Real-time VWAP (volume-weighted average price) calculation engine.
To use this repository, the make utility is required. In addition, if you'd like to build the project and execute locally on your machine, Go v1.17+ and golangci-lint are required. However, building and executing can also be done through Docker.
To build the project locally, run:
make install
To build the project using Docker, run:
make docker/build
The vwap calculation engine logs in real time the VWAP values for each trading pair of interest, in a window of configurable size. To start the engine, run the following command:
| Local | Docker |
|---|---|
make run |
make docker/run |
For both cases, the following environment variables can be passed to customize the engine:
- FEED_ENDPOINT: WebSocket endpoint to read trading pair match data from, e.g.,
wss://endpoint.company.com. - TRADING_PAIRS: Comma-separated list of trading pairs of interest to calculate VWAP for, e.g.,
BTC-USD,ETH-BTC. - WINDOW_SIZE: Size of the sliding window to use when calculating VWAP. This has to be at least
1.
This section presents design choices and implementation details for the project.
All WebSocket messages are read as JSON objects. These objects are simple map[string]interface{}, so that any message can be read using the same underlying type. Since, for the VWAP calculation, we are only interested in the match or last_match message types, these maps are parsed to the Match structure when we observe these messages.
In order to allow for increased throughput of incoming WebSocket messages, one goroutine is spawned for reading messages, and another one is spawned for handling them. This way, the reader goroutine reads messages and place them in a buffered channel. The handler goroutine then feeds from this channel to handle new messages.
The VWAP of a product over a window of n data points is defined as SUM(P_i * Q_i) / SUM(Q_i), where P_i and Q_i are the price and quantity of a given data point i. Since we would like to calculate the VWAP for a sliding window, for every new data point, we also want to minimize the cost of performing this calculation, as it will be executed repeatedly.
By keeping track of the numerator and denominator for the fraction that calculates the VWAP, we can incorporate a new data point j into the calculation by adding its contribution to both the numerator and to the denominator. In addition, if the sliding window has reached its limit size, it will discard its oldest entry, meaning that we have to subtract its contribution from the numerator and from the denominator. Hence, we have that VWAP_j = (SUM(P*Q) - (P*Q)_old + (P*Q)_j) / (SUM(Q) - Q_old + Q_j).
This calculation method has been implemented using a channel, which behaves as a queue for the sliding window, allowing us to easily pop the oldest data point and push the new one.
Unit and integration tests have been added for the project. To run them, execute the following command:
make test
The coverage.html file can be opened in a browser to display details about the test coverage for each file.