Perform service multiplexing with Apache Thrift and TServlet - java

I have a system (Java with Spring Framework) that exposes 7 different Apache Thrift servlets over HTTP using the TServlet class. Currently they all need their own Servlets, ServletMappings, Processors, Handlers etc. so implementing clients have to also keep an internal list of all the various URLs for the different services.
I understand that Apache Thrift supports multiplexing when using TServer and its derivatives by using TMultiplexingProcessor, however since I am using Spring and my Servlet, Handler and Processor are all Spring Beans that get autowired into one another, I'm unsure how to proceed.
Here's an example of how one of the services gets wired up:
UserServiceHandler.java
#Component
public class UserServiceHandler implements UserService.Iface {
#Override
public User getUser(String userId) throws TException {
// implementation logic goes here
}
}
UserServiceProcessor.java
#Component
public class UserServiceProcessor extends UserService.Processor<UserServiceHandler> {
private UserServiceHandler handler;
#Autowired
public UserServiceProcessor(UserServiceHandler iface) {
super(iface);
handler = iface;
}
public UserServiceHandler getHandler() {
return handler;
}
public void setHandler(UserServiceHandler handler) {
this.handler = handler;
}
}
UserServiceServlet.java
#Component
public class UserServiceServlet extends TServlet {
private UserServiceProcessor processor;
#Autowired
public UserServiceServlet(UserServiceProcessor p) {
super(p, new TBinaryProtocol.Factory());
processor = p;
}
}
Servlet Registration
ServletRegistration.Dynamic userService = servletContext.addServlet("UserServiceServlet", (UserServiceServlet) ctx.getBean("userServiceServlet"));
userService.setLoadOnStartup(1);
userService.addMapping("/api/UserService/*");
// This same block repeated 7 times for each *ServiceServlet with different mappings
I would like to have all 7 service handlers map to a single URL like /api/*. Is this even possible? I suppose I would have to create a single servlet and processor, but I'm unsure what they should look like. My processors extend UserService.Processor and the like.

OK, figured it out. Might not be the best way, so I welcome criticism.
Here were my rough steps:
Keep the handler classes the way they were.
Create a new class that extends TMultiplexedProcessor
Create a new class that extends TServlet
All Processors (e.g. the UserServiceProcessor have a handler property and a corresponding getter and setter
Here is my ApiMultiplexingProcessor:
#Component
public class ApiMultiplexingProcessor extends TMultiplexedProcessor {
UserServiceHandler userServiceHandler;
ReportServiceHandler reportServiceHandler;
// ... more service handlers can go here
#Autowired
public ApiMultiplexingProcessor(UserServiceProcessor userServiceProcessor, ReportServiceProcessor reportServiceProcessor) {
this.registerProcessor("UserService", userServiceProcessor);
this.registerProcessor("ReportService", reportServiceProcessor);
// add more registerProcessor lines here for additional services
userServiceHandler = userServiceProcessor.getHandler();
reportServiceHandler = reportServiceProcessor.getHandler();
// set any additional service handlers here
}
// getters and setters for the handlers
public UserServiceHandler getUserServiceHandler() {
return userServiceHandler;
}
public void setUserServiceHandler(UserServiceHandler userServiceHandler) {
this.userServiceHandler = userServiceHandler;
}
public ReportServiceHandler getReportServiceHandler() {
return reportServiceHandler;
}
public void setReportServiceHandler(ReportServiceHandler reportServiceHandler) {
this.reportServiceHandler = reportServiceHandler;
}
}
So to explain the above a bit, if you add any additional services, you need to add the *ServiceHandler classes as fields on this class, and create the getters and setters etc.
So now that we have that, we can create a new single servlet that will be added to the servlet context.
Here is my ApiServlet:
#Component
public class ApiServlet extends TServlet {
private ApiMultiplexingProcessor processor;
#Autowired
public ApiServlet(ApiMultiplexingProcessor p) {
super(p, new TBinaryProtocol.Factory());
processor = p;
}
}
And then you just add this servlet to the servlet context (from a bean) as before:
ServletRegistration.Dynamic api = servletContext.addServlet("ApiServlet", (ApiServlet) ctx.getBean("apiServlet"));
api.setLoadOnStartup(1);
api.addMapping("/api/*");
// yay now we have a single URL and a single servlet
This all could be helpful to someone else in my situation, so enjoy!
P.S. make sure when adapting your clients you use the TMultiplexedProtocol so that you can pass the service name through when talking to the server e.g.
TTransport transport = new THttpClient(new Uri("https://myapp.com/api/"));
TProtocol protocol = new TBinaryProtocol(transport);
TMultiplexedProtocol mp = new TMultiplexedProtocol(protocol, "UserService");
UserService.Client userServiceClient = new UserService.Client(mp);

