FailedProxy/Varaverk
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
Files
613634473af277273bb579b2a9a89e1c603da526
Varaverk/Deployment/master.conf.template
T
Gmer4Lfe 613634473a Let General Chat search the web, and only General Chat 
Search is the one capability that fits the profile holding none: every other capability either
reads this installation or changes it, and chat's whole contract is that it can do neither. The
assistant deliberately does not get it — its contract is that answers come from this install's
own documents, and a web result there is an answer that looks sourced and is not. A chat question
about this machine hands off to the assistant before the search would run, so it never reaches
the internet even with the box ticked.

Off by default, and not because it is dangerous: searching sends the operator's question outside
the house, which is theirs to decide. Asked for per turn as well as enabled in conf.

Provider-agnostic, as asked — searxng, brave, tavily. Only whichever is configured here can be
verified; all three read every field by name, so a shape that changes underneath yields no
results rather than wrong ones.

The explain fixtures asserted chat holds no capabilities at all, which is exactly the guarantee
worth keeping. caps=only: now states the set rather than its emptiness, so the check still fails
the day something else is granted there.
2026-08-09 22:42:43 -04:00

1865 lines
120 KiB
Bash
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
#!/bin/bash
# ==============================================================================================
# ================================= MASTER CONFIGURATION =======================================
# ==============================================================================================
# Shared configuration for the unRAID script ecosystem.
# Contains all settings that apply to every server — thresholds, toggles, profiles, job lists.
#
# ── HOW THE THREE-FILE SYSTEM WORKS ──────────────────────────────────────────────────────────
# Scripts source all three files at startup:
# source master.conf ← shared config (this file)
# source host1.conf ← HOST1 credentials, shares, container lists
# source host2.conf ← HOST2 credentials, shares, container lists
#
# Sparse checkout (git) ensures each server only pulls its own host*.conf.
# HOST2 never sees HOST1 credentials. HOST1 never sees HOST2 credentials.
#
# What belongs here: thresholds, toggles, intervals, profiles, job lists
# What belongs in HOST*: hostnames, SSH keys, API keys, passwords, share paths,
# container names, failover lists, watchdog containers
#
# ── HOW THIS FILE WORKS ───────────────────────────────────────────────────────────────────────
# Every script sources all three conf files and common.sh at startup.
# Change a value here and it affects all scripts that use it — no hunting through files.
# To disable something: comment it out with # rather than deleting it.
# To add a new rsync profile: add a key to each PROFILE_* array in the RSYNC section.
# To add or remove orchestrator jobs: edit the arrays in the ORCHESTRATORS section.
#
# ── INDEX ─────────────────────────────────────────────────────────────────────────────────────
#
# Section Description
# ───────────────────────────────────────────────────────────────────────────────────────────
# SHARED HOST CONFIGURATION DATA_DIR and shared connection settings
# PARTNERSHIP Mirror relationship lifecycle — onboard/offboard/transfer
# LOGGING Enable or disable verbose logging
# NOTIFICATIONS unRAID native and Discord webhook settings
# GIT / REPO Gitea repository and SSH settings
#
# ── ORCHESTRATORS ──────────────────────────────────────────────────────────────────────────
# ARRAY START Scripts launched at array start (array_started.sh)
# ARRAY STOP Scripts run at planned shutdown (array_stopping.sh)
# WATCHDOG ORCHESTRATOR Per-minute watchdog runner (watchdog_orchestrator.sh)
# SYSTEM WATCHDOG Sub-scripts called by watchdog_orchestrator.sh
# CRITICAL SYNC MAINTENANCE 30-minute jobs + sync shares + partnership