Levandor

Session handover — 2026-08-09/10 (overnight — AirPlay, Frigate watchdog, UPS hardening, master agent, C302 FaceTime cam, Drive pipeline)

5 min read

Session handover — 2026-08-09/10

Single entry point for the 2026-08-09/10 overnight session on telep-mainframe + the tailnet. Every thread below already has a detailed dated note — this ties them together, states current status per subsystem, and lists what still needs a human. Canonical living runbook: telep-mainframe-handover. Rolling context: LOG / TOPICS / SESSION-HANDOVER.

Access

/usr/bin/ssh levander@telep-mainframe (bare ssh is broken on this Mac). Router root@telep-router over Tailscale. Box auto-powers-on after mains events (BIOS “Restore on AC Power Loss” = ON).

Status at a glance

SubsystemState
AirPlay receiver (telep-tv)✅ fixed — avahi locked to enp5s0, verified cast
Cameras / Frigate✅ healthy — new FPS watchdog auto-heals both jam modes
Camera OSD clocks✅ correct — DST-aware ONVIF push, self-corrects across DST
Power / UPS resilience✅ hardened — full power-off now self-recovers on boot
Knowledge base✅ live — was crash-looping (Qdrant shed), restarted + hardened
Master control-plane agent✅ LIVE on host — phone-drivable, permissions ACTIVE
Stirling-PDF✅ deployed on own tag:telep node — ⚠️ no vault note (doc gap)
C302 tailnet FaceTime cam✅ working — go2rtc P2P + WebRTC → OBS, air-gap restored on Tapo cams
Top képek → Drive pipeline✅ live — self-serve crop + Drive push in the picker
North camera .139 (cam3/cam4)⚠️ dropping frames — weak 2.4 GHz link, needs Ethernet
HP LaserJet M203dw⚠️ OFF the network — needs power-on / Ethernet
Needs you⚠️ see below

Open items — needs you

Human action required

  • Printer — power it on / put it on Ethernet (it’s offline: no DHCP lease / ARP / mDNS; it drops WiFi). CUPS is healthy — once it’s on the network it just works. No software fix.
  • DHCP reservations to add on telep-router — the printer (so it stops dropping off) and optionally the C302 (192.168.30.168).
  • Ethernet to the north camera (.139) — the real fix for its frame drops (−65 dBm vs −40 dBm on .119; diagnosed as link, not software).
  • go2rtc-cam WebUI — move to its own tag:telep node per tailnet-service-exposure-convention (currently on the shared node via tailscale serve --https=8450).
  • Stirling-PDF — no dedicated vault note yet (doc gap — see below).
  • Cosmetic/low: master-agent app session display-name; top-kép archive numbering only partial (~24/72).
  • Longstanding (from prior handover): 3rd-GPU PCIe cable; a dedicated mains circuit for the space.

What happened tonight

Fixes

  • AirPlay dead after the UPS switch — avahi was broadcasting telep-tv on docker/veth/loopback interfaces (bogus records) + a DHCP IP-flap on boot dropped the mDNS group. Locked avahi to enp5s0; verified a real cast. → 2026-08-04-telep-mainframe-airplay-receiver-uxplay (2026-08-08 fix section).
  • Frigate detect-jam + a frozen camera feed (twice) — manual docker restart frigate each time, then built frigate-fps-watchdog.service that auto-heals BOTH modes (frozen feed camera_fps<1.5 + detection jam process_fps<1.0) with debounce/cooldown and a 3-per-hour circuit breaker. → 2026-08-08-frigate-fps-watchdog.
  • Knowledge base crash-looping — kb-qdrant was shed overnight on battery and never restarted; restarted it, then hardened the whole UPS shed/restore cycle (boot-time reconcile power-restore-reconcile.service, kb-qdrant --restart unless-stopped, KB wait-qdrant drop-in) so a full power-off recovers. → 2026-08-08-ups-shed-restore-hardening.
  • Camera OSD clock 1 h off — set-cam-time.py sent UTC + a DST-aware TZ the Tapo silently ignores; rewrote it to push DST-aware Europe/Budapest local time with GMT0 so the cam mirrors it verbatim. Self-corrects across DST. → 2026-08-08-camera-osd-clock-drift-onvif-timesync.
  • .139 (cam3/cam4) dropping frames — diagnosed as a weak 2.4 GHz link (−65 dBm vs −40 dBm on .119), NOT software. Standing recommendation: run Ethernet to the north camera. Context: 2026-07-31-camwall-lag-24ghz-wifi-bufferbloat.
  • camwall bottom-left showed the birdseye — re-rolled with systemctl restart camwall.
  • HP LaserJet M203dw won’t print — it’s simply off the network (no DHCP lease / ARP / mDNS; drops WiFi). CUPS is healthy. Needs power-on / Ethernet. No software fix.

New services / deployments

  • Master control-plane agent — always-on claude remote-control on the host, phone-drivable, systemd-autostarted; vault cloned to /home/levander/obsidian with obsidian-sync.timer; ported historian + obsidian-documenter agent defs. Permissions stay ACTIVE (no --dangerously-skip-permissions); only trust dialogs pre-seeded. Write deploy key added to private wowjeeez/obsidian. → 2026-08-08-telep-master-agent-design + 2026-08-08-telep-master-agent-impl.
  • Stirling-PDF — self-hosted on its own tagged tag:telep node pdf.taild4189d.ts.net (login disabled, Hungarian default, update-nag off); home-portal tile added. No dedicated note yet — flagged as a doc gap.
  • Xiaomi C302 as a tailnet FaceTime cam — go2rtc P2P + WebRTC → OBS, red presence banner, firewall scoped to just its IP, relay power-cut. → 2026-08-09-xiaomi-c302-tailnet-facetime-cam.
  • Top képek → Drive curation/print pipeline — picker jump-to-time + Drive buttons + bulk helpers. → 2026-08-10-top-kepek-drive-curation-print-pipeline.
  • Tailnet exposure convention — a dedicated tagged node per service. → tailnet-service-exposure-convention.
  • Home-portal tile swap — Szemantika → Képosztályozás. → 2026-07-24-global-dashboard.

Router (telep-router) changes

  • DHCP reservation for telep-mainframe (192.168.1.123, MAC d8:5e:d3:a7:05:d6) — stops the AirPlay mDNS flap.
  • Camera VLAN — temporarily opened cams→wan/cams→lan for the C302, then scoped to only 192.168.30.168 via per-IP rules, restoring the full air-gap for the Tapo cams.
  • telep-cc SSID unhidden for C302 onboarding, then re-hidden.
  • Router SSH host key changed (2026-08-03 factory-reset aftershock) — clear + re-accept known_hosts.

Graph View

Backlinks