Related

Spring use different property files depending on request params

Background:
I am working on a java Spring REST microservice that needs to work with multiple identical back-end systems and multiple identical databases depending on the request parameters.
Basically I have 3 "brands". For each brand there is a set of downstream services and a database. I have no control over those.
My spring service will receive brand as a part of request and will need to call the right downstream services and use the correct database.
Previously I would deal with this by having a separate instance of the spring service for each of the brands. There would be a single property file for each brand and spring would use it to wire up beans. I would have separate URL's for each brand and there was no problem.
Some of my beans need to know about "brand" during creation as they are wrappers around connections downstream services. I.e. once the bean is created there won't be a way to switch it to be a "different brand".
Problem:
I would like to change this so that a single instance of my service can handle requests for any brand.
Requirements:
I was thinking about the following solution:
Have a general property file for non-branded stuff. Spring would wire any non-branded beans and keep them as singleton beans.
Have a property file with brand specific urls etc for each of the brands
Spring would create set of singleton beans for each of the brand using appropriate property file.
Next when the request comes in spring would read the request params and use bean specific for that brand.
Performance is important to me so I would like to reuse the beans as much as possible.
I would like to make this thing as transparent as possible so that people creating new beans don't have to worry about doing anything outside standard configuration/context class.
Does anyone know what would be the best solution to achieve this?
I think you can solve the problem injecting the service in every request with the right set of configurations and beans; possibly already existing in your Application Context.
Given:
$ curl http://localhost:8080/greetings/rodo && echo
Hi from brand1, rodo
$ curl -H "x-brand-name: brand1" http://localhost:8080/greetings/rodo
Hi from brand1, rodo
$ curl -H "x-brand-name: brand2" http://localhost:8080/greetings/rodo && echo
Hi from brand2, rodo
The following code would work:
-- application.yml --
brand1:
greetingPrefix: Hi from brand1,
brand2:
greetingPrefix: Hi from brand2,
-- DemoApplication.java --
#SpringBootApplication
public class DemoApplication {
public static void main(String[] args) {
SpringApplication.run(DemoApplication.class, args);
}
#Configuration
class ServiceConfig {
#Bean
public GreetingService greetingServiceBrand1(Brand1Config config) {
return new GreetingService(config);
}
#Bean
public GreetingService greetingServiceBrand2(Brand2Config config) {
return new GreetingService(config);
}
}
#Configuration
class WebConfig implements WebMvcConfigurer {
#Autowired
private ApplicationContext applicationContext;
#Override
public void addArgumentResolvers(List<HandlerMethodArgumentResolver> resolvers) {
resolvers.add(greetingServiceResolver());
}
private GreetingServiceResolver greetingServiceResolver() {
GreetingService greetingServiceBrand1 = applicationContext.getBean("greetingServiceBrand1", GreetingService.class);
GreetingService greetingServiceBrand2 = applicationContext.getBean("greetingServiceBrand2", GreetingService.class);
return new GreetingServiceResolver(greetingServiceBrand1, greetingServiceBrand2);
}
}
}
#RestController
#RequestMapping("/greetings")
class GreetingController {
#GetMapping("/{name}")
public String get(GreetingService greetingService, #PathVariable String name) {
return greetingService.sayHi(name);
}
}
class GreetingServiceResolver implements HandlerMethodArgumentResolver {
private final GreetingService greetingServiceBrand1;
private final GreetingService greetingServiceBrand2;
public GreetingServiceResolver(GreetingService greetingServiceBrand1, GreetingService greetingServiceBrand2) {
this.greetingServiceBrand1 = greetingServiceBrand1;
this.greetingServiceBrand2 = greetingServiceBrand2;
}
#Override
public boolean supportsParameter(MethodParameter parameter) {
return parameter.getParameterType().equals(GreetingService.class);
}
#Override
public Object resolveArgument(
MethodParameter methodParameter,
ModelAndViewContainer modelAndViewContainer,
NativeWebRequest nativeWebRequest,
WebDataBinderFactory webDataBinderFactory
) throws Exception {
String brand = nativeWebRequest.getHeader("x-brand-name");
return resolveGreetingService(brand);
}
private GreetingService resolveGreetingService(String brand) {
if ("brand2".equals(brand)) {
return greetingServiceBrand2;
}
return greetingServiceBrand1; // default
}
}
class GreetingService {
private BaseConfig config;
public GreetingService(BaseConfig config) {
this.config = config;
}
public String sayHi(String name) {
return config.getGreetingPrefix() + " " + name;
}
}
abstract class BaseConfig {
private String greetingPrefix;
public String getGreetingPrefix() {
return greetingPrefix;
}
public void setGreetingPrefix(String greetingPrefix) {
this.greetingPrefix = greetingPrefix;
}
}
#Configuration
#ConfigurationProperties("brand1")
class Brand1Config extends BaseConfig {
}
#Configuration
#ConfigurationProperties("brand2")
class Brand2Config extends BaseConfig {
}
As you can see, it's fundamental to pass the service to each controller method, write a resolver and inject the right set of dependencies depending on a parameter passed to the request, in this case via header.
Since your property files need to be declared statically anyway, you can just write all your different brand stuff in the same property file, like in a key-value format, that Spring can pick up as a list of configurations.
brandConfigs:
- brand: foo
property: foos
- brand2: bar
porperty: bars
Load all your connection beans to your downstream services on startup and just route to them according to your request param. Imo this seems to be the most straight forward and performant way. If some of these downstreams are used very rarely you can lazy load the beans on-demand, but probably this wouldn't make a sense unless you have thousands of different downstream routes.

