proxmox-infra/docs/CHANGELOG.md

# Changelog

> **Purpose**: Historical record of all significant infrastructure changes

## 2025-12-18

### Service Additions
- **Pi-hole**: Added network-wide ad blocker with recursive DNS
  - LXC on pm2
  - IP: 10.4.2.129
  - Domain: pihole.kavcorp.com
  - Unbound configured for recursive DNS resolution
  - Deployed via ProxmoxVE community helper script
  - Tagged: adblock, dns

## 2025-12-15

### Frigate Migration & Upgrade
- **Frigate**: Migrated from source install (LXC 111) to Docker-based (LXC 128)
  - Old: LXC 111 on pm3 (source install, 0.14.1)
  - New: LXC 128 on pm3 (Docker, 0.17.0-beta1)
  - IP: 10.4.2.8
  - Domain: frigate.kavcorp.com
  - Privileged LXC required for USB device passthrough (Coral TPU)
  - Coral USB TPU successfully passed through
  - NFS mount for media storage: `/mnt/pve/KavNas/frigate-media`

- **Frigate Configuration Updates**:
  - Enabled built-in authentication (port 8971)
  - Updated MQTT to correct Home Assistant IP (10.4.2.199)
  - Consolidated camera configs using global defaults
  - Fixed garage stream bug (was using wrong ffmpeg source)
  - Added stationary car filtering (stops tracking after 30 seconds)

- **Traefik Updates**:
  - Updated Frigate route to use HTTPS backend (port 8971)
  - Added serversTransport for self-signed cert (insecureSkipVerify)
  - Fixed disk full issue (removed 903MB old access log)
  - Added logrotate config: 50MB max, 3 rotations, daily

### Service Recovery
- **Power Outage Recovery**: Started all stopped LXCs on pm2, pm3, pm4
- **VM 109 (docker-pm3)**: Fixed missing onboot setting

### Infrastructure Notes
- LXC 111 (old Frigate) pending deletion after new setup confirmed
- Port 5000 on Frigate remains available for Home Assistant integration (unauthenticated)
- Admin credentials logged on first auth-enabled startup

## 2025-12-08

### Service Configuration
- **Shinobi (LXC 103)**: Configured NVR storage and Traefik endpoint
  - Added to Traefik reverse proxy: shinobi.kavcorp.com
  - Traefik config: `/etc/traefik/conf.d/shinobi.yaml`
  - Created NFS storage on elantris (`/el-pool/shinobi`) - 11TB available
  - Added Proxmox NFS storage: `elantris-shinobi`
  - Mounted NFS to LXC 103: `/opt/Shinobi/videos`
  - Coral USB TPU device passed through to container
  - Coral object detection plugin attempted but blocked by TensorFlow Lite unavailability for Ubuntu 24.04/Python 3.12
  - Motion detection available and working

### Notes
- Coral TPU native plugin requires building TensorFlow Lite from source, which is complex for Ubuntu 24.04
- Basic motion detection works out of the box for event recording
- Object detection may require alternative approach (Frigate, or CPU-based detection)

## 2025-12-07

### Service Additions
- **Vaultwarden**: Created new password manager LXC
  - LXC 125 on pm4
  - IP: 10.4.2.212
  - Domain: vtw.kavcorp.com
  - Traefik config: `/etc/traefik/conf.d/vaultwarden.yaml`
  - Tagged: community-script, password-manager

- **Immich**: Migrated from Docker (dockge LXC 107 on pm3) to native LXC
  - LXC 126 on pm4
  - IP: 10.4.2.24:2283
  - Domain: immich.kavcorp.com
  - Traefik config: `/etc/traefik/conf.d/immich.yaml`
  - Library storage: NFS mount from elantris (`/el-pool/downloads/immich/`)
  - 38GB photo library transferred via rsync
  - Fresh database (version incompatibility: old v1.129.0 → new v2.3.1)
  - Services: immich-web.service, immich-ml.service
  - Tagged: community-script, photos

- **Gitea**: Added self-hosted Git server
  - LXC 127 on pm4
  - IP: 10.4.2.7:3000
  - Domain: git.kavcorp.com
  - Traefik config: `/etc/traefik/conf.d/gitea.yaml`
  - Config: `/etc/gitea/app.ini`
  - Push-to-create enabled for users and orgs
  - Initial repo: `proxmox-infra` (infrastructure documentation)
  - Tagged: community-script, git

### Infrastructure Maintenance
- **Traefik (LXC 104)**: Fixed disk full issue
  - Truncated 895MB access log that filled 2GB rootfs
  - Added logrotate config to prevent recurrence (50MB max, 7 day rotation)
  - Cleaned apt cache and journal logs

