I'm working on a legacy Spring-Boot application where I would like to use dependency injection with some code that exists outside the application context. One a part of the application comes as a separate JAR-file and cannot be modified. But I am able to modify some classes that are instantiated in that part. Here how I'm planning to do this:
class ServiceHolder {
private static FooService fooService;
public static FooService getFooService() { return fooService; }
public static void setFooService(FooService service) { fooService = service; }
}
#Bean
#Profile("production")
FooService fooService() {
var service = new ProductionFooService();
ServiceHolder.setFooService(service);
return service;
}
public class LegacyPojo {
private final FooService fooService;
public LegacyPojo() {
fooService = ServiceHolder.getFooService();
}
//.. some business logic
}
I'm worried about possible visibility problems when different requests in separate threads will call new LegacyPojo() and reach for FooService instance.
So my question is: should I declare ServiceHolder#getFooService and ServiceHolder#setFooService synchronized or not?
There is a lot of others things you can do that with security, dont you think that you could have the instance of FooService into LegacyPojo passing it by constructor, it will be less coupled.
Other thing that you can do is to control the instances of FooService, you may do it as a singleton, declaring it as a static property on ServiceHolder and not to have a setMethod. I think, the way you told, you want a single instance of FooService.
Even LegacyPojo being a Pojo, you dont need to create a getter for FooService.
Once you use ServiceHolder.setFooService(service); you may do a implementation like this:
class ServiceHolder{
private static FooService fooService;
public static void setFooService(FooService newFooService){
if(fooService== null){
fooService = newFooService;
}
}
}
So that way, you will set only the first instance of FOoService and it will not be changed, of course you can do any condition to setFooService in ServiceHolder
It would work without any synchronization because singleton bean will be instantiated in a critical section inside synchronized block. In DefaultSingletonBeanRegistry class there is a method getSingleton which, according to the doc:
/**
* Return the (raw) singleton object registered under the given name,
* creating and registering a new one if none registered yet.
* ...
*/
And at the very beginning of this method, the critical section starts synchronized (this.singletonObjects). So the effect of ServiceHolder.setFooService(service) call will be visible to all threads after leaving the critical section.
Related
I have a Spring Boot application where I would like to ensure that a list of decorators are verified to be executed. These decorators all extend from the same Abstract class, which in turn extend from the same interface, and they are autowired into a service class as a list of decorators. I would have thought that providing the #SpyBean(MyDecorator.class) at the class level of the test would have done the trick, but I got the error specifying that the decorator is not a spy. It looks like the MockitoPostProcessor class expects that we provide the individual concrete classes in the annotation as so #SpyBean(classes = {decorator1.class,decorator2.class}). I tried the latter, and it worked.
However, the issue that I have with this is that we have to add to this list every time we create a new decorator, which is not ideal. This is why I thought it makes sense to have the interface type be checked as well. Please let me know if there is a better way of doing this, or if I missed something. A thought that crossed my mind was to define my own post processor to wrap any bean from a defined type in a mockito spy, but I would like to check here first. Here is a skeleton definition of the classes to help you understand my dilemma.
MyDecorator.java
public interface MyDecorator{
public void decorate(SomeObject obj);
}
AbstractDecorator.java
public class AbstractDecorator implements MyDecorator{
//common decorator logic
}
Decorator1.java
#Component
public class Decorator1 extends AbstractDecorator{
public void decorate(SomeObject obj){
//decoration logic
}
}
Decorator2.java
#Component
public class Decorator2 extends AbstractDecorator{
public void decorate(SomeObject obj){
//decoration logic
}
}
DecorationService.java
#Service
public class DecorationService implements Service{
#Autowired
private List<MyDecorator> decoratorList;
public void processDecorators(){
//go through list of decorators and process some object
}
}
DecoratorServiceTest.java
#Runwith(SpringRunner.class)
#SpringBootTest
#ActiveProfiles("test")
//#SpyBean(MyDecorator.class) //<-- This doesn't wrap the classes in a spy and errors out
#SpyBean(classes = {Decorator1.class, Decorator2.class}) //<-- This works
public class DecoratorServiceTest{
#Autowired
private List<MyDecorator> decoratorList;
#Test
public void testProcessDecorator(){
//verify that each decorator was processed
}
}
I posted a spring boot github issue here. Hopefully we would either see an improvement on it or we get an explanation as to why it is designed in this way.
I have a workaround in place that I'm using which is I've created a class that implements Spring's BeanPostProcessor interface, and I override the postProcessAfterInitialization method, and I check if the class is what I'm expecting, then I would wrap it in a mockito spy. Also, you would need to define the spring bean.
Here is a snippet of the class that I created.
public class SpyBeanPostProcessor<T> implements BeanPostProcessor{
/**
* The class type to spy on.
*/
private Class<T> typeToSpy;
/**
* Construct a SpyBeanPostProcessor with a class type to wrap
* as a {#link org.mockito.Spy}
* #param typeToSpy The class type to spy on.