How to create a dynamic proxy of a already proxyed class in spring

I'm relativly new to spring/spring boot.
At the moment I'm using a spring boot rest application which provides an FeignClient to be included in other projects. Now, I want those FeignClients be wrapped by a CircuitBreaker.
The best solution I came up with, is that I dynamically create a proxy which includes the CircuitBreaker implementation which itself calls the created FeignClient.
So let's assume I have the following interface which describes the RestController:
#RequestMapping("/")
public interface MyWebService {
#GetMapping("name")
public String getName();
}
Now, I have the interface for the FeignClient:
#FeignClient("app")
public interface WebServiceClient extends WebService {
}
So.. My goal would be to achieve something like I have another annotation e. g. #WithCircuitBreaker which I then will be scanned for and dynamically create a proxy bean which will be injected instead of the FeignClient bean.
At the moment my code looks like this:
#FeignClient("app")
#WithCircuitBreaker
public interface WebServiceClient extends WebService {
}
As far as I know, I can now create a #Configuration Class which will look like this:
#Configuration
public class WithCircuitBreakerConfiguration implements ImportAware {
private AnnotationMetadata annotationMetadata;
private AnnotationAttributes annotationAttributes;
#Override
public void setImportMetadata(AnnotationMetadata importMetadata) {
this.annotationMetadata = importMetadata;
Map<String, Object> annotatedClasses = importMetadata.getAnnotationAttributes(WithCircuitBreaker.class.getName());
this.annotationAttributes = AnnotationAttributes.fromMap(annotatedClasses);
}
What else to import to create the proxy and inject it?
}
Now I'm at the point, which I don't know how to continue. How to dynamically create a proxy class which does something like this:
public class PorxyedWebService {
private WebService feignClientProxy;
#Autowired
public ProxyedWebService(WebService feignClientProxy) {
this.feignClientProxy = feignClientProxy;
}
public String getName() {
...
<some circuitbreaker stuff>
....
return this.feignClientProxy.getName();
}
}
and then return this proxy instead of the proxy generated from Feign as soon as someone autowires the WebService interface.
I am not a Spring user, but I do know that Spring does not create proxies recursively if e.g. multiple Spring AOP aspects are applied to the same object. Instead, additional interceptors (or advices in AOP language) are registered upon the same proxy. I think you want to use that infrastructure in order to achieve whatever your objective is.
You can just use the resilience4j Spring Boot2 starter.
You can combine the #CircuitBreaker annotation with the #FeignClient annotation at interface level.
You can then use it as follows:
#FeignClient(name = DUMMY_FEIGN_CLIENT_NAME)
#CircuitBreaker(name = DUMMY_FEIGN_CLIENT_NAME)
public interface DummyFeignClient {
String DUMMY_FEIGN_CLIENT_NAME = "dummyFeignClient";
#GetMapping(path = "/api/{param}")
void doSomething(#PathVariable(name = "param") String param);
}

