Observability

Logging

• Format: JSON structured
• Fields: timestamp, level, service, trace_id, message, context
• Aggregation: Loki или ELK

Пример:

{
  "level": "info",
  "service": "game-runtime",
  "trace_id": "abc123",
  "message": "action applied",
  "game_id": "...",
  "action_type": "choose_cell",
  "duration_ms": 42
}

Metrics (Prometheus)

Metric	Type	Labels
http_request_duration_seconds	histogram	service, path, status
ws_connections_active	gauge	instance
sessions_active	gauge	—
game_actions_total	counter	plugin_id, action_type
nats_publish_errors	counter	topic

Dashboards (Grafana)

1. Overview — RPS, error rate, active sessions
2. Game — actions/min, apply latency p95
3. Realtime — WS connections, event lag
4. Bot — callback latency, edit_message errors

Tracing (OpenTelemetry)

• Propagate traceparent HTTP header
• Span: action.apply, media.render, ws.push
• Export: OTLP → Jaeger/Tempo

Alerting

Alert	Condition	Severity
HighErrorRate	5xx > 1% 5m	critical
WSDown	health fail 2m	critical
SlowActions	p95 > 500ms 10m	warning
RedisDown	ping fail	critical

Audit log

• Table game_actions — immutable player moves (payload, result_state, channel)
• Table game_events — lifecycle and rejected actions (game.started, game.finished, action.rejected, etc.)
• Retention 90 days (NFR-027)
• Export for dispute resolution via admin panel (/admin/games/{id}/log)

Admin log viewer

Platform operator accesses /admin in the web client after Telegram OTP login. See admin-panel.md.

NFR

NFR-024, NFR-025, NFR-026, NFR-027