What happens when Istio's meets the reality of 100,000 requests per second? This talk shares hard-won lessons from a high-scale migration, focusing on the tricky parts the documentation doesn't cover.
This session dissects the solutions for unpredictable latency and dropped connections, detailing the importance of edge connection management and targeted kernel tuning using Kubernetes sysctl support. It further explores methods to achieve graceful connection termination to prevent user-facing errors during deployments and node refreshes. Finally, the talk demonstrates a method for integrating A/B experimentation at the edge, a technique that allows for real-time, data-driven comparison of user experience between old and new infrastructure.