check (critical_sync_maintenance.sh)
# INTERMEDIATE SYNC MAINTENANCE Arr sync + optional mid-day rsync (intermediate_sync_maintenance.sh)
# DAILY SYNC MAINTENANCE Job list + media shares (daily_sync_maintenance.sh)
# WEEKLY SYNC MAINTENANCE Job list + sync shares + update toggles (weekly_sync_maintenance.sh)
# MONTHLY MAINTENANCE Uptime-triggered heavy tasks — ZFS scrub, SMART tests (monthly_maintenance.sh)
#
# ── RSYNC ──────────────────────────────────────────────────────────────────────────────────
# RSYNC ENABLE/DISABLE Two-tier toggle system — global gate + per-orchestrator
# RSYNC DEFAULTS Global fallback rsync options and limits
# REMOTE HEALTH CHECKS Rootfs threshold for pre-flight abort
# RSYNC PROFILE SYSTEM Per-profile overrides for appdata syncs
#
# ── FALLBACK ───────────────────────────────────────────────────────────────────────────────
# FALLBACK Mutual container failover shared settings
# FALLBACK TEST Simulated outage settings for fallback_test.sh
#
# ── DOCKER ESSENTIALS ──────────────────────────────────────────────────────────────────────
# DOWNLOADERS RESET Retention and thresholds for slskd, SABnzbd, qBittorrent
# DOCKER DAILY RESTART Containers restarted daily
# DOCKER WEEKLY RESTART Containers restarted weekly
# DOCKER WATCHDOG Thresholds and toggles for container monitoring
# DOCKER NETWORK CONNECT Networks to ensure + containers to connect
#
# ── UNRAID ESSENTIALS ──────────────────────────────────────────────────────────────────────
# INOTIFY TUNING inotify limits — raised at array start by inotify_tuning.sh
# SYSTEM TUNING MONITOR Tracks inotify + php-fpm usage over time
# REBOOT User warning delay before scheduled reboot
# MOVER Mover stop timeout
# SYSLOG FILTER Docker veth noise filter file path
# PHP-FPM PHP-FPM max children config
# CLEAR LOGS System log file paths
# WEBGUI WATCHDOG WebGUI nginx + emhttp monitoring and restart
#
# ── MEDIA ──────────────────────────────────────────────────────────────────────────────────
# MEDIA PERMISSIONS Share list, mode and owner for permissions script
# MEDIA CLEANER Anime and media folder lists and file patterns
# ARR CLEANUP Lidarr, Sonarr, Radarr orphan file cleanup settings
# ARR FAILED/STALLED RECOVERY Auto blocklist + re-search settings
#
# ── TRANSCODES ─────────────────────────────────────────────────────────────────────────────
# TRANSCODE MANAGER Ramdisk and SSD fallback transcode management
# TRANSCODE SERVER ARRAY Multi-server session monitoring (Emby, Jellyfin, Plex)
#
# ── MONITORS ───────────────────────────────────────────────────────────────────────────────
# CERTIFICATE MONITOR SSL certificate expiry monitoring
# BACKUP VERIFY Random sample checksum verification against remote
# SMART HEALTH Drive SMART attribute monitoring
# ZFS MEMORY SNAPSHOT Weekly ZFS health and memory diagnostic report
# BANDWIDTH MONITOR Daily rsync transfer logging and weekly summary
# HEALTH DIGEST Aggregated system health digest — always/smart/weekly
# EMBY SESSION REPORT Weekly Emby usage statistics via API
#
# ── SYSTEM WATCHDOG ────────────────────────────────────────────────────────────────────────
# SYSTEM WATCHDOG Continuous system health monitoring — last line of defense
#
# ==============================================================================================
# ==============================================================================================
# ── HOST IDENTITIES ───────────────────────────────────────────────────────────────────────────
# ==============================================================================================
# Hostnames for every server in the ecosystem — not credentials, safe for all machines.
# Must match the exact unRAID hostname AND Tailscale device name (case sensitive).
# detect_hosts() in common.sh matches the local hostname against these to set MY_ID / REMOTE_ID.
# Tailscale IP resolution uses these names — no hardcoded IPs needed.
# To add a new server: add HOST3="unRAID-NewServer" here + create host3.conf.
HOST1="" # this server's Unraid hostname — must match exactly (case-sensitive)
