringBean.class) to avoid the cast itself, but I still have the type info in 2 places.
It seems that Spring can get the class info (ApplicationContext.getType) or by getting the type from the bean itself, but no way to automatically cast the type without programmer intervention.

I agree with Sii, you should avoid calling getBean as much as you can. Just wire your beans to classes that depends on them.
Still, if you have a single class that holds the application context, you can provide a wrapper generic method like the following:
class MyContextHolder{
ApplicationContext appContext;
......
#SuppressWarnings("unchecked")
public static <T> T getBean(String beanName)
{
return (T)appContext.getBean(beanName);
}
}
Then you can call it without casting
MyClass mc = MyContextHolder.getBean("myClassBean");

The answer is you shouldn't be using ApplicationContext.getBean() at all if it's possible, and bear with the one place you have to in the bootstrap code. (Generally, you should never need to use getBean() outside of your application's entry points.)
Also, what you're asking is likely impossible in the Java language at all. Casting is a compile-time feature, combined with a runtime check. The return type of getBean() simply must be known at compile time. Even if Spring can determine the type of an object, it can't change its own method signatures at runtime.
Another thing is that even if this were possible, the feature wouldn't be all that useful. Because Spring AOP is implemented using dynamic proxies, you nearly always want Spring to hand you an instance of an interface the bean implements (which could be an AOP proxy), not of the implementation class.

Another approach I also use is autowiring the bootstrapping class using:
public class Main {
#Autowired FooFacade foo;
#Autowired BarFacade bar;
public static void main(String[] args) {
ApplicationContext ctx = new ClassPathXmlApplicationContext("appCtx.xml");
AutowireCapableBeanFactory bf = ctx.getAutowireCapableBeanFactory();
Object main = bf.createBean(Main.class,
AutowireCapableBeanFactory.AUTOWIRE_AUTODETECT,
false);
((Main) main).run();
}
private void run() {
foo.doBootstrapStuff();
bar.doMoreBootstrapStuff();
}
}
(Code done from memory. Might only work if you have the Spring context configured to process wiring annotations, in that case making setters for foo and bar should work.)

The main reason for getBean being untyped is the compatibility of Spring (up to version 2.5.x) with Java 1.4. Spring 3.0 will drop that and thus offer typed getBean method then.
Nevertheless you should avoid looking up beans directly and minimize its usage as far as possible.

What if I use Spring as an object factory that my application uses extensively.
That is, instead of writing a bunch of classes that already inherit or wrap around
known Java classes I just decide to move all that into an xml file to cut down on
lines of Java code. This will mean many lines of xml but I won't need to make
skeleton Java classes which I inject with Spring or autowire. Thus making the lines
of Java code less.
For this reason and still as I am new to Spring I have just as stated in previous
posts used static Java methods which wrap around the getBeans().
I have worked with Spring but it is still new to me so forgive my question.

"Because Spring AOP is implemented using dynamic proxies, you nearly always want Spring to hand you an instance of an interface the bean implements (which could be an AOP proxy), not of the implementation class"
So there is no way to get a dynamic proxy using getBean(), then whats the best practice if there are no interfaces and its a stand alone driver class to be executed?