Sunday, December 10, 2017

Spring Webflux - Writing Filters

Spring Webflux is the new reactive web framework available as part of Spring 5+.  The way filters were written in a traditional Spring MVC based application(Servlet Filter, HandlerInterceptor) is very different from the way a filter is written in a Spring Webflux based application and this post will briefly go over the WebFlux approach to Filters.

Approach 1 - WebFilter

The first approach using WebFilter affects all endpoints broadly and covers Webflux endpoints written in a functional style as well the endpoints that are written using an annotation style. A WebFilter in Kotlin look like this:

    @Bean
    fun sampleWebFilter(): WebFilter {
        return WebFilter { e: ServerWebExchange, c: WebFilterChain ->
            val l: MutableList<String> = e.getAttributeOrDefault(KEY, mutableListOf())
            l.add("From WebFilter")
            e.attributes.put(KEY, l)
            c.filter(e)
        }
    }

The WebFilter adds a request attribute with the value being a collection where the filter is just putting in a message that it has intercepted the request.

Approach 2 - HandlerFilterFunction


The second approach is more focused and covers only endpoints written using functional style. Here specific RouterFunctions can be hooked up with a filter, along these lines:

Consider a Spring Webflux endpoint defined the following way:

@Bean
fun route(): RouterFunction<*> = router {
    GET("/react/hello", { r ->
        ok().body(fromObject(
                Greeting("${r.attribute(KEY).orElse("[Fallback]: ")}: Hello")
        ))
    POST("/another/endpoint", TODO())
        
    PUT("/another/endpoint", TODO())
})
        
}

A HandlerFilterFunction which intercepts these API's alone can be added in a highly focused way along these lines:

fun route(): RouterFunction<*> = router {
    GET("/react/hello", { r ->
        ok().body(fromObject(
                Greeting("${r.attribute(KEY).orElse("[Fallback]: ")}: Hello")
        ))
    })
    
    POST("/another/endpoint", TODO())
    
    PUT("/another/endpoint", TODO())
    
}.filter({ r: ServerRequest, n: HandlerFunction<ServerResponse> ->
    val greetings: MutableList<String> = r.attribute(KEY)
            .map { v ->
                v as MutableList<String>
            }.orElse(mutableListOf())

    greetings.add("From HandlerFilterFunction")

    r.attributes().put(KEY, greetings)
    n.handle(r)
})

Note that there is no need to be explicit about the types in Kotlin, I have added it just to be clear about the types in some of the lambda expressions


Conclusion

The WebFilter approach and the HandlerFilterFunction are very different from the Spring WebMVC based approach of writing filters using Servlet Specs or using HandlerInterceptors and this post summarizes the new approaches - I have samples available in my git repo which goes over these in more detail.

Friday, December 1, 2017

Annotated controllers - Spring Web/Webflux and Testing

Spring Webflux and Spring Web are two entirely different web stacks. Spring Webflux, however, continues to support an annotation-based programming model

An endpoint defined using these two stacks may look similar but the way to test such an endpoint is fairly different and a user writing such an endpoint has to be aware of which stack is active and formulate the test accordingly.

Sample Endpoint

Consider a sample annotation based endpoint:


import org.springframework.web.bind.annotation.PostMapping
import org.springframework.web.bind.annotation.RequestBody
import org.springframework.web.bind.annotation.RequestMapping
import org.springframework.web.bind.annotation.RestController


data class Greeting(val message: String)

@RestController
@RequestMapping("/web")
class GreetingController {
    
    @PostMapping("/greet")
    fun handleGreeting(@RequestBody greeting: Greeting): Greeting {
        return Greeting("Thanks: ${greeting.message}")
    }
    
}


Testing with Spring Web

If Spring Boot 2 starters were used to create this application with Spring Web as the starter, specified using a Gradle build file the following way:

compile('org.springframework.boot:spring-boot-starter-web')

then the test of such an endpoint would be using a Mock web runtime, referred to as Mock MVC:

import org.junit.Test
import org.junit.runner.RunWith
import org.springframework.beans.factory.annotation.Autowired
import org.springframework.boot.test.autoconfigure.web.servlet.WebMvcTest
import org.springframework.test.context.junit4.SpringRunner
import org.springframework.test.web.servlet.MockMvc
import org.springframework.test.web.servlet.request.MockMvcRequestBuilders.post
import org.springframework.test.web.servlet.result.MockMvcResultMatchers.content


@RunWith(SpringRunner::class)
@WebMvcTest(GreetingController::class)
class GreetingControllerMockMvcTest {

    @Autowired
    lateinit var mockMvc: MockMvc

    @Test
    fun testHandleGreetings() {
        mockMvc
                .perform(
                        post("/web/greet")
                                .content(""" 
                                |{
                                |"message": "Hello Web"
                                |}
                            """.trimMargin())
                ).andExpect(content().json("""
                    |{
                    |"message": "Thanks: Hello Web"
                    |}
                """.trimMargin()))
    }
}


Testing with Spring Web-Flux

If on the other hand Spring-Webflux starters were pulled in, say with the following Gradle dependency:

compile('org.springframework.boot:spring-boot-starter-webflux')

then the test of this endpoint would be using the excellent WebTestClient class, along these lines:

import org.junit.Test
import org.junit.runner.RunWith
import org.springframework.beans.factory.annotation.Autowired
import org.springframework.boot.test.autoconfigure.web.reactive.WebFluxTest
import org.springframework.http.HttpHeaders
import org.springframework.test.context.junit4.SpringRunner
import org.springframework.test.web.reactive.server.WebTestClient
import org.springframework.web.reactive.function.BodyInserters


@RunWith(SpringRunner::class)
@WebFluxTest(GreetingController::class)
class GreetingControllerTest {

    @Autowired
    lateinit var webTestClient: WebTestClient

    @Test
    fun testHandleGreetings() {
        webTestClient.post()
                .uri("/web/greet")
                .header(HttpHeaders.CONTENT_TYPE, "application/json")
                .body(BodyInserters
                        .fromObject(""" 
                                |{
                                |   "message": "Hello Web"
                                |}
                            """.trimMargin()))
                .exchange()
                .expectStatus().isOk
                .expectBody()
                .json("""
                    |{
                    |   "message": "Thanks: Hello Web"
                    |}
                """.trimMargin())
    }
}


Conclusion

It is easy to assume that since the programming model looks very similar using Spring Web and Spring Webflux stacks, that the tests for such a legacy test using Spring Web would continue over to Spring Webflux, this is however not true, as a developer we have to be mindful of the underlying stack that comes into play and formulate the test accordingly. I hope this post clarifies how such a test should be crafted.

Monday, November 20, 2017

Using Micrometer with Spring Boot 2

This is a very quick introduction to using the excellent Micrometer library to instrument a Spring Boot 2 based application and recording the metrics in Prometheus


Introduction

Micrometer provides a Java based facade over the client libraries that the different monitoring tools provide.

As an example consider Prometheus, if I were to integrate my Java application with Prometheus, I would have used the client library called Prometheus Client Java, used the data-structures(Counter, Gauge etc) to collect and provide data to Prometheus. If for any reason the monitoring system is changed, the code will have to be changed for the new system.

Micrometer attempts to alleviate this by providing a common facade that the applications use when writing code, binding to the monitoring system is purely a runtime concern and so changing Metrics system from Prometheus to say Datadog just requires changing a runtime library without needing any code changes.



Instrumenting a Spring Boot 2 Application

Nothing special needs to be done to get Micrometer support for a Spring Boot 2 based app, adding in the actuator starters pulls in Micrometer as a transitive dependency:

for eg. in a gradle based project this is sufficient:

dependencies {
    compile('org.springframework.boot:spring-boot-starter-actuator')
    ...
}

Additionally since the intention is to send the data to Prometheus a dependency has to be pulled in which provides the necessary Micrometer SPI's.


dependencies {
    ...
    runtime("io.micrometer:micrometer-registry-prometheus")
    ...
}

By default Micrometer provides a set of intelligent bindings which instruments the Spring based Web and Webflux endpoints and adds in meters to collect the duration, count of calls. Additionally it also provides bindings to collect JVM metrics - memory usage, threadpool, etc.