HOST2="" # partner server hostname — leave blank if running standalone
# ==============================================================================================
# ── SHARED HOST CONFIGURATION ─────────────────────────────────────────────────────────────────
# ==============================================================================================
# Requirement: internal NVMe/SSD boot drive. Everything lives in the plugin folder on flash —
# scripts, state, and data are all available before the array mounts.
# Both directories are created automatically if they don't exist.
#
# DATA_DIR is the one on-disk root. Everything Varaverk persists lives under it, in a
# subdirectory named for what the files are. Move DATA_DIR and the whole tree follows.
#
# It used to be two roots plus two strays: DATA_DIR beside State_Files/ as siblings, with the
# conf-cache backup off in SCRIPTS_DIR/.cache/vv/d and the arr cache backups loose in DATA_DIR's
# root. Nothing was wrong with any one of those decisions; together they meant no single place
# answered "what does Varaverk keep on disk". State is data — it is the data that happens to
# describe right now — so it belongs under the same root as the rest.
#
# db/ — statistics, histories, counters, blocklists. Things that accumulate.
# state/ — runtime state for every script: watchdogs, fallback, transcode, setup.
# Requirement: ALL state files MUST use $STATE_DIR. No /tmp, no repo root.
# ai/ — the retrieval index, operator memory, token ledger, filed bugs, saved chats.
# cache/ — persistent backups of the tmpfs caches, and ONLY those. A file belongs here when
# losing it costs a re-fetch and nothing else; anything that is a source of truth
# belongs in db/ or state/.
# logs/ — retained log output. Live logging still goes to LOG_DIR (/var/log/varaverk).
#
# STATE_DIR keeps its name and changes only its value, which is why this restructure did not
# touch the 15 conf entries, 18 shell paths and 23 PHP paths that build on it.
#
# The tmpfs caches are NOT here and must not be moved here — see VV_CACHE_ROOT below. These are
# on flash; those are read every second by the WebGUI and rewritten by the hundred megabytes.
DATA_DIR="/boot/config/plugins/varaverk/data"
DB_DIR="${DATA_DIR}/db"
STATE_DIR="${DATA_DIR}/state"
AI_DATA_DIR="${DATA_DIR}/ai"
CACHE_BACKUP_DIR="${DATA_DIR}/cache"
LOG_ARCHIVE_DIR="${DATA_DIR}/logs"
BACKUP_DIR="${DATA_DIR}/Backups"
CONF_BACKUP_DIR="${BACKUP_DIR}/Confs" # 0700 — holds credentials
PERSISTENT_CONF_CACHE="${CACHE_BACKUP_DIR}/conf"
ARR_CACHE_BACKUP_DIR="${CACHE_BACKUP_DIR}/arr"
