How can you keep routine events out of an urgent-error log without losing them elsewhere? Give log records severity levels, then set thresholds on the logger and its handlers. This makes a growing log easier to search when something goes wrong. section 12.2 section 12.2.2
Python’s logging levels rise in severity: DEBUG, INFO, WARNING, ERROR, and CRITICAL. They are integer constants valued 10 through 50 in steps of 10. A logger set to WARNING accepts WARNING, ERROR, and CRITICAL records, but not INFO or DEBUG records. Use CRITICAL when a failure affects functionality essential to normal operation. section 12.2.1 section 12.2.4
Keep a broad warning log and a narrow critical log
Set the logger to DEBUG so it can accept all five standard levels, then give its two file handlers different thresholds: import logging logger = logging.getLogger(__name__) logger.setLevel(logging.DEBUG) warning_handler = logging.FileHandler("warnings.log") warning_handler.setLevel(logging.WARNING) logger.addHandler(warning_handler) critical_handler = logging.FileHandler("critical.log") critical_handler.setLevel(logging.CRITICAL) logger.addHandler(critical_handler) logger.info("Task created") logger.warning("Task update delayed") logger.critical("Cannot update task status") The warning file receives the warning and critical messages; the critical file receives only the critical message. Neither receives the INFO message. section 12.2.2
The logger checks a record before sending it to any handler. A handler set to DEBUG therefore cannot recover an INFO record rejected by a logger set to WARNING. If handlers need different views, keep the logger’s threshold inclusive enough for both. Format retained records with a timestamp, level, module name, and message so important events are easier to find. section 12.2.5 section 12.2.3
