When using Spring it is rather cumbersome to incorporate runtime data in bean construction. While there are techniques to circumvent this separation, I have a feeling that it was put in place for a reason. My question is whether this is a known paradigm, and if there is any litterature discussing it. Personally I find that it has both advantages and drawbacks, depending on the dynamicity of the app.
You have at least five well known methods to pass runtime data to beans configuration:
Use ApplicationContextInitializer to add PropertySources to the Enviroment.
Use SPEL to inject dependencies.
Use FactoryBeans.
Use the factory-bean and factory-method attributes.
If you write the class that need the runtime data, you only need to inject the
collaborator that provide it.
For example
<bean id="requestAttributeReader" class="example.RequestAttributeReader" />
<bean id="requestInjectedBean" class="example.RequestInjectedBean" scope="request">
<property name="a" value="#{requestAttributeReader.a}" />
</bean>
Class RequestAttributeReader {
public String getA() {
return RequestContextHolder.getAttributes().getAttribute("a");
}
}
EDIT
The bean description files of an IoC container lets you to configure implementors on application beans. This is normally a static definition of the implementation classes that you want to use for a concrete configuration, so xml it's good for it.
If you need to choose an implementor based on runtime then you need to write code to choose them and then inform the container.
for example, using PropertySources and PropertyPlaceholderConfigurer:
String service = "example.NormalService";
if (BOSS_USERNAME.equals(System.getProperty("user.name")))
service = "example.BossService";
ctx.getEnvironment().getPropertySources().addFirst(new PropertiesPropertySource("service", service));
<bean id="service" class="${service}" />
The same could be done with a ServiceFactoryBean, a external ServiceFactory, SPEL and so on...
Maybe, you are interested on replacing implementations at runtime, ie changing the Service implementation in all beans that depends on when the container is already refreshed. (without destroy and refresh).
As far as I know, the framework don't provides a clear way to do it.
Sounds like you should look at spring binding, eg:
public String create(#Valid Market market, BindingResult bindingResult, Model uiModel, HttpServletRequest httpServletRequest) {
So this will take request params that match the fields in the Market object and set them in that object. It will also validated the params/object since there is the #Valid annotation.
This binding can be customised with PropertyEditors or Converters.
If the market object is annotated #Configurable, it can use #Autowired and #Value annotations to configure the bean when spring creates it.
Related
I'm sure this has been asked a hundred times before but I can't seem to find the question so feel free to refer me to other stackoverflow answers.
What do most Spring users do for objects that are non-singleton beans that require injection? For example, I have classes like Customer where I want to instantiate a new one each time. Lets say it is an entity and I want to inject listeners to iterate through in #PreRemove or somewhere else. The usual solution is to use #Configurable but that almost seems a workaround and I was wonder if there was a more canonical way to handle these.
The only thing I can think of is to create a factory newCustomer instance method in my CustomerRepository class which IS a managed bean. Then instead of injecting listeners into Customer (the most natural place) I inject them into the CustomerRepository and specify them as an explicit constructor argument to Customer ala new Customer( injectedListeners ).
Do people tend to just use Configurable or is there a better way to inject non-singleton instances? Or do most users create a factory method as above? The entity example is just an example, I have other objects that are non-singleton, are typically new'd but require injection.
Would this be handled differently in something like Guice? How would you do it just using JSR-330 features?
You can make beans non-singletons if you like. Depends on whether you are ok with the XML:
<bean id="beanA" class="misc.BeanClass" scope="prototype">
<property ... />
</bean>
Which will give you a new instance every time. Each instance will be initialized with injected values.
You can use this annotation too:
#Scope("prototype")
I have one small issue with creating a new bean. Basically as per request, I get some parameters, which needs to be passed to a bean. Below I am instantiating ControllerService for each request. Rather I would like it to be a bean with scope=protype. So that I get a fresh object for every request.
But then how do i set the 2 properties (kpiName, kpiInput) that I am sending via constructors in the bean??
#Autowired
#Qualifier("serviceManager")
Cleanser serviceManager;
#RequestMapping(method = RequestMethod.POST)
public #ResponseBody
String getKPIResult(#RequestParam("kpiName") String kpiName,
#RequestParam("kpiInput") String kpiInput) {
return serviceManager.checkAndExecute(new ControllerService(kpiName, kpiInput));
}
In situations like this where you're going against the grain of Spring, I'd suggest that perhaps you're doing something in a way that's not considered best practice. Without more context it's hard to see though.