# ── Cache Roots ──
# Everything Varaverk keeps in RAM, under one root, defined once.
#
# These used to be four literals in load_config.sh and three more in the PHP layer, spread over
# /tmp/vv_cache, /tmp/arr_cache, /tmp/.cache/vv/d, /tmp/.cache/vv/ai and /tmp/varaverk_ai_jobs —
# five naming schemes, and no single place that could tell you what Varaverk had in tmpfs. PHP
# could not read load_config.sh, so it restated the paths it needed and the two layers were kept
# in agreement by hand. They live here because master.conf is the one file both layers actually
# read: bash sources it, and the PHP conf parser resolves ${VAR} against it the same way.
#
# ONE ROOT, NOT ONE DIRECTORY. The subdirectories are deliberately separate and must stay that
# way — they have genuinely different rules:
#
# conf/ holds partner credentials, is chmod 700, and IS snapshotted to PERSISTENT_CONF_CACHE
# so it survives a reboot.
# ai/ holds non-secret token counters that are worthless when stale and must NOT be
# preserved across a reboot — losing them means "not collected here" until the next
# sync, which is the honest answer.
# jobs/ holds in-flight work handed off to a detached worker — an AI answer being generated,
# a container action being applied. chmod 700: each is readable by anyone who can guess
# its token, which is why the tokens are random_bytes and not sequential.
# arr/ is the large one (100s of MB) and is restored from a DATA_DIR backup on demand.
# api/ is the WebGUI payload cache, and is the only one safe to delete at any moment.
#
# Collapsing those into a single directory would give the credentials the token cache's
# persistence rules, or the reverse. The win here is one definition, not one folder.
#
# Point VV_CACHE_ROOT somewhere else and everything follows. It must be on a filesystem that is
# cleared or safe to clear on boot — every consumer treats a missing cache as a cold start, and
# nothing here is a source of truth for anything.
VV_CACHE_ROOT="/tmp/varaverk"
VV_CACHE_DIR="${VV_CACHE_ROOT}/api" # WebGUI payload cache — monitor, arrs, ai
CONF_RAM_CACHE_DIR="${VV_CACHE_ROOT}/conf" # partner host*.conf — chmod 700, snapshotted
ARR_CACHE_DIR="${VV_CACHE_ROOT}/arr" # arr payloads — restored from DATA_DIR on demand
AI_TOKEN_CACHE_DIR="${VV_CACHE_ROOT}/ai" # partner token ledgers — never preserved
AI_JOB_DIR="${VV_CACHE_ROOT}/jobs/ai" # in-flight AI answers — chmod 700
DOCKER_JOB_DIR="${VV_CACHE_ROOT}/jobs/docker" # in-flight container actions — chmod 700
# ── Version Parity ──
# Controls behaviour when local and remote unRAID versions differ.
# Major version mismatch always aborts regardless of this setting.
# "warn" — log warning and continue (safe for minor/patch differences)
# "abort" — refuse to continue (strict — ensures both sides always match)
UNRAID_VERSION_MISMATCH_ACTION="warn"
# ── Conf Sync ──
# conf_sync.sh pulls partner host confs into /tmp/.vv/ RAM cache on array start and after
# any conf save — makes partner vars (HOST2_*, etc.) available without touching git.
CONF_SYNC_ENABLED=true
# ── Arr Sync ──
# arr_sync.sh syncs Lidarr/Sonarr/Radarr libraries across all nodes bidirectionally.
# Runs before rsync — all nodes agree on tracked library before files are transferred.
# Remote API keys are read live from each node's config.xml via SSH — never stored here.
ARR_SYNC_ENABLED=true
ARR_SYNC_BLOCKLIST="${DB_DIR}/arr_sync_blocklist.tsv"
ARR_SYNC_CONNECT_TIMEOUT=10 # seconds — SSH connect timeout per node
ARR_SYNC_API_TIMEOUT=60 # seconds — curl timeout for library fetches
DOCKER_APPDATA_BASE="/mnt/user/appdata"