An application property needs to be enabled to expose an endpoint which Prometheus will use to scrape the metrics data:

endpoints:
  prometheus:
    enabled: true

If the application is brought up at this point, the "/applications/prometheus" endpoint should be available showing a rich set of metrics, the following is a sample on my machine:


The default metrics is very rich and should cover most of the common set of metrics requirements of an application, if additional metrics is required it can easily added in as shown in the following code snippet:

class MessageHandler {
    
    private val counter = Metrics.counter("handler.calls", "uri", "/messages")
    
    fun handleMessage(req: ServerRequest): Mono<ServerResponse> {
        return req.bodyToMono<Message>().flatMap { m ->
            counter.increment()
            ...
...
}

Integrating with Prometheus

Prometheus can be configured to scrape data from the endpoint exposed by the Spring Boot2 app, a snippet of Prometheus configuration looks like this:

scrape_configs:
  - job_name: 'myapp'
    metrics_path: /application/prometheus
    static_configs:
      - targets: ['localhost:8080']

This is not really a production configuration, in a production setting it may better to use a Prometheus Push Gateway to broker the collection of metrics.

Prometheus provides a basic UI to preview the information that it scrapes, it can be accessed by default at port 9090. Here is a sample graph with the data produced during a load test:



Conclusion

Micrometer makes it very easy to instrument an application and collect a good set of basic metrics which can be stored and visualized in Prometheus. If you are interested in following this further, I have a sample application using Micrometer available here - https://github.com/bijukunjummen/boot2-load-demo

Sunday, October 22, 2017

Raw performance numbers - Spring Boot 2 Webflux vs Spring Boot 1

Summary

Spring Boot 2 with Spring Webflux based application outperforms a Spring Boot 1 based application by a huge margin for IO heavy workloads. The following is a summarized result of a load test - Response time for a IO heavy transaction with varying concurrent users:



When the number of concurrent users remains low (say less than 1000) both Spring Boot 1 and Spring Boot 2 handle the load well and the 95 percentile response time remains milliseconds above a expected value of 300 ms.

At higher concurrency levels, the Async Non-Blocking IO and reactive support in Spring Boot 2 starts showing its colors - the 95th percentile time even with a very heavy load of 5000 users remains at around 312ms! Spring Boot 1 records a lot of failures and high response times at these concurrency levels.

Details



My set-up for the performance test is the following:



The sample applications expose an endpoint(/passthrough/message) which in-turn calls a downstream service. The request message to the endpoint looks something like this:

{
  "id": "1",
  "payload": "sample payload",
  "delay": 3000
}

The downstream service would delay based on the "delay" attribute in the message (in milliseconds).


Spring Boot 1 Application

I have used Spring Boot 1.5.8.RELEASE for the Boot 1 version of the application. The endpoint is a simple Spring MVC controller which in turn uses Spring's RestTemplate to make the downstream call. Everything is synchronous and blocking and I have used the default embedded Tomcat container as the runtime. This is the raw code for the downstream call:

public MessageAck handlePassthrough(Message message) {
    ResponseEntity<MessageAck> responseEntity = this.restTemplate.postForEntity(targetHost 
                                                            + "/messages", message, MessageAck.class);
    return responseEntity.getBody();
}

Spring Boot 2 Application

Spring Boot 2 version of the application exposes a Spring Webflux based endpoint and uses WebClient, the new non-blocking, reactive alternate to RestTemplate to make the downstream call - I have also used Kotlin for the implementation, which has no bearing on the performance. The runtime server is Netty:

import org.springframework.http.HttpHeaders
import org.springframework.http.MediaType
import org.springframework.web.reactive.function.BodyInserters.fromObject
import org.springframework.web.reactive.function.client.ClientResponse
import org.springframework.web.reactive.function.client.WebClient
import org.springframework.web.reactive.function.client.bodyToMono
import org.springframework.web.reactive.function.server.ServerRequest
import org.springframework.web.reactive.function.server.ServerResponse
import org.springframework.web.reactive.function.server.bodyToMono
import reactor.core.publisher.Mono

class PassThroughHandler(private val webClient: WebClient) {

    fun handle(serverRequest: ServerRequest): Mono<ServerResponse> {
        val messageMono = serverRequest.bodyToMono<Message>()

        return messageMono.flatMap { message ->
            passThrough(message)
                    .flatMap { messageAck ->
                        ServerResponse.ok().body(fromObject(messageAck))
                    }
        }
    }

    fun passThrough(message: Message): Mono<MessageAck> {
        return webClient.post()
                .uri("/messages")
                .header(HttpHeaders.CONTENT_TYPE, MediaType.APPLICATION_JSON_VALUE)
                .header(HttpHeaders.ACCEPT, MediaType.APPLICATION_JSON_VALUE)
                .body(fromObject<Message>(message))
                .exchange()
                .flatMap { response: ClientResponse ->
                    response.bodyToMono<MessageAck>()
                }
    }
}



Details of the Perfomance Test


The test is simple, for different sets of concurrent users (300, 1000, 1500, 3000, 5000), I send a message with the delay attribute set to 300 ms, each user repeats the scenario 30 times with a delay between 1 to 2 seconds between requests. I am using the excellent Gatling tool to generate this load.

Results

These are the results as captured by Gatling:

300 concurrent users:
Boot 1 Boot 2



1000 concurrent users:
Boot 1 Boot 2




1500 concurrent users:
Boot 1 Boot 2




3000 concurrent users:
Boot 1 Boot 2




5000 concurrent users:
Boot 1 Boot 2




Reference

The sample application and the load scripts are available in my github repo - https://github.com/bijukunjummen/boot2-load-demo.

Thursday, October 5, 2017

Kata - implementing a functional List data structure in Kotlin

I saw an exercise in chapter 3 of the excellent Functional Programming in Scala book which deals with defining functional data structures and uses the linked list as an example on how to go about developing such a datastructure. I wanted to try this sample using Kotlin to see to what extent I can replicate the sample.

A scala skeleton of the sample is available in the companion code to the book here and my attempt in Kotlin is heavily inspired (copied!) by the answerkey in the repository.

Basic

This is what a basic List representation in Kotlin looks like:

sealed class List<out A> {

    abstract val head: A

    abstract val tail: List<A>
}

data class Cons<out T>(override val head: T, override val tail: List<T>) : List<T>()

object Nil : List<Nothing>() {
    override val head: Nothing
        get() {
            throw NoSuchElementException("head of an empty list")
        }

    override val tail: List<Nothing>
        get() {
            throw NoSuchElementException("tail of an empty list")
        }
}

the List has been defined as a sealed class, this means that all subclasses of the sealed class will be defined in the same file. This is useful for pattern matching on the type of an instance and will come up repeatedly in most of the functions.

There are two implementations of this List -
1. Cons a non-empty list consisting of a head element and a tail List,
2. Nil an empty List

This is already very useful in its current form, consider the following which constructs a List and retrieves elements from it:

val l1:List<Int> = Cons(1, Cons(2, Cons(3, Cons(4, Nil))))
assertThat(l1.head).isEqualTo(1)
assertThat(l1.tail).isEqualTo(Cons(2, Cons(3, Cons(4, Nil))))


val l2:List<String> = Nil


Pattern Matching with "when" expression

Now to jump onto implementing some methods of List. Since List is a sealed class it allows for some good pattern matching, say to get the sum of elements in the List:

fun sum(l: List<Int>): Int {
    return when(l) {
        is Cons -> l.head + sum(l.tail)
        is Nil -> 0
    }
}

The compiler understands that Cons and Nil are the only two paths to take for the match on a list instance.

A little more complex operation, "drop" some number of elements from the beginning of the list and "dropWhile" which takes in a predicate and drops elements from the beginning matching the predicate:

fun drop(n: Int): List<A> {
    return if (n <= 0)
        this
    else when (this) {
        is Cons -> tail.drop(n - 1)
        is Nil -> Nil
    }
}

val l = list(4, 3, 2, 1)
assertThat(l.drop(2)).isEqualTo(list(2, 1))

fun dropWhile(p: (A) -> Boolean): List<A> {
    return when(this) {
        is Cons -> if (p(this.head)) this.tail.dropWhile(p) else this
        is Nil -> Nil
    }
}

val l = list(1, 2, 3, 5, 8, 13, 21, 34, 55, 89)
assertThat(l.dropWhile({e -> e < 20})).isEqualTo(list(21, 34, 55, 89))

These show off the power of pattern matching with the "when" expression in Kotlin.


Unsafe Variance!

