A real-time VWAP calculation engine using Coinbase's Websocket data feed.

1. This engine subscribes to the Coinbase Websocket matches channel and retrieves data for the following trading pairs:

   • BTC-USD
   • ETH-USD
   • ETH-BTC

2. It then calculates the VWAP per trading pair using a sliding window of 200 (maximum) data points.

3. The VWAP formula used is based on the following resources:

   • https://academy.binance.com/en/articles/volume-weighted-average-price-vwap-explained
   • https://en.wikipedia.org/wiki/Volume-weighted_average_price

4. The sliding window is implemented as a slice of data points. When a new data point is pushed into the window, we check if the window size has been exceeded, and pop the oldest data point from the queue.

5. Rather than recalculate the Product * Quantity and Quantity summations each time a new data point is added, the PQ and Q of the oldest data point is first deducted from a running total in O(N) constant time and O(1) space.

For the purposes of this exercise, this engine takes the KISS+YAGNI approach. Therefore, the following assumptions have been made:

1. No client authentication. The engine uses the publicly available websocket feed, and takes rate limiting into consideration. Client authentication can be trivially added if required.

2. The engine only handles the matches channel. This could be further extended to handle the other channel types in the Coinbase websocket client.

3. Only the Coinbase websocket feed is supported. The packages may need to be structured differently to handle other exchanges, and client types.

4. Under normal circumstances, the data feed service would be separate from the vwap calculation engine. I have opted to combine it in a single service due to time constraints.

To run the engine...

make run

This builds the docker image and runs the docker container.

make run-cli

This runs the command go run ./cmd/server/main.go

To run the tests:

make test

Unit tests only. No integration test.

Things to do if I had more time:

1. Improve test coverage
2. Split the websocket data feed and VWAP calculation engine into two separate service implementions. We use the Dogma framework to build event-sourcing services like this.
3. Handle websocket timeouts and disconnects.

1. The sandbox environment operates differently to the live environment. Firstly, it only accepts BTC-USD. All other product_id results in a Failed to subscribe error.
2. Secondly, it sends a single last_match response and nothing else for an extended period. It intermittently sends match responses in short bursts, again at extended intervals. It could be a quirk with the sandbox environment, as it seems to work fine on the live feed.