ARR_SYNC_LIDARR_PORT=8686
ARR_SYNC_SONARR_PORT=8989
ARR_SYNC_RADARR_PORT=7878
# ── Remote Docker Daemon ──
# Strike system for remote Docker daemon health checks.
# Scripts that issue remote container commands check the remote daemon first.
# Below limit → skip operation this run | At limit → notify critical + exit
REMOTE_DOCKER_STRIKE_LIMIT=3 # consecutive failures before critical notify
# ==============================================================================================
# ── PARTNERSHIP ───────────────────────────────────────────────────────────────────────────────
# ==============================================================================================
# Manages the relationship lifecycle between two unRAID servers.
# HOST1 is always the owner (source of truth) — HOST2 is always the mirror.
# PARTNERSHIP_OWNER_HOST flips to "HOST2" after a --transfer operation.
# All identity vars (hostnames, SSH keys) live in host*.conf.
# Hostnames already match Tailscale device names — IP resolution is automatic.
#
# State files on /boot/config — survives reboots, available before array starts:
# /boot/config/partnership_HOST1.db ← HOST1 writes only
# /boot/config/partnership_HOST2.db ← HOST2 writes only
# Propagated via SSH — no rsync needed
#
# critical_sync_maintenance.sh runs --check every 30min:
# Reads both state files via SSH
# Detects offboard requests → finalises from owner side
# Increments offline counter → auto-offboards after threshold
# Silent when healthy ✅
#
# See README-Partnership.md for full lifecycle documentation.
PARTNERSHIP_ENABLED=false
PARTNERSHIP_OWNER_HOST="HOST1" # "HOST1" or "HOST2" — flips on --transfer
# Auth containers reconfigured on onboard/offboard — defined per host in host*.conf.
# Format: "ContainerName|WebUIPort"
# HOST1_PARTNERSHIP_AUTH_WEBUIS / HOST2_PARTNERSHIP_AUTH_WEBUIS
# On onboard → WebUI pointed at owner's Tailscale IP
# On offboard → WebUI pointed back at localhost
# Paths to collect during the grace window after offboard — defined per host in host*.conf.
# HOST1_PARTNERSHIP_MIRROR_BACKUPS / HOST2_PARTNERSHIP_MIRROR_BACKUPS
# Notified on offboard — no auto-deletion, manual collection.
# Timing — single var controls both Tailscale removal and backup access expiry.
# Both expire at the same time — keeping backups accessible beyond Tailscale removal is pointless.
PARTNERSHIP_GRACE_HOURS=6 # hours after offboard before Tailscale removal
# backup access expires at the same time
PARTNERSHIP_OFFLINE_THRESHOLD=30 # days either server unreachable before auto-offboard
# works both directions independently
# Partnership and setup state files — all in STATE_DIR per the project requirement.
# Scripts use these variables; do not hardcode /boot/config paths in scripts.
PARTNERSHIP_BLOCKLIST_FILE="$STATE_DIR/partnership_blocklist.db"
VARAVERK_SETUP_FILE="$STATE_DIR/varaverk_setup.db"
# Tailscale removal on offboard.
PARTNERSHIP_REMOVE_TAILSCALE=true # remove mirror from Tailscale tailnet on offboard
# false = skip removal (manual or testing)
# Tailscale API — required when PARTNERSHIP_REMOVE_TAILSCALE=true.
# Stays in shared conf — only owner uses it, and owner is always running this script.
# API key: https://login.tailscale.com/admin/settings/keys → Devices write scope
TAILSCALE_API_KEY="" # tskey-api-...
TAILSCALE_TAILNET="" # your tailnet name (e.g. yourname.github)
# Transfer safety.
PARTNERSHIP_TRANSFER_CONFIRM="i-understand-this-transfers-ownership"
PARTNERSHIP_TRANSFER_STRIKES=3 # consecutive health checks required
PARTNERSHIP_TRANSFER_MAX_ATTEMPTS=20 # max health check attempts before giving up
# Onboard settings.
PARTNERSHIP_ONBOARD_VERIFY=true # verify WebUI reachable after reconfiguration
PARTNERSHIP_ONBOARD_NOTIFY=true # notify both servers on completion
PARTNERSHIP_SYNC_INTERVAL=15 # minutes — informational, actual schedule in cron
# SSH key lifecycle — managed by Partnership/ssh_setup.sh.
# Key named after this server: hostname lowercased, unraid- prefix stripped.
# unRAID-Gmer4Lfe → /root/.ssh/gmer4lfe_rsync_automation
# Run Partnership/partnership_onboard.sh to generate, copy, and update conf automatically.
SSH_MAX_STRIKES=5 # consecutive SSH auth failures before critical notify
SSH_STRIKE_RESET_HRS=24 # hours since last failure before strike counter resets
# FolderView3 plugin integration — chodeus/VladoPortos/scolcipitato.
# Creates a "{Partner}-Failover" folder on onboard, removes + cleans containers on offboard.
# Folder name derived from remote hostname (strip unraid- prefix case-insensitively).
# Plugin config: /boot/config/plugins/folder.view3/docker.json
PARTNERSHIP_FOLDERVIEW3=true # create/remove FolderView3 folder on onboard/offboard
PARTNERSHIP_FOLDERVIEW3_URL="" # CA plugin URL — leave empty to skip auto-install
# ==============================================================================================
# ── LOGGING ───────────────────────────────────────────────────────────────────────────────────
# ==============================================================================================
# Controls verbose [LOG] output across all scripts.
# true = show detailed [LOG] lines — useful for debugging or first-time setup
# false = show only user-facing output — cleaner for scheduled runs
ENABLE_LOGGING=false
# ==============================================================================================
# ── NOTIFICATIONS ─────────────────────────────────────────────────