To touch on a wrinkle, see how the List is defined with a type parameter that is declared as "out T", this is called the "declaration site variance" which in this instance makes List co-variant on type T. Declaration site variance is explained beautifully with the Kotlin documentation. With the way List is declared, it allows me to do something like this:

val l:List<Int> = Cons(1, Cons(2, Nil))
val lAny: List<Any> = l

Now, consider an "append" function which appends another list:

fun append(l: List<@UnsafeVariance A>): List<A> {
    return when (this) {
        is Cons -> Cons(head, tail.append(l))
        is Nil -> l
    }
}

here a second list is taken as a parameter to the append function, however Kotlin would flag the parameter - this is because it is okay to return a co-variant type but not to take it as a parameter. However since we know the List in its current form is immutable, I can get past this by marking the type parameter with "@UnsafeVariance" annotation.

Folding

Folding operations allow the list to be "folded" into a result based on some aggregation on individual elemnents in it.

Consider foldLeft:

fun <B> foldLeft(z: B, f: (B, A) -> B): B {
    tailrec fun foldLeft(l: List<A>, z: B, f: (B, A) -> B): B {
        return when (l) {
            is Nil -> z
            is Cons -> foldLeft(l.tail, f(z, l.head), f)
        }
    }

    return foldLeft(this, z, f)
}

If a list were to consist of elements (2, 3, 5, 8) then foldLeft is equivalent to "f(f(f(f(z, 2), 3),5),8)"

With this higher order function in place, the sum function can expressed this way:

val l = Cons(1, Cons(2, Cons(3, Cons(4, Nil))))
assertThat(l.foldLeft(0, {r, e -> r + e})).isEqualTo(10)


foldRight looks like the following in Kotlin:

fun <B> foldRight(z: B, f: (A, B) -> B): B {
    return when(this) {
        is Cons -> f(this.head, tail.foldRight(z, f))
        is Nil -> z
    }
}
If a list were to consist of elements (2, 3, 5, 8) then foldRight is equivalent to "f(2, f(3, f(5, f(8, z))))"

This version of the foldRight, though cooler looking is not tail recursive, a more stack friendly version can be implemented using the previously defined tail recursive foldLeft by simply reversing the List and calling foldLeft internally the following way:

fun reverse(): List<A> {
    return foldLeft(Nil as List<A>, { b, a -> Cons(a, b) })
}

fun <B> foldRightViaFoldLeft(z: B, f: (A, B) -> B): B {
    return reverse().foldLeft(z, { b, a -> f(a, b) })
}

map and flatMap

map is a function which transforms the element of this list:

fun <B> map(f: (A) -> B): List<B> {
    return when (this) {
        is Cons -> Cons(f(head), tail.map(f))
        is Nil -> Nil
    }
}

An example of using this function is the following:
val l = Cons(1, Cons(2, Cons(3, Nil)))
val l2 = l.map { e -> e.toString() }
assertThat(l2).isEqualTo(Cons("1", Cons("2", Cons("3", Nil))))

A variation of map where the transforming function returns another list, and the final results flattens everything, best demoed using an example after the implementation:

fun <B> flatMap(f: (a: A) -> List<@UnsafeVariance B>): List<B> {
    return flatten(map { a -> f(a) })
}

companion object {
    fun <A> flatten(l: List<List<A>>): List<A> {
        return l.foldRight(Nil as List<A>, { a, b -> a.append(b) })
    }
}


val l = Cons(1, Cons(2, Cons(3, Nil)))

val l2 = l.flatMap { e -> list(e.toString(), e.toString()) }

assertThat(l2)
        .isEqualTo(
                Cons("1", Cons("1", Cons("2", Cons("2", Cons("3", Cons("3", Nil)))))))


This covers the basics involved in implementing a functional list datastructure using Kotlin, there were a few rough edges when compared to the scala version but I think it mostly works. Admittedly the sample can likely be improved drastically, if you have any observations on how to improve the code please do send me a PR at my github repo for this sample or as comment to this post.

Tuesday, September 19, 2017

Testing time based reactor core streams with Virtual time

Reactor Core implements the Reactive Streams specification and deals with handling a (potentially unlimited) stream of data. If it interests you, do check out the excellent documentation it offers. Here I am assuming some basic familiarity with the Reactor Core libraries Flux and Mono types and will cover Reactor Core provides an abstraction to time itself to enable testing of functions which depend on passage of time.

For certain operators of Reactor-core, time is an important consideration - for eg, a variation of "interval" function which emits an increasing number every 5 seconds after an initial "delay" of 10 seconds:

val flux = Flux
        .interval(Duration.ofSeconds(10), Duration.ofSeconds(5))
        .take(3)

Testing such a stream of data depending on normal passage of time would be terrible, such a test would take about 20 seconds to finish.

Reactor-Core provides a solution, an abstraction to time itself - Virtual time based Scheduler, that provides a neat way to test these kinds of operations in a deterministic way.

Let me show it in two ways, an explicit way which should make the actions of Virtual time based scheduler very clear followed by the recommended approach of testing with Reactor Core.

import org.assertj.core.api.Assertions.assertThat
import org.junit.Test
import reactor.core.publisher.Flux
import reactor.test.scheduler.VirtualTimeScheduler
import java.time.Duration
import java.util.concurrent.CountDownLatch


class VirtualTimeTest {
    
    @Test
    fun testExplicit() {
        val mutableList = mutableListOf<Long>()

        val scheduler = VirtualTimeScheduler.getOrSet()
        val flux = Flux
                .interval(Duration.ofSeconds(10), Duration.ofSeconds(5), scheduler)
                .take(3)

        val latch = CountDownLatch(1)
        
        flux.subscribe({ l -> mutableList.add(l) }, { _ -> }, { latch.countDown() })
        
        scheduler.advanceTimeBy(Duration.ofSeconds(10))
        assertThat(mutableList).containsExactly(0L)
        
        scheduler.advanceTimeBy(Duration.ofSeconds(5))
        assertThat(mutableList).containsExactly(0L, 1L)
        
        scheduler.advanceTimeBy(Duration.ofSeconds(5))
        assertThat(mutableList).containsExactly(0L, 1L, 2L)

        latch.await()
    }
    
}

1. First the scheduler for "Flux.interval" function is being set to be the Virtual Time based Scheduler.

2. The stream of data is expected to be emitted every 5 seconds after a 10 second delay

3. VirtualTimeScheduler provides an "advanceTimeBy" method to advance the Virtual time by a Duration, so the time is being first advanced by the delay time of 10 seconds at which point the first element(0) is expected to be emitted

4. Then it is subsequently advanced by 5 seconds twice to get 1 and 2 respectively.

This is deterministic and the test completes quickly. This version of the test is ugly though, it uses a list to collect and assert the results on and a CountDownLatch to control when the test terminates. A far cleaner approach for testing Reactor-Core types is using the excellent StepVerifier class and a test which makes use of this class looks like this:

import org.junit.Test
import reactor.core.publisher.Flux
import reactor.test.StepVerifier
import reactor.test.scheduler.VirtualTimeScheduler
import java.time.Duration

class VirtualTimeTest {

    @Test
    fun testWithStepVerifier() {

        VirtualTimeScheduler.getOrSet()
        val flux = Flux
                .interval(Duration.ofSeconds(10), Duration.ofSeconds(5))
                .take(3)

        StepVerifier.withVirtualTime({ flux })
                .expectSubscription()
                .thenAwait(Duration.ofSeconds(10))
                .expectNext(0)
                .thenAwait(Duration.ofSeconds(5))
                .expectNext(1)
                .thenAwait(Duration.ofSeconds(5))
                .expectNext(2)
                .verifyComplete()
    }
 }

This new test with StepVerifier reads well with each step advancing time and asserting on what is expected at that point.



Friday, September 1, 2017

Spring Webflux - Kotlin DSL - a walkthrough of the implementation

In a previous blog post I had described how Spring Webflux, the reactive programming support in Spring Web Framework, uses a Kotlin based DSL to enable users to describe routes in a very intuitive way. Here I wanted to explore a little of the underlying implementation.


