Giving Away Your Secrets: Opening Metrics Up to Users
summary
Bloomberg shares about 200 metrics and publishes roughly 1 billion datapoints a day (peaks ~1M/min) by ingesting component metrics into a Prometheus-like central store, normalizing with an indirect access adapter, and distributing via Kafka-backed dashboards and a streaming API. Four approaches are compared: status dashboards, direct access (easy but leaks internal tags and prevents backwards-incompatible changes), bespoke metrics (duplication and drift), and indirect access (sanitizes and augments). Recommendation: start with nothing, add only meaningful, actionable fields, and prefer adapters over exposing raw internal data.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 1:58Sharing internal metrics
- 7:12Status dashboards
- 9:17Direct metrics access
- 12:51Bespoke metrics
- 15:41Indirect access via adapters
- 18:11What metrics to share and why
- 21:28Meaningful data
- 28:16Bloomberg’s case
- 37:24Key takeaways
related talks
48mMetrics All the Way: Data Driven Operations
01h 01mFinding Metrics that matter and using them safely
41mMetrics Applied
30mThe Challenges of Managing Open Source Infrastructure at Bloomberg
48mAPI Observability by Design: Building Systems That Tell Their Own Stories
36mMetrics - What's your code actually doing?
27mEasy App Metrics
29mMonitoring: Follow the Data from the Darkest Corners