Production

Create one base logger for the process, derive child loggers for stable application context, and let the observer add run-specific fields.

const appLogger = createPinoLogger({
  name: 'support-worker',
  level: process.env.LOG_LEVEL === 'debug' ? 'debug' : 'info',
})

const tenantLogger = appLogger.child({ tenantId })
const observer = createLoggerObserver({ logger: tenantLogger })

For process-local newline-delimited JSON with crash-resistant synchronous writes:

const fileLogger = createPinoLogger({
  filePath: './var/log/anvia.ndjson',
})

await fileLogger.flush()

sync and parent-directory creation default to true; files are appended by default. Set sync: false for higher throughput and call flush() during graceful shutdown.

Operational checklist

• Send JSON logs to stdout unless the hosting platform requires another destination.
• Keep payload capture disabled in production by default.
• Set retention, sampling, and transport backpressure in the logging platform.
• Do not treat successful log writes as proof that a run completed; observe the run-end record.
• Use trace identifiers to correlate logs with an external tracing system.
• Test log serialization with real error causes and large tool arguments.

Logging methods return void; the observer does not wait for a remote sink. Both built-in factories return a FlushableLogger. Custom Logger implementations may expose the optional flush() method; durable delivery remains the selected transport's and process lifecycle's responsibility.