A sample DSL describing a set of endpoints looks like this:

package sample.routes

import org.springframework.context.annotation.Bean
import org.springframework.context.annotation.Configuration
import org.springframework.http.MediaType.APPLICATION_JSON
import org.springframework.web.reactive.function.server.router
import sample.handler.MessageHandler

@Configuration
class AppRoutes(private val messageHandler: MessageHandler) {

    @Bean
    fun apis() = router {
        (accept(APPLICATION_JSON) and "/messages").nest {
            GET("/", messageHandler::getMessages)
            POST("/", messageHandler::addMessage)
            GET("/{id}", messageHandler::getMessage)
            PUT("/{id}", messageHandler::updateMessage)
            DELETE("/{id}", messageHandler::deleteMessage)
        }
    }

}


To analyze the sample let me start with a smaller working example:

import org.junit.Test
import org.springframework.test.web.reactive.server.WebTestClient
import org.springframework.web.reactive.function.server.ServerResponse.ok
import org.springframework.web.reactive.function.server.router

class AppRoutesTest {

    @Test
    fun testSimpleGet() {
        val routerFunction = router {
            GET("/isokay", { _ -> ok().build() })
        }

        val client = WebTestClient.bindToRouterFunction(routerFunction).build()

        client.get()
                .uri("/isokay")
                .exchange()
                .expectStatus().isOk
    }
}

The heart of the route definition is the "router" function:

import org.springframework.web.reactive.function.server.router
...
val routerFunction = router {
    GET("/isokay", { _ -> ok().build() })
}

which is defined the following way:

fun router(routes: RouterFunctionDsl.() -> Unit) = RouterFunctionDsl().apply(routes).router()

The parameter "routes" is a special type of lambda expression, called a Lambda expression with a receiver. This means that in the context of the router function, this lambda expression can only be invoked by instances of "RouterFunctionDsl" which is what is done in the body of the function using apply method, this also means in the body of the lambda expression "this" refers to an instance of "RouterFunctionDsl". Knowing this opens up access to the methods of "RouterFunctionDsl" one of which is GET that is used in the example, GET is defined as follows:

fun GET(pattern: String, f: (ServerRequest) -> Mono<ServerResponse>) {
  ...
}

There are other ways express the same endpoint:

GET("/isokay2")({ _ -> ok().build() })

implemented in Kotlin very cleverly as:

fun GET(pattern: String): RequestPredicate = RequestPredicates.GET(pattern)

operator fun RequestPredicate.invoke(f: (ServerRequest) -> Mono<ServerResponse>) {
 ...
}

Here GET with the pattern returns a "RequestPredicate" for which an extension function has been defined (in the context of the DSL) called invoke, which is in turn a specially named operator.

Or a third way:

"/isokay" { _ -> ok().build() }

which is implemented by adding an extension function on String type and defined the following way:

operator fun String.invoke(f: (ServerRequest) -> Mono<ServerResponse>) {
  ...
}


I feel that the Spring Webflux makes an excellent use of the Kotlin DSL in making some of these route definitions easy to read while remaining concise.

This should provide enough primer to explore the source code of Routing DSL in Spring Webflux .

My samples are available in a github repository here - https://github.com/bijukunjummen/webflux-route-with-kotlin

Monday, August 21, 2017

Gradle Kotlin DSL

Gradle build scripts can now be written using a dsl with Kotlin Language. All the concepts that work with traditional gradle build translate to a very intuitive dsl in Kotlin and have two additional features - it is typesafe and the script has excellent IDE support using Intellij IDEA.

My experience with the Gralde Kotlin DSL is fairly limited - all of one build script which is the subject of this article.

If you want to simply see how a sample script looks, I have a sample github repo with just that here - https://github.com/bijukunjummen/cf-show-env


Just to compare:

1. Consider the way different plugins are applied with gradle:

plugins {
 id "com.github.pivotalservices.cf-app" version "1.0.9"
}

apply plugin: 'kotlin"
apply plugin: 'java'
apply plugin: 'org.springframework.boot'
apply from: 'gradle/gatling.gradle'


An equivalent kotlin dsl is the following:

plugins {
    id("com.github.pivotalservices.cf-app").version("1.0.9")
}

apply {
    plugin("kotlin")
    plugin("java")
    plugin("org.springframework.boot")    
    from("gradle/gatling.gradle")
}



2. Adding project dependencies:

dependencies {
    compile('org.springframework.boot:spring-boot-starter-actuator')
    compile('org.springframework.boot:spring-boot-devtools')
    compile('org.springframework.boot:spring-boot-starter-thymeleaf')
    compile('org.springframework.boot:spring-boot-starter-web')
    compile('com.google.guava:guava:19.0')
    compile("org.webjars:bootstrap:3.3.7")
    compile("org.webjars:jquery:3.1.1")
    compile("io.prometheus:simpleclient:${prometheus_client_version}")
    compile("io.prometheus:simpleclient_spring_boot:${prometheus_client_version}")
    compile('nz.net.ultraq.thymeleaf:thymeleaf-layout-dialect')
    testCompile('org.springframework.boot:spring-boot-starter-test')
}

an equivalent code using kotlin DSL:

dependencies {
    val prometheus_client_version = "0.0.21"

    compile("org.springframework.boot:spring-boot-starter-actuator")
    compile("org.springframework.boot:spring-boot-devtools")
    compile("org.springframework.boot:spring-boot-starter-thymeleaf")
    compile("org.springframework.boot:spring-boot-starter-web")
    compile("com.google.guava:guava:19.0")
    compile("org.webjars:bootstrap:3.3.7")
    compile("org.webjars:jquery:3.1.1")
    compile("io.prometheus:simpleclient:${prometheus_client_version}")
    compile("io.prometheus:simpleclient_spring_boot:${prometheus_client_version}")
    compile("nz.net.ultraq.thymeleaf:thymeleaf-layout-dialect")
    testCompile("org.springframework.boot:spring-boot-starter-test")
}

3. Configuring Plugins - I have a plugin which helps deploy applications to Cloud Foundry, and works off a configuration which looks like this, when expressed using normal gradle build:

cfConfig {
    //CF Details
    ccHost = "api.local.pcfdev.io"
    ccUser = "admin"
    ccPassword = "admin"
    org = "pcfdev-org"
    space = "pcfdev-space"

    //App Details
    name = "cf-show-env"
    hostName = "cf-show-env"
    filePath = "build/libs/cf-show-env-0.1.3-SNAPSHOT.jar"
    path = ""
    domain = "local.pcfdev.io"
    instances = 2
    memory = 1024
    timeout = 180

    //Env and services
    buildpack = "https://github.com/cloudfoundry/java-buildpack.git"


    environment = ["JAVA_OPTS": "-Djava.security.egd=file:/dev/./urandom", "SPRING_PROFILES_ACTIVE": "cloud"]

    cfService {
        name = "p-mysql"
        plan = "512mb"
        instanceName = "test-db"
    }
    
 
    
    cfUserProvidedService {
        instanceName = "mydb1"
        credentials = ["jdbcUri": "someuri1"]
    }
}

this can now be configured in a typesafe way with full auto-completion support in IntelliJ the following way using Kotlin DSL:

configure< CfPluginExtension> {
    //CF Details
    ccHost = "api.local.pcfdev.io"
    ccUser = "admin"
    ccPassword = "admin"
    org = "pcfdev-org"
    space = "pcfdev-space"

    //App Details
    name = "cf-show-env"
    hostName = "cf-show-env"
    filePath = "build/libs/cf-show-env-1.0.0-M1.jar"
    path = ""
    domain = "local.pcfdev.io"
    instances = 2
    memory = 1024
    timeout = 180

    //Env and services
    buildpack = "https://github.com/cloudfoundry/java-buildpack.git"

    environment = mapOf(
            "JAVA_OPTS" to "-Djava.security.egd=file:/dev/./urandom", 
            "SPRING_PROFILES_ACTIVE" to "cloud"
    )

    cfService(closureOf<CfService> {
        name = "p-mysql"
        plan = "512mb"
        instanceName = "test-db"
    })
    
    cfUserProvidedService(closureOf<CfUserProvidedService> { 
        instanceName = "myups"
        credentials = mapOf(
                "user" to "someuser",
                "uri" to "someuri"
        )
    })

}

4. And finally a straight task:
task "hello-world" {
    doLast {
        println("Hello World")
    }
}

task showAppUrls(dependsOn: "cf-get-app-detail") << {
    print "${project.cfConfig.applicationDetail}"
}

looks more or less the same in Kotlin DSL:

task("hello-world") {
    doLast {
        println("Hello World")
    }
}


task("showAppUrls").dependsOn("cf-get-app-detail").doLast {
       println(cfConfig);
}


I am excited about using Kotlin DSL to configure my gradle builds, there are a few quirks to keep in mind though - the Intellij support tends to be a little flaky, it took a few tries for the IDEA to start helping with the auto-completions, also I needed to google quite a bit and look at some of the sample projects in gradle kotlin dsl, all in all though this has an awesome potential.

