How We Moved Spotify To a Proxyless gRPC Service Mesh
Talkat KubeCon + CloudNativeCon Europe 2025 · on Thursday, 2025-04-03· at 14:15· Level 1 | Hall Entrance N10 | Room E·
summary
Spotify replaces DNS-based discovery with a proxyless gRPC mesh using XDS, scaling to 1.5 million Kubernetes pods, and builds traffic splitting, zone-aware locality load balancing, and a live service call graph. Control plane integrates XDS into Nameless (Shameless) and Argo CD/Rollouts, XDS/ADS is implemented from scratch, and clients run dual channels with a DNS TXT flag to gate rollout. Gradual onboarding moves hundreds of services over ~6 weeks, rollback resets via panic-button DNS fallback in ~3–4 minutes; bootstrapping flakiness and dual-system complexity remain.
AI-written summary · about
chapters (10)
- 0:00Spotify's service mesh migration
- 3:52DNS-based discovery limits
- 5:17Evaluating service mesh solutions
- 8:49XDS for service discovery
- 12:25XDS features
- 13:06Nameless redesign for XDS
- 17:28XDS implementation
- 21:21XDS rollout strategy
- 26:39Rollout challenges and lessons
- 28:49Future plans
related talks
24mgRPC Proxyless Service Mesh
36mHow Spotify Migrated Ingress HTTP Systems to Envoy
24mHow Spotify Migrated Ingress HTTP Systems to Envoy
25mAdapting Our XDS Control Plane for Proxyless gRPC
35mIntelligent Load Balancing in Kubernetes
18mHigh Performance Service Meshes with gRPC Proxyless
23mxDS Support in gRPC
22mDos and Don’ts: Designing an XDS Backend for the LinkedIn Scale