Architecture¶

System Diagram¶

Python (TradingNode)
  ├── 5 Strategies  ──── strategy.log.debug(json payload) ──► ./logs/*.json
  └── DatabaseActor ──── self.log.debug(json payload)     ──► ./logs/*.json
                                                                   │
                                            Grafana Alloy (tail & parse)
                                                                   │
                                                   stage.json  →  stage.labels  →  stage.match
                                                                   │              │
                                                            metric.gauge    metric.counter
                                                                   │              │
                                            Mimir (Prometheus remote_write, persisted on disk)
                                                                   │
                                                      Grafana Dashboard (port 3000)

FastAPI (dashboard/)
  └── REST API ──► PostgreSQL

Service Breakdown¶

engine/¶

The NautilusTrader node is the core trading runtime. It hosts:

5 strategies — MeanReversionStrategy, HourlyBuyStrategy, and 3 latency-instrumented variants
DatabaseActor — persists fills, orders, and signals to PostgreSQL
BudgetActor — enforces per-strategy capital caps

Each strategy emits structured JSON log lines consumed by Alloy:

{
  "component_name": "EMACross",
  "metric_type": "gauge",
  "metric_name": "strategy_latency_us",
  "metric_value": 42.5,
  "strategy_id": "mean_reversion"
}

dashboard/¶

FastAPI server at port 8000. Provides REST endpoints for:

Trading data (orders, fills, positions, signals)
Budget management (per-strategy capital allocation)
Audit logs
Docker service management

Observability Stack¶

Service	Port	Role
Grafana Alloy	12345	Tail logs, parse JSON, emit Prometheus metrics
Mimir	9009	Prometheus-compatible metrics storage
Grafana	3000	Dashboards and alerting
Loki	3100	Log aggregation
cAdvisor	8080	Container CPU/memory/network metrics

Metric Flow¶

Log line (JSON in message field)
  └── Alloy stage.json  →  extracts metric_type / metric_name / metric_value / strategy_id
      └── stage.labels  →  promotes to stream labels
          ├── {metric_type="gauge"}   →  loki_process_custom_nautilus_latency_us
          └── {metric_type="counter"} →  loki_process_custom_nautilus_metrics_this_run_total

Note

Counters reset when Alloy restarts. Use increase(metric[duration]) in PromQL for windowed counts. For authoritative totals, query the database directly.

Docker Compose Files¶

File	Purpose
`docker-compose.dev.yml`	Local dev: PostgreSQL 18 + Redis only
`docker-compose.yml`	Observability stack: Mimir, Alloy, Grafana, Loki, cAdvisor
`docker-compose.prod.yml`	Production overlay: adds `trader` container