Backend Development › Backend Basics · also in Observability
Log Levels
DEBUG, INFO, WARN and ERROR, and when to use each.
Also known as: logging levels, DEBUG INFO WARN ERROR, severity levels, log severity
Log levels label each log message with its importance, so you can filter. In production you might keep only the important ones; when debugging, you turn on the detailed ones.
| Level | Meaning | Example |
|---|---|---|
| DEBUG | Detail for developers diagnosing a problem | Fetched 3 rows for user 42 |
| INFO | Normal, notable events | Order 917 created, Server started on :8000 |
| WARNING (WARN) | Something unexpected, but handled; might need attention | Payment provider slow, retrying (attempt 2) |
| ERROR | An operation failed | Could not save order 917: database timeout |
| CRITICAL (FATAL) | The application can’t continue | Cannot connect to the database; exiting |
(Names vary slightly by language: Python has WARNING and CRITICAL, others WARN and FATAL.)
import logging
log = logging.getLogger(__name__)
log.debug("cache miss for key %s", key)
log.info("order %s created", order.id)
log.warning("retrying payment (attempt %d)", attempt)
log.error("failed to save order %s", order.id, exc_info=True)Choosing a level
- ERROR should mean “someone may need to look at this.” If errors are constant noise, people stop reading them, and the real ones get lost. A user typing a wrong password isn’t an ERROR. It’s normal (INFO or WARNING, or nothing at all).
- A failure you handled (a retry that succeeded) is a WARNING, not an ERROR.
- INFO should tell the story of the app: lifecycle events and business actions, without every detail.
- DEBUG can be verbose, because it’s off in production by default.
Configuring
Set the minimum level through configuration, such as LOG_LEVEL=INFO
(configuration). Anything below it is dropped. That lets you raise detail temporarily on a
problem without redeploying code.
Don’t log secrets or personal data at any level. Levels are also what alerts are often built on, for example “alert on ERROR rate”. See logging and structured logging.