apidays LIVE Australia 2020 - Strangling the monolith with a reactive GraphQL gateway by Martin Varga
0 likes105 views
The document outlines the implementation of a reactive GraphQL gateway aimed at modernizing software development practices, particularly through the migration from monoliths to microservices. It discusses the use of technologies like Kotlin, Spring Boot, and Project Reactor, demonstrating the architecture for a sample application involving animal facts APIs. Key takeaways emphasize adopting GraphQL for APIs, building resilience to prevent error propagation, and the benefits of reactive programming for I/O heavy applications.
![type Query {
board(id: ID): Board
}
type Board {
id: ID
name: String
issues: [Issue]
}
type Issue {
id: ID
key: String
}
SCHEMA](https://image.slidesharecdn.com/apidaysaustraliaonline2020martinvarga-200919014514/85/apidays-LIVE-Australia-2020-Strangling-the-monolith-with-a-reactive-GraphQL-gateway-by-Martin-Varga-45-320.jpg)
![{
"data": {
"board": {
"name": "Software board”
"issues": [
{
"key": "KEY-1"
}
]
}
}
}
RESPONSE](https://image.slidesharecdn.com/apidaysaustraliaonline2020martinvarga-200919014514/85/apidays-LIVE-Australia-2020-Strangling-the-monolith-with-a-reactive-GraphQL-gateway-by-Martin-Varga-47-320.jpg)
![@PostMapping(
path = ["/graphql"],
consumes = [MediaType.APPLICATION_JSON_VALUE],
produces = [MediaType.APPLICATION_JSON_VALUE]
)
fun graphql(
@RequestBody body: Map<String, Any>
): Mono<Map<String, Any>> {
val query: String = body["query"] as String? ?: "{}"
return graphQLService.processQuery(query)
.map { it.toSpecification() }
}
REACTIVE API](https://image.slidesharecdn.com/apidaysaustraliaonline2020martinvarga-200919014514/85/apidays-LIVE-Australia-2020-Strangling-the-monolith-with-a-reactive-GraphQL-gateway-by-Martin-Varga-66-320.jpg)
![{
"data": {
"dog": {…},
"cat": null
},
"errors": [
{
"message": "Exception while fetching data (/cat) :
Operation timed out",
"path": [
"cat"
]
}
]
}](https://image.slidesharecdn.com/apidaysaustraliaonline2020martinvarga-200919014514/85/apidays-LIVE-Australia-2020-Strangling-the-monolith-with-a-reactive-GraphQL-gateway-by-Martin-Varga-71-320.jpg)
![{
"data": {
"dog": {…},
"cat": null
},
"errors": [
{
"message": "Exception while fetching data (/cat) :
Did not observe any item or terminal signal within 5000ms
in 'map' (and no fallback has been configured)",
"path": [
“cat"
]
}
]
}](https://image.slidesharecdn.com/apidaysaustraliaonline2020martinvarga-200919014514/85/apidays-LIVE-Australia-2020-Strangling-the-monolith-with-a-reactive-GraphQL-gateway-by-Martin-Varga-74-320.jpg)
![{
"data": {
"dog": {…},
"cat": null
},
"errors": [
{
"message": "Exception while fetching data (/cat) :
CircuitBreaker 'CatCCB' is OPEN and does not permit
further calls",
"path": [
"cat"
]
}
]
}](https://image.slidesharecdn.com/apidaysaustraliaonline2020martinvarga-200919014514/85/apidays-LIVE-Australia-2020-Strangling-the-monolith-with-a-reactive-GraphQL-gateway-by-Martin-Varga-79-320.jpg)
apidays LIVE Australia 2020 - Strangling the monolith with a reactive GraphQL gateway by Martin Varga
- 2. MARTIN VARGA | SOFTWARE DEVELOPER | @MARTINTEEVARGA Strangling the monolith with a reactive GraphQL gateway
- 3. STRANGLING THE MONOLITH WITH A REACTIVE GRAPHQL GATEWAY
- 4. MONOLITH?
