Infrastructure Ops

Infrastructure Operations Skill

Purpose

Understand and modify the operational infrastructure that connects market data to order execution. Covers the event loop, WebSocket management, rate limiting, reconnection, recovery, metrics, and order execution.

When to Use

• Working on orchestrator/, infra/, messages/, core/, fills/, or execution/ modules
• Debugging connectivity, order placement, or feed issues
• Adding new message handlers or data sources
• Understanding the event loop and quote cycle
• Investigating "orders not sending" or "stale data"

Module Map

All paths relative to src/market_maker/:

orchestrator/
  event_loop.rs         # Main event loop — sync, startup, recovery, dispatch
  handlers.rs           # WS message dispatching (L2 book, trades, fills, user events)
  quote_engine.rs       # Quote generation — warmup, estimator validation, strategy
  reconcile.rs          # ReconciledOrder — order diff/reconciliation
  order_ops.rs          # Place, cancel, modify order abstractions
  recovery.rs           # Safety sync, state refresh, stuck order recovery
  event_accumulator.rs  # EventAccumulator — event-driven quote triggering

infra/
  connection_supervisor.rs  # Proactive WS health monitoring
  reconnection.rs           # Feed reconnection state machine
  data_quality.rs           # Market data anomaly detection & validation
  margin.rs                 # MarginTracker — margin-aware position sizing
  ws_executor.rs            # WS POST executor with REST fallback
  arena.rs                  # Arena allocator for quote cycle latency
  capacity.rs               # Pre-allocation sizes
  orphan_tracker.rs         # Prevents false orphan detection
  logging.rs                # Structured multi-stream logging
  recovery.rs               # Stuck position recovery
  reconciliation.rs         # Event-driven position sync
  sync_health.rs            # Feed sync health tracking
  dashboard_ws.rs           # Real-time dashboard WebSocket
  rate_limit/
    mod.rs                  # Rate limit orchestrator
    proactive.rs            # Predict rate limits before rejection
    rejection.rs            # React to rate limit rejections
    error_type.rs           # RateLimitError types
  metrics/
    mod.rs                  # PrometheusMetrics aggregator
    atomic.rs               # Lock-free metric updates
    dashboard.rs            # Signal snapshots, diagnostics
    fields.rs               # All tracked metric fields
    output.rs               # Prometheus text format
    summary.rs              # Aggregated statistics
    updates.rs              # Event-driven metric recording

messages/
  all_mids.rs           # Mid-price message processing
  l2_book.rs            # L2 order book message processing
  trades.rs             # Trade message processing
  processors.rs         # Message processor traits
  context.rs            # Message context (timestamps, sequence)

core/
  components.rs         # Core component initialization
  state.rs              # CoreState — shared mutable state

fills/
  consumer.rs           # Fill event consumption
  dedup.rs              # Fill deduplication
  processor.rs          # Fill processing pipeline

execution/
  fill_tracker.rs       # Fill tracking and reporting
  order_lifecycle.rs    # Order state machine (pending -> open -> filled/cancelled)

Event Loop Architecture

Startup Sequence

1. sync_open_orders()    — fetch all orders from exchange
2. cancel_all()          — clean slate
3. sync_position()       — reconcile local vs exchange position
4. sync_limits()         — fetch exchange-enforced limits
5. enter main loop

Main Loop

loop {
    1. Receive WS message (L2 book, trades, fills, user events)
    2. Dispatch to handler (handlers.rs)
    3. Accumulate events (event_accumulator.rs)
    4. If enough events accumulated -> update quotes (quote_engine.rs)
    5. Periodic: recovery check, safety audit, metrics flush
}

Event Accumulator

Prevents excessive quote updates. Triggers when:

• L2 book changed significantly
• Trade occurred
• Fill received
• Timer elapsed (max staleness)

Quote Generation Pipeline

quote_engine.rs drives each quote cycle:

