gRPC Kotlin Coroutines, Protobuf DSL, Scripting for Protoc
Community Contributions are Welcomed
ℹ️ | Docs are being expanded and moved to Readme.io
Quick Start: gRPC Coroutines
Run the following command to get started with a preconfigured template project. (kotlin-coroutines-gRPC-template)
git clone https://github.com/marcoferrer/kotlin-coroutines-gRPC-template && \
cd kotlin-coroutines-gRPC-template && \
./gradlew run
Getting Started
Code Generators
- There are several built in code generators that each accept unique configuration options.
- Configuration Setup
Proto Builder Generator (Message DSL)
Setup & Documentation
This generator creates lambda based builders for message types
val starPlatinum = Stand {
name = "Star Platinum"
powerLevel = 500
speed = 550
attack {
name = "ORA ORA ORA"
damage = 100
range = Attack.Range.CLOSE
}
}
val attack = Attack {
name = "ORA ORA ORA"
damage = 100
range = Attack.Range.CLOSE
}
// Copy extensions
val newAttack = attack.copy { damage = 200 }
// orDefault() will return the messages default instance when null
val nullableAttack: Attack? = null
nullableAttack.orDefault()
// Plus operator extensions
val mergedAttack = attack + Attack { name = "Sunlight Yellow Overdrive" }
gRPC Coroutines Client & Server 
This option requires the artifact kroto-plus-coroutines as a dependency.
Configuration Options
Client / Server Examples
Method Signature Option Support
- Design
- Back pressure is supported via Manual Flow Control
- Related Reading: Understanding Reactive gRPC Flow Control
- Cooperative cancellation across network boundaries.
- Back pressure is supported via Manual Flow Control
- Client Stubs
- Designed to work well with Structured Concurrency
- Cancellation of the client
CoroutineScopewill propagate to the server. - Cancellations can now be propagated across usages of a specific stub instance.
- Rpc methods are overloaded with inline builders for request types
- The request parameter for rpc methods defaults to
RequestType.defaultsInstance
// Creates new stub with a default coroutine context of `EmptyCoroutineContext`
val stub = GreeterCoroutineGrpc.newStub(channel)
// Suspends and creates new stub using the current coroutine context as the default.
val stub = GreeterCoroutineGrpc.newStubWithContext(channel)
// An existing stub can replace its current coroutine context using either
stub.withCoroutineContext()
stub.withCoroutineContext(coroutineContext)
// Stubs can accept message builder lambdas as an argument
stub.sayHello { name = "John" }
// For more idiomatic usage in coroutines, stubs can be created
// with an explicit coroutine scope using the `newGrpcStub` scope extension function.
launch {
// Using `newGrpcStub` makes it clear that the resulting stub will use the receiving
// coroutine scope to launch any concurrent work. (usually for manual flow control in streaming apis)
val stub = newGrpcStub(GreeterCoroutineGrpc.GreeterCoroutineStub, channel)
val (requestChannel, responseChannel) = stub.sayHelloStreaming()
...
}
- Service Base Impl
- Rpc calls are wrapped within a scope initialized with the following context elements.
CoroutineNameset toMethodDescriptor.fullMethodNameGrpcContextElementset toio.grpc.Context.current()
- Base services implement
ServiceScopeand allow overriding the initialcoroutineContextused for each rpc method invocation. - Each services
initialContextdefaults toEmptyCoroutineContext - A common case for overriding the
initialContextis for setting up application specificThreadContextElementorCoroutineDispatcher, such asMDCContext()ornewFixedThreadPoolContext(...)
- Rpc calls are wrapped within a scope initialized with the following context elements.
Cancellation Propagation
- Client
- Both normal and exceptional coroutine scope cancellation will cancel the underlying call stream. See
ClientCall.cancel()in io.grpc.ClientCall.java for more details. - In the case of service implementations using coroutines, this client call stream cancellation will cancel the coroutine scope of the rpc method being invoked on the server.
- Both normal and exceptional coroutine scope cancellation will cancel the underlying call stream. See
- Server
- Exceptional cancellation of the coroutine scope for the rpc method will be mapped to an instance of
StatusRuntimeExceptionand returned to the client. - Normal cancellation of the coroutine scope for the rpc method will be mapped to an instance of
StatusRuntimeExceptionwith a status ofStatus.CANCELLED, and returned to the client. - Cancellation signals from the corresponding client will cancel the coroutine scope of the rpc method being invoked.
