Why Observability Is Hard and Expensive: Cardinality

Observability: the present and future, with Charity Majors

Topic 6 of 14

Published Jan 22, 2025 · Clip 26:55–38:34 (11:39)

Charity admits she always hated monitoring, then explains why observability is hard: software is hard, and writing observability requires imagining what future-you will need at 2 a.m. On cost, she cites system complexity, the multiplier effect of storing data in many tools, and cardinality: traditional tools force engineers to govern high-cardinality fields like request IDs, which are paradoxically the most valuable debugging data. She explains metrics-based time series storage's inability to handle cardinality and argues the solution is structured wide data in columnar stores—'fewer logs but wider logs'—enabling tools like Bubble Up to automatically identify outliers.