I would like to be able to change the Guice injections at runtime to support multiple injections based on user input. This is what I would like to achieve:
public interface IDao {
public int someMethod();
}
public class DaoEarth implements IDao {
#Override
public int someMethod(){ ... }
}
public class DaoMars implements IDao {
#Override
public int someMethod(){ ... }
}
public class MyClass {
#Inject
private IDao myDao;
public int myMethod(String domain) {
//If Domain == Earth, myDao should be of the type DaoEarth
//If Domain == DaoMars, myDao should be of the type DaoMars
}
}
I was thinking of writing my own Provider, but I don't know how to use that provider to change my bindings at runtime. Any input is welcome and appreciated :)!
Update
Here's what I currently came up with, it's not as pretty as I'd like, so I'm still looking for feedback
public class DomainProvider {
#Inject #Earth
private IDaoProvider earthDaoProvider;
#Inject #Mars
private IDaoProvider marsDaoProvider;
public IDaoProvider get(Domain domain){
switch (domain){
case EARTH:
return earthDaoProvider;
case MARS:
return marsDaoProvider;
}
}
public IDaoProvider get(String domain){
Domain parsedDomain = Domain.valueOf(domain.toUpperCase());
return get(parsedDomain);
}
}
//MarsDaoProvider would be equivalent
public class EarthDaoProvider implements IDaoProvider {
#Inject #Earth
private IDao earthDao;
public IDao getDao() {
return earthDao;
}
}
// This means that in "MyClass", I can do:
public class MyClass {
#Inject
private DomainProvider domainProvider;
public int myMethod(String domain) {
IDaoProvider daoProvider = domainProvider.get(domain);
IDao dao = daoProvider.getDao();
//Now "dao" will be of the correct type based on the domain
}
}
//Of course elsewhere I have the bindings set like
bind(IDao.class).annotatedWith(Earth.class).to(EarthDao.class);
Your version is almost perfect as it is: You're going to need to inject some kind of object that returns one or the other based on code you write, and don't need assisted injection or anything like that. That said, you can skip some of the boilerplate:
public class DomainProvider {
// Just inject Providers directly without binding them explicitly.
#Inject #Earth Provider<IDao> earthDaoProvider;
#Inject #Mars Provider<IDao> marsDaoProvider;
public Provider<IDao> get(Domain domain){
switch (domain){
case EARTH:
return earthDaoProvider;
case MARS:
return marsDaoProvider;
}
}
public Provider<IDao> get(String domain){
Domain parsedDomain = Domain.valueOf(domain.toUpperCase());
return get(parsedDomain);
}
}
Your MyClass in that case would be exactly identical. Here, Provider is either the one-method generic interface com.google.inject.Provider, or the equivalent builtin javax.inject.Provider that it extends. Read more about Guice Providers on the relevant Guice wiki topic.
bind(IDao.class).annotatedWith(Earth.class).to(EarthDao.class);
// You can now inject "#Earth IDao" and also "#Earth Provider<IDao>".
Basically, if you bind a key Foo (to a class, provider, #Provides method, or instance), you automatically get to inject either a Foo or Provider<Foo> with no additional work. Providers are also a great way to ensure that you get a new instance with every call to get, if that's what you want; with your original, you'll always get the same instance of EarthDao or MarsDao for any given DomainProvider you inject. (If you have a scoped binding like #Singleton, Guice will respect that too; Provider just lets Guice get involved, rather than reusing a plain old Java reference.)
This means you can skip your custom EarthDaoProvider and MarsDaoProvider, unless you really need to perform any external initialization on them—at which point you'd probably be better off calling bind(EarthDao.class).toProvider(EarthDaoProvider.class) so the preparation also happens when injecting EarthDao directly. You could also just have DomainProvider return an IDao instance directly by calling get on the appropriate Provider, and be assured that it'll be a new instance every time.
Related
My current situation:
I want to inject the following class into my application:
public interface IConfigAccessor<T extends IConfig> {
...
