So you think you can mesh? Scaling ING's service mesh
summary
ING scales a service mesh from 1 to 10,000 running instances over eight years, supporting 2,000 microservices across containers and VMs; operational pieces include a control plane for API endpoint discovery, an API marketplace with JWT subscriptions enforcing fine‑grained authorization, mutual TLS for authentication, and a choice of library vs sidecar for heterogeneous runtimes. A regression test suite verifies Envoy rollouts and TLS behavior, disaster‑recovery steering between regions is automated, and the approach reports clear production success.
AI-written summary · about
chapters (14)
- 0:00Introduction
- 1:05What is a service mesh
- 3:13Monolith to microservices
- 7:53Service mesh architecture
- 10:42ING governance and API marketplace
- 15:13Auth, authorization, subscriptions
- 17:41Control plane runtime config
- 18:45Data plane security observability
- 20:54Library vs sidecar choices
- 21:34Adoption with docs and community
- 23:37Automating certificates
- 25:03Evolving service mesh
- 27:08Service mesh abstraction
- 27:51Lessons learned and closing
related talks
45mSo you think you can mesh? Scaling ING's service mesh
33mHow ING Accelerated Kubernetes Deployments Across 500+ Apps
41mService Mesh beyond single clusters
34mService Mesh for Developers
34mContainer Hosting Journey
43mService Mesh and the future of networking
11mWelcome & Opening Remarks
01h 03mJust what is a "service mesh", and if I get one will it make everything OK?