*/
public SpyBeanPostProcessor(Class<T> typeToSpy) {
this.typeToSpy = typeToSpy;
}
#Override
public Object postProcessAfterInitialization(Object bean, String beanName) throws BeansException {
if (typeToSpy.isAssignableFrom(bean.getClass())){
return Mockito.spy(bean);
}else{
return bean;
}
}
}
I also needed to create a new spring bean that loads the BeanPostProcessor as shown below.
#Bean
public static SpyBeanPostProcessor decoratorSpyBeanPostProcessor(){
return new SpyBeanPostProcessor(MyDecorator.class);
}
I have this class:
#ApplicationScoped
public class BookService {
public boolean doorClosed;
public boolean isDoorClosed() {
return doorClosed;
}
}
And I have this test class:
#QuarkusTest
public class BookServiceTest {
#Inject BookService bookService;
#Test
public void testBookServiceDoor() throws InterruptedException {
bookService.doorClosed=true;
assertTrue(bookService.doorClosed);
assertTrue(bookService.isDoorClosed());
}
}
I am surprised that the last test assertion fails. The first passes but the last fails. It almost seems like that the getter/setters are using different variables than the one I am accessing directly.
I did the same test with Spring Boot and got the two assertions passing:
#Service
public class BookService {
public boolean doorClosed;
public boolean isDoorClosed() {
return doorClosed;
}
}
And the test:
#SpringBootTest
public class BookServiceTest {
#Autowired BookService bookService;
#Test
public void testBookServiceDoor() throws InterruptedException {
bookService.doorClosed=true;
assertTrue(bookService.doorClosed);
assertTrue(bookService.isDoorClosed());
}
}
Using the #ApplicationScoped annotation makes the class a normal scoped bean. This means that the container never injects an actual instance of the class; instead, you get a proxy that will lookup the correct instance on each method invocation.
This is most commonly useful for example with #RequestScoped beans, where you get a single proxy, which will dispatch method invocations to instances that belong to the "current request" (typically an HTTP request). However, there are good reasons why you might want this for application scoped beans as well (e.g. when you want lazy initialization).
The rule is: never access fields directly on normal scoped beans, only call methods.
If you want, you can make the class #Singleton. That is a pseudo scope, and you get no proxy, you get the actual instance that you can work with directly.
These days im working on a Web project and i just want to clarify couple of things regarding Spring bean scopes and best practices for Spring based developments. Here i am using a scenario using a sample code
I have a Web Controller as below
#Controller
Public class JobController{
private JobService jobService;
#Autowired
public void setJobService(JobService jobService ) {
this.jobService = jobService ;
}
public void run(){
Job job = new Job();
-- Setting the properties for the Object
jobService.run(job);
}
}
Then I have the Service as below
#Service
Public class JobService {
public void run(Job job){
-- perform the business logic
}
}
In Here i want to make the JobService class stateless so i can define JobService as singleton hence reduce the unnecessary object creation. As per my understanding in-order make a class stateless we do not want to keep instance properties.In This scenario i pass different Job objects to the service. Does this make this JobService statefull because JObservice process different different job objects? Can you please help me to understand
Thanks,
Keth
Passing different objects does not make your service stateful.
Consider this for example.
#Service
Public class JobService {
private Job currentJob;
public void setJob(Job job) {
currentJob = job;
}
public void run(){
-- perform the business logic on currentJob
}
}
This would make the bean 'stateful' and cause unexplained behavior.
The execution of the method in your singleton by multiple controller/threads will not collide and can be assumed to be safe.
We have a web service that one of its parameters is called origin and this origin is always validated against a code in the database.
For each one of our services I have to validate this code. This code does not change so I want to keep it in a constant, but I still have to validate it to prevent clients from sending a wrong code.
Basically what I want is this:
#Service
public class Service {
#Autowired
private LogBS logBS;
// I know this cannot be used in a static context.
public static final Long CODE = this.logBS.retrieveLogWebServiceCode("webServiceName");
public void validateOriginCode(final Long origin) {
if (!origin.equals(CODE)) {
throw new ServiceException("Wrong origin code!");
}
}
}
I know something similar can be done with Spring caching, but is it possible to do it with a constant?
I would rather go with this:
#Service
public class CodeValidatorService {
private LogBS logBS;
private Long CODE;
#Autowired
public CodeValidatorService(LogBS logBS){
this.logBS = logBS;
CODE = this.logBS.retrieveLogWebServiceCode("webServiceName");
if (CODE == null){
throw new ServiceException("Code cannot be read from DB!");
}
}
public void validateOriginCode(final Long origin) {
if (!origin.equals(CODE)) {
throw new ServiceException("Wrong origin code!");
}
}
}
Just as a code review, I prefer injecting dependencies in the constructor rather than using #Autowired in the field directly, it makes the service testable. You could also try to read the code in a #PostConstruct method, but I think it's better to do it in the constructor so you always have the service in a ready-to-go state.
