Autoscaling Multi-Cluster Observability with Thanos and Linkerd
summary
Linkerd provides zero-code metrics by injecting a Linkerd injected Rust proxy into each pod that exports request rate, success rate, latency; Prometheus scrapes those, Thanos uploads immutable blocks to object storage via a Thanos sidecar and exposes them through a Thanos querier. Demo runs four clusters (AWS, DO, AKS, GKE), each 100 rps, one cluster spiked to 2,000 rps causing global P99 to jump ~3→62 ms; Kubernetes HPAs (via Prometheus adapter) scale replicas (up to 10) and latency falls toward the 10 ms target, though not fully below during the demo.
AI-written summary · about
chapters (10)
- 0:00Start
- 0:21What is a service mesh?
- 4:15Linkerd Prometheus Endpoint
- 4:47Quick Prometheus intro
- 6:06Typical monitoring setup
- 8:28Anatomy of Thanos
- 10:39Thanos receive
- 13:32Store API is a building block
- 14:39Demo time! bit.ly/thanos-linkerd
- 16:01Autoscaling
related talks
25mMulti-Cluster Observability: What you need to survive with a fleet
26mPrometheus That Scales
36mRunning the Observability As a Service For Your Teams With Thanos
15mScaling Observability Stack for AIOPS
31mHandling Billions of Metrics with Prometheus and Thanos
24mTaming Observability at Scale in a Multi-Cluster Kubernetes Platform at Bloom...
41mHigh Available + Scalable Prometheus with Thanos in Alibaba
10mUpstream-First, High Scale Prometheus Ecosystem