Linkerd Salon
summary
Linkerd runs production traffic at NCBI across about 2,000 services and handles roughly half of external traffic from ~17 requests/sec, using a host-level proxy and a DNS-like dtab routing scheme to enable dynamic per-request routing (change a single request to a different service version via a header for safe testing). HomeAway runs Linkerd for ~12 months across hundreds of instances and thousands of apps on three continents. Tabula uses a ~10-line Linkerd config for gRPC TensorFlow Serving. Conduit builds a Kubernetes control plane with a Rust proxy, Prometheus telemetry, tap, and iptables-based sidecar injection.
AI-written summary · about
chapters (7)
- 0:00Conference welcome
- 1:03NCBI’s Linkerd in production
- 12:01HomeAway’s Linkerd migration
- 20:59Tabula’s Linkerd journey
- 24:57Conduit architecture deep dive
- 44:51Linkerd and Conduit AMA
- 1:02:17Closing remarks
related talks
37mLinkerd End User Panel: Case Studies from Production
35mDeep Dive: Linkerd
32mLinkerd and the Service Mesh
35mIntro: Linkerd
36mOverview and State of Linkerd
19mReal-time Kubernetes: How Entain Australia 10x'd throughput with Linkerd
39mService Mesh: There and Back Again
09mLinkerd vs. COVID-19: Addressing the global Pandemic with a Service Mesh