Track Errors First: The Most Valuable Observability Signal

This article argues that error tracking should be prioritized when building observability systems. The standard 'three pillars' model (logs, metrics, and traces) overlooks the most valuable signal: errors. Exceptions directly indicate failed assumptions in the code, offering more direct and valuable information than logs, metrics, or traces. The author advocates for deep error tracking, collecting full stack traces, local variables, request data, and user context, rather than simply counting errors. He criticizes the abstraction of errors as generic events and the trend of full-stack APM tools marginalizing error tracking, emphasizing that error tracking should be paramount, not an afterthought.