A systematic audit of the lab.bitsmasher.net subnet mesh revealed that approximately 40% of hosts have become unreachable or lost key-based authentication within weeks of initial provisioning. This decay is driven by stale host keys, password fallback chains on Jetson devices, and DNS resolution gaps for decommissioned nodes. Documenting this decay curve and its root causes provides a template for infrastructure self-healing in heterogeneous lab environments where automated recovery is not yet available.

The Decay Curve: 24 Hosts Down to 7 Active

TIMELINE OF HOST STATUS (as of Aug 20, 2026):

 Jul 31: ~24 hosts discovered via /etc/hosts inventory
 Aug  1: SSH Role Audit — active working set established
 Aug 20: Current audit — active working set reduced to 7

HOSTS BY STATUS:
 [ACTIVE]     chonk, skynet, wonderland, time, ldap/bbb1, kdc1, node1
 [AUTH FAIL]  stargate (key auth failing), blowfish (no matching key)
 [UNREACH]    dream-machine, edge-t, netlab1/2, ns1, node3/5/6
              thelio, head2, femputer, bbb1, server3
 [REMOVED]   snowy (disk removed), node902

The decay is not linear. Most losses occurred in a concentrated window during July-to-August host reconfigurations, where SSH key mismatches cascaded across nodes that shared common key deployment procedures.

Failure Mode Taxonomy: Bad Keys, Auth Denied, Unreachable, Removed

DECAY CATEGORIES:

 BAD HOST KEY (TCP open, auth rejected):
   • dream-machine (10.10.8.1) — host key changed without update
   • edge-t (10.10.16.4) — re-provisioned with new host key

 AUTH DENIED (valid connection, wrong credentials):
   • blowfish (10.10.14.85) — no matching SSH key in authorized_keys
   • music (direct, 10.10.8.2) — unreachable from chonk, reachable via skynet

 UNREACHABLE (no TCP response):
   • netlab1/2 (10.10.8.3-4) — powered off or firewalled
   • ns1 (10.10.12.12) — DNS server unreachable
   • node5/6, thelio, head2, femputer, bbb1

 REMOVED FROM SERVICE:
   • snowy — hard disk physically removed
   • node902 — powered off permanently

Jetson node90x Cluster Forensics: Revoked Keys and Zero GPU Access

The three-node Jetson Nano cluster (node900-903) presents a distinct failure pattern. All devices are NVIDIA Jetson Nano 2GB Developer Kits (Tegra T210 SoC, Cortex-A57 quad-core) with ed25519 keys revoked and password-only authentication remaining:

node900: 10.10.12.90 — franklin/123 — eMMC 63%, uptime ~106d
 node901: 10.10.12.91 — franklin/123 — eMMC 23%, uptime ~106d  
 node903: 10.10.12.93 — franklin/123 — eMMC 50%, uptime ~19d

 GPU STATUS: No nvidia-smi, no OpenCL, no pytorch. CUDA not installed.
 NOTE: Tegra integrated GPU exists on SoC but is inaccessible from host OS.

The password fallback chain (franklin/123) is a critical security vulnerability — these devices have no ed25519 keyauth and rely on a shared weak password. This makes them high-value targets for lateral movement if any single node is compromised. GPU access was never materialized because the NVIDIA stack (driver, CUDA toolkit) was never installed on the Jetson OS images.

stargate Key Restoration: The Single Point of Cascading Failure

Among auth-failed hosts, stargate is the highest priority. It runs nfs-kernel-server and exports /mnt/clusterfs2 — it's the reference architecture for the lab's NFS landscape. SSH key authentication failed as of Aug 20 2026; password auth is disabled. The authorized_keys file needs verification and restoration.

A successful key restore on stargate would immediately improve three downstream services: the Ansible Molecule test harnesses, NFS clusterfs2 management, and general lab automation that depends on stargate as an intermediate hop.

Ansible Molecule Testing as Decay Prevention

The existing Ansible Molecule testing infrastructure covers 37 roles with test harnesses running on stargate. The cron job fires at 03:00 MDT but times out (600s is insufficient). DNS role tests confirmed working; NFS role has hostname gate issues that need fixing.

CURRENT MOLECULE STATUS:
 Coverage: 37 roles on stargate
 Schedule: daily at 03:00 MDT (times out at 600s)
 Working: DNS role confirmed
 Broken: NFS role hostname gate issues

 REMEDIATION: Increase timeout to 1200s, fix NFS hostname gates.
 Deploy as heartbeat check target for ongoing decay detection.

Molecule test results should be wired into the heartbeat monitoring pipeline (currently tracked in HEARTBEAT.md) so host connectivity failures trigger immediate alerts rather than waiting for manual inventory audits.

Remediation Playbook: Priority Order, DNS Cleanup, and Recovery Windows

PRIORITY ORDER FOR RECOVERY:

 P1 (today):    stargate key restore — enables Ansible molecule fix
                 DNS cleanup for removed hosts (snowy)
                 node90x cluster: deploy ed25519 keys, install nvidia-utils

 P2 (this week): blowfish key deployment (auth denied)
                 ns1/edge-t host key regeneration
                 Node reachability sweep via skynet (known-good SSH hop)

 P3 (ongoing):  Molecule test timeout fix + heartbeat integration
                 Automated key rotation across all working hosts
                 Decommissioned host DNS entry removal

The most effective single action is the stargate key restoration, which restores the chain: stargate → Ansible molecule → automated host verification → early decay detection before it compounds.