}
ConfigAccessors are a proxy-objects, created dynamically at runtime. The creation of these object works as follows:
public class ConfigFactory implements IConfigFactory {
private final IConfigUpdater updater;
#Inject
public ConfigFactory(IConfigUpdater updater) {
this.updater = updater;
}
#Override
public <T extends IConfig> IConfigAccessor<T> register(final String configKey, final Class<T> configClass) {
ConfigCache<T> configCache = new ConfigCache<>(new SomeOtherThings(), configKey, configClass);
updater.register(configCache);
return new ConfigAccessor<>(configCache, configKey, configClass);
}
}
As you can see, to create these objects, I need to inject the ConfigUpdater and other depdencies. This means, that guice needs to be fully configured already.
To get the instance out of Guice, I use the following code:
IConfigFactory configClient = injector.getInstance(IConfigFactory.class);
IConfigAccessor<ConcreteConfig> accessor = configClient.register("key", ConcreteConfig.class)
How I want to inject them via Guice:
Currently, I can get the requried objects, but I have to manually pass them around in my application.
Instead, what I want to have is the following:
public class SomeClass {
#Inject
public SomeClass(#Config(configKey="key") IConfigAccessor<ConcreteConfig> accessor) {
// hurray!
}
}
What's the correct approach/technology to get this working?
After a lot of research, I'm feeling a bit lost on how to approach this topic. There are a lot of different things Guice offers, including simple Providers, custom Listeners which scan classes and identify custom annotations, FactoryModuleBuilders and more.
My problem is quite specific, and I'm not sure which of these things to use and how to get it working. I'm not even sure if this is even possible with Guice?
Edit: What I have so far
I have the following annotation which I want to use inside constructor paramters:
#Target({ ElementType.FIELD, ElementType.PARAMETER })
#Retention(RetentionPolicy.RUNTIME)
public #interface InjectConfig {
String configKey();
}
Inside the module, I can bind a provider to IConfigAccessor (with the above annotation) as such:
bind(IConfigAccessor.class).annotatedWith(InjectConfig.class)
.toProvider(new ConfigProvider<>());
However, there are two problems whith this:
The provider cannot provide IConfigAccessor. To create such an instance, the provider would need an IConfigUpdater, but since I use 'new' for the provider, I can't inject it.
Inside the provider, there is no way to find out about the configKey used in the Annotation.
Second approach:
Let's assume that I already know all configurations and configKeys I want to inject during startup. In this case, I could loop over all possible configKeys and have the following binding:
String configKey = "some key";
final Class<? extends IConfig> configClass =...;
bind(IConfigAccessor.class).annotatedWith(Names.named(configKey))
.toProvider(new ConfigProvider<>(configKey, configClass));
However, problem (1) still resides: The provider cannot get an IConfigUpdater instance.
The main problem here is that you cannot use the value of the annotation in the injection. There is another question which covers this part:
Guice inject based on annotation value
Instead of binding a provider instance, you should bind the provider class, and get the class by injecting a typeliteral.
That way, your config factory can look like that:
public class ConfigFactory<T extends IConfig> implements IConfigFactory {
#Inject private final IConfigUpdater updater;
#Inject private TypeLiteral<T> type;
#Override
public IConfigAccessor<T> register(final String configKey) {
Class<T> configClass = (Class<T>)type.getRawType();
ConfigCache<T> configCache = new ConfigCache<>(new SomeOtherThings(), configKey, configClass);
updater.register(configCache);
return new ConfigAccessor<>(configCache, configKey, configClass);
}
}
And then SomeClass:
public class SomeClass {
#Inject
public SomeClass(ConfigFactory<ConcreteConfig> accessor) {
ConcreteConfig config = accessor.register("key");
}
}
Since SomeClass needs to know "key" anyway, this is not too much a change information-wise. The downside is that the SomeClass API now gets a factory instead of the concrete config.
[EDIT]
And here is someone who actually did inject annotated values using custom injection.
I am trying to define a custom DeltaSpike ConfigSource. The custom config source will have the highest priority and check the database for the config parameter.