Monday, August 14, 2017

Concourse caching for Java Maven and Gradle builds

Concourse CI 3.3.x has introduced the ability to cache paths between task runs. This feature helps speed up tasks which cache content in specific folders - here I will demonstrate how this feature can be used for speeding up maven and gradle based java builds.

The code and the pipeline that I am using for this post is available at my github repo here - https://github.com/bijukunjummen/ci-concourse-caching-sample

Let me start with the gradle build, if I were to build the project using a gradle wrapper using the following command:

./gradlew clean build

then gradle would download the dependent libraries into a ".gradle" folder in the users home folder by default. This location of this folder can be changed using a "GRADLE_USER_HOME" environment variable, which is what I will be using in a concourse task to control the location of a cached path.

A concourse task which builds my project looks like this:

---
platform: linux
image_resource:
  type: docker-image
  source:
    repository: openjdk
    tag: 8-jdk
inputs:
  - name: repo
outputs:
  - name: out
run:
  path: /bin/bash
  args:
    - repo/ci/tasks/build.sh

caches:
  - path: .gradle/
  - path: .m2/

params:
  PROJECT_TYPE: 

See the caches parameter is specified as ".gradle" above. So all I have to do now is to ensure that Gradle uses this location as its home folder, which I would do in my build script:

export ROOT_FOLDER=$( pwd )
export GRADLE_USER_HOME="${ROOT_FOLDER}/.gradle"


The process to cache maven resources for a maven build is along the same lines, maven caches the dependent jars in a location that can be specified in a variety of ways, the one I have used is to specify this location via a dynamically generated settings.xml file the following way:

M2_HOME=${HOME}/.m2
mkdir -p ${M2_HOME}

M2_LOCAL_REPO="${ROOT_FOLDER}/.m2"

mkdir -p "${M2_LOCAL_REPO}/repository"

cat > ${M2_HOME}/settings.xml <<EOF

<settings xmlns="http://maven.apache.org/SETTINGS/1.0.0"
      xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
      xsi:schemaLocation="http://maven.apache.org/SETTINGS/1.0.0
                          https://maven.apache.org/xsd/settings-1.0.0.xsd">
      <localRepository>${M2_LOCAL_REPO}/repository</localRepository>
</settings>

EOF

which is quite a bit of bash scripting, all it is doing is generating a settings.xml with a localRepository tag set to ".m2/repository" folder which is relative to the temporary folder created by concourse for the build and thus can be cached.

With these changes in place, the behavior is that the downloads happen for the first run of the task but then get cached for subsequent runs. In my local concourse set-up a gradle build taking about 2 mins for a first time build takes about 20 seconds for a subsequent build !

You can try out this feature in my demo project here - https://github.com/bijukunjummen/ci-concourse-caching-sample



Friday, July 28, 2017

Kotlintest and property based testing

I was very happy to see that Kotlintest, a port of the excellent scalatest in Kotlin, supports property based testing.

I was introduced to property based testing through the excellent "Functional programming in Scala" book.

The idea behind property based testing is simple - the behavior of a program is described as a property and the testing framework generates random data to validate the property. This is best illustrated with an example using the excellent scalacheck library:


import org.scalacheck.Prop.forAll
import org.scalacheck.Properties

object ListSpecification extends Properties("List") {
  property("reversing a list twice should return the list") = forAll { (a: List[Int]) =>
    a.reverse.reverse == a
  }
}

scalacheck would generate a random list(of integer) of varying sizes and would validate that this property holds for the lists. A similar specification expressed through Kotlintest looks like this:

import io.kotlintest.properties.forAll
import io.kotlintest.specs.StringSpec


class ListSpecification : StringSpec({
    "reversing a list twice should return the list" {
        forAll{ list: List<Int> ->
            list.reversed().reversed().toList() == list
        }
    }
})

If the generators have to be a little more constrained, say if we wanted to test this behavior on lists of integer in the range 1 to 1000 then an explicit generator can be passed in the following way, again starting with scalacheck:

import org.scalacheck.Prop.forAll
import org.scalacheck.{Gen, Properties}

object ListSpecification extends Properties("List") {
  val intList = Gen.listOf(Gen.choose(1, 1000))
  property("reversing a list twice should return the list") = forAll(intList) { (a: List[Int]) =>
    a.reverse.reverse == a
  }
}

and an equivalent kotlintest code:

import io.kotlintest.properties.Gen
import io.kotlintest.properties.forAll
import io.kotlintest.specs.StringSpec

class BehaviorOfListSpecs : StringSpec({
    "reversing a list twice should return the list" {
        val intList = Gen.list(Gen.choose(1, 1000))

        forAll(intList) { list ->
            list.reversed().reversed().toList() == list
        }
    }
})

Given this let me now jump onto another example from the scalacheck site, this time to illustrate a failure:

import org.scalacheck.Prop.forAll
import org.scalacheck.Properties

object StringSpecification extends Properties("String") {

  property("startsWith") = forAll { (a: String, b: String) =>
    (a + b).startsWith(a)
  }

  property("concatenate") = forAll { (a: String, b: String) =>
    (a + b).length > a.length && (a + b).length > b.length
  }

  property("substring") = forAll { (a: String, b: String, c: String) =>
    (a + b + c).substring(a.length, a.length + b.length) == b
  }
}

the second property described above is wrong - if two strings are concatenated together they are ALWAYS larger than each of the parts, this is not true if one of the strings is blank. If I were to run this test using scalacheck it correctly catches this wrongly specified behavior:

+ String.startsWith: OK, passed 100 tests.
! String.concatenate: Falsified after 0 passed tests.
> ARG_0: ""
> ARG_1: ""
+ String.substring: OK, passed 100 tests.
Found 1 failing properties.

An equivalent kotlintest is the following:

import io.kotlintest.properties.forAll
import io.kotlintest.specs.StringSpec

class StringSpecification : StringSpec({
    "startsWith" {
        forAll { a: String, b: String ->
            (a + b).startsWith(a)
        }
    }

    "concatenate" {
        forAll { a: String, b: String ->
            (a + b).length > a.length && (a + b).length > b.length
        }
    }

    "substring" {
        forAll { a: String, b: String, c: String ->
            (a + b + c).substring(a.length, a.length + b.length) == b
        }
    }
})

on running, it correctly catches the issue with concatenate and produces the following result:

java.lang.AssertionError: Property failed for

Y{_DZ<vGnzLQHf9|3$i|UE,;!%8^SRF;JX%EH+<5d:p`Y7dxAd;I+J5LB/:O)

 at io.kotlintest.properties.PropertyTestingKt.forAll(PropertyTesting.kt:27)

However there is an issue here, scalacheck found a simpler failure case, it does this by a process called "Test Case minimization" where in case of a failure it tries to find the smallest test case that can fail, something that the Kotlintest can learn from.


There are other features where Kotlintest lags with respect to scalacheck, a big one being able to combine generators:

case class Person(name: String, age: Int)

val genPerson = for {
  name <- Gen.alphaStr
  age <- Gen.choose(1, 50)
} yield Person(name, age)

genPerson.sample

However all in all, I have found the DSL of Kotlintest and its support for property based testing to be a good start so far and look forward to how this library evolves over time.

If you want to play with these samples a little more, it is available in my github repo here - https://github.com/bijukunjummen/kotlintest-scalacheck-sample

Friday, July 14, 2017

Cloud Foundry Application manifest using Kotlin DSL

I had a blast working with and getting my head around the excellent support for creating DSL's in Kotlin Language.
Kotlin DSL is now being used for creating gradle build files, for defining routes in Spring Webflux, for creating html templates using kotlinx.html library.

Here I am going to demonstrate creating a kotlin based DSL to represent a Cloud Foundry Application Manifest content.