1. Warmup check     — enough data for reliable estimates?
2. Estimator check  — kappa, sigma, regime all valid?
3. Risk check       — risk system allows quoting?
4. Strategy compute — GLFT formula -> optimal bid/ask
5. Ladder generate  — multi-level quote ladder
6. Reconcile        — diff against current orders
7. Execute          — place/cancel/modify minimum set

Order Reconciliation (reconcile.rs)

Minimizes exchange API calls by diffing desired vs current:

• Keep: same price and size -> no action
• Modify: same price, different size -> modify
• Cancel+Place: different price -> cancel old, place new
• Cancel: existing not in new set -> cancel
• Place: new not in existing set -> place

WebSocket Execution

ws_executor.rs:

• Primary: WS POST for lowest latency
• Fallback: REST API when WS degraded
• Automatic retry with exponential backoff

Order Lifecycle (execution/order_lifecycle.rs)

Pending -> Open -> Filled | Cancelled | Expired
                -> PartialFill -> Filled | Cancelled

Rate Limiting

Three-tier defense:

1. Proactive (rate_limit/proactive.rs)

• Sliding window rate tracking
• Predicts when limit will be hit
• Throttles before rejection

2. Reactive (rate_limit/rejection.rs)

• Handles RateLimitError from exchange
• Exponential backoff on rejection

3. Budget-based

• Token bucket for statistical impulse control
• Distributes order budget across cycles

Reconnection & Recovery

Feed Reconnection (infra/reconnection.rs)

Connected -> Disconnected -> Reconnecting (with backoff) -> Connected

• Exponential backoff with jitter
• Separate tracking per feed (Hyperliquid L2, trades, Binance)
• Connection supervisor monitors all feeds

Recovery (orchestrator/recovery.rs)

After reconnection:

1. Safety sync: verify position matches exchange
2. State refresh: re-fetch margins, limits, open orders
3. Stuck orders: detect and cancel non-responsive orders
4. Orphan cleanup: reconcile local tracking vs exchange

Data Quality

data_quality.rs validates incoming market data:

• Timestamp freshness
• Price sanity (no negative, no >10% jumps without explanation)
• Book consistency (bid < ask, levels sorted)
• Sequence number tracking (detect gaps)

Validation failures trigger data staleness warnings in risk system.

Metrics

Architecture

Event -> AtomicMetric (lock-free) -> MetricsSummary -> PrometheusOutput

Key Metrics

Arena Allocator

arena.rs — per-quote-cycle allocation:

• Pre-allocates memory for hot path
• Avoids heap allocation during quote cycle
• Reset between cycles (near-zero cost)
• Capacity constants in capacity.rs

Common Debugging

"Orders not sending"

1. Rate limiter — throttled?
2. WS executor — connection alive?
3. Risk system — quotes gated?
4. Order lifecycle — stuck in Pending?
5. Margin — enough for new orders?

"Stale quotes / not updating"

1. Event accumulator — triggering updates?
2. Data quality — feed data fresh?
3. Connection supervisor — all feeds connected?
4. Warmup — estimator has enough data?

"Position mismatch"

1. Safety auditor — SafetyAuditor::audit()
2. Orphan orders — fills from unknown orders?
3. Fill dedup — double-counted fills?
4. Reconciliation — local vs exchange position

"High latency"

1. quote_cycle_us metric
2. Arena allocator — capacity sufficient?
3. Blocking in event loop — no sync I/O!
4. ws_roundtrip_us metric

Key Rules

1. Never block the event loop — all I/O must be async
2. Arena allocator on hot path — no heap allocation in quote cycle
3. Rate limiting is defense — never rely on exchange not rejecting
4. Reconciliation over creation — minimize order operations
5. Data quality gates quotes — stale data = no quotes
6. Recovery is automatic — system must self-heal from disconnections

Adding a New Message Handler

1. Define message type in messages/
2. Implement MessageProcessor trait
3. Add dispatch case in handlers.rs
4. Register event type in event_accumulator.rs
5. Update metrics in metrics/updates.rs
6. Add data quality checks in data_quality.rs

Supporting Files

Converted and distributed by TomeVault — claim your Tome and manage your conversions.