venue-debug

Venue Debug Workflow

Use this skill when any get_ticker / get_ohlcv / get_order_book / get_trades call fails, returns stale data, or surfaces reason codes that do not match expectations. The goal is to pinpoint where in the L2 chain the fault is, not to blindly retry.

Symptoms → starting points

Symptom	Start with
"provider unavailable" / timeout	Step 1 (health) → Step 3 (HTTP pool)
staleness != "fresh"	Step 2 (cache) → Step 4 (sync worker)
rate_limited reason code	Step 3 (rate limiters)
intermittent silent failures	Step 5 (metrics)

The chain (reminder)

LoggingDecorator
  └── InMemoryCachingDecorator (ticker_ttl=5s, ohlcv_ttl=60s, order_book_ttl=3s)
      └── RateLimitDecorator (TokenBucket per venue)
          └── RetryDecorator (3 attempts, exponential backoff)
              └── MetricsDecorator (provider_calls_total, duration_ms)
                  └── base provider (CCXT or CoinGecko HTTP)

Step 1 — Check venue health

Read cryptozavr://venue_health (or run /cryptozavr:health). Verify the state and last_checked_ms fields for the venue in question.

• down → skip to Step 3.
• degraded → HealthMonitor saw at least one probe failure; go to Step 5.
• healthy but tool still fails → go to Step 2.

Step 2 — Check the cache

Look at the tool response reason_codes:

• cache:hit with unexpected stale data → a realtime invalidation did not fire. Check CacheInvalidator.on_ticker_change logs.
• cache:miss → cache was cold; next step is whether the upstream call succeeded. Go to Step 3.

Step 3 — Rate limits + HTTP pool

• rate_limited outcome in venue_health_check_total counter → the venue exceeded its TokenBucket. Confirm bucket config in src/cryptozavr/mcp/bootstrap.py (RateLimiterRegistry.register).
• Transport failures (error or timeout outcome without rate-limit cause) → inspect HTTP client pool in src/cryptozavr/infrastructure/providers/http.py.

Step 4 — TickerSyncWorker / RealtimeSubscriber

If data staleness persists through cache TTL:

• Verify RealtimeSubscriber.subscriptions() returns the expected (venue, symbol) pairs.
• Confirm TickerSyncWorker.is_running is True in the running lifespan.
• Tail logs for ticker sync failed for ... warnings.

Step 5 — Metrics snapshot

Aggregate counter outcomes from the shared MetricsRegistry:

provider_calls_total{venue=kucoin, endpoint=fetch_ticker, outcome=rate_limited}
provider_calls_total{venue=kucoin, endpoint=fetch_ticker, outcome=error}
venue_health_check_total{venue=kucoin, outcome=timeout}

Compare against provider_call_duration_ms histogram — a sudden bucket shift above 1000ms is a transport-level regression.

Recap template

When reporting to the user, use this 5-line format:

Venue: <kucoin|coingecko>
Symptom: <exact reason codes or error>
Chain layer: <cache|rate_limit|retry|provider>
Root cause: <one sentence>
Fix: <config change | restart | downstream incident>