A sample manifest looks like this when represented as a yaml file:
applications:
 - name: myapp
   memory: 512M
   instances: 1
   path: target/someapp.jar
   routes:
     - somehost.com
     - antother.com/path
   envs:
    ENV_NAME1: VALUE1
    ENV_NAME2: VALUE2

And here is the kind of DSL I am aiming for:

cf {
    name = "myapp"
    memory = 512(M)
    instances = 1
    path = "target/someapp.jar"
    routes {
        +"somehost.com"
        +"another.com/path"
    }
    envs {
        env["ENV_NAME1"] = "VALUE1"
        env["ENV_NAME2"] = "VALUE2"
    }
}


Getting the basic structure


Let me start with a simpler structure that looks like this:


cf {
    name = "myapp"
    instances = 1
    path = "target/someapp.jar"
}

and want this kind of a DSL to map to a structure which looks like this:

data class CfManifest(
        var name: String = "",
        var instances: Int? = 0,
        var path: String? = null
)

It would translate to a Kotlin function which takes a Lambda expression:

fun cf(init: CfManifest.() -> Unit) {
 ...
}


The parameter which looks like this:
() -> Unit
is fairly self-explanatory, a lambda expression which does not take any parameters and does not return anything.

The part that took a while to seep into my mind is this modified lambda expression, referred to as a Lambda expression with receiver:

CfManifest.() -> Unit

It does two things the way I have understood it:

1. It defines in the scope of the wrapped function an extension function for the receiver type - in my case the CfManifest class
2. this within the lambda expression now refers to the receiver function.

Given this, the cf function translates to :

fun cf(init: CfManifest.() -> Unit): CfManifest {
    val manifest = CfManifest()
    manifest.init()
    return manifest
}

which can be succinctly expressed as:

fun cf(init: CfManifest.() -> Unit) = CfManifest().apply(init)

so now when I call:
cf {
    name = "myapp"
    instances = 1
    path = "target/someapp.jar"
}

It translates to:
CFManifest().apply {
  this.name = &quot;myapp&quot;
  this.instances = 1
  this.path = &quot;target/someapp.jar&quot;
}

More DSL

Expanding on the basic structure:

cf {
    name = "myapp"
    memory = 512(M)
    instances = 1
    path = "target/someapp.jar"
    routes {
        +"somehost.com"
        +"another.com/path"
    }
    envs {
        env["ENV_NAME1"] = "VALUE1"
        env["ENV_NAME2"] = "VALUE2"
    }
}

The routes and the envs in turn become methods on the CfManifest class and look like this:

data class CfManifest(
        var name: String = "",
        var path: String? = null,
        var memory: MEM? = null,
        ...
        var routes: ROUTES? = null,
        var envs: ENVS = ENVS()
) {

    fun envs(block: ENVS.() -> Unit) {
        this.envs = ENVS().apply(block)
    }

    ...

    fun routes(block: ROUTES.() -> Unit) {
        this.routes = ROUTES().apply(block)
    }
}

data class ENVS(
        var env: MutableMap<String, String> = mutableMapOf()
)

data class ROUTES(
        private val routes: MutableList<String> = mutableListOf()
) {
    operator fun String.unaryPlus() {
        routes.add(this)
    }
}

See how the routes method takes in a Lambda expression with a receiver type of ROUTES, this allows me to define an expression like this:

cf {
    ...
    routes {
        +"somehost.com"
        +"another.com/path"
    }
    ...
}

Another trick here is way a route is being added is using :

+"somehost.com"

which is enabled using a Kotlin convention which translates specific method names to operators, here the unaryPlus method. The cool thing for me is that this operator is visible only in the scope of ROUTES instance!


Another feature of the DSL making use of Kotlin features is the way a memory is specified, there are two parts to it - a number and the modifier, 2G, 500M etc.
This is being specified in a slightly modified way via the DSL as 2(G) and 500(M).

The way it is implemented is using another Kotlin convention where if a class has an invoke method then instances can call it the following way:

class ClassWithInvoke() {
    operator fun invoke(n: Int): String = "" + n
}
val c = ClassWithInvoke()
c(10)

So implementing invoke method as an extension function on Int in the scope of the CFManifest class allows this kind of a DSL:

data class CfManifest(
        var name: String = "",
        ...
) {
    ...
    operator fun Int.invoke(m: MemModifier): MEM = MEM(this, m)
}


This is pure experimentation on my part, I am both new to Kotlin as well as Kotlin DSL's so very likely there are a lot of things that can be improved in this implementation, any feedback and suggestions are welcome. You can play with this sample code at my github repo here

Tuesday, June 27, 2017

Spring Webflux - Kotlin DSL

Spring Webflux has introduced a feature for defining functional application endpoints using a very intuitive Kotlin based DSL

This post will be to simply show a contrasting api defined using a Java based fluent api and a Kotlin based DSL


A functional way to define a CRUD based Spring Webflux endpoint in Java would look like this:


RouterFunction<?> apis() {
    return nest(path("/hotels"), nest(accept(MediaType.APPLICATION_JSON),
            route(
                    GET("/"), messageHandler::getMessages)
                    .andRoute(POST("/"), messageHandler::addMessage)
                    .andRoute(GET("/{id}"), messageHandler::getMessage)
                    .andRoute(PUT("/{id}"), messageHandler::updateMessage)
                    .andRoute(DELETE("/{id}"), messageHandler::deleteMessage)
    ));
}

The details of the endpoint is very clear and is defined in a fluent manner with just a few keywords - route, nest and the HTTP verbs.

These endpoints can be expressed using a Kotlin based DSL(and some clever use of Kotlin extension functions) the following way:

@Bean
fun apis() = router {
    (accept(APPLICATION_JSON) and "/messages").nest {
        GET("/", messageHandler::getMessages)
        POST("/", messageHandler::addMessage)
        GET("/{id}", messageHandler::getMessage)
        PUT("/{id}", messageHandler::updateMessage)
        DELETE("/{id}", messageHandler::deleteMessage)
    }
}

I feels that this reads a little better than the Java based DSL. If the API is more complicated, as demonstrated in the excellent samples by S├ębastien Deleuze with multiple levels of nesting, the Kotlin based DSL really starts to shine.


In the next post, I will delve into how this support has been implemented.

This sample is available in my github repo here

Sunday, June 11, 2017

Spring Boot Web Slice test - Sample

Spring Boot introduced test slicing a while back and it has taken me some time to get my head around it and explore some of its nuances.

Background


The main reason to use this feature is to reduce the boilerplate. Consider a controller that looks like this, just for variety written using Kotlin.

@RestController
@RequestMapping("/users")
class UserController(
        private val userRepository: UserRepository,
        private val userResourceAssembler: UserResourceAssembler) {

    @GetMapping
    fun getUsers(pageable: Pageable, 
                 pagedResourcesAssembler: PagedResourcesAssembler<User>): PagedResources<Resource<User>> {
        val users = userRepository.findAll(pageable)
        return pagedResourcesAssembler.toResource(users, this.userResourceAssembler)
    }

    @GetMapping("/{id}")
    fun getUser(id: Long): Resource<User> {
        return Resource(userRepository.findOne(id))
    }
}


A traditional Spring Mock MVC test to test this controller would be along these lines:

@RunWith(SpringRunner::class)
@WebAppConfiguration
@ContextConfiguration
class UserControllerTests {

    lateinit var mockMvc: MockMvc

    @Autowired
    private val wac: WebApplicationContext? = null

    @Before
    fun setup() {
        this.mockMvc = MockMvcBuilders.webAppContextSetup(this.wac).build()
    }

    @Test
    fun testGetUsers() {
        this.mockMvc.perform(get("/users")
                .accept(MediaType.APPLICATION_JSON))
                .andDo(print())
                .andExpect(status().isOk)
    }

    @EnableSpringDataWebSupport
    @EnableWebMvc
    @Configuration
    class SpringConfig {

        @Bean
        fun userController(): UserController {
            return UserController(userRepository(), UserResourceAssembler())
        }

        @Bean
        fun userRepository(): UserRepository {
            val userRepository = Mockito.mock(UserRepository::class.java)
            given(userRepository.findAll(Matchers.any(Pageable::class.java)))
                    .willAnswer({ invocation ->
                        val pageable = invocation.arguments[0] as Pageable
                        PageImpl(
                                listOf(
                                        User(id = 1, fullName = "one", password = "one", email = "one@one.com"),
                                        User(id = 2, fullName = "two", password = "two", email = "two@two.com"))
                                , pageable, 10)
                    })
            return userRepository
        }
    }
}

There is a lot of ceremony involved in setting up such a test - a web application context which understands a web environment is pulled in, a configuration which sets up the Spring MVC environment needs to be created and MockMvc which is handle to the testing framework needs to be set-up before each test.