I have a ConfigParameter entity, that simply has a key and a value.
#Entity
#Cacheable
public class ConfigParameter ... {
private String key;
private String value;
}
I have a #Dependent DAO that finds all config parameters.
What I am trying to do now, is define a custom ConfigSource, that is able to get the config parameter from the database. Therefore, I want to inject my DAO in the ConfigSource. So basically something like
#ApplicationScoped
public class DatabaseConfigSource implements ConfigSource {
#Inject
private ConfigParameterDao configParameterDao;
....
}
However, when registering the ConfigSource via META-INF/services/org.apache.deltaspike.core.spi.config.ConfigSource, the class will be instantiated and CDI will not work.
Is there any way to get CDI working in this case?
Thanks in advance, if you need any further information, please let me know.
The main problem is, that the ConfigSource gets instantiated very early on when the BeanManager is not available yet. Even the JNDI lookup does not work at that point in time. Thus, I need to delay the injection/lookup.
What I did now, is add a static boolean to my config source, that I set manually. We have a InitializerService that makes sure that the system is setup properly. At the end of the initialization process, I call allowInitialization() in order to tell the config source, that the bean is injectable now. Next time the ConfigSource is asked, it will be able to inject the bean using BeanProvider.injectFields.
public class DatabaseConfigSource implements ConfigSource {
private static boolean allowInit;
#Inject
private ConfigParameterProvider configParameterProvider;
#Override
public int getOrdinal() {
return 500;
}
#Override
public String getPropertyValue(String key) {
initIfNecessary();
if (configParameterProvider == null) {
return null;
}
return configParameterProvider.getProperty(key);
}
public static void allowInitialization() {
allowInit = true;
}
private void initIfNecessary() {
if (allowInit) {
BeanProvider.injectFields(this);
}
}
}
I have a request-scoped bean that holds all my config variables for type-safe access.
#RequestScoped
public class Configuration {
#Inject
#ConfigProperty(name = "myProperty")
private String myProperty;
#Inject
#ConfigProperty(name = "myProperty2")
private String myProperty2;
....
}
When injecting the Configuration class in a different bean, each ConfigProperty will be resolved. Since my custom DatabaseConfigSource has the highest ordinal (500), it will be used for property resolution first. If the property is not found, it will delegate the resolution to the next ConfigSource.
For each ConfigProperty the getPropertyValue function from the DatabaseConfigSource is called. Since I do not want to retreive the parameters from the database for each config property, I moved the config property resolution to a request-scoped bean.
#RequestScoped
public class ConfigParameterProvider {
#Inject
private ConfigParameterDao configParameterDao;
private Map<String, String> configParameters = new HashMap<>();
#PostConstruct
public void init() {
List<ConfigParameter> configParams = configParameterDao.findAll();
configParameters = configParams.stream()
.collect(toMap(ConfigParameter::getId, ConfigParameter::getValue));
}
public String getProperty(String key) {
return configParameters.get(key);
}
}
I could sure change the request-scoped ConfigParameterProvider to ApplicationScoped. However, we have a multi-tenant setup and the parameters need to be resolved per request.
As you can see, this is a bit hacky, because we need to explicitly tell the ConfigSource, when it is allowed to be instantiated properly (inject the bean).
I would prefer a standarized solution from DeltaSpike for using CDI in a ConfigSource. If you have any idea on how to properly realise this, please let me know.
Even though this post has been answered already I'd like to suggest another possible solution for this problem.
I managed to load properties from my db service by creating an #Signleton #Startup EJB which extends the org.apache.deltaspike.core.impl.config.BaseConfigSource and injects my DAO as delegate which I then registered into the org.apache.deltaspike.core.api.config.ConfigResolver.
