Logging saves output such as errors, warnings and event information to persistent storage for debugging purposes.
Runtime exceptions that prevent code from running are important to log to investigate and fix the source of the problems. Informational and debugging logging also helps to understand how the application is performing even if code is working as intended.
Logging is often grouped into several categories:
Logging errors that occur while a web framework is running is crucial to understanding how your application is performing.
When you are running your application on several servers, it is helpful to have a monitoring tool called a "logging aggregator". You can configure your application to forward your system and application logs to one location that provides tools for viewing, searching, and monitoring logging events across your cluster.
Another advantage of log aggregation tools is they allow you to set up custom alerts and alarms so you can get notified when error rates breach a certain threshold.
Sentry started as a Django-only exception handling service but now has separate logging clients to cover almost all major languages and frameworks. It still works really well for Python-powered web applications and is often used in conjunction with other monitoring tools. Raven is open source Python client for Sentry.
Graylog2 provides a central server for log aggregation as well as a GUI for browsing and searching through log events. There are libraries for most major languages, including python. Saves data in Elasticache.
Logstash Similar to Graylog2, logstash offers features to programmatically configure log data workflows.
Scribe A project written by Facebook to aggregate logs. It's designed to run on multiple servers and scale with the rest of your cluster. Uses the Thrift messaging format so it can be used with any language.
Loggly is a third party cloud based application that aggregates logs. They have instructions for every major language, including python. It includes email alerting on custom searches.
Splunk offers third party cloud and self hosted solutions for event aggregation. It excels at searching and data mining any text based data.
Papertrail is similar to both Loggly and Splunk and provides integration with S3 for long term storage.
Raygun logs errors and provides immediate notification when issues arise.
Scalyr provides log aggregation, dashboards, alerts and search in a user interface on top of standard logs.
There is a hosted version of Sentry in case you do not have the time to set up the open source project yourself.
This intro to logging presents the Python logging module and how to use it.
A Brief Digression About Logging is a short post that gets Python logging up and running quickly.
Taking the pain out of Python logging shows a logging set up with uWSGI.
Good logging practice in Python shows how to use the standard library to log data from your application. Definitely worth a read as most applications do not log nearly enough output to help debuggin when things go wrong, or to determine if something is going wrong.
Django's 1.3 release brought unified logging into project configurations. This post shows how to set up logging in a project's settings.py file. Caktus Group also has a nice tutorial on central logging with graypy and Graylog2.
Django Logging Configuration: How the Default Settings Interfere with Yours explains a problem with the default Django logging configuration and what to do about in your project.
Exceptional Logging of Exceptions in Python shows how to log errors more accurately to pinpoint the problem instead of receiving generic exceptions in your logs.
Read how to integrate logging into your web application framework.
Ensure errors and anomalous results are logged. While these logs can be stored in monitoring solutions, it's best to have your own log storage location to debug issues as they arise to complement other monitoring systems.
Integrate logging for system events you may need to use for debugging purposes later. For example, you may want to know the return values on functions when they are above a certain threshold.