- 5. Plan, track and release software. JIRA SOFTWARE
- 7. The Story of Jira From the developer’s perspective
- 8. Server Cloud 1001001 0 0 1 0 0 1 0 0 1 0 0 1 1101010 1010101 0 0 1 0 0 1 1110001001 0011110010 1100101011 01111110001 Single Code Base
- 9. Server Cloud 1001001 0 0 1 0 0 1 0 0 1 0 0 1 1101010 1010101 0 0 1 0 0 1
- 10. 1101010 1010101 0 0 1 0 0 1 1101010 1010101 0 0 1 0 0 1
- 11. 1101010 1010101 0 0 1 0 0 1
- 18. 1101010101010100 1001010100100101 0101110101010101 010010010101001 0010101011101010 1010101001001010 1001001010101110 1 0 1 0 1 0 1010111 001001
- 19. 1101010101 010100100 101010010 0101010111 0 1 0 1 0 1 1011100 1 0 0 1 0 1 1 0 1 0 1 0 1010111 001001
- 20. 1101010 1010101 001001 0 1 0 1 0 1 1011100 1 0 0 1 0 1 1 0 1 0 1 0 1010111 001001 0 1 0 1 0 1 1011100 1 0 0 1 0 1
- 21. 1101010 1010101 001001 0 1 0 1 0 1 1011100 1 0 0 1 0 1 1 0 1 0 1 0 1010111 001001 0 1 0 1 0 1 1011100 1 0 0 1 0 1
- 23. How Atlassian builds services Making developers productive with platform as a service
- 24. PaaS that allows devs to create and deploy a service in matter of minutes. MICROS
- 25. MICROS
- 26. name: "animalfacts" compose: animalfacts: image: internal.docker.repo/animalfacts tag: 'PROJECT_VERSION' ports: - 8080:8080 MICROS DESCRIPTOR
- 27. links: healthcheck: uri: heartbeat port: 8080 source: url: "git@bitbucket.org:atlassianlabs/animal-facts- graphql-gateway-demo.git" notifications: email: "animalfactsowner@atlassian.com" resources: - type: dynamo-db name: facts attributes: ReadWriteCapacityMode: ON_DEMAND
- 28. Jira Software Tech Stack DynamoDB or PostgreSQL Depends on the application. Project Reactor Mature reactive library. Kotlin For all new services. Spring Boot Custom wrapper that makes monitoring etc. easier.
- 29. A case for a reactive gateway A real world example
- 34. Strangler Isolate front-end and back-end changes, dark traffic, comparisons. Minimum business logic Keep it light, no-code configuration could be enough. Fully stateless Scales (almost) indefinitely.
- 35. Operators Useful for modifying and combining sources. Allows composition of libraries that add resilience, tracing etc. Async Calls a service, then does mostly nothing and waits. Perfect fit. Project reactor Clean code No callback hell.
- 36. val currentBoard = “5”.toMono() MONO 5
- 37. val recentBoards = listOf("0", "1", "2").toFlux() FLUX 0 1 2
- 38. val currentBoard = "5".toMono() val recentBoards = listOf("0", "1", "2").toFlux() currentBoard.concatWith(recentBoards) COMPOSITION & CLEAN CODE 5 0 1 2 0 1 25
- 39. val currentBoard = "5".toMono() val recentBoards = listOf("0", "1", "2").toFlux() currentBoard.concatWith(recentBoards) .map { boardService.getName(it) } .take(3) .subscribe { println(it) } COMPOSITION & CLEAN CODE
- 40. Surfboard Skateboard Ironing board COMPOSITION & CLEAN CODE
- 41. currentBoard.concatWith(recentBoards) .map { boardService.getName(it) } .timeout(Duration.ofMillis(2500)) .onErrorReturn("Recent board”) .take(3) .subscribe { println(it) } COMPOSITION & CLEAN CODE
- 42. Surfboard Recent board Ironing board COMPOSITION & CLEAN CODE
- 43. val currentBoard = currentBoardServiceMono() val recentBoards = recentBoardsServiceFlux() return currentBoard.concatWith(recentBoards) .map { boardService.getName(it) } .timeout(Duration.ofMillis(2500)) .onErrorReturn("Recent board”) .take(3) REAL WORLD EXAMPLE
- 44. Get exactly what you want Query multiple sources at once, specify what you want. Strongly typed schema Schema is a living documentation. Mature web clients Clients like Apollo let us remove dependency on Redux in ReactJs web apps. GraphQL
- 45. type Query { board(id: ID): Board } type Board { id: ID name: String issues: [Issue] } type Issue { id: ID key: String } SCHEMA
- 46. POST /jsw/graphql query { board(id: 42) { name issues { key } } } QUERY
- 47. { "data": { "board": { "name": "Software board” "issues": [ { "key": "KEY-1" } ] } } } RESPONSE
- 48. Project Reactor + GraphQL Java OUR SOLUTION
- 49. Let’s write some code Fully working GraphQL gateway
- 50. GET https://catfact.ninja/fact { "fact": "The leopard is the most widespread of all big cats.", "length": 51 } CAT FACTS API
- 51. GET https://some-random-api.ml/facts/dog { "fact": "A dogs' first sense to develop is touch." } DOG FACTS API
- 53. WHY NOT BOTH?