#Startup
#Singleton
public class DatabaseConfigSourceBean extends BaseConfigSource {
private static final Logger logger = LoggerFactory.getLogger(DatabaseConfigSourceBean.class);
private #Inject PropertyService delegateService;
#PostConstruct
public void onStartup() {
ConfigResolver.addConfigSources(Collections.singletonList(this));
logger.info("Registered the DatabaseConfigSourceBean in the ConfigSourceProvider ...");
}
#Override
public Map<String, String> getProperties() {
return delegateService.getProperties();
}
#Override
public String getPropertyValue(String key) {
return delegateService.getPropertyValue(key);
}
#Override
public String getConfigName() {
return DatabaseConfigSourceBean.class.getSimpleName();
}
#Override
public boolean isScannable() {
return true;
}
}
I know that creating an EJB for this purpose basically produces a way too big overhead, but I think it's a bit of a cleaner solution instead of handling this problem by some marker booleans with static accessors ...
DS is using the java se spi mechanism for this which is not CD'Injectable'. One solution would be to use the BeanProvider to get hold of your DatabaseConfigSource and delegate operations to it.
This is a simplified version of what I am trying to achieve.
I have multiple implementations of the same interface. Based on the user input at runtime I want to pick the correct implementation.
For example suppose I an interface called Color. There are many classes that implement this interface, the Red class, the Blue class, the Green class and so on.
At run time I need to pick implementations based on the user input. One way to achieve this would be something like this
#Autowired
#Qualifier("Red")
private Color redColor;
#Autowired
#Qualifier("Green")
private Color greenColor;
private Color getColorImplementation()
{
if(userInput=="red")
{
return redColor;
}
else if(userInput=="green")
{
return greenColor;
}
else
{
return null;
}
}
But the problem with this is that everytime a new implementation is added, I would have to update the code that picks the implementation, which beats the whole purpose of inversion of control part of spring. What is the right way to do this using spring?
You could autowire all implementations of the interface in question and then decide based on properties provided by interface which to use.
#Autowired
private List<Color> colors;
public void doSomething(String input) {
colors.stream().filter(c -> c.getName().contains(input)).findFirst().ifPresent(c -> {
// something
}
}
This is also less magical and more in line with OO principles. Dependency injection is to wire up things initially, not for dynamic switching at runtime.
You want to Autowire the ApplicationContext, then you can get all the Color beans with Map<String, Color> colors = appContext.getBeansOfType(Color.class);. This presumes that the userInput and the bean name are identical.
If that isn't the case, a solution would be to add a getName() to the Color interface; then you can autowire a List<Color> and construct the Map yourself.
Can't you make the Color an Enum?
The Spring ServiceLocatorFactoryBean (scroll down to the middle) API was built just for this purpose:
Create a dummy interface (ColorFactory) that provides a single method such as Color getColor(String color)
Create the proxy bean instance for org.springframework.beans.factory.config.ServiceLocatorFactoryBean passing ColorFactory as the serviceLocatorInterface parameter
Define beans for all of your color implementations with names matching the parameter you'd like to pass to getColor
Inject the factory into the collaborators and invoke getColor as needed
You could contrive this with similar APIs on the ApplicationContext, but the advantage of this approach is that it abstracts Spring from your Java implementation (for XML configured projects).
Same issue happen in my implementation where in, the scenario was based on user input, where the respective interface implementation needs to be invoked.
This solve my problem:
**Base Interface**
#Service
public interface ParentInterface {
public String doThis(ClassA param);
}
**First Implementation**
#Component("FirstImp")
public class FirstServiceImp implements ParentInterface {
public String doThis(ClassA param){
}
**Second Implementation**
#Component("SecondImp")
public class SecondServiceImp implements ParentInterface {
public String doThis(ClassA param){
}
**Factory**
#Service
public class ServiceResolver {
#Autowired
#Qualifier("FirstImp")
private ParentInterface firstImpl;
#Autowired
#Qualifier("SecondImp")
private ParentInterface secondImpl;
public ParentInterface getInstance(String condition){
switch(condition) {
case "X": return firstImpl;
case "Y": return secondImpl;
default:
throw new IllegalArgumentException(condition);
}
}
}
**Controller**
#RestController
public class UserController {
#Resource
private ServiceResolver serviceresolver;
#PostMapping("/userbase/{inp1}/messages/{inptype}")
public ResponseEntity<String> sendData(#PathVariable String
inp1,#PathVariable String inptype, #RequestBody XYZBean msg)
{
for(ABC data : msg.getSubData())
serviceresolver.getInstance(data.getType()).doThis(msg);
return new ResponseEntity<String>("created",HttpStatus.OK);
}
}
For my application, I have a Scale interface and multiple classes implementing this interface, for example NormalizedScale, LogScale, etc. In one of my Services, I need to create many Scales, and I want to use Spring to define which implementation of the Scale it should create. How would I implement something like this?
