Keeping Envoy Gateway On, Even When Bad Things Happen
summary
Envoy Gateway keeps proxies serving by preserving the XDS cache, using Leader Election, Active-Passive Failover for auxiliary services, and Panic Recovery so a control-plane panic doesn’t drop the data plane. It prioritizes availability over consistency, supports HPA for control and data plane scaling, exposes metrics for panic alerts, and runs a resilience test suite in CI. Proposal: resource load shedding at ~80% memory; discussion tracked at issue 3860.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 0:42Envoy Gateway role
- 3:13Failure scenarios
- 3:36Leader pod failures
- 5:23Kubernetes API failures
- 7:10High availability for XRock
- 8:13Kubernetes upgrades and node failures
- 9:40Control plane crash prevention
- 10:45Misconfiguration handling
- 11:49Future improvements
- 12:33Resilience features
- 12:54Closing remarks
related talks
21mMaking Envoy Resilient to Sudden Increases in Load
33mGet to know Envoy - the modern proxy for cloud-native infrastructure - Baptiste Collard
25mBuilding Low Latency Topologies with Envoy
37mEvolution of Envoy AI Gateway
26mIntegrating Stateful Load Balancing At Scale In Envoy
15mFail Open, Fail Fast: Improving Envoy Resilience in Latency-Critical Systems
33mRunning Envoy as an Edge Proxy
06mSponsored Keynote: Rescue Agents From Prototype Purga...