- 54. POST https://animalfactsgateway.com/grapqhl query { dog { fact length } cat { fact length } } ANIMAL FACTS GRAPHQL API
- 55. { "data": { "dog": { "fact": "Dog fact", "length": 8 }, "cat": { "fact": "Cat fact", "length": 8 } } }
- 57. package com.atlassian.jsw.factsdemo import o.s.b.autoconfigure.SpringBootApplication import o.s.b.runApplication @SpringBootApplication class DemoApplication fun main(args: Array<String>) { runApplication<DemoApplication>(*args) } APPLICATION
- 58. type Query { cat: Fact dog: Fact } type Fact { fact: String length: Int } GRAPHQL SCHEMA
- 59. interface FactFetcher: DataFetcher<CompletionStage<Fact?>> FETCHERS public interface DataFetcher<T> { T get(DataFetchingEnvironment env) throws Exception; }
- 60. data class Fact(val fact: String, val length: Int) FACT CLASS
- 61. @Component class CatFactFetcher(val webClient: WebClient) : FactFetcher { override fun get(environment: DataFetchingEnvironment): CompletionStage<Fact?> = webClient.get().uri("https://catfact.ninja/fact") .retrieve() .bodyToMono(CatFact::class.java) .map { Fact(it.fact, it.length) } .toFuture() } data class CatFact(val fact: String, val length: Int) CAT FACTS FETCHER
- 62. @Component class DogFactFetcher(val webClient: WebClient) : FactFetcher { override fun get(environment: DataFetchingEnvironment): CompletionStage<Fact?> = webClient.get().uri("https://.../facts/dog") .retrieve() .bodyToMono(DogFact::class.java) .map { Fact(it.fact, it.fact.length) } .toFuture() } data class DogFact(val fact: String) DOG FACTS FETCHER
- 63. @Configuration class GraphqlConfig { @Bean fun schema(wiring: RuntimeWiring): GraphQLSchema = SchemaGenerator().makeExecutableSchema( parseSchema(“/graphql/schema.graphqls"), wiring ) } GRAPHQL CONFIGURATION
- 64. @Configuration class GraphqlConfig { @Bean fun factsWiring( dogFactFetcher: DogFactFetcher, catFactFetcher: CatFactFetcher ): RuntimeWiring = RuntimeWiring.newRuntimeWiring() .type( newTypeWiring("Query") .dataFetcher("dog", dogFactFetcher) .dataFetcher("cat", catFactFetcher) ) .build() }
- 65. override fun processQuery( query: String ): Mono<ExecutionResult> = Mono.fromFuture( GraphQL .newGraphQL(graphQLSchema) .build() .executeAsync( ExecutionInput.newExecutionInput() .query(query) .build() ) ) SERVICE METHOD
- 66. @PostMapping( path = ["/graphql"], consumes = [MediaType.APPLICATION_JSON_VALUE], produces = [MediaType.APPLICATION_JSON_VALUE] ) fun graphql( @RequestBody body: Map<String, Any> ): Mono<Map<String, Any>> { val query: String = body["query"] as String? ?: "{}" return graphQLService.processQuery(query) .map { it.toSpecification() } } REACTIVE API
- 67. { "data": { "dog": { "fact": "“Him” and “Her” were the names of President Lyndon Johnson's two beagles.", "length": 73 }, "cat": { "fact": "Jaguars are the only big cats that don't roar.", "length": 46 } } } RESULT
- 68. Resilience Protect the downstream services from the upstream.
