The Internals of Metric Store
summary
Metric Store is a durable time‑series DB that writes points to disk using the InfluxDB 1.x storage engine, serves Prometheus endpoints with a PromQL-compatible query engine, and enforces access via the CF Auth Proxy with offline token validation and cached CC lookups. Ingress uses a nozzle from Loggregator (internal process comms use lightweight TLV frames, not gRPC), timers are downsampled (reducing noisy timer data by ~90%), shards are one day with a default 42‑day retention, and multi‑node HA/replication work is underway.
AI-written summary · about
chapters (12)
- 0:00Intro
- 1:24Metric Store Overview
- 4:14Metric Store Processes
- 5:34Nozzle
- 8:35What does a Point look like?
- 9:48Auth Proxy
- 12:09Metric Store - Persistence
- 14:57Metric Store - Indexing
- 17:18Metric Store - Querying
- 19:04Prometheus Interfaces - Storage
- 23:56Available Query Endpoints
- 25:36Future Work
related talks
27mMetric Store
40mInfluxDB storage engine internals
31mHow the Metrics Backend Works at Datadog
26mIntegrating Prometheus and InfluxDB
36mObservability with Prometheus and Beyond: Leveraging Cloud-native Technologies Outside of Cloud
14mMetrics Without Limits
30mScaling Time-Series Data to Infinity: A Kubernetes-Powered Solution with Envoy
30mIntro: Prometheus