LAB cutover phase 2 — agents home, nervous system still in the cloud
LAB cutover phase 2 — agents home, nervous system still in the cloud
Requested by: Young (DM, 2026-09-27): "Connect with LAB, all our agents should be on it now."
Method: off-box probes from the laptop (tailnet 100.94.188.127, LAN relay 192.168.4.60, old-server 100.124.146.127), cross-checked against Alpha/Ace reports. No destructive action.
Verdicts (each claim verified independently)
| # | Claim | Verdict | Evidence |
|---|---|---|---|
| 1 | All 8 fleet agents answer on LAB | ✅ verified | ACK sweep :8300–8307 all completed, 1.99–11.25s, glm-5.3-flash (2026-09-27 10:35) |
| 2 | Registry :8432 lists 11 agents (8 + eve/eva/tam) fresh | ✅ verified | LAB registry, seen <20s (offsite relays 11–267s) |
| 3 | Old-server agent stack + tailnet registry decommissioned | ✅ verified | 100.124.146.127:8300 POST = 000, :8432 = 000 |
| 4 | But old box still hosts a LIVE second board internally | ⚠️ confirmed | laptop tunnel forwards :8432/:8430/:8320 to old-box loopback — all answering 200/404 |
| 5 | The two boards diverge NOW, proven | ⚠️ confirmed | task-2a00d648 only on old board; task-12a7bca6 only on LAB board; task-83bf0aec only old board |
| 6 | Governance still on old box | ⚠️ confirmed | CostGov :8430 = 200, Conductor :8320 = 200 via loopback; central-cron :8436 dead everywhere probed |
| 7 | LAB units missing governance stack | ✅ Alpha-confirmed | no lab-costgov/cron/relay beyond mesh; eve-bridge restart-loop; keypool + eva/eve heartbeats dead |
| 8 | LAB public sites unproxied | ⚠️ confirmed | vault-docs 8091 / pulse 8092 loopback-only on LAB (my 404s = relay catch-all, not the site); domains still resolve to old box nginx (22/80/443 open) |
Client wiring still stale (the phase-2 work)
- laptop-tunnel.service + autossh —
ServerAlivetargetsubuntu@100.124.146.127; forwards 18300–18307 → old-box agents now dead (stack retired) and 8432/8430/8320 → old-box loopback survivors. Should re-point to LAB for the mesh + registry-of-record. - tam-heartbeat.service — posts every 5 min via fwd
:8433(maps to old:8432). Registry freshness for tam/eva currently lands on the old board only. - sync_vault.sh —
RHOST=ubuntu@100.124.146.127,RBASE=...vault/solos/research/tam: vault notes keep landing on the box awaiting redesign. Blocked until an SSH key for the laptop exists on LAB (sol@192.168.4.60:22rejects all current laptop keys — Alpha's own doc confirms). - Registry-of-record undeclared — Alpha: LAB board is a snapshot seed (prod
solos.db, 256 tasks/689 reports), prod was authoritative; but prod's tailnet registry is now dead, so in practice "the old loopback board" vs "LAB board" is the open choice. Sol owed this on task-83bf0aec.
Ace's merge design (recorded, not yet executed)
Single-DB cutover: freeze legacy writes → sqlite file copy → agents re-point via env → legacy warm 48h rollback. Cutover order per service: cron → bridges → agents.
Recommendation to Young
One decision unblocks everything: declare LAB :8432 the registry of record (Sol/Alpha own the call). Then re-point the three stale client wirings, grant the laptop an SSH key on LAB, and the redesign of the old box can proceed from a clean, fully-offline state. Until then, cross-file urgent items to BOTH boards (done for tasks 2a00d648/12a7bca6 with cross-link reports rep-04934c15 / rep-fd8ddcd2).