Trading System

A data acquisition pipeline for exchange orderbook data, written in C++23.

It connects to Kraken's L2 book channel and, on command from a frontend client, records pairs into per-pair binary files or streams their books live to the client over a websocket.

Layout

• data_feed/ — the acquisition program (CMake project).
• example_scripts/ — standalone helper scripts (e.g. fetching a Kraken websocket token from Python).
• third_party/ — vendored headers (rapidjson) and the vcpkg install tree.

What it records

Nothing is subscribed at start. A client connects to the websocket on 127.0.0.1:8765 and sends protobuf ClientCommands (proto/market_data.proto): ListPairs for the pairs a venue can stream, Start / Stop / List for recordings and live feeds, Watch / Unwatch to stream a live feed's books. See data_feed/ARCHITECTURE.md.

Each subscription is set up as follows:

Setting	Value
Venue	wss://ws.kraken.com/v2
Channel	book
Symbols	any Kraken pair, one subscription per pair
Depth	one of Kraken's book depths (10, 25, 100, 500, 1000)
Recording	one binary file per pair per venue connection, named by its capture window (e.g. BTCUSD-<start>-<end>.bin)
Live feed	protobuf Books to every client watching it

Each pair gets its own writer lane on its own thread; a write failure only takes down that pair's lane (and triggers an unsubscribe for it) — the rest of the feed keeps running. The connection status message and subscription acknowledgements are consumed during Connect() and printed to the console, not written to the files.

Requirements

• CMake 4.0+, Ninja, MinGW-w64 GCC (see data_feed/CMakePresets.json)
• vcpkg with VCPKG_ROOT set; dependencies are pinned in vcpkg.json
• Kraken API credentials in the environment:
  • KRAKEN_API_KEY
  • KRAKEN_PRIVATE_KEY

Build and run

From data_feed/:

cmake --preset mingw
cmake --build --preset debug      # or: --preset release
./build/Debug/data_feed.exe

Stop with Ctrl+C.