Web Slice Test

A web slice test when compared to the previous test is far simpler and focuses on testing the controller and hides a lot of the boilerplate code:

@RunWith(SpringRunner::class)
@WebMvcTest(UserController::class)
class UserControllerSliceTests {

    @Autowired
    lateinit var mockMvc: MockMvc

    @MockBean
    lateinit var userRepository: UserRepository

    @SpyBean
    lateinit var userResourceAssembler: UserResourceAssembler

    @Test
    fun testGetUsers() {

        this.mockMvc.perform(get("/users").param("page", "0").param("size", "1")
                .accept(MediaType.APPLICATION_JSON))
                .andDo(print())
                .andExpect(status().isOk)
    }

    @Before
    fun setUp(): Unit {
        given(userRepository.findAll(Matchers.any(Pageable::class.java)))
                .willAnswer({ invocation ->
                    val pageable = invocation.arguments[0] as Pageable
                    PageImpl(
                            listOf(
                                    User(id = 1, fullName = "one", password = "one", email = "one@one.com"),
                                    User(id = 2, fullName = "two", password = "two", email = "two@two.com"))
                            , pageable, 10)
                })
    }
}

It works by creating a Spring Application context but filtering out anything that is not relevant to the web layer and loading up only the controller which has been passed into the @WebTest annotation. Any dependency that the controller requires can be injected in as a mock.


Coming to some of the nuances, say if I wanted to inject one of the fields myself the way to do it is have the test use a custom Spring Configuration, for a test this is done by using a inner static class annotated with @TestConfiguration the following way:

@RunWith(SpringRunner::class)
@WebMvcTest(UserController::class)
class UserControllerSliceTests {

    @Autowired
    lateinit var mockMvc: MockMvc

    @Autowired
    lateinit var userRepository: UserRepository

    @Autowired
    lateinit var userResourceAssembler: UserResourceAssembler

    @Test
    fun testGetUsers() {

        this.mockMvc.perform(get("/users").param("page", "0").param("size", "1")
                .accept(MediaType.APPLICATION_JSON))
                .andDo(print())
                .andExpect(status().isOk)
    }

    @Before
    fun setUp(): Unit {
        given(userRepository.findAll(Matchers.any(Pageable::class.java)))
                .willAnswer({ invocation ->
                    val pageable = invocation.arguments[0] as Pageable
                    PageImpl(
                            listOf(
                                    User(id = 1, fullName = "one", password = "one", email = "one@one.com"),
                                    User(id = 2, fullName = "two", password = "two", email = "two@two.com"))
                            , pageable, 10)
                })
    }

    @TestConfiguration
    class SpringConfig {

        @Bean
        fun userResourceAssembler(): UserResourceAssembler {
            return UserResourceAssembler()
        }

        @Bean
        fun userRepository(): UserRepository {
            return mock(UserRepository::class.java)
        }
    }

}


The beans from the "TestConfiguration" adds on to the configuration which the Slice tests depend on and don't completely replace it.

On the other hand, if I wanted to override the loading of the main "@SpringBootApplication" annotated class then I can pass in a Spring Configuration class explicitly, but the catch is that I have to now take care of all of loading up the relevant Spring Boot features myself (enabling auto-configuration, appropriate scanning etc), so a way around it to explicitly annotate the configuration as a Spring Boot Application the following way:

@RunWith(SpringRunner::class)
@WebMvcTest(UserController::class)
class UserControllerExplicitConfigTests {

    @Autowired
    lateinit var mockMvc: MockMvc

    @Autowired
    lateinit var userRepository: UserRepository

    @Test
    fun testGetUsers() {

        this.mockMvc.perform(get("/users").param("page", "0").param("size", "1")
                .accept(MediaType.APPLICATION_JSON))
                .andDo(print())
                .andExpect(status().isOk)
    }

    @Before
    fun setUp(): Unit {
        given(userRepository.findAll(Matchers.any(Pageable::class.java)))
                .willAnswer({ invocation ->
                    val pageable = invocation.arguments[0] as Pageable
                    PageImpl(
                            listOf(
                                    User(id = 1, fullName = "one", password = "one", email = "one@one.com"),
                                    User(id = 2, fullName = "two", password = "two", email = "two@two.com"))
                            , pageable, 10)
                })
    }

    @SpringBootApplication(scanBasePackageClasses = arrayOf(UserController::class))
    @EnableSpringDataWebSupport
    class SpringConfig {

        @Bean
        fun userResourceAssembler(): UserResourceAssembler {
            return UserResourceAssembler()
        }

        @Bean
        fun userRepository(): UserRepository {
            return mock(UserRepository::class.java)
        }
    }

}


The catch though is that now other tests may end up finding this inner configuration which is far from ideal!, so my learning has been to depend on bare minimum slice testing, and if needed extend it using @TestConfiguration.


I have a little more detailed code sample available at my github repo which has working examples to play with.

Tuesday, May 30, 2017

Ratio based routing to a legacy and a modern app - Netflix Zuul via Spring Cloud

A very common requirement when migrating from a legacy version of an application to a modernized version of the application is to be able to migrate the users slowly over to the new application. In this post I will be going over this kind of a routing layer written using support for Netflix Zuul through Spring Cloud . Before I go ahead I have to acknowledge that most of the code demonstrated here has been written in collaboration with the superlative Shaozhen Ding


Scenario

I have a legacy service which has been re-engineered to a more modern version(assumption is that as part of this migration the uri's of the endpoints have not changed). I want to migrate users slowly over from the legacy application over to the modern version.


Implementation using Spring Cloud Netflix - Zuul Support


This can be easily implemented using Netflix Zuul support in Spring Cloud project.

Zuul is driven by a set of filters which act on a request before(pre filters), during(route filters) and after(post filters) a request to a backend. Spring Cloud adds it custom set of filters to Zuul and drives the behavior of these filters by configuration that looks like this:

zuul:
  routes:
    ratio-route:
      path: /routes/**
      strip-prefix: false

This specifies that Zuul will be handling a request to Uri with prefix "/routes" and this prefix will not be stripped from the downstream call. This logic is encoded into a "PreDecorationFilter". My objective is to act on the request AFTER the PreDecorationFilter and specify the backend to be either the legacy version or the modern version. Given this a filter which acts on the request looks like this:

import com.netflix.zuul.ZuulFilter;
import com.netflix.zuul.context.RequestContext;
...

@Service
public class RatioBasedRoutingZuulFilter extends ZuulFilter {

    public static final String LEGACY_APP = "legacy";
    public static final String MODERN_APP = "modern";
    
    private Random random = new Random();
    
    @Autowired
    private RatioRoutingProperties ratioRoutingProperties;

    @Override
    public String filterType() {
        return "pre";
    }

    @Override
    public int filterOrder() {
        return FilterConstants.PRE_DECORATION_FILTER_ORDER + 1;
    }

    @Override
    public boolean shouldFilter() {
        RequestContext ctx = RequestContext.getCurrentContext();
        return ctx.containsKey(SERVICE_ID_KEY)
                && ctx.get(SERVICE_ID_KEY).equals("ratio-route");
    }

    @Override
    public Object run() {
        RequestContext ctx = RequestContext.getCurrentContext();

        if (isTargetedToLegacy()) {
            ctx.put(SERVICE_ID_KEY, LEGACY_APP);
        } else {
            ctx.put(SERVICE_ID_KEY, MODERN_APP);
        }
        return null;
    }

    boolean isTargetedToLegacy() {
        return random.nextInt(100) < ratioRoutingProperties.getOldPercent();
    }
}

The filter is set to act after the "PreDecorationFilter" by overriding the filterOrder() method. The routing logic is fairly naive but should work for most cases. The serviceId being set in the Zuul context has a value of "legacy" or "modern" and represents a "named" Ribbon client, a handle using which the details of the backend can be set. So with Spring Cloud, my named clients are mapped to the legacy and modern versions of the app the following way:


legacy:
  ribbon:
    listOfServers: http://localhost:8081

modern:
  ribbon:
    DeploymentContextBasedVipAddresses: modern-app

Here just for a little more variation I am making a direct call to an endpoint for the legacy app and making a call via Eureka for the modern version of the application.


