PFVCluster/questions-09022206.md

Working ask-file; knowledge docs live on Discourse: https://community.turnsys.com/c/vp-techops (audit log: https://community.turnsys.com/t/298)

questions-09022206.md — consolidated open asks (2026-09-02)

Merged from questions-v2.md (repo root) + HomeAssistant/needfromcharles.md per founder request; both source files since removed (history in git). Answer inline; agent checks before asking again.

HA / monitoring (was needfromcharles)

1. Minisplit (Midea/Carrier) — blocks #614

Login charles@turnsys.com rejected by all three Midea cloud backends.

• Exact app name on your phone that controls the minisplit:
• Account email that app uses:

2. Govee — blocked on hardware path (#620/#683)

Integration live (key+account+MQTT); account has ZERO cloud devices = BLE-only.

• Sensor model numbers (decides BLE dongle vs hub):
• API key — received + configured (2026-09-02)

3. HomeKit codes — blocks #619 pairing (ecobee multicast-reachable NOW)

• ecobee 8-digit code:
• Lock brand/model: (verdict 2026-09-02: BLE-only — DHCP can't name it)
• Lock HomeKit code:

4. Emporia Vue 3 — blocks #623 (BTU loop)

• Ordered? ETA:

9. iDRAC — RESOLVED 2026-09-02

[x] Creds worked; both SNMP agents live; tsys6 public accepted by founder.

NEW (2026-09-02)

10. DVR → MotionEye (#711)

• DVR login (any user):
• Channel count: (Night Owl; RTSP format documented on #711)

11. DIRIGERA Matter code (#712) — when you're back at the house

• IKEA app → hub → integrations → share via Matter (11-digit code)

12. WiZ bulb→room mapping (#712) — Friday onsite

2 bulbs reserved (pfv-wiz-bulb-1/2 @ 192.168.1.150/.151); map MACs→rooms, then HA WiZ entries by host + we identify the other ~11 lights' integration.

General (was questions-v2)

Q7. Close #167 (old garage UPS rebuild, superseded)? — still open

Q9. Session-start gate relaxation — superseded by scope-alignment gate

Q10. Git flow: keep direct-push for PFVCluster? — still open (default: keep)

(Q8 Friday order — moot, that Friday passed)

Founder decisions pending

• #626 Cloudron HA access architecture (a/b/c on ticket — recommend proxy)
• #684 PBS session scheduling (+ drive-rotation target design)
• TasmoAdmin uninstall (stopped+disabled 2026-09-02 — full uninstall?)
• Govee disabled dup entry: one-click delete in HA UI

13. Cloudron API token (#685/#727) — unblocks API-first access

The #685 perf investigation ran over raw SSH (root@my.knownelement.com) because no panel API token exists. A token (read scope minimum) lets the agent use the Cloudron API (apps/configs/backups) instead of shell — cleaner audit trail and the path for AWX/Ansible modules. Also tracked in KNEL/cloudron questions-v1.md (https://git.knownelement.com/KNEL/cloudron).

• Create token (panel: my.knownelement.com → profile → API access) — received 2026-09-02
• Material value stored → ~/.creds/cloudron.env (0600; API verified: profile mrcharles, apps 200)

14. docker daemon.json bounce window — my.knownelement.com (#727)

One planned dockerd restart (ALL ~300 containers bounce; live-restore=false today, so no zero-downtime path) to apply nofile 262144 defaults + log rotation + live-restore. Script ready: KNEL/cloudron perf/docker-daemon.json.planned. Until then the postgres nofile bridge is ephemeral (lost on postgres container recreate). Detail: t/316.

• Pick a window — DONE 2026-09-02 ~19:30 CDT: applied (live-restore + nofile 262144 defaults); first-attempt log-driver conflict incident documented on #731

CMDB change-control rulings needed (#705 P0 live → P2/P3)

C1. CR itemtype for the 7-state workflow

GLPI natively has both Ticket and Change. The t/319 workflow (Draft→Review→Approved→Scheduled→Implemented→Verified→Closed) maps most cleanly onto GLPI Changes (validation/plan/approval fields built in), with Tickets for break-fix. Recommend: Changes for CRs, Tickets stay for incidents; both linked to the CI they touch.

C2. Write access for the cmdb agent in the CR phase

Agent is Read-Only today (verified). If CRs live in GLPI, the agent needs to CREATE/transition them — recommend adding the Self-Service profile alongside Read-Only (can open + follow own items, still can't edit CIs). Alternative: CRs enter via Redmine only and GLPI stays read-only for now.

C3. #307 → GLPI refresh cadence

from-inventory.sh regenerates the CSV from the Discourse #307 raw; to-glpi.sh is idempotent by name. Recommend folding "refresh #307 → regenerate → import → report diff" into the daily 09:00 C2 housekeeping. Related data-quality: #307 §5 has pfvsvrki.knel.net typo; 14 stale-dns-record CIs need reconcile-or-delete ruling.

Agent-stack rulings needed (#767 — after TSGCOO tree discovery 2026-09-03)

C4. Credentials policy — one statement, please

Three coexisting models: Bitwarden (TSGCOO's bw-run.sh design — replaces ~/.creds for agent/system creds), /.creds 0600 (current house rules), Vaultwarden (your app-secrets ruling). Proposed: Bitwarden = agent + system API credentials going forward (/.creds deprecated as #478 ingests them), Vaultwarden stays for app-internal secrets. Confirm?

C5. #767 scope split

This chat owns the CR escalation ladder + GLPI Changes wiring + mechanical gates (ticket-gate CR check, X-Consumer attribution); TSGCOO's lane owns identity provisioning execution (#442/#478). Confirm the seam?

C6. Resume TSGCOO's blockers?

Q1-Q6 in org-buildout/questions-v1.md + #478 stage-2 re-gate. I can clear Q1 (docker group for TSGCOO) now — say the word. Invite links (Q3) + Gitea token (Q4) are yours.

C7. Leaked vptechops Gitea token

Listed in a KNELSecretsManager remote URL — rotate at your convenience (your console, not my lane).

C8. GLPI Changes vs Tickets (was C1)

Still the blocking ruling for the whole CR wiring — see C1 above (recommend Changes).