Deploy Parca Agent as a Kubernetes DaemonSet for continuous eBPF-based whole-system profiling and query the collected profiles via the Parca server's API/UI in pprof format

Steps

1. Deploy the Parca server into a dedicated namespace (kubectl create namespace parca; kubectl apply -f https://github.com/parca-dev/parca/releases/download/<version>/kubernetes-manifest.yaml) and port-forward service/parca on port 7070 to reach its UI/API.
2. Deploy Parca Agent as a DaemonSet in the same namespace via kubectl apply -f https://github.com/parca-dev/parca-agent/releases/download/<version>/kubernetes-manifest.yaml; the agent container must run privileged with access to /sys/kernel/debug and /sys/fs/bpf to load its eBPF programs.
3. Point the agent at the server by setting --remote-store-address=parca.<namespace>.svc:7070 (add --remote-store-insecure and --remote-store-insecure-skip-verify for plaintext/no-TLS test setups) and --node=$(NODE_NAME) so per-node identity matches the actual Kubernetes node name.
4. Verify the agent is discovering targets by port-forwarding its own HTTP server (default port 7071) and checking the 'active profilers' and 'processes' pages it exposes.
5. Optionally narrow what gets profiled by adding --config-path pointing at a ConfigMap-mounted parca-agent.yaml with Prometheus-style relabel_configs (note label names are sanitized, e.g. app.kubernetes.io/name becomes app_kubernetes_io_name).
6. Query collected profiles in the Parca web UI (http://<parca-svc>:7070) by picking a profile type such as CPU Samples and a label-selector query; use Merge for an aggregate-over-time flame graph or Compare for a differential flame graph, and export/download the underlying profile in pprof format via the QueryService API (gRPC, also exposed as REST via grpc-gateway) for use with go tool pprof or other pprof-compatible tooling.