Jersey: Detect when Controller Class is Created

I've implemented a JAX-RS server application using Jersey 2.24.
I use the Guice-HK2 bridge so that the controller classes (those annotated with #Path) are injected with dependencies from Guice, not Jersey/HK2.
However, HK2 still creates instances of the #Path annotated classes itself.
Is there a way I can plug into Jersey/HK2 so that I'm notified when a #Path annotated class is created? Like some sort of lifecycle listener? Every time a #Path annotated class is created by Jersey/HK2 I want to do some registering/logging of that class.
If Guice were doing the actual creation of the #Path annotated class I think I could do it using a generic Provider but that's not available in this case, since Jersey/HK2 is creating the actual instance.
Thank you!!
I think the least intrusive way would be to just use AOP. HK2 offers AOP. What you can do is create a ConstructorInterceptor. Something like
public class LoggingConstructorInterceptor implements ConstructorInterceptor {
private static final Logger LOG
= Logger.getLogger(LoggingConstructorInterceptor.class.getName());
#Override
public Object construct(ConstructorInvocation invocation) throws Throwable {
Constructor ctor = invocation.getConstructor();
LOG.log(Level.INFO, "Creating: {0}", ctor.getDeclaringClass().getName());
// returned instance from constructor invocation.
Object instance = invocation.proceed();
LOG.log(Level.INFO, "Created Instance: {0}", instance.toString());
return instance;
}
}
Then create a InterceptorService to only use the interceptor for classes annotated with #Path
public class PathInterceptionService implements InterceptionService {
private static final ConstructorInterceptor CTOR_INTERCEPTOR
= new LoggingConstructorInterceptor();
private final static List<ConstructorInterceptor> CTOR_LIST
= Collections.singletonList(CTOR_INTERCEPTOR);
#Override
public Filter getDescriptorFilter() {
return BuilderHelper.allFilter();
}
#Override
public List<MethodInterceptor> getMethodInterceptors(Method method) {
return null;
}
#Override
public List<ConstructorInterceptor> getConstructorInterceptors(Constructor<?> ctor) {
if (ctor.getDeclaringClass().isAnnotationPresent(Path.class)) {
return CTOR_LIST;
}
return null;
}
}
Then just register the InterceptionService and ConstructorInterceptor with the DI system
new ResourceConfig()
.register(new AbstractBinder(){
#Override
public void configure() {
bind(PathInterceptionService.class)
.to(InterceptionService.class)
.in(Singleton.class);
bind(LoggingConstructorInterceptor.class)
.to(ConstructorInterceptor.class)
.in(Singleton.class);
}
});
See complete example in this Gist
See Also:
HK2 documentation on AOP

Architecture for a RESTful web service with a single thread executor?

