Wide Events

Wide events (also called canonical log lines) are Junando’s primary observability strategy. Every meaningful unit of work produces exactly one structured JSON line instead of scattered logger.info() calls.

Philosophy

Traditional logging scatters information across many lines: one for “dedup started”, one for “dedup finished”, one for “LLM request sent”, one for “notification delivered”. Correlating these requires matching timestamps or request IDs across multiple log streams.

Wide events invert this pattern. Instead of logging activity, you log results:

• One line per cluster at the end of processing
• Each pipeline stage writes its result (not its activity) into a shared builder
• The final event captures the complete chain: dedup → traces → LLM → notify

Benefits

Concern	Scattered Logs	Wide Events
Query cost	N LogQL queries per trace	1 LogQL query, filter by requestId
Missing context	Implicit — log lines are disconnected	Explicit — one event has the full chain
Volume	Unbounded per processing unit	Exactly 1 per unit (tail-sampled)
Debugging	Follow the breadcrumbs	Every answer is in one JSON object

Schema

interface WideEvent {
  requestId: string;
  component: Component;
  timestamp: string;
  correlationId?: string;
  outcome?: Outcome;
  durationMs?: number;

  cluster?: ClusterSection;
  dedup?: DedupSection;
  llm?: LlmSection;
  notify?: NotifySection;
  error?: ErrorSection;
}

Deep Detail

For complete documentation including the full schema, field taxonomy, tail sampling configuration, PII redaction, how to add new wide events, and LogQL query examples, see:

https://github.com/GerMoren/junando/blob/main/docs/WIDE-EVENTS.md

Feature Flag

Wide events can be disabled via the WIDE_EVENTS_ENABLED environment variable. Set it to false and redeploy to suppress wide event emission without changing code. The pipeline logic (dedup, LLM, notifier) runs identically either way.