## 2025-11-20

### Service Changes
- **AMP**: Added to Traefik reverse proxy
  - LXC 124 on elantris (10.4.2.26:8080)
  - Domain: amp.kavcorp.com
  - Traefik config: `/etc/traefik/conf.d/amp.yaml`
  - Purpose: Game server management via CubeCoders AMP

## 2025-11-19

### Service Changes
- **LXC 123 (elantris)**: Migrated from Ollama to llama.cpp
  - Removed Ollama installation and service
  - Built llama.cpp from source with CURL support
  - Downloaded TinyLlama 1.1B Q4_K_M model (~667MB)
  - Created systemd service for llama.cpp server
  - Server running on port 11434 (OpenAI-compatible API)
  - Model path: `/opt/llama.cpp/models/tinyllama-1.1b-chat-v1.0.Q4_K_M.gguf`
  - Service: `llama-cpp.service`
  - Domain remains: ollama.kavcorp.com (pointing to llama.cpp now)

- **LXC 124 (elantris)**: Created new AMP (Application Management Panel) container
  - IP: 10.4.2.26
  - Resources: 4 CPU cores, 4GB RAM, 16GB storage
  - Storage: local-lvm on elantris
  - OS: Ubuntu 24.04 LTS
  - Purpose: Game server management via CubeCoders AMP
  - Tagged: gaming, amp

## 2025-11-17

### Service Additions
- **Ollama**: Added to Traefik reverse proxy
  - LXC 123 on elantris
  - IP: 10.4.2.224:11434
  - Domain: ollama.kavcorp.com
  - Traefik config: `/etc/traefik/conf.d/ollama.yaml`
  - Downloaded Qwen 3 Coder 30B model

- **Frigate**: Added to Traefik reverse proxy
  - LXC 111 on pm3
  - IP: 10.4.2.215:5000
  - Domain: frigate.kavcorp.com
  - Traefik config: `/etc/traefik/conf.d/frigate.yaml`

- **Foundry VTT**: Added to Traefik reverse proxy
  - LXC 112 on pm3
  - IP: 10.4.2.37:30000
  - Domain: vtt.kavcorp.com
  - Traefik config: `/etc/traefik/conf.d/foundry.yaml`

### Infrastructure Changes
- **SSH Access**: Regenerated SSH keys on pm2 and distributed to all cluster nodes
  - pm3 SSH service was down, enabled and configured
  - All nodes (pm1, pm2, pm3, pm4, elantris) now accessible from pm2 via Proxmox web UI

### Service Configuration
- **NZBGet**: Fixed file permissions
  - Set `UMask=0000` in nzbget.conf to create files with 777 permissions
  - Fixed permission issues causing Sonarr import failures

- **Sonarr**: Enabled automatic permission setting
  - Media Management → Set Permissions → chmod 777
  - Ensures imported files are accessible by Jellyfin

- **Jellyseerr**: Fixed Traefik routing
  - Corrected IP from 10.4.2.20 to 10.4.2.18 in media-services.yaml

- **Jellyfin**: Fixed LXC mount issues
  - Restarted LXC 121 to activate media mounts
  - Media now visible in `/media/tv`, `/media/movies`, `/media/anime`

### Documentation
- **Major Reorganization**: Consolidated scattered docs into structured system
  - Created `README.md` - Documentation index and guide
  - Created `INFRASTRUCTURE.md` - All infrastructure details
  - Created `CONFIGURATIONS.md` - Service configurations
  - Created `DECISIONS.md` - Architecture decisions and patterns
  - Created `TASKS.md` - Current and pending tasks
  - Created `CHANGELOG.md` - This file
  - Updated `CLAUDE.md` - Added documentation policy

## 2025-11-16

### Service Deployments
- **Home Assistant**: Added to Traefik reverse proxy
  - Domain: hass.kavcorp.com
  - Configured trusted proxies in Home Assistant

- **Frigate**: Added to Traefik reverse proxy
  - Domain: frigate.kavcorp.com

- **Proxmox**: Added to Traefik reverse proxy
  - Domain: pm.kavcorp.com
  - Backend: pm2 (10.4.2.6:8006)

- **Recyclarr**: Configured TRaSH Guides automation
  - Sonarr and Radarr quality profiles synced
  - Dolby Vision blocking implemented
  - Daily sync schedule via cron

### Configuration Changes
- **Traefik**: Removed Authelia from *arr services
  - Services now use only built-in authentication
  - Simplified access for Sonarr, Radarr, Prowlarr, Bazarr, Whisparr, NZBGet

### Issues Encountered
- Media organization script moved files incorrectly
- Sonarr database corruption (lost TV series tracking)
- Permission issues with NZBGet downloads
- Jellyfin LXC mount not active after deployment

### Lessons Learned
- Always verify file permissions (777 required for NFS media)
- Backup service databases before running automation scripts
- LXC mounts may need container restart to activate
- Traefik auto-reloads configs, no restart needed

## Earlier History

*To be documented from previous sessions if needed*