Spring Social uses a request scope bean to embody a repository for a specific user. I've now idea why as it's a horribly inefficient way of doing things, and much less understandable IMHO.
<bean id="connectionRepository" factory-method="createConnectionRepository"
factory-bean="usersConnectionRepository" scope="request">
<constructor-arg
value="#{T(org.springframework.security.core.context.SecurityContextHolder).getContext().getAuthentication().getPrincipal()}" />
<aop:scoped-proxy proxy-target-class="false" />
</bean>
You can see here the use of factory-bean and factory-method to declare a class/method to call when wanting an instance of your class. The constructor argument is passed using SpEL. I'm not quite sure how you'd achieve this with Spring MVC responding to web requests, but I'm fairly sure you could use Spring Integration to pass a message and use SpEL to grab headers/payload form that message to pass to the constructor.
Again though, I'd really question your design pattern here - a more usual SOA idiom is to create services on startup, and have them as stateless as possible from there-on in, rather than create an instance with specific state for each request. Best of luck!
Don't. The Controller as it's intended in Spring MVC is largely derived from the old Java servlet, which should be stateless by specification.
In fact, Controller-objects are hard-cached inside the handler mapping framework and not fetched from the bean context on each request. Setting scope to "prototype" would, effectively, do nothing as the handler (controller) is really only gotten once.
Just a quick question: In an OOP MVC application, one key principle is the seperation of responsibilities. I therefor think that a model and the object that fetches the model from a database, file, xml, webservice, etc. should be seperated from the model itself. This can for example be done by implementing a datamapper.
However, what do I do when I have a model that can be loaded from different sources? Should the model be in charge of the datasource, or is this the responsibility of the controller?
An simple example could be a config class that can be loaded from a database or a file. Should the controller instruct the datasource, or should the model know when to load the config info from a database or a file?
have used frameworks were the datasource is informed by the controller MachII, Model-Glue (Coldfusion frameworks) and also from the model layer (ColdSpring) - like Spring in Java.
I think the key thing is to use what makes more sense to you, keep the coupling to a minimum and be consistent, meaning don't put datasource or object dependencies in multiple places.
You could also consider using a service type object to abstract the datasource and have it serve either who it likes.
That IOC file could look like this:
<?xml version="1.0" encoding="UTF-8"?>
<beans>
<bean id="chartShareObj" class="model.charts.ChartShared" autowire="byType" />
<bean id="trendChartObj" class="model.charts.TrendChart" autowire="byType" />
<bean id="adminRightsDA0" class="org.datamentor.institution.RightsDAO">
<constructor-arg name="dsn">
<value>${dsn_dm}</value>
</constructor-arg>
</bean>
<bean id="assessmentManager" class="model.assessment.Manager">
<constructor-arg name="dsn">
<value>${dsn_au}</value>
</constructor-arg>
</bean>
</beans>
You can see the different datasources specified by args via args defined in a controller.
Based on your situation and response I would suggest looking into dependency injection. You can then let it handle determining which data source to use based on whatever set of variables you want to let it determine things by. This is what I use when I have multiple data sources and want to have the data source determined by some predetermined factors I have chosen.
http://en.wikipedia.org/wiki/Dependency_injection
As to who should handle the injection, I leave that to a repository factory and simply ask for an interface in the controller. The factory then determines based on the dependency injection which repository to provide.
Example:
Dependency Injection in an global Infrastructure class:
Bind<INewsArticleRepository>().ToMethod(context => NewsRepositoryFactory.Create((NewsRepositoryFactory.RepositoryType)Enum.Parse(typeof(NewsRepositoryFactory.RepositoryType), ConfigurationManager.AppSettings["NewsArticleRepositoryProvider"])));
Repository Factory
public static INewsArticleRepository Create(RepositoryType type)
{
switch (type)
{
case RepositoryType.Mock:
return new MockNewsArticlesRepository();
case RepositoryType.Sql:
return new SqlNewsArticleRepository();
default:
throw new NotImplementedException();
}
}
Call in the controller for a repository
private INewsArticleRepository newsItemRepository;
public NewsController(INewsArticleRepository newsItemRepository)
{
this.newsItemRepository = newsItemRepository;
}
They way I do it in Coldbox is using CB's INJECT method in the model. In the cfargument of the constructor, I specify:
<cfargument name="dsn" type="any" inject="coldbox:datasource:dsn">
And that's from specifying the dsn in the coldbox.cfc file, and calling it "dsn". I keep it generic so I can copy this stuff to other projects and only have to change the DSN name in the coldbox.cfc.
