VWAP Engine

VWAP Calculator for Go.

Go implementation of real-time VWAP (volume-weighted average price) calculation engine. vwap is calculated in real time utilizing the Coinbase websocket stream "wss://ws-feed.pro.coinbase.com". For each trading pair, the calculated VWAP will be logged to Stdout. The default trading pairs are BTC-USD, ETH-USD, and ETH-BTC, but you can define your owns via ENV variables

Workflow

• Subscribe to the matches channel and retrieve data feed from the coinbase websocket. Default trading pairs' data: BTC-USD, ETH-USD, and ETH-BTC.
• Calculate the VWAP per trading pair using a sliding window(default 200) data points. VWAP is cached for each trading pair.
• When a new data point arrives through the websocket feed the oldest data point will fall off and the new one will be added such that no more than 200 data points are included in the calculation.

Architecture

This application main components comprises:

Tunnel interface

A coinbase websocket stream client to receive data from coinbase websocket server.

1. It subscribes(Tunnel.Subscribe) to the coinbase channel's websocket using trading pairs (productIDs).
2. Read (Tunnel.Read) real-time data points from the coinbase and passes them to the receiver channel.

VWAP interface:

• Represents a queue of DataPoints and their VWAPs.
• DataPoints is fast circular fifo data structure (aka., queue) with a specific limit.
• The arrays allocated in memory are never returned. Therefor A dynamic doubly Linked list structure, is better to be used for a long-living queue.
• For every new coinbase entry, it pushes an item onto the queue and calculates the new VWAP.
• When Limit is reached, will delete the first one. Every time a new data point is added to the queue and saved for each trading pair, the VWAP computation is updated accordingly. For performance, and to avoid exponential complexity, the computation is cached for VWAP, CumulativeQuantity, and CumulativePriceQuantity for existing data points and updated with new entries.

Two implementations:

1. Doubly linked-list queue: Manipulation with LinkedList is faster than ArrayList because it uses a doubly linked list, so no bit shifting is required in memory.
2. Array-backed queue: Manipulation with ArrayList is slow because it internally uses an array. If any element is removed from the array, all the other elements are shifted in memory.

Main

The core entry point into the app. will setup the config, run the App context, It is resilient tolerant. It will gracefully shutdown and can receive an interrupt signal and safely close the connexio.

App

Setup the config, run the App context, subscribe to the ws, and initiate the vwap storage and calculation for the trading pairs. It is resilient tolerant.

Helm & K8S

Helm charts to deploy this micro-service in a Kubernetes platform We generate the container image and reference it in a Helm chart

Project layout

This layout is following pattern:

vwap-engine
└───
    ├── .env // optional
    ├── .github
    │   └── workflows
    │     └── go.yml
    ├── api
    │   └── models
    │     └── coinbase.go
    ├── cmd
    │   └── main.go
    ├── internal
    │   └── app
    │     └── app.go
    │     └── setup.go
    │     └── context.go
    │   └── tunnel
    │     └── receiver.go
    │     └── tunnel.go
    │   └── storage
    │     └── vwap.go
    │     └── linked-list
    │         └── vwap_linked_list.go
    │     └── queue
    │         └── vwap_queue.go
    ├── build
    │   └── Dockerfile
    ├── helm
    │   └── <helm chart files>
    ├── Makefile
    ├── Jenkinsfile
    ├── README.md
    ├── VERSION
    └── <source packages>

Setup

The app is configurable via the ENV variables or Helm values for cloud-native deployment Config parameters:

• TRADING_PAIRS: a list of coinbase product IDS. Example: BTC-USD,ETH-USD,ETH-BTC
• WEBSOCKET_URL: coinbase websocket server. Example: wss://ws-feed.pro.coinbase.com
• WINDOW_SIZE: Data points sliding window for VWAP computation.

Getting started

Vwap engine is available in github Vwap-engine

go get github.com/reactivejson/storage-engine

Run

go run cmd/main.go

Build

make build

Testing

make test

Build docker image:

make docker-build

This will build this application docker image so-called vwap-engine

Deploy with Helm chart in a Kubernetes environment

 helm upgrade --namespace neo --install vwap-engine chart/vwap-engine -f your-custom-values.yml

Test coverage

Test coverage is checked as a part of test execution with the gotestsum tool.

Test coverage is checked for unit tests and integration tests.

Coverage report files are available and stored as *coverage.txt and are also imported in the SonarQube for easier browsing.

golangci-lint

In the effort of reducing errors and improving the overall quality of code, golangci-lint is run as a part of the pipeline. Linting is run for the services and packages that have changes since the previous green build (in master) or previous commit (in local or review).

Any issues found by golangci-lint for the changed code will lead to a failed build.

golangci-lint rules are configured in .golangci.yml.

Sonar scan

A Sonar scan is run as a part of the pipeline.

Requirements

• Go 1.15 or newer https://golang.org/doc/install
• Docker 18.09.6 or newer
• docker-compose 1.24.0 on newer

Variable names

Commonly used one letter variable names:

• i for index
• r for reader
• w for writer
• c for client
• vwap for Volume Weighted Average Price

License

Apache 2.0, see LICENSE.