The goal of this article is to compare the performance between Spring MVC (Non-Reactive) and Spring WebFlux (Reactive). For this task we have taken into account three differents implementations:

• Using Spring MVC synchronous
• Using Spring MVC asynchronous
• Using Spring WebFlux

Implementations

Using Spring MVC synchronous

For this implementation the next endpoints have been defined:

1
2
3
4
5
6
7
8
9
@RequestMapping("/mvcsync/{id}")
public Message findById(@PathVariable(value = "id") String id) {
    return nonReactiveRepository.findById(id).orElse(null);
}
    
@PostMapping("/mvcsync")
public Message post(@Valid @RequestBody Message message) {
    return nonReactiveRepository.save(message);
}

Using Spring MVC asynchronous

Java 8 CompletableFuture library has been used to get asynchronous methods.

1
2
3
4
5
6
7
8
9
@RequestMapping("/mvcasync/{id}")
public CompletableFuture<Message> findById(@PathVariable(value = "id") String id) {
    return CompletableFuture.supplyAsync(() -> nonReactiveRepository.findById(id).orElse(null));
}
    
@PostMapping("/mvcasync")
public CompletableFuture<Message> post(@Valid @RequestBody Message message) {
    return CompletableFuture.supplyAsync(() -> nonReactiveRepository.save(message));
}

Using Spring WebFlux

The reactive version including MongoDB Reactive Repository.

1
2
3
4
5
6
7
8
9
@RequestMapping("/reactive/{id}")
public Mono<Message> findByIdReactive(@PathVariable(value = "id") String id) {
    return reactiveRepository.findById(id);
}
    
@PostMapping("/reactive")
public Mono<Message> postReactive(@Valid @RequestBody Message message) {
    return reactiveRepository.save(message);
}

The Test

Using JMeter we have created a test with the next characteristics:

• Each iteration will be composed by one create operation (POST) and one find operation (GET) to MongoDB.
• For each test, we are defining the number of concurrent users/threads

Hardware:

• MacBook Pro (15-inch, 2016)
• 2,6 GHz Intel Core i7
• 16 GB 2133 MHz LPDDR3

Results

The results shown in the following table refer to a one-minute execution with a one-second ramp-up period.

Now, we can extract the following conclusions based in our tests:

• Spring MVC Sync vs Async are similar in performance but with the asynchronous implementation we have a less error rate at beginning of the execution
• Spring WebFlux allows an increment in performance greater than 60% respect to Spring MVC

References

[1] Link to the project in Github