Reactive microservice end to end from Rx Java

Reactive microservice end to end from Rx. Java to the wire with g. RPC Ryan Michela @ Salesforce

Let’s poll the room g. RP C

Let’s poll the room Rx. Jav a

Let’s poll the room g. RP C Rx. Jav a

M g. RPC E About me Rx. Jav a Microservices practitioner Reactive programming enthusiast

What’s in this talk 1. The story of reactive-grpc 2. An overview of g. RPC and Rx. Java 3. Some great demos 4. A deep dive into backpressure and flow control 5. An even better demo

Where did Reactive-g. RPC come from? 20 year old monolith to Microservices Bespoke REST APIs Apache Thrift g. RPC + Mesh

What is g. RPC? § High-performance, open source, RPC framework § Protobuf and HTTP/2 § Big features - Contract-first microservices - Polyglot code generation - Unary and bi-directional streaming § https: //grpc. io/

What is Rx. Java? § API for asynchronous programming with observable streams § Everything is a stream (Observable) § Combine, filter, transform § Subscribe (Observer) § http: //reactivex. io/

What is Rx. Java? § API for asynchronous programming with observable streams § Everything is a stream (Observable) § Combine, filter, transform Java. Fx. Observable. events. Of(scene, § Subscribe (Observer) Mouse. Event. CLICK) . debounde(1, Time. Unit. SECONDS). map(event -> new Robot. Command(event)). subscribe(command -> robot. do(command));

1+1=3 https: //imgur. com/gallery/c 7 NJRa 2

g. RP C Reactive g. RPC Rx. Jav a Protobuf compiler plugin + new stubs

Questions so far?

Hello World demo - What will we see? § Only g. RPC § Reactive g. RPC § Protobuf IDL § Protoc code generator § Single<T> and Flowable<T>

Demo summary § Hello. proto defines operations and data structures § Protoc compiler generates java stubs from proto during build § Client and server build on generated stubs § Reactive-g. RPC generates Rx. Java stubs, extends protoc compiler § Rx. Java makes writing streaming services easy

Questions so far?

Backpressure § How downstream consumers signal to producers to slow down https: //cheezburger. com/7944963840

Three Backpressure Strategies Reactive g. RPC

Producer? Consumer?

HTTP/2 backpressure and flow control § Stream-based flow control – per stream and per connection § Initial flow control window size 65, 535 bytes § Producer stops when window closes § Consumer asks for more bytes by sending a WINDOW_UPDATE frame Flow Control Window Producer BYTES WINDOW_UPDATE Flow Control Window Consumer

g. RPC backpressure and flow control § Message-based flow control over HTTP/2 § Producer - On. Ready. Handler produces and serializes proto messages until HTTP/2 flow control window is exhausted or there are no more messages to send - WINDOW_UPDATE frames from consumer re-trigger On. Ready. Handler Producer On. Ready Handler on. Next(message) Ready? HTTP/2 Flow Control Window BYTES WINDOW_UPDATE Flow Control Window

g. RPC backpressure and flow control § Message-based flow control over HTTP/2 § Consumer - Consumer calls request() - g. RPC deserializes messages and calls on. Next() - g. RPC sends WINDOW_UPDATE frames to receive more bytes Producer On. Ready Handler on. Next(message) Consumer on. Next(message) Ready? HTTP/2 Flow Control Window BYTES WINDOW_UPDATE Flow Control Window request()

Rx. Java backpressure and flow control § Message-based flow control over Java method invocations § Consumer calls request() § Producer calls on. Next() § Intermediate operators propagate both calls Producer Observable request() on. Next() Operator Map request() on. Next() Operator Reduce request() on. Next() Consumer Observer

Reactive-g. RPC backpressure and flow control § Message-based flow control over Rx. Java over g. RPC over HTTP/2 Rx. Java Observer Observable on. Next() request() Operator on. Next() request() g. RPC On. Ready Handler on. Next() Consumer on. Next() Ready? HTTP/2 Flow Control Window request()

Reactive-g. RPC backpressure and flow control § Message-based flow control over Rx. Java over g. RPC over HTTP/2 Rx. Java Observer Observable on. Next() request() Operator on. Next() request() g. RPC On. Ready Handler on. Next() Consumer on. Next() Ready? HTTP/2 Flow Control Window https: //imgur. com/gallery/k. W 0 tz. FN request()

Questions so far?

Backpressure demo – What will we see? § Slow consumer § Fast producer

Demo summary § Fast producer § Slow consumer § Producer outpaces consumer § Producer slows down § Backpressure is free (for you)

Questions so far?

Applied Reactive-g. RPC demo – What will we see? § IRC-like Chat § Command line § Java. FX Desktop § Android Mobile

Applied Reactive-g. RPC demo – What will we see? Server § Chat. Message ➜ Empty § Empty ➜ stream Chat. Message

Demo summary § Reactive-g. RPC make chat server trivial to write § Three different clients, nearly identical code - Command line - Java. FX - Android

To Conclude § g. RPC makes microservices easy -. proto IDL - Code generation § Rx. Java is a different way of thinking - Streams of events - Function composition § Reactive-g. RPC bridges both technologies

Why Reactive-g. RPC? Async by default

Why Reactive-g. RPC? Unified programming model

Why Reactive-g. RPC? Simplified programming model

Why Reactive-g. RPC? Consistent programming model

Thank you! § Resources - https: //grpc. io/ - http: //reactivex. io/ - https: //github. com/salesforce/reactive-grpc - �� Demo code Questions?