If you are interested in exploring the entirety of the application it is available in my github repo


With the entire set-up in place, a small test with the legacy handling 20% of the traffic confirms that the filter works effectively:

Conclusion

Spring Cloud support for Netflix Zuul makes handling such routing scenarios a cinch and should be a good fit for any organization having these kinds of routing scenarios that they may want to implement.

Friday, May 19, 2017

Cloud Foundry Custom User Provided Services(CUPS) and tagging

Custom User Provided Services or CUPS for short is a way to deliver credentials for external services to an application hosted on Cloud Foundry.

Consider a set of credentials represented as a json of the following form:

{
 "hostname": "mysql-broker.local.pcfdev.io",
 "jdbcUrl": "jdbc:mysql://mysql-broker.local.pcfdev.io:3306/somedb?user=someuser\u0026password=somepass",
 "name": "somedb",
 "password": "somepass",
 "port": 3306,
 "uri": "mysql://someuser:somepass@mysql-broker.local.pcfdev.io:3306/somedb?reconnect=true",
 "username": "someuser"
}

I could create a user provided service out of these values using cf-cli. The following is highly bash shell specific, so on a different shell the mileage is likely to vary:

CUPS_PARAM=$(cat <<-'EOF'
{
 "hostname": "mysql-broker.local.pcfdev.io",
 "jdbcUrl": "jdbc:mysql://mysql-broker.local.pcfdev.io:3306/somedb?user=someuser\u0026password=somepass",
 "name": "somedb",
 "password": "somepass",
 "port": 3306,
 "uri": "mysql://someuser:somepass@mysql-broker.local.pcfdev.io:3306/somedb?reconnect=true",
 "username": "someuser"
}
EOF
)

cf create-user-provided-service mycups -p ''"$CUPS_PARAM"''

This Custom User provided service can be bound to an app:

cf bind-service myapp mycups

and the application can retrieve the credentials via an environment variable called VCAP_SERVICES at runtime.


Issue

There is one small issue with the Custom User provided services over normal services created via Service Brokers on Cloud Foundry - there is no simple way to tag a Custom User Provided service. Tags are sometimes useful in getting a little more information about the service bound to an app and is extensively used by Spring Cloud Connectors to connect to services.


Solution

I have written a custom service broker called the CUPS tagging broker using which a service can be created with all the parameters normally passed to create the CUPS, additionally since it is a normal service it can be tagged.

Assuming that the "CUPS tagging broker" has been installed using the instructions here, an equivalent user provided service with tags can be created the following way, with two tags attached to the service:

cf create-service cups-tagging-service default my-cups-tagged -c ''"$CUPS_PARAM"'' -t "tag1, tag2"

If I were to bind this service to an app, the VCAP_SERVICES environment variable of the app would be along these lines:

{"cups-tagging-service":[{
  "credentials": {
    "hostname": "mysql-broker.local.pcfdev.io",
    "jdbcUrl": "jdbc:mysql://mysql-broker.local.pcfdev.io:3306/somedb?user=someuser&password=somepass",
    "name": "somedb",
    "password": "somepass",
    "port": 3306,
    "uri": "mysql://someuser:somepass@mysql-broker.local.pcfdev.io:3306/somedb?reconnect=true",
    "username": "someuser"
  },
  "syslog_drain_url": null,
  "volume_mounts": [

  ],
  "label": "cups-tagging-service",
  "provider": null,
  "plan": "default",
  "name": "my-cups-tagged",
  "tags": [
    "cups-tag",
    "tag1",
    "tag2"
  ]
}]}

See how the two additional tags show up.

That is all there is to the CUPS tagging Service Broker!





Thursday, May 4, 2017

Integrating Gatling into a Gradle build - Understanding SourceSets and Configuration

I recently worked on a project where we had to integrate the excellent load testing tool Gatling into a Gradle based build. There are gradle plugins available which make this easy, two of them being this and this, however for most of the needs a simple execution of the command line tool itself suffices, so this post will go into some details of how gatling can be hooked up into a gradle build and in the process understand some good gradle concepts.


SourceSets and Configuration


To execute the gatling cli I need to do two things, I need a location for the source code and related content of the Gatling simulations, and I need a way to get the gatling libraries. This is where two concepts of Gradle(SourceSets and Configuration) come into play.

Let us start with the first one - SourceSets.

SourceSets


SourceSets are simply a logical grouping of related files and are best demonstrated with an example. If I were to add a "java" plugin to a gradle build:

apply plugin: 'java'


sourceSets property would now show up with two values "main" and "test" and if I wanted to find details of the these sourceSets, a gradle task can be used for printing the details:

task sourceSetDetails {
    doLast {
        sourceSets {
            main {
                println java.properties
                println resources.properties
            }
        
            test {
                println java.properties
                println resources.properties
            }
        }
    }
}

Coming back to gatling, I can essentially create a new sourceSet to hold the gatling simulations:

sourceSets {
    simulations
}

This would now expect the gatling simulations to reside in "src/simulations/java" and the resources related to it in "src/simulations/resources" folders, which is okay, but ideally I would want to keep it totally separate from the project sources. I would want my folder structure to be with load simulations in "simulations/load" and resources in "simulations/resources" folder. This can be tweaked by first applying the "scala" plugin, which brings in scala compilation support to the project and then modifying the "simulations" source set along these lines:

apply plugin: 'scala'

sourceSets {
    simulations {
        scala {
            srcDirs = ['simulations/load']
        }
        resources {
            srcDirs = ['simulations/resources']
        }
    }
}

With these set of changes, I can now put my simulations in the right place, but the dependency of gatling and scala has not been pulled in, this is where the "configuration" feature of gradle comes in.

Configuration


Gradle Configuration is a way of grouping related dependencies together. If I were to print the existing set of configurations using a task:

task showConfigurations  {
    doLast {
        configurations.all { conf -> println(conf) }
    }
}

these show up:

configuration ':archives'
configuration ':compile'
configuration ':compileClasspath'
configuration ':compileOnly'
configuration ':default'
configuration ':runtime'
configuration ':simulationsCompile'
configuration ':simulationsCompileClasspath'
configuration ':simulationsCompileOnly'
configuration ':simulationsRuntime'
configuration ':testCompile'
configuration ':testCompileClasspath'
configuration ':testCompileOnly'
configuration ':testRuntime'
configuration ':zinc'

"compile" and "testCompile" should be familiar one's, that is where a normal source dependency and a test dependency is typically declared like this:

dependencies {
    compile 'org.slf4j:slf4j-api:1.7.21'
    testCompile 'junit:junit:4.12'   
}

However, it also looks like there is now configuration for "simulations" sourceset also available - "simulationsCompile" and "simulationsRuntime" etc, so with this I can declare the dependencies required for my gatling simulations using these configurations, however my intention is to declare a custom configuration just to go over the concept a little more, so let us explicitly declare one:

configurations {
    gatling
}

and use this configuration for declaring the dependencies of gatling:
dependencies {
    gatling 'org.scala-lang:scala-library:2.11.8'
    gatling 'io.gatling.highcharts:gatling-charts-highcharts:2.2.5'
}

Almost there, now how do we tell the sources in simulations source set to use dependency from gatling configuration..by tweaking the sourceSet a little:

sourceSets {
    simulations {
        scala {
            srcDirs = ['simulations/load']
        }
        resources {
            srcDirs = ['simulations/resources']
        }

        compileClasspath += configurations.gatling
    }
}


Running a Gatling Scenario

With the source sets and the configuration defined, all we need to do is to write a task to run a gatling simulation, which can be along these lines:

task gatlingRun(type: JavaExec) {
    description = 'Run gatling tests'
    new File("${buildDir}/reports/gatling").mkdirs()

    classpath = sourceSets.simulations.runtimeClasspath + configurations.gatling

    main = "io.gatling.app.Gatling"
    args = ['-s', 'simulations.SimpleSimulation',
            '-sf', 'simulations/resources',
            '-df', 'simulations/resources',
            '-rf', "${buildDir}/reports/gatling"
    ]
}

See how the compiled sources of the simulations and the dependencies from the gatling configuration is being set as the classpath for the "JavaExec" task


A good way to review this would be to look at a complete working sample that I have here in my github repo - https://github.com/bijukunjummen/cf-show-env