- Exceptional cancellation of the coroutine scope for the rpc method will be mapped to an instance of
Examples
Unary
Client: Unary calls will suspend until a response is received from the corresponding server. In the event of a cancellation or the server responds with an error the call will throw the appropriate StatusRuntimeException
val response = stub.sayHello { name = "John" }
Server: Unary rpc methods can respond to client requests by either returning the expected response type, or throwing an exception.
override suspend fun sayHello(request: HelloRequest): HelloReply {
if (isValid(request.name))
return HelloReply { message = "Hello there, ${request.name}!" } else
throw Status.INVALID_ARGUMENT.asRuntimeException()
}
Client Streaming
Client: requestChannel.send() will suspend until the corresponding server signals it is ready by requesting a message. In the event of a cancellation or the server responds with an error, both requestChannel.send() and response.await(), will throw the appropriate StatusRuntimeException.
val (requestChannel, response) = stub.sayHelloClientStreaming()
launchProducerJob(requestChannel){
repeat(5){
send { name = "name #$it" }
}
}
println("Client Streaming Response: ${response.await()}")
Server: Client streaming rpc methods can respond to client requests by either returning the expected response type, or throwing an exception. Calls to requestChannel.receive() will suspend and notify the corresponding client that the server is ready to accept a message.
override suspend fun sayHelloClientStreaming(
requestChannel: ReceiveChannel<HelloRequest>
): HelloReply = HelloReply {
message = requestChannel.toList().joinToString()
}
Server Streaming
Client: responseChannel.receive() will suspend and notify the corresponding server that the client is ready to accept a message.
val responseChannel = stub.sayHelloServerStreaming { name = "John" }
responseChannel.consumeEach {
println("Server Streaming Response: $it")
}
Server: Server streaming rpc methods can respond to client requests by submitting messages of the expected response type to the response channel. Completion of service method implementations will automatically close response channels in order to prevent abandoned rpcs.
Calls to responseChannel.send() will suspend until the corresponding client signals it is ready by requesting a message. Error responses can be returned to clients by either throwing an exception or invoking close on responseChannel with the desired exception.
For an example of how to implement long lived response streams please reference MultipleClientSubscriptionsExample.kt.
override suspend fun sayHelloServerStreaming(
request: HelloRequest,
responseChannel: SendChannel<HelloReply>
) {
for(char in request.name) {
responseChannel.send {
message = "Hello $char!"
}
}
}
Bi-Directional Streaming
Client: requestChannel.send() will suspend until the corresponding server signals it is ready by requesting a message. In the event of a cancellation or the server responds with an error, both requestChannel.send() and response.await(), will throw the appropriate StatusRuntimeException.
val (requestChannel, responseChannel) = stub.sayHelloStreaming()
launchProducerJob(requestChannel){
repeat(5){
send { name = "person #$it" }
}
}
responseChannel.consumeEach {
println("Bidi Response: $it")
}
Server: Bidi streaming rpc methods can respond to client requests by submitting messages of the expected response type to the response channel. Completion of service method implementations will automatically close response channels in order to prevent abandoned rpcs.
Calls to responseChannel.send() will suspend until the corresponding client signals it is ready by requesting a message. Error responses can be returned to clients by either throwing an exception or invoking close on responseChannel with the desired exception.
For an example of how to implement long lived response streams please reference MultipleClientSubscriptionsExample.kt.
override suspend fun sayHelloStreaming(
requestChannel: ReceiveChannel<HelloRequest>,
responseChannel: SendChannel<HelloReply>
) {
requestChannel.mapTo(responseChannel){
HelloReply {
message = "Hello there, ${it.name}!"
}
}
}
gRPC Stub Extensions
Configuration Options
This modules generates convenience extensions that overload the request message argument for rpc methods with a builder lambda block and a default value. It also supports generating overloads based off (google.api.method_signature) method options. More info available here
//Kroto+ Generated Extension
val response = serviceStub.myRpcMethod {
id = 100
name = "some name"
}
//Original Java Fluent builders
val response = serviceStub.myRpcMethod(ExampleServiceGrpc.MyRpcMethodRequest
.newBuilder()
.setId(100)
.setName("some name")
.build())
For unary rpc methods, the generator will create the following extensions
//Future Stub with default argument
fun ServiceBlockingStub.myRpcMethod(request: Request = Request.defaultInstance): ListenableFuture<Response>
//Future Stub with builder lambda
inline fun ServiceFutureStub.myRpcMethod(block: Request.Builder.() -> Unit): ListenableFuture<Response>
//Blocking Stub with default argument
fun ServiceBlockingStub.myRpcMethod(request: Request = Request.defaultInstance): Response
//Blocking Stub with builder lambda
inline fun ServiceBlockingStub.myRpcMethod(block: Request.Builder.() -> Unit): Response
Coroutine Support
In addition to request message arguments as builder lambda rpc overloads, coroutine overloads for rpc calls can also be generated. This provides the same functionality as the generated coroutine stubs. Usage is identical to the client examples outlined in Coroutine Client Examples.
- This is accomplished by defining extension functions for async service stubs.
