Configuring the logging module once, at the entry point

Goal

Make every module's log records reach the same, deliberately configured destination, with levels that operators can change without editing code.

Prerequisites

An application entry point (main function, ASGI factory) that runs before any module logs.

Steps

1. In every module: logger = logging.getLogger(__name__) and nothing else; no handlers, no basicConfig.
2. At the entry point, configure once: a handler (stream or file), a formatter (plain line or JSON), the root or project logger level, and propagate decisions for noisy third-party loggers.
3. Choose levels by audience: DEBUG for developers, INFO for normal operation events, WARNING for things that need attention, ERROR for failed operations.
4. Read the desired level from configuration so that operators can raise it temporarily.
5. Use logger.exception(...) inside except blocks to include the traceback; use lazy formatting (logger.info("user %s", user_id)) so that arguments are only rendered when the record is emitted.

Expected result

One line per event in one format; third-party libraries' records appear at the level you chose; no duplicated lines from handlers added twice.

Limits and test basis

The logging module is process-wide; two frameworks configuring it (for example, an SDK's root handler) must be reconciled explicitly. Structured output needs a formatter or a small custom handler. Guidance follows the cited HOWTO.