But then after doing that, you get the dsn like this:
variables.dsn = arguments.dsn.getName();
I hope this helps, at least a little.
Rob
This might be an obvious but I'm having a hard time understanding why we need to define the class of a bean in two places....
From the spring reference manual...
...
<bean id="petStore"
class="org.springframework.samples.jpetstore.services.PetStoreServiceImpl">
<property name="accountDao" ref="accountDao"/>
<property name="itemDao" ref="itemDao"/>
<!-- additional collaborators and configuration for this bean go here -->
</bean>
// retrieve configured instance
PetStoreServiceImpl service = context.getBean("petStore", PetStoreServiceImpl.class);
Shouldn't the xml fine be enough for the container to know the class of petStore?
You can use the following method:
context.getBean("petStore")
However, as this returns a java.lang.Object, you'd still need to have a cast:
PetStoreServiceImpl petstore = (PetStoreServiceImpl)context.getBean("petStore");
However, this could lead to problems if your "petStore" bean is not actually a PetStoreServiceImpl, and to avoid casts (which since the advent of Generics are being seen as a bit dirty), you can use the above method to infer the type (and let's spring check whether the bean you're expecting is really of the right class, so hence you've got:
PetStoreServiceImpl service = context.getBean("petStore", PetStoreServiceImpl.class);
Hope that helps.
EDIT:
Personally, I would avoid calling context.getBean() to lookup methods as it goes against the idea of dependency injection. Really, the component that uses the petstore bean should have a property, which can then be injected with the correct component.
private PetStoreService petStoreService;
// setter omitted for brevity
public void someotherMethod() {
// no need for calling getBean()
petStoreService.somePetstoreMethod();
}
Then you can hook up the beans in the application context:
You could also do away with the configuration via XML and use annotation to wire up your beans:
#Autowired
private PetStoreService petStoreService;
As long as you've got
in your spring context, the "petStore" bean defined in your application context will automatically be injected. If you've got more than one bean with the type "PetStoreService", then you'd need to add a qualifier:
#Autowired
#Qualifier("petStore")
private PetStoreService petStoreService;
There's no requirement to specify the class in the getBean() method. It's just a question of safety. Note there's also a getBean() that takes only a class so that you can just look up beans by type instead of needing to know the name.
Assume I have a bean DialogBox, with properties for height and width:
public class DialogBox {
int x;
int y;
...
}
In my applicationContext.xml I would define properties as reasonable defaults:
<bean id="dialogbox" class="DialogBox">
<property name="x" value="100"/>
<property name="y" value="100"/>
</bean>
We have multiple clients that use the dialogBox bean, and each wants a custom value for x and y. One route we have discusses is having multiple properties files, one for each client, and have the client id map to the proper file, for example client 123 would map to dialogbox_123.properties:
dialogbox_123.properties:
x=200
y=400
Then at runtime when the bean is requested, spring would look to see if a custom properties file exists for the client, and use those properties, otherwise use the defaults. I am aware of the PropertyOverrideConfigurer, but AFAIK this only works when the context is started so will not work for our purposes. Is there an existing facility in spring to achieve this, or can someone recommend another way?
Use FactoryBean (as already suggested) to customize instantiation.
set scope="prototype" on the bean, so that each time an instance is required, a new one should be created.
In case you want to inject the prototype bean into a singleton bean, use lookup-method (Search for lookup-method here)
I'm not sure if this would fit your case though. Another suggestion would be:
In #PostConstruct methods of your various "clients" set the properties as desired on the already injected dialog window. Like:
public class MyDialogClient {
#Autowired
private Dialog dialog;
#PostConstruct
public void init() {
dialog.setWidth(150); //or read from properties file
dialog.setHeight(200);
}
...
}
Again, in this case, you can play with the scope atrribute.
Take a look at the Spring OSGi Compendium services, they've got a property manager called "managed-properties", which allows you not only to update the properties at runtime, but while the application is running if you select the "container-managed" update strategy.
If I understood the question, you can use a FactoryBean to customize bean creation logic in Spring.