Logging

Logging Guidelines

This project follows the Wide Events pattern. Logs are optimized for querying, not writing.

Core Principle

Emit one comprehensive event per request per service containing all contextual information. Do not scatter log statements throughout your code.

Wrong mental model: Log what your code is doing Correct mental model: Log what happened to this request

Wide Event Structure

Each event should include:

{
  // Request metadata
  requestId: string,
  traceId: string,
  timestamp: Date,
  service: string,

  // HTTP details
  method: string,
  path: string,
  statusCode: number,
  durationMs: number,

  // User context
  userId?: string,
  subscriptionTier?: string,

  // Business context
  pollId?: string,
  spaceId?: string,
  featureFlags: Record<string, boolean>,

  // Error details (when applicable)
  errorType?: string,
  errorCode?: string,
  errorMessage?: string,
  isRetriable?: boolean,

  // Performance metrics
  dbQueryCount?: number,
  dbQueryDurationMs?: number
}

Implementation Pattern

Use middleware to build the event throughout the request lifecycle:

1. Initialize event at request start with request/service metadata
2. Enrich with user context after authentication
3. Add business context as processing occurs
4. Emit single event in finally block after request completes

Anti-Patterns

Never do these:

• Scattered console.log statements throughout code
• String-based log messages without structured data
• Low-context logs with only timestamp, level, and message
• Multiple log lines for a single request

Example of what NOT to do:

// BAD - scattered, low-context logs
console.log("Starting request");
console.log("User authenticated");
console.log("Fetching poll");
console.log("Request complete");

Example of correct approach:

// GOOD - single wide event with full context
logger.info({
  requestId: ctx.requestId,
  userId: ctx.user?.id,
  pollId: params.pollId,
  method: "GET",
  path: "/api/polls/:id",
  statusCode: 200,
  durationMs: 45,
  dbQueryCount: 2,
  dbQueryDurationMs: 12
});

Tail Sampling

When implementing log sampling, make decisions after request completion:

Always capture 100%:

• All errors (5xx status, exceptions)
• Requests exceeding p99 latency
• Pro/enterprise user requests
• Requests with feature flags under rollout

Sample remaining traffic: 1-5% of successful, fast requests

High-Cardinality Data

Include high-cardinality fields (userId, pollId, requestId). Modern observability tools handle this efficiently. Aim for 50+ fields per event. Include:

• All IDs involved in the request
• All feature flags evaluated
• All external service calls made
• Timing breakdown for significant operations
• User tier and account metadata