Profile every process on a node with one DaemonSet and no code changes. How eBPF profilers work, which tool to pick, a tested Parca deployment, and the privileges, storage and trace-linking limits you need to plan around.

For profiling and network observability on a modern Linux kernel, yes. Cilium, Parca, Pixie and Tetragon all run in production. The caveats are kernel version and BTF support, the privileges the agent needs, and application-level tracing, which still trails what an SDK gives you.

Tracing tells you which path a request took and how long each hop took. Profiling tells you what your CPU was actually doing during those hops. They're complementary — use both.