Lightning Talk: Protecting Envoy: Overload Manager
summary
Envoy’s overload manager measures memory pressure and triggers concrete actions: scale timeouts (configured 60s→5s between 60% and 90% memory), drop expensive HTTP/2 streams, disable listeners, probabilistically return 503s, tell clients to disconnect, and ask tcmalloc to return memory to the OS. Under a Slowloris test (60 KB headers then 1 byte every 15s) static timeouts cause crashes around ~18k concurrent clients; with scale timeouts memory levels off and the proxy holds ~16k clients. Results work but require careful tuning per workload.
AI-written summary · about
chapters (7)
- 0:00Start
- 0:16Envoy Edge Deployments
- 0:41Envoy Multi-Tenant Deployments
- 3:37Stop Accepting Request
- 4:11Disabling Keep Alive: Fleet perspective co
- 5:18Experiment: Static Timers vs Slowloris
- 8:13Overload Manager: Caveats
related talks
10mLightning Talk: Envoy @ Netflix
25mBuilding Low Latency Topologies with Envoy
06mLivin' on the Edge (of your cluster)
26mRunning Envoy at the Edge
26mIntegrating Stateful Load Balancing At Scale In Envoy
30mLightning in the Cloud: A Study of Transient Bottlenecks on n-Tier Web Application
01mLightning Talk: Golden Rules of Resource Isolation in Istio
36mFine Grained Mesh Metrics for Better Visibility With Native Performance