--
I was thinking to create a factory ScaleFactory, like in the Abstract Factory Pattern, which I could call ScaleFactory.getScale() to get a Scale of whichever implementation I configured in the Spring XML:
class ScaleFactory {
Class<? extends Scale> scaleImplClass;
public static Scale getScale() {
return scaleImplClass.newInstance();
}
}
Scale myScale = ScaleFactory.getScale();
But with that approach, how could I configure which implementation the ScaleFactory should use from Spring XML?
--
An alternative would be to make the ScaleFactory a #Service, and then autowire the ScaleFactory into my service:
#Autowired
ScaleFactory scaleFactory;
...
Scale myScale = scaleFactory.getScale();
Then I can use an autowired property in the ScaleFactory to define the scaleImplClass. But that seems weird because my Factory is also a Service and I have an instance of that factory.
--
Another approach would be to have the Class scaleImplementationClass property in my service instead of the ScaleFacotry and use the ScaleFactory like so:
#Value("${scaleImplementationClass}")
Class scaleImplementationClass
...
Scale myScale = ScaleFactory.getScale(scaleImplementationClass);
But then the factory is quite pointless because I could also just as well run scaleImplementationClass.newInstance().
There are a couple of different Spring-like ways you can handle this. The approach I have personally gone for looks a bit like this:
public interface ScaleFactory {
public Scale newInstance();
public String type();
}
public class FirstScaleFactory implements ScaleFactory {
public Scale newInstance() {
return new FirstScale();
}
public String type() {
return "first";
}
}
public class SecondScaleFactory implements ScaleFactory {
public Scale newInstance() {
return new SecondScale();
}
public String type() {
return "second";
}
}
public class ScaleManager {
private final Map<String, ScaleFactory> factories;
#Autowired
public ScaleManager(List<ScaleFactory> factories) {
this.factories = factories.stream()
.collect(Collectors.toMap(f -> f.type(), Function::identity));
}
public Scale newInstance(String type) {
return Optional.ofNullable(factories.get(type))
.map(factory -> factory.newInstance())
.orElseThrow(IllegalArgumentException::new);
}
}
With this approach, your ScaleManager is a standard Spring bean that can be wired into any class that needs a scale instance. At initialization time, it gets all ScaleFactories that are defined in the Spring context, and autowires them in as a List<ScaleFactory>, which is then converted to a Map (where the ScaleFactory type is the key). This avoids you needing to worry about class names of Scale, and gives your the ability to change them later (as long as you keep the type key consistent)`
Your ScaleFactory implementations can then do whatever they need to do. For example, if you have one type of Scale that you know is immutable, you can have the factory return the same instance every time. Alternatively you can have every invocation return a separate instance - the instantiation of the Scale is up to the implementation-dependent factory.
You can simply use "Qualifiers" which is basically going to point to a specific "named" bean. By default the bean names are the name of your classes, with the first letter in lower case (MyClass -> myClass). If you want to define your own names you can do as follow :
#Service("customizedBeanName")
You would end up doing something like this :
#Autowired
#Qualifier("logScale")
private Scale logScale;
#Autowired
#Qualifier("anotherScale")
private Scale anotherScale;
As for spring 5.x there's a simpler and cleaner way of doing this. I have decided to use #ConditionalOnProperty annotation but you may choose any #Conditional* of your preference.
