Building a network topology of a Kubernetes application in a non-intrusive way

In Kubernetes, applications are separated into computational units (pods) and access points (services), where services handle load balancing using iptables or IPVS at L4 with basic round-robin, but for intricate request tracing and debugging, eBPF tools like gala-gopher can gather non-intrusive metrics; however, full request tracing requires augmenting the kernel with eBPF to insert TCP header options, as shown in FlowTracer, leveraging sock ops programs to dynamically enable address injection, revealing client and server correlatives without modifying application code, although adding and managing this capability demands careful handling of connection lifecycle events.