Istio in Production: Expected Value, Results, and Effort
summary
GitHub runs Istio across its platform (about 130 clusters, ~250k pods) and a github.com mesh of 21 clusters with ~110k pods, enforcing mTLS and centralized authN/authZ. Engineers build automation CRDs and Kubernetes controllers to onboard workloads without code changes, and offer custom ingress gateways per team for high-volume traffic. Traffic management enables staged canaries (staff → ramped users) with instant rollback; least-connection load balancing trims ~15 ms latency. A hybrid topology yields ~70% control-plane load reduction; an observability ingress cut dropped 15 million spans/min lost to zero.
AI-written summary · about
chapters (9)
- 0:00Istio at GitHub
- 1:03Kubernetes and mesh scale
- 3:26Istio solves networking gaps
- 7:00Designing Istio mesh
- 11:58Abstracting Istio for teams
- 14:29Istio performance gains
- 17:20Challenges scaling Istio
- 21:14Future of Istio at GitHub
- 23:19Q&A
related talks
38mIstio in Production: Real Customers, Real Solutions
28mIstio in Action: Taking Kubernetes Traffic Control to the Next Level
36mExperiences From Running Istio in a K8s Production Environment
35mIstio: The Past, Present and Future of the Project and Communi...
01h 01mMeshing Around with Istio
36mIstio: Real world scenarios
43mService Mesh for developers
25mIstio for Node Developers