- new compose templates + READMEs for the three LXC apps (110-112)
- Makefile deploy targets (rackula pre-owns its data dir for uid 1001)
- Caddy vhosts for all four, incl. Hermes /auth/login redirect workaround
- firewall docs: CADDY->TRUSTED rules for LXCs 110-113
The official grampsweb image works under /app, not /opt/gramps-web —
media/index/cache paths pointed at unmounted directories. OIDC is now
opt-in via GRAMPS_OIDC_ENABLED and all optional placeholders have
defaults so rendering works without an OIDC/email config.
Bind data/media/consume/export/trash from PAPERLESS_DATA_ROOT with
create_host_path: false so the stack refuses to start when the NFS
share is not mounted. Wire PAPERLESS_TRASH_DIR.
- Add compose stacks for apps already proxied by Caddy: bento-pdf,
it-tools, paperless-ngx, and beszel-hub (with Makefile targets)
- Remove the unused Vault stack and the drawio Caddyfile block
- Fix Caddyfile gramps upstream to port 80 to match the compose file
- Bring immich and gramps-web up to the shared template pattern
(no-new-privileges, log rotation, memory limits, healthchecks)
- Single-quote vaultwarden SSO_AUDIENCE_TRUSTED so regex values like
^\d{18}$ render as valid YAML
- Move root docs into kebab-case topic folders, rebuild the docs index
to cover every doc, and fix all broken links
- Add docker-compose-files/vaultwarden/ with template.yaml, .env.sample,
and README (LXC assigned 10.0.10.16, port 8800)
- Reorganise Caddyfile site blocks into VLAN 1 / VLAN 10 / VLAN 30 sections
matching the TRUSTED / IoT zone layout from docs/network/firewall-rules.md
- Sort VLAN 10 entries by IP; split into physical devices vs LXC Apps sub-sections
- Move zigbee from LAN section to VLAN 30 where it actually lives (10.0.30.5)
Replaced light.kids_lambs group entity with explicit list of mpho_lamb
and hloni_lamb in sunset and 20:30 automations. Group multicast was
causing hloni to silently miss Zigbee commands; unicast to each bulb
individually ensures reliable delivery.
HA 2026 rejects color_temp (mireds) via the service API with 400 Bad
Request, silently breaking automation actions. Switch all color
parameters to color_temp_kelvin: 6500 (cool white) and 2200 (warm white).
Dims mpho_lamb to 10% warm at 02:00 as a night light, and replaces the
fixed 06:00 off trigger with a sun sunrise event so the light goes off
at actual daybreak year-round.
Creates a light group combining mpho_lamb and hloni_lamb, and sets up
four daily automations: on at sunset/18:00 (full cool white), warm dim
at 20:30, hloni off at 22:00, mpho off at 06:00 — all with fade transitions.
setup-homeassistant.sh now deploys automations.yaml alongside configuration.yaml.
- Add SLZB06_HOST and CADDY_HOST_IP stubs to root .env.sample so cp -n covers all required vars
- Remove dead :-10.0.15.5 default from ha_configuration.yaml (CADDY_HOST_IP is required; default was unreachable)
- Genericise CADDY_HOST_IP comment in ha_configuration.yaml (removed hardcoded personal IP)
- Default zigbee2mqtt pan_id and network_key to GENERATE (remove personal network credentials from public blueprint)
- Remove hardcoded device list from zigbee2mqtt_configuration.yaml (Z2M builds this on pairing)
- Clarify README installation steps: root .env path, missing-var behaviour for SLZB06_HOST and CADDY_HOST_IP
Documents the FolderSync/PhotoSync → UNAS Pro → Immich external library
pattern as the workaround for immich-app/immich#4282 (deleted photos
re-uploading on next mobile sync). Includes all steps: UNAS share creation,
Proxmox NFS mount, LXC bind mount, docker-compose volume, Immich library
import path update, and per-app config for Android (FolderSync) and iOS
(PhotoSync).