Here's the thing, I've have simplified to extreme:
public interface MyService {}
#Service
#ConditionalOnProperty(prefix = "myService", name = "Impl", havingValue = "Some")
public class SomeService implements MyService {}
#Service
#ConditionalOnProperty(prefix = "myService", name = "Impl", havingValue = "Foo")
public class FooService implements MyService {}
#Service
public class SimpleService {
#Autowired
SimpleService(MyService service) {
// service instance will depend on configuration
}
}
I'm using springboot so I've decided to use application.properties in order to set values via environment variables like this:
myService.Impl=${MY_SERVICE_IMPL}
Then, I have a fully dynamic injection based on environment variables that may be passed to a docker container for instance.
Recently, when I played around with Google Guice I was trying to do something like this:
#Override
protected void configure() {
...
bind(Provider.class).to(ViewFactory.class);
...
}
Where ViewFactory was:
public class ViewFactory implements Provider<SomeType> {...}
Of course, Guice didn't let me do that returing error:
1) Binding to Provider is not allowed.
{stacktrace}
What is the reason why it is not possible to bind to provider?
I guess it is because Provider interface is very special to Guice. In fact, all its internal machinery is implemented in term of providers.
Moreover, this could create ambiguities. If bindings to providers were possible:
bind(SomeClass.class).to(SomeClassImpl1.class);
bind(new TypeLiteral<Provider<SomeClass>>() {}).to(() -> new SomeClassImpl2());
then what should Guice inject here?
#Inject
OtherClass(Provider<SomeClass> someClassProvider) { ... }
Should it be a provider which returns SomeClassImpl1 (because of the first binding; remember, direct injections and provider injections are interchangeable in Guice) or should it be a provider which returns SomeClassImpl2 (because of the second binding)?
It really is redundant. Because you can inject SomeClass or Provider<SomeClass> regardless of the actual binding, you can bind the class itself to its provider:
bind(SomeClass.class).toProvider(() -> new SomeClassImpl());
// Either of the following will work
#Inject
OtherClass1(Provider<SomeClass> someClassProvider) { ... }
#Inject
OtherClass2(SomeClass someClass) { ... }
Provider is a special case. Guice does a lot of things behind the scenes with Provider, so they just ban binding to the Provider class entirely. One example is with scoping: your custom Provider might call new every single time, but if you create the provider in the Singleton scope, that should not happen. So Guice doesn't actually inject your provider, it injects a wrapped version. Things like that is why they ban binding to Provider.class directly. Here's a code example:
import com.google.inject.*;
import com.google.inject.name.*;
public class ProviderBindExample {
public static class ProvModule extends AbstractModule {
#Override
protected void configure() {
bind(Foo.class).toProvider(FooProvider.class);
bind(Foo.class).annotatedWith(Names.named("singleton"))
.toProvider(FooProvider.class)
.in(Singleton.class);
}
}
public static interface Foo { }
public static class FooProvider implements Provider<Foo> {
#Override
public Foo get() {
return new Foo() {};
}
}
public static class SomeClass {
#Inject public Provider<Foo> provider;
#Inject #Named("singleton") public Provider<Foo> singletonProvider;
}
public static void main(String... args) {
Injector inj = Guice.createInjector(new ProvModule());
SomeClass s = inj.getInstance(SomeClass.class);
System.out.println("Provider class = " + s.provider.getClass());
System.out.println("Singleton provider class = " + s.singletonProvider.getClass());
Foo first = s.provider.get();
Foo second = s.provider.get();
System.out.printf("regular scope: objects are %s%n", first == second ? "the same" : "different");
first = s.singletonProvider.get();
second = s.singletonProvider.get();
System.out.printf("singleton scope: objects are %s%n", first == second ? "the same" : "different");
}
}
Output:
Provider class = class com.google.inject.internal.InjectorImpl$4
Singleton provider class = class com.google.inject.internal.InjectorImpl$4
regular scope: objects are different
singleton scope: objects are the same