Scheduling 68k COVID Tests with Linkerd
summary
Penn State builds a microservice scheduling system, sends 68,000 invites and hits a real load spike; observability from Linkerd plus Grafana and Prometheus shows high outbound latency to an on‑prem dependency. Load testing with k6 reproduces ~50–60s p95/p99 responses; root cause is an unnecessary synchronous check to a single‑threaded service (on‑prem) that scaling helps but removing the check drops p99 to ~2s and restores acceptable user latency.
AI-written summary · about
chapters (9)
- 0:00Intro
- 4:52Cluster Layout
- 6:35Terminal Setup
- 9:45Monitoring
- 14:47Testing
- 17:00Troubleshooting
- 18:59Final Changes
- 20:37Dashboards
- 23:52Summary
related talks
40mDeep Dive: Linkerd
44mDeep Dive talk | Resilience in Microservices with Congestion Control
34mFrom Monolith to Microservices with Kubernetes and Linkerd
39mGrow with Less Pains - Meshing From Monolith to Microservices
53mHow do we build high‑load microservices
36mWhat WePay Learned From Processing Billions of Dollars on GKE Using Linkerd
25mUse Envoy + Knative to auto scale Java RPC micro service
01h 01mBreaking your REST APIs to make them stronger!!