Boost Kubernetes Ingress Performance: Tuning Nginx Keep‑Alive for Double QPS

Background: A business application running in Kubernetes delivers over 100,000 QPS when exposed through a Service NodePort, but only about 50,000 QPS when exposed via an Ingress controller.

Parameter optimization: The Ingress controller is based on Nginx; improving Nginx’s keep‑alive settings can close the performance gap. See the official Nginx ConfigMap documentation for reference.

Key Nginx upstream parameters to adjust:

keep-alive: "100"
keep-alive-requests: "110"
upstream-keepalive-connections: "20000"
upstream-keepalive-requests: "110"
upstream-keepalive-timeout: "100"

What is Keep‑Alive mode? In HTTP, the non‑Keep‑Alive mode creates a new TCP connection for each request/response pair, while Keep‑Alive (persistent connection) reuses the same connection for subsequent requests, eliminating the overhead of repeatedly establishing connections.

Reason for the performance boost: Enabling the upstream keep‑alive parameters makes Nginx reuse TCP connections to the Kubernetes Service, avoiding connection‑setup costs and allowing much higher request throughput.

How to achieve high availability for nginx‑ingress‑controller

The ingress controller is deployed as a DaemonSet with a nodeSelector on two worker nodes, sharing the host’s IP address. Combining Nginx keep‑alive with LVS (Linux Virtual Server) provides high availability for the ingress controller.