- This option requires the artifact
kroto-plus-coroutinesas a dependency. - If using rpc interceptors or other code that relies on
io.grpc.Contextthen you need to be sure to add aGrpcContextElementto yourCoroutineContextwhen launching a coroutine. Child coroutines will inherit thisThreadContextElementand the dispatcher will ensure that your grpc context is present on the executing thread.
Context.current().withValue(MY_KEY, myValue).attach()
val myGrpcContext = Context.current()
val job = launch( GrpcContextElement() ) { //Alternate usage: myGrpcContext.asContextElement()
launch {
assertEquals(myGrpcContext, Context.current())
}
GlobalScope.launch{
assertNotEquals(myGrpcContext, Context.current())
}
}
Mock Service Generator
Configuration Options
This generator creates mock implementations of proto service definitions. This is useful for orchestrating a set of expected responses, aiding in unit testing methods that rely on rpc calls. Full example for mocking services in unit tests. The code generated relies on the kroto-plus-test artifact as a dependency. It is a small library that provides utility methods used by the mock services.
- If no responses are added to the response queue then the mock service will return the default instance of the response type.
- Currently only unary methods are being mocked, with support for other method types on the way
@Test fun `Test Unary Response Queue`(){
MockStandService.getStandByNameResponseQueue.apply {
//Queue up a valid response message
addMessage {
name = "Star Platinum"
powerLevel = 500
speed = 550
addAttacks {
name = "ORA ORA ORA"
damage = 100
range = StandProto.Attack.Range.CLOSE
}
}
//Queue up an error
addError(Status.INVALID_ARGUMENT)
}
val standStub = StandServiceGrpc.newBlockingStub(grpcServerRule.channel)
standStub.getStandByName { name = "Star Platinum" }.let{ response ->
assertEquals("Star Platinum",response.name)
assertEquals(500,response.powerLevel)
assertEquals(550,response.speed)
response.attacksList.first().let{ attack ->
assertEquals("ORA ORA ORA",attack.name)
assertEquals(100,attack.damage)
assertEquals(StandProto.Attack.Range.CLOSE,attack.range)
}
}
try{
standStub.getStandByName { name = "The World" }
fail("Exception was expected with status code: ${Status.INVALID_ARGUMENT.code}")
}catch (e: StatusRuntimeException){
assertEquals(Status.INVALID_ARGUMENT.code, e.status.code)
}
}
Extendable Messages Generator (Experimental)
Configuration Options
Generated code relies on the kroto-plus-message artifact. This generator adds tagging interfaces to the java classes produce by protoc. It also adds pseudo companion objects to provide a way to access proto message APIs in a non static manner. The following is a small example of how to write generic methods and extensions that resolve both message and builders type.
inline fun <reified M, B> M.copy( block: B.() -> Unit ): M
where M : KpMessage<M, B>, B : KpBuilder<M> {
return this.toBuilder.apply(block).build()
}
// Usage
myMessage.copy { ... }
inline fun <reified M, B> build( block: B.() -> Unit ): M
where M : KpMessage<M, B>, B : KpBuilder<M> {
return KpCompanion.Registry[M::class.java].build(block)
}
// Usage
build<MyMessage> { ... }
inline fun <M, B> KpCompanion<M, B>.build( block: B.() -> Unit ): M
where B : KpBuilder<M>,M : KpMessage<M,B> {
return newBuilder().apply(block).build()
}
// Usage
MyMessage.Companion.build { ... }
User Defined Generator Scripts
Users can define kotlin scripts that they would like to run during code generation. For type completion, scripts can be couple with a small gradle build script, although this is completely optional. Samples are available in the kp-script directory of the example project.
There are two categories of scripts available.
-
Insertion Scripts
- Configuration Options
- Using the insertion api from the java protoc plugin, users can add code at specific points in generated java classes.
- This is useful for adding code to allow more idiomatic use of generated java classes from Kotlin.
- The entire
ExtendableMessagesgenerator can be implemented using an insertion script, an example can be in the example script extendableMessages.kts. - Additional information regarding the insertion api can be found in the official docs
-
Generator Scripts
- Configuration Options
- These scripts implement the
Generatorinterface used by all internal kroto+ code generators. - Generators rely on the
GeneratorContext, which is available via the propertycontext. - The
contextis used for iterating over files, messages, and services submitted by protoc. - Example usage can be found in the kp-script directory of the example project, as well as inside the
generatorspackage of theprotoc-gen-kroto-plusartifact.
Community Scripts
Community contributions for scripts are welcomed and more information regarding guidelines will be published soon.