- 70. CAT FACTS BROKEN 🐈🚫🐕250ms 30s 30s
- 71. { "data": { "dog": {…}, "cat": null }, "errors": [ { "message": "Exception while fetching data (/cat) : Operation timed out", "path": [ "cat" ] } ] }
- 73. class CatFactFetcher( val webClient: WebClient ) : FactFetcher { override fun get(environment: DataFetchingEnvironment): CompletionStage<Fact?> = webClient.get().uri("https://catfact.ninja/fact") .retrieve() .bodyToMono(CatFact::class.java) .map { … } .timeout(Duration.ofMillis(5000)) .toFuture() } CAT FACTS WITH TIMEOUT
- 74. { "data": { "dog": {…}, "cat": null }, "errors": [ { "message": "Exception while fetching data (/cat) : Did not observe any item or terminal signal within 5000ms in 'map' (and no fallback has been configured)", "path": [ “cat" ] } ] }
- 77. class CatFactFetcher( val webClient: WebClient, val catBreaker: CircuitBreaker ) : FactFetcher { override fun get(environment: DataFetchingEnvironment): CompletionStage<Fact?> = webClient.get() … .timeout(Duration.ofMillis(2500)) .transform( CircuitBreakerOperator.of(catBreaker) ) .toFuture() } CAT FACTS FETCHER WITH CIRCUIT BREAKER
- 78. @Bean fun catCircuitBreaker(): CircuitBreaker { val circuitBreakerConfig = CircuitBreakerConfig.custom() .ringBufferSizeInClosedState(3) .ringBufferSizeInHalfOpenState(3) .waitDurationInOpenState(Duration.ofMillis(5000)) .failureRateThreshold(0.5f) .build() return circuitBreakerRegistry.circuitBreaker( "CatCCB", circuitBreakerConfig ) }
- 79. { "data": { "dog": {…}, "cat": null }, "errors": [ { "message": "Exception while fetching data (/cat) : CircuitBreaker 'CatCCB' is OPEN and does not permit further calls", "path": [ "cat" ] } ] }
- 80. CIRCUIT BREAKER 🐈🚫🐕250ms 0ms 250ms But we might miss some cat facts.
- 81. Takeaways Even if you forget everything else I said, please try…
- 82. GraphQL For APIs consumed from the front-end. Resilience To prevent propagating errors downstream. Reactive For I/O heavy apps. Takeaways
- 83. Resources Write your own reactive and resilient GraphQL gateway in Kotlin
- 84. 1 2 3 Animal Facts GraphQL Gateway https://bitbucket.org/atlassianlabs/animal-facts-graphql-gateway-demo/src/ master/ graphql-java https://www.graphql-java.com/ Resilience4j https://github.com/resilience4j/resilience4j Resources / Code
- 86. 1 2 Engineering stateless, high-availability cloud services at Atlassian Atlassian Engineering Blog: https://www.atlassian.com/blog/technology/aws-scaling-multi-region-low-latency- service More about Micros Slightly outdated, but still relevant. https://www.youtube.com/watch?v=dc2nqzgqH24 Resources / Further information
- 87. 1 Why not both? https://knowyourmeme.com/memes/why-not-both-why-dont-we-have-both Resources / Other 2 We are hiring! Jira Software is looking for engineers, Java and Kotlin http://go.atlassian.com/jsw-hiring
- 88. MARTIN VARGA | SOFTWARE DEVELOPER | @MARTINTEEVARGA Strangling the monolith with a reactive GraphQL gateway