eBPF Integration for Kubernetes Workloads

I’m extending our cloud-native platform with eBPF so we can observe what really happens inside Kubernetes without shipping sidecars or kernel modules. The goal is to load a custom eBPF program at runtime, attach it to selected hooks (syscalls, kprobes, tc, or XDP depending on the use-case we decide together), and surface the collected data through a REST/GRPC API that other services in the cluster can consume.

Scope of work
• Design the eBPF logic, write it in C (or Rust) and build it with clang/LLVM and libbpf.
• Provide a lightweight loader in Go that runs as a Kubernetes DaemonSet or Pod, handles feature probing, and safely pins maps.
• Wire the program’s metrics/events to an API endpoint: JSON over HTTP or GRPC is fine—whichever you prefer—as long as it can be queried from within the cluster.
• Supply Helm (or Kustomize) manifests so the whole stack can be deployed with a single command.
• Document kernel/version prerequisites and any tunables we need to set on the nodes.

Nice-to-have but not mandatory: experience with Kubernetes CNI, Prometheus/OpenTelemetry export, and writing CO-RE compatible eBPF.

If this sounds interesting, tell me how you’d approach the hook selection and what tooling you favour (BCC, bpftool, self-built CO-RE, etc.). I look forward to collaborating with someone who loves kernel-level problem solving as much as I do.