Service Mesh: There and Back Again
Talkat KubeCon + CloudNativeCon North America 2019 · on Tuesday, 2019-11-19· at 10:55· Ballroom Sec 20CD - San Diego Convention Center, Upper Level·
summary
Nordstrom evaluates Linkerd on Kubernetes, builds a custom CNI plugin so proxies don’t need NET_ADMIN, and bakes that plugin into node images for reliability. Clusters hold ~6,680 pods (spiking to >12,000 and 36 TB RAM), one service reached 3,000 pods. Benchmarks show p90 latency ~3.70 ms at 500 RPS and ~4.23 ms at 5,000 RPS, pod memory stays near baseline. Identity is tied into AWS Private CA (cert-manager work planned); the mesh runs in production.
AI-written summary · about
chapters (14)
- 0:00Intro
- 1:28TABLE OF CONTENTS
- 1:54WHY SERVICE MESH?
- 5:16INVESTIGATIONS AND FINDINGS
- 9:52CHALLENGE ACCEPTED
- 10:54OUT OF THE FRYING PAN INTO THE FIRE
- 14:57THE FRONT GATE
- 15:59LINKERD INSTALL
- 17:45INSIDE INFORMATION
- 18:19CHALLENGES
- 21:01THE RETURN JOURNEY
- 21:12OPEN SOURCE COMMUNITY
- 25:58THE LAST STAGE
- 34:10THE END
related talks
01h 10mService Mesh Introduction
37mThe Service Mesh Wars: A New Hope for Kubernetes
36mOverview and State of Linkerd
12mService Mesh & Kubernetes Multicluster Communication With Linkerd
16mWhat Happens when Service Mesh Maintainers Get a Taste of Their Own Mesh?
34mService Mess to Service Mesh
39mDo you need a Service Mesh?
32mLinkerd and the Service Mesh