I would like to know what could be the best architecture for a RESTful web service with a single thread executor.
My goal :
Call a RESTful web service
The web service add a task in a thread queue and execute all the task 1 per 1.
The life cyle of instanciated object is really important (there must be only one thread queue). I know that a RESTful web service life cycle is "per request" (similar to #RequestScoped I think), so I see 2 options :
Option 1 :
public class RestService {
protected final static Executor executor;
protected final static Implementation1 impl1;
protected final static Implementation2 impl2;
static {
executor = Executors.newSingleThreadExecutor();
impl1 = new Implementation1();
impl2 = new Implementation2();
}
}
#Path("/servicename")
public class MyService extends RestService {
#POST
#Path("/compute")
public void compute(){
executor.execute(new Runnable(){
public void run(){
impl1.compute();
}
});
}
}
Option 2 :
#Singleton
public class RestService {
private Executor executor;
private Implementation1 impl1;
private Implementation2 impl2;
public RestService () {
executor = Executors.newSingleThreadExecutor();
impl1 = new Implementation1();
impl2 = new Implementation2();
}
public void execute(Runnable run){
executor.execute(run);
}
public Implementation1 getImplementation1(){
return impl1;
}
public Implementation2 getImplementation2(){
return impl2;
}
}
#Path("/servicename")
public class MyService {
#Inject
private RestService rs;
#POST
#Path("/compute")
public void compute(){
rs.execute(new Runnable(){
public void run(){
rs.getImplementation1().compute();
}
});
}
}
For option 1 I'm not sure about the "life cycle" about a static field. Which option should I use ? How would you do that ?
Thanks
EDIT :
Option 3 (thread handled by EJB Container) and "ordering" is not important :
#Singleton
public class RestService {
private final Executor executor;
private final Implementation1 impl1;
private final Implementation2 impl2;
public RestService () {
executor = Executors.newSingleThreadExecutor();
impl1 = new Implementation1();
impl2 = new Implementation2();
}
public void compute1(){
executor.execute(new Runnable(){
public void run(){
impl1.compute();
}
});
}
public void compute2(){
executor.execute(new Runnable(){
public void run(){
impl2.compute();
}
});
}
}
#Path("/servicename")
public class MyService {
#Inject
private RestService rs;
#POST
#Path("/compute1")
public void compute1(){
rs.compute1();
}
#POST
#Path("/compute2")
public void compute2(){
rs.compute2();
}
}
I think Option 3 is still better than Option 1 & 2.
I think it's a bad idea to do this. Threading ought to be handled by your container, not your code.
If you're deploying on a Java EE app server, you should let it handle the threading.
If you're deploying on a non-blocking I/O server like Netty or vert.x, you should let it handle the threading.
Under no circumstances should you be managing threads.
Static fields are instantiated when the .class loads. They don't have a "lifecycle" the way instances do. They won't be cleaned up until the class loader removes the .class file.
If you must have this kind of behavior, I'd either use a JMS queue to order the processing or a producer/consumer deque to manage it. You want the processing to be asynch. Let the REST service return a token or receipt to the client and have them come back to see when the processing is done. If the line is long, you'll have no way to know when their shirt is ready. The receipt lets them come back and check when their result is available.
If you used Spring, and then separated your business logic out into separate components (beans) and injected these beans into your service class you could effectively control the threading model of your implementation simply by changing the 'scope' attribute of your business logic beans.
Here's a description of Spring bean scopes.
This approach would allow you to easily modify/experiment with your solution and discover the best alternative for your situation.
As far as architecture is concerned I would recommend that you layer your system(loose coupling, high cohesion, separation of concerns, etc.). In this regard, you would probably have a service layer (REST), a business layer, and a data (dao) layer minimum. The layers interact with each other via interfaces. Use Spring to wire things together (inject dependencies). This gives you the flexibility to inject different implementations. Example would be to inject mock DAOs for your business classes for unit tests. The service layer where your REST services live would perform the translation of requests/responses, some validation, determine which business components to invoke, and invoke them.

Accessing main class from Jersey listener without the main class being a singleton object

I often use Jersey simple server as a webservice in my Java applications. Whenever I do this I only seem to be able to fetch properties from my application (main class) if it is setup as a singleton. So, I only seem to be able to fetch the information with MainClass.getInstance().
I know I can make the Jersey listener some sort of singleton in Jersey but this only means that the listener object will be instantiated once within Jersey in stead of for every single request: The listener is not a real singleton because it still needs a public constructor.
Is there a way to access my application class without it being a singleton object itself?
Here is the relevant code:
public class JerseyService {
private MainClass mainClass;
private int port = 8080;
private DefaultResourceConfig config;
private Closeable jerseyServer;
public JerseyService() {
config = new DefaultResourceConfig(JerseyListener.class); // Attach the listener (resource) to the Jersey server
jerseyServer = SimpleServerFactory.create("http://0.0.0.0:" + jerseyPort, config);
}
public String getData() {
return "somedata";
}
}
#Path("/")
public class JerseyListener {
#Path("getData")
#GET
#Produces("application/json")
public Response getData() {
// How can i call getData() from JerseyService
// without JerseyService being a singleton??
return Response.status(Response.Status.OK).entity("{<jsondata>}").build();
}
}

Categories