Reference application developed in the Functional event-driven architecture: Powered by Scala 3 book.
The web application allows users to subscribe/unsubscribe to/from symbol alerts such as EURUSD, which are emitted in real-time via Web Sockets.
It is written in Elm and can be built as follows.
$ nix build .#elm-webapp
$ open result/index.html # or specify browserThere's also a development shell handy for local development.
$ nix develop .#elm
$ cd web-app
$ elm make src/Main.elm --output=Main.js
$ open index.html # or specify browserIf Nix is not your jam, you can install Elm by following the official instructions and then compile as usual.
$ cd web-app
$ elm make src/Main.elm --output=Main.js
$ xdg-open index.html # or specify browserThere is also a replica of the Elm application written in Scala using the Tyrian framework that can be built as follows.
$ sbt 'webapp/fullLinkJS'You can then run it via Nix as shown below (it requires flakes).
$ nix run .#tyrian-webapp
Using cache dir: /home/gvolpe/workspace/trading/modules/ws-client/nix-parcel-cache
Server running at http://localhost:1234
✨ Built in 7msNOTICE: The nix run command will create a directory for the Parcel cache, which needs write permissions.
We use fullLinkJS to create a fully optimized JS file. However, we can use fastLinkJS for faster iterations.
For such cases, it may be more convenient to use yarn directly.
$ nix develop .#tyrian
$ cd modules/ws-client
$ yarn install
$ yarn build
$ yarn start
yarn run v1.22.17
parcel index.html --no-cache --dist-dir dist --log-level info
Server running at http://localhost:1234
✨ Built in 1.82sHowever, this is not fully reproducible and can't be guaranteed this will work in the future.
Without Nix, you need to install yarn and parcel, and use yarn as shown above.
Here's an overview of all the components of the system.
- Dotted lines: Pulsar messages such as commands and events.
- Bold lines: read and writes from / to external components (Redis, Postgres, etc).
The back-end application is structured as a mono-repo, and it requires both Apache Pulsar and Redis up and running. To make things easier, you can use the provided docker-compose.yml file.
The docker-compose file depends on declared services to be published on the local docker server. All docker builds are handled within the build.sbt using sbt-native-packager.
To build all of the service images, we have a few options.
The first one via the given Dockerfile.
$ docker build -t jdk17-curl modules/The second one via Nix, from where we can build a slim image also based on openjdk:17-slim-buster.
$ nix build .#slimDocker -o result-jre
$ docker load -i result-jreThe third one also via Nix, though building a layered image based on the same JDK we use for development.
$ nix build .#docker -o result-jre
$ docker load -i result-jreThe main difference between these three options is the resulting image size.
$ docker images | rg jdk17
jdk17-curl latest 0ed94a723ce3 10 months ago 422MB
jdk17-curl-nix latest c28f54e42c21 52 years ago 557MB
jdk17-curl-slim latest dbe24e7a7163 52 years ago 465MBAny image is valid. Feel free to pick your preferred method.
NOTE: As of September 2022, the Docker image resulting from nix build .#docker is no longer compatible with sbt-native-packager, so either go for nix build (defaults to the slim image) or build it directly via Docker with the given Dockerfile.
Once the base jdk17-curl image has been built, we can proceed with building all our services' images.
$ sbt docker:publishLocal$ docker-compose up -d pulsar redis
To run the Kafka Demo (see more below in X Demo), kafka.yml should be used instead.
$ docker-compose -f kafka.yml upIf we don't specify any arguments, then all the containers will be started, including all our services (except feed), Prometheus, Grafana, and Pulsar Manager.
$ docker-compose up
Creating network "trading_app" with the default driver
Creating trading_pulsar_1 ... done
Creating trading_redis_1 ... done
Creating trading_ws-server_1 ... done
Creating trading_pulsar-manager_1 ... done
Creating trading_alerts_1 ... done
Creating trading_processor_1 ... done
Creating trading_snapshots_1 ... done
Creating trading_forecasts_1 ... done
Creating trading_tracing_1 ... done
Creating trading_prometheus_1 ... done
Creating trading_grafana_1 ... doneIt is recommended to run the feed service directly from sbt whenever necessary, which publishes random data to the topics where other services are consuming messages from.
The back-end application consists of 9 modules, from which 5 are deployable applications, and 3 are just shared modules. There's also a demo module and a web application.
modules
├── alerts
├── core
├── domain
├── feed
├── forecasts
├── it
├── lib
├── processor
├── snapshots
├── tracing
├── ws-client
├── ws-server
├── x-demo
└── x-qa
Capability traits such as Logger, Time, GenUUID, and potential library abstractions such as Consumer and Producer, which abstract over different implementations such as Kafka and Pulsar.
Commands, events, state, and all business-related data modeling.
Core functionality that needs to be shared across different modules such as snapshots, AppTopic, and TradeEngine.
Generates random TradeCommands and ForecastCommands followed by publishing them to the corresponding topics. In the absence of real input data, this random feed puts the entire system to work.
Registers new authors and forecasts, while calculating the author's reputation.
The brain of the trading application. It consumes TradeCommands, processes them to generate a TradeState and emitting TradeEvents via the trading-events topic.
It consumes TradeEvents and recreates the TradeState that is persisted as a snapshot, running as a single instance in fail-over mode.
The alerts engine consumes TradeEvents and emits Alert messages such as Buy, StrongBuy or Sell via the trading-alerts topic, according to the configured parameters.
It consumes Alert messages and sends them over Web Sockets whenever there's an active subscription for the alert.
A decentralized application that hooks up on multiple topics and creates traces via the Open Tracing protocol, using the Natchez library and Honeycomb.
All unit tests can be executed via sbt test. There's also a small suite of integration tests that can be executed via sbt it/test (it requires Redis to be up).
It contains all the standalone examples shown in the book. It also showcases both KafkaDemo and MemDemo programs that use the same Consumer and Producer abstractions defined in the lib module.
To run the Pulsar CDC Demo, you need Postgres and Pulsar (make sure no other instances are running). Before running them, we need to download the connector NAR file.
$ mkdir -p pulsarconf/connectors && cd pulsarconf/connectors
$ wget https://archive.apache.org/dist/pulsar/pulsar-2.10.1/connectors/pulsar-io-debezium-postgres-2.10.1.nar
$ docker-compose -f pulsar-cdc.yml upOnce both instances are up and healthy, we can run the Pulsar Debezium connector.
$ docker-compose exec -T pulsar bin/pulsar-admin source localrun --source-config-file /pulsar/conf/debezium-pg.yamlYou should see this in the logs.
Snapshot step 3 - Locking captured tables [public.authors]It contains the smokey project that models the smoke test for trading.
JVM stats are provided for every service via Prometheus and Grafana.
Two Pulsar topics can be compacted to speed-up reads on startup, corresponding to Alert and TradeEvent.Switch.
To compact a topic on demand (useful for manual testing), run these commands.
$ docker-compose exec pulsar bin/pulsar-admin topics compact persistent://public/default/trading-alerts
Topic compaction requested for persistent://public/default/trading-alerts.
$ docker-compose exec pulsar bin/pulsar-admin topics compact persistent://public/default/trading-switch-events
Topic compaction requested for persistent://public/default/trading-switch-eventsIn production, one would configure topic compaction to be triggered automatically at the namespace level when certain threshold is reached. For example, to trigger compaction when the backlog reaches 10MB:
$ docker-compose exec pulsar bin/pulsar-admin namespaces set-compaction-threshold --threshold 10M public/default