Method Signature Options Support
Usage of (google.api.method_signature) method option is now supported. This allows users to customize the method parameters outputted in generated clients as well as stub extensions. To config your rpc methods, first add the google common proto dependency to your build
dependencies{
compileOnly "com.google.api.grpc:proto-google-common-protos:1.16.0"
}
Then add the following import to your proto definition.
import "google/api/client.proto";
Now the method option should be available for usage in your method definition
// Sends a greeting
rpc SayHello (HelloRequest) returns (HelloReply){
option (google.api.method_signature) = "name";
};
This will result in the following method signature being outputed from gRPC and stub extension code generators.
fun GreeterStub.sayHello(name: String): HelloReply{
val request = HelloRequest.newBuilder()
.setName(name)
.build()
return sayHello(request)
}
Getting Started With Gradle
Repositories
- Available on
jcenter()ormavenCentral() - SNAPSHOT
repositories {
maven { url 'https://oss.jfrog.org/artifactory/oss-snapshot-local' }
}
- Bintray
// Useful when syncronization to jcenter or maven central are taking longer than expected
repositories {
maven { url 'https://dl.bintray.com/marcoferrer/kroto-plus/' }
}
Configuring Protobuf Gradle Plugin
plugins{
id 'com.google.protobuf' version '0.8.6'
}
protobuf {
protoc { artifact = "com.google.protobuf:protoc:$protobufVersion"}
plugins {
kroto {
artifact = "com.github.marcoferrer.krotoplus:protoc-gen-kroto-plus:$krotoPlusVersion"
}
}
generateProtoTasks {
def krotoConfig = file("krotoPlusConfig.asciipb") // Or .json
all().each{ task ->
// Adding the config file to the task inputs lets UP-TO-DATE checks
// include changes to configuration
task.inputs.files krotoConfig
task.plugins {
kroto {
outputSubDir = "java"
option "ConfigPath=$krotoConfig"
}
}
}
}
}
Getting Started With Maven
Repositories
- Available on
jcenterormavenCentral - SNAPSHOT
<repository>
<id>oss-snapshot</id>
<name>OSS Snapshot Repository</name>
<url>https://oss.jfrog.org/artifactory/oss-snapshot-local</url>
<snapshots>
<enabled>true</enabled>
</snapshots>
</repository>
- Bintray
<!-- Useful when syncronization to jcenter or maven central are taking longer than expected-->
<repository>
<id>kroto-plus-bintray</id>
<name>Kroto Plus Bintray Repository</name>
<url>https://dl.bintray.com/marcoferrer/kroto-plus/</url>
</repository>
Configuring Protobuf Maven Plugin
<plugin>
<groupId>org.xolstice.maven.plugins</groupId>
<artifactId>protobuf-maven-plugin</artifactId>
<version>0.6.1</version>
<configuration>
<protocArtifact>com.google.protobuf:protoc:3.6.1:exe:${os.detected.classifier}</protocArtifact>
</configuration>
<executions>
<execution>
<goals><goal>compile</goal></goals>
</execution>
<execution>
<id>grpc-java</id>
<goals><goal>compile-custom</goal></goals>
<configuration>
<pluginId>grpc-java</pluginId>
<pluginArtifact>io.grpc:protoc-gen-grpc-java:1.17.1:exe:${os.detected.classifier}</pluginArtifact>
</configuration>
</execution>
<execution>
<id>kroto-plus</id>
<goals>
<goal>compile-custom</goal>
</goals>
<configuration>
<pluginId>kroto-plus</pluginId>
<pluginArtifact>com.github.marcoferrer.krotoplus:protoc-gen-kroto-plus:${krotoPlusVersion}:exe:${os.detected.classifier}</pluginArtifact>
<pluginParameter>ConfigPath=${project.basedir}/krotoPlusConfig.asciipb</pluginParameter>
</configuration>
</execution>
</executions>
</plugin>
Add generated sources to Kotlin plugin
<plugin>
<artifactId>kotlin-maven-plugin</artifactId>
<groupId>org.jetbrains.kotlin</groupId>
<version>${kotlin.version}</version>
<executions>
<execution>
<id>compile</id>
<goals>
<goal>compile</goal>
</goals>
<configuration>
<sourceDirs>
<sourceDir>${project.basedir}/target/generated-sources/protobuf/kroto-plus</sourceDir>
</sourceDirs>
</configuration>
</execution>
</executions>
</plugin>
Configuring Generators
Credit
This project relies on Kotlin Poet for building Kotlin sources. A big thanks to all of its contributors.
2 Nov 29, 2021
19 Sep 11, 2022
7 Nov 28, 2022
18 Apr 21, 2022
2 May 26, 2022
0 Nov 15, 2021
5 Nov 14, 2022
5 Dec 6, 2022
8 Sep 14, 2022
74 Dec 26, 2022
17 Sep 12, 2022
24 Nov 8, 2022
0 Jan 3, 2022
379 Jan 7, 2023
50 Dec 30, 2022
151 Jan 1, 2023
176 Jan 3, 2023
2.2k Jan 4, 2023
2.1k Jan 3, 2023