System architecture

Components and ownership

Component	Owns	Does not own
MongoDB	Strategy and market-data deployments, versioned configuration, and provenance	Live process state or market data
dyson-market-data	Loading one feed bundle and supervising its publisher processes	Venue parsing or strategy decisions
Imported venue publishers	Venue connections, native feed handling, archival output, and optional live NATS publication	Strategy lifecycle
NATS	Live market data, strategy commands, status, and runtime events	Durable strategy configuration or analytics
Strategy binary	Concrete strategy bootstrap, subjects, Kuru connection, and post-trade wiring	Shared lifecycle machinery
dyson-runtime	Lifecycle, serialized dispatch, NATS transport, health, metrics, and shutdown	Strategy policy or venue protocol details
Algo-trading submodule	Signals, actors, strategy engine, canonical actions, and feed primitives	Dyson process lifecycle or Kubernetes state
Venue adapters	Packet normalization and venue execution (Kuru orderbook or vault, Lighter API/WS)	Strategy policy
InfluxDB	Structured strategy, order, fill, and health projections	Runtime commands or configuration
Control panel	One-deployment monitoring and the Start action	Runtime truth or general lifecycle administration
Strategy API	Liveness, MongoDB-backed readiness, and OpenAPI	Strategy CRUD or lifecycle control at this revision
dyson-deployment	Kubernetes and platform desired state	Application behavior or image builds

Market-data flow

flowchart TD
    CFG["MongoDB<br/>feed deployment + version"]
    SUP["dyson-market-data supervisor"]
    PUB["Imported venue publisher processes<br/>one per venue"]
    ARC["Local archive"]
    NATS["NATS live publication<br/>(optional)"]

    CFG --> SUP --> PUB
    PUB --> ARC
    PUB --> NATS

    classDef store fill:#2e7d32,stroke:#1b5e20,color:#fff
    class CFG,ARC,NATS store

A feed bundle may contain multiple publishers, with one publisher per venue and multiple markets inside that publisher's native configuration. If one child publisher exits, the supervisor stops the bundle rather than leaving a partial feed set running.

Strategy flow

flowchart TD
    CFG["MongoDB<br/>strategy deployment + version"]
    BIN["Strategy binary"]
    RT["dyson-runtime<br/>serialized dispatch"]
    IN["NATS<br/>market data + commands"]
    ENG["Algo-trading engine<br/>or custom strategy backend"]
    ADP["Venue adapters"]
    VEN["Venues"]
    OUT["NATS<br/>status and events"]
    INF["InfluxDB<br/>strategy projections"]

    CFG --> BIN --> RT
    IN --> RT
    RT --> ENG --> ADP --> VEN
    RT --> OUT
    ENG --> INF

    classDef store fill:#2e7d32,stroke:#1b5e20,color:#fff
    class CFG,IN,OUT,INF store

Each strategy is a separate process. The runtime reconciles desired state, processes one input at a time, and keeps venue-specific translation outside the shared lifecycle core. A multi-venue strategy fans the adapter stage out across legs — see Multi-venue strategy.

Operator flow

flowchart LR
    BR["Browser"] --> CP["Control-panel server"]
    CP --> NATS["NATS"]
    CP --> INF["InfluxDB"]
    CP --> KWS["Kuru public WebSocket"]
    API["Strategy API"] --> MDB["MongoDB<br/>readiness check"]

    classDef store fill:#2e7d32,stroke:#1b5e20,color:#fff
    class NATS,INF,KWS,MDB store

The control panel does not call the Strategy API. The two are separate operator surfaces with different current responsibilities.

Current boundaries

• Live execution covers Kuru and, for the multi-venue strategy, Lighter; market-data publishers cover additional venues.
• The control panel publishes Start only, even though the runtime supports a wider lifecycle command set.
• NATS is the live bus, not a replacement for MongoDB configuration or InfluxDB projections.
• Application behavior changes in dyson; release and cluster state change in dyson-deployment.