For using it in the rest of your services, inject the CodeValidatorService instance on them:
#Service
public class OtherService {
private CodeValidatorService codeValidatorService;
#Autowired
public OtherService(CodeValidatorService codeValidatorService){
this.codeValidatorService = codeValidatorService;
}
public void performAction(final Long origin) {
codeValidatorService.validateOriginCode(origin);
//do the rest of your logic here
}
}
See also:
Spring Beans and dependency injection
Setter injection versus constructor injection
You can have a constantsProvider class
#Component
public class ConstantsProvider {
#Autowired
private LogBS logBS;
private String CODE;
#PostConstruct
public void init() {
CODE = this.logBS.retrieveLogWebServiceCode("webServiceName");
}
public String getCode() {
return CODE;
}
}
Add this snippet of code to Service class
#Autowired
private ConstantsProvider constantsProvider;
You can use constantsProvider.getCode() in your services. This way CODE is going to be immutable and not defined in a static context.
Note: If you have more constants similar to this, there is a better way to define the ConstantsProvider class. If there is only one, I would stick to the above implementation.
Edit 1:
If you need it in all the service classes, make the constantsProvider a spring bean and initialize the CODE there itself. Updated the answer
This is an Spring MVC project with Hibernate.
I'm, trying to make a Logger class that, is responsible for inputting logs into database.
Other classes just call proper methods with some attributes and this class should do all magic.
By nature it should be a class with static methods, but that causes problems with autowiring dao object.
public class StatisticLogger {
#Autowired
static Dao dao;
public static void AddLoginEvent(LogStatisticBean user){
//TODO code it god damn it
}
public static void AddDocumentEvent(LogStatisticBean user, Document document, DocumentActionFlags actionPerformed){
//TODO code it god damn it
}
public static void addErrorLog(Exception e, String page, HashMap<String, Object> parameters){
ExceptionLogBean elb=new ExceptionLogBean();
elb.setStuntDescription(e);
elb.setSourcePage(page);
elb.setParameters(parameters);
if(dao!=null){ //BUT DAO IS NULL
dao.saveOrUpdateEntity(elb);
}
}
How to make it right? What should I do not to make dao object null?
I know that I could pass it as a method parameter, but that isn't very good.
I'm guessing that autowired can't work on static objects, because they are created to early to autowiring mechanism isn't created yet.
You can't #Autowired a static field. But there is a tricky skill to deal with this:
#Component
public class StatisticLogger {
private static Dao dao;
#Autowired
private Dao dao0;
#PostConstruct
private void initStaticDao () {
dao = this.dao0;
}
}
In one word, #Autowired a instance field, and assign the value to the static filed when your object is constructed. BTW, the StatisticLogger object must be managed by Spring as well.
Classical autowiring probably won't work, because a static class is not a Bean and hence can't be managed by Spring. There are ways around this, for example by using the factory-method aproach in XML, or by loading the beans from a Spring context in a static initializer block, but what I'd suggest is to change your design:
Don't use static methods, use services that you inject where you need them. If you use Spring, you might as well use it correctly. Dependency Injection is an Object Oriented technique, and it only makes sense if you actually embrace OOP.
I know this is an old question but just wanted to share what I did,
the solution by #Weibo Li is ok but the problem it raises Sonar Critical alert about assigning non static variable to a static variable
the way i resolved it with no sonar alerts is the following
I change the StatisticLogger to singlton class (no longer static)
like this
public class StatisticLogger {
private static StatisticLogger instance = null;
private Dao dao;
public static StatisticLogger getInstance() {
if (instance == null) {
instance = new StatisticLogger();
}
return instance;
}
protected StatisticLogger() {
}
public void setDao(Dao dao) {
this.dao = dao;
}
public void AddLoginEvent(LogStatisticBean user){
//TODO code it god damn it
}
public void AddDocumentEvent(LogStatisticBean user, Document document, DocumentActionFlags actionPerformed){
//TODO code it god damn it
}
public void addErrorLog(Exception e, String page, HashMap<String, Object> parameters){
ExceptionLogBean elb=new ExceptionLogBean();
elb.setStuntDescription(e);
elb.setSourcePage(page);
elb.setParameters(parameters);
if(dao!=null){
dao.saveOrUpdateEntity(elb);
}
}
I created a service(or Component) that autowire the service that i want and set it in the singlton class
This is safe since in spring it will initialize all the managed beans before doing anything else and that mean the PostConstruct method below is always called before anything can access the StatisticLogger
something like this
#Component
public class DaoSetterService {
#Autowired
private Dao dao0;
#PostConstruct
private void setDaoValue () {
StatisticLogger.getInstance().setDao(dao0);
}
}
Instead of using StatisticLogger as static class I just use it as StatisticLogger.getInstance() and i can access all the methods inside it
You can pass the DAO to StatisticLogger from where you call it.
public static void AddLoginEvent(LogStatisticBean user, DAO dao){
dao.callMethod();
}
It might be too late to put an answer to this question, especially when a question is already having an accepted answer. But it might help others in case they face the same issue.
inside the StatisticLogger class create an instance of the Dao service.
public static Dao daoService = new Dao();
then, auto-wire the service instance through the constructor of the StatisticLogger class.
#Autowired
public functionName(Dao daoService0) {
this.daoService = daoService0;
}
//use this service as usual in static class
daoService.fun();
I think this is the simplest solution for the problem.