Many of Heroku's internal components make heavy use of logfmt to log information about what's going on in production. The format is hugely valuable in that it allows us to retroactively analyze what happened during any arbitrary request to our components, query our log traces in very flexible ways, and combined with Splunk, easily generate arbitrary metrics on historical data. It's unquestionably been an invaluable tool for fixing countless bugs, tracking down the root cause of many production incidents, and assessing usage in ways that would have been difficult otherwise.
Subscribe to the full-text RSS feed for Brandur Leach.