Logging

Pattern: Explicit Constructor Injection

Loggers are passed explicitly via constructor injection throughout the codebase. There is no global/ambient logger.

import type { Logger } from '@repo/shared';

class MyService {
  constructor(private logger: Logger) {}

  async doWork(context: WorkContext) {
    const childLogger = this.logger.child({ operation: 'work' });
    childLogger.info('Working', { context });
  }
}

Child loggers

Use logger.child({ ... }) to attach structured context that will appear on every log line from that child. Prefer child loggers at the boundary of a unit of work (request, operation, session) rather than re-passing context on every call.

Configuration

Two environment variables, both read once at startup:

Use json in production (machine-parseable) and pretty for local dev.

In Tests

Use createNoOpLogger() from @repo/shared to silence logging in tests:

import { createNoOpLogger } from '@repo/shared';

const service = new MyService(createNoOpLogger());

Don't construct real loggers in unit tests — they add noise and can mask real failures with log output.

When Adding Logs

• Log at info for significant lifecycle events (operation started/completed)
• Log at debug for fine-grained tracing (request bodies, intermediate state)
• Log at warn for recoverable anomalies
• Log at error for failures that surface to the caller; include the error object as structured context: logger.error('Failed', { err })
• Pass structured context as the second argument, not via string interpolation