eBPF Telemetry Pipeline and Jetson Node Discovery

eBPF Telemetry Pipeline Design

The lab topology requires a lightweight telemetry layer that doesn't add external agent overhead. An eBPF-based approach on chonk (10.10.8.60) provides host-level system call tracing with minimal CPU impact. The pipeline collects open(), connect(), and execve() events for all processes, aggregating connection metadata by PID namespace — this gives visibility into outbound DNS lookups and SSH sessions without requiring per-host agent deployment.

Jetson node90x Discovery Status

The four node900–node903 Jetson Nano 2GB hosts remain partially accessible. Today's connectivity audit:

• node900: SSH reachable on password fallback (franklin/123). Key material was revoked and not replaced.
• node901: Port 22 timeout — no response. Physical link check required.
• node902: Offline since 2026-08-27 per infrastructure notes. Port 22 timed out; node requires physical power/link inspection on the Jetson chassis.
• node903: SSH reachable but key authentication failing — password fallback works. Key state needs reconciliation.

GPU Inaccessibility Confirmed

As noted in the prior audit, all four Jetson hosts have fully inaccessible integrated GPUs: no nvidia-smi output, no OpenCL platforms detected, and no CUDA runtime available. The Tegra T210 SoC driver stack is either missing or blacklisted in the current kernel configuration. Remediation requires a kernel-level GPU module load check during ansible-test common role validation.

Log Aggregation Considerations

With Rsyslog already deployed on multiple lab hosts, the eBPF telemetry output can flow through existing log paths rather than requiring a separate collector. The BIND9 and ISC DHCP auditing work from yesterday converges here — centralized logs enable automated anomaly detection for both DNS queries and DHCP lease churn across the `10.10.x.0/21` subnet range.