- Adds docs/immich/mobile-sync.md with full step-by-step guide
- Adds mobile_photos to share map in docs/unas/nfs-mounts.md
- Adds IMMICH_MOBILE_SYNC_DIR volume (/mnt/mobile:ro) to template.yaml
- Documents IMMICH_MOBILE_SYNC_DIR in README.md and .env.sample
- immich-server: add runtime: nvidia + NVIDIA_DRIVER_CAPABILITIES=video,
compute,utility to enable NVENC hardware H.264 transcoding
- immich-machine-learning: change image tag to :release-cuda so
CUDAExecutionProvider is selected over CPUExecutionProvider for CLIP,
face detection, and OCR
- Update GPU Verification section in README with verification commands
and note on resetting QSV → NVENC in the admin UI
- Adds OIDC disable-password-login workflow to README (enforce SSO after
confirming Zitadel PKCE works; clarifies client secret is not needed)
- Expands external library section with user UUID lookup and warning about
Immich blocking paths under the upload volume
- Adds Allow Caddy to Immich rule (10.0.15.5 → 10.0.10.15:2283) to
firewall-rules.md
Adds IMMICH_EXTERNAL_LIBRARY volume (mounted read-only at /mnt/external
in the container) so existing photos stored outside the upload folder
can be imported as an external library without triggering Immich's
upload-folder restriction.
Documents the API-based external library creation workflow in README.
- qbittorrent memory limit 1G → 3G: piece-verification buffers for 25 GB+
torrents exceed 1G and trigger cgroup OOM kill mid-download
- bazarr, prowlarr, seerr limits 512M → 1G: all three were at 40-67% at
idle with insufficient headroom for subtitle batch events, RSS floods,
and Emby library syncs respectively
- Add *media-data-mount and storage-guard dependency to decluttarr so
detect_deletions can access /data/media paths
- Fix decluttarr auth ban loop: add Docker bridge 172.18.0.0/16 to
qBittorrent WebUI auth subnet whitelist (applied live via API)
- Update README mount table (Decluttarr now gets /data/media), memory
limit table with per-service rationale sized for 25 GB average files
- Add TROUBLESHOOTING sections 5e-5j covering qBittorrent path issues,
OOM kill pattern, and decluttarr auth ban loop with fixes
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
UniFi devices validate the Origin header on WebSocket upgrades and reject
connections where it doesn't match their own hostname, causing real-time
dashboard stats to show N/A. Add unifi-upstream snippet that strips Origin
before forwarding, and apply it to both udm and unas proxy blocks.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Both use self-signed TLS backends so they get the office-admin-upstream
snippet (tls_insecure_skip_verify). Update firewall backend policy table
in network docs to match.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- switchlite8poe moved to 10.0.1.6
- tplink16de removed (device no longer exists)
- speedtest removed (no longer needed)
- Update firewall backend policy table in network docs to match
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Four issues encountered during the arr-stack LXC 101 deployment that were not
covered or not clear enough in existing docs:
PROXMOX-NVIDIA-LXC.md — expanded Section 4 into four explicit sub-steps:
4a: copy installer into LXC via pct push (it lives on the Proxmox host)
4b: install userspace libs with --no-kernel-module then run ldconfig
4c: install Container Toolkit (must come after libs, not before)
4d: switch runtime mode auto→legacy with clear explanation of why CDI fails
in an LXC even after the libraries are installed
TROUBLESHOOTING.md — new section 5c for the stale gluetun namespace error:
"joining network namespace of container: No such container"
Cause: gluetun recreated with new container ID, qbittorrent holds old ID
Fix: docker rm -f qbittorrent && docker compose up -d qbittorrent
Renumbered old 5c (HTTPS 502) to 5d
arr-stack README.md — AirVPN preshared key warning in the Installation section:
AirVPN generates per-server preshared keys; the built-in airvpn provider
selects a server dynamically so the key never matches → silent VPN failure.
Documents when and how to use VPN_SERVICE_PROVIDER=custom with a pinned server.
DEPLOYMENT.md — two new sections:
Section 9: Set a Static IP — via Proxmox .conf net0 line, not inside the LXC
Section 10: Rename a Container (VMID change) — lvrename + conf file swap
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Add optional WIREGUARD_PUBLIC_KEY, WIREGUARD_ENDPOINT_IP, WIREGUARD_ENDPOINT_PORT
to gluetun template so custom provider renders cleanly without manual patching
- Make SERVER_COUNTRIES optional (:-) so re-renders don't fail for custom provider
- Document the required mode = "legacy" change in /etc/nvidia-container-runtime/config.toml
for NVIDIA Container Toolkit in a privileged LXC; the default "auto" CDI mode fails
until userspace libraries are installed and the mode is explicitly set to legacy
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Correct DEPLOYMENT.md fstab entry (vers=3, UUID path, 10.0.10.25) and
add No Root Squash requirement note to the chown step. Fix arr-stack
README to document host-side chown from a privileged LXC instead of
the incorrect unprivileged-LXC warning. Also carries forward nfs-mounts.md
from the previous session.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
