Following a packet journey through the networking stack in OpenShift
summary
A single-node OpenShift cluster with two pods (HTTP client and server behind a Service) is traced end-to-end using Retis to capture function-level events; the trace shows Open vSwitch producing a miss→upcall→flow insert roundtrip into userspace until a kernel data-path flow is cached, after which packets traverse without upcalls. The 10 KB response appears as only three–four visible frames thanks to GSO condensing many segments, and a network policy proves drops happen via an explicit OVS drop action.
AI-written summary · about
chapters (10)
- 0:00Packet journey in OpenShift
- 6:04TCP SYN packet tracing
- 10:45Open vSwitch miss events
- 14:22Packets through user space
- 18:45Subsequent TCP packets
- 22:37Config to kernel flows
- 24:05GSO packet optimization
- 25:50Network policy enforcement
- 27:17Key takeaways
- 29:55Q&A
related talks
35mWho Dropped the Packet? Solving K8s Network Mysteries in Real-Time
32mTracing the Linux networking stack with Retis
46mNetworking in Kubernetes: can we make it clearer?
33mWhose Packet Is It Anyway? Life of a Packet Through a Service Mesh
19mThe Easy Button to Kubernetes Networking with Open vSwitch
29mFinding the Needle in a Haystack: Tenant-Level Network...
26mRun your Kubernetes cluster on OpenStack in Production
51mHow Kubernetes Networking Really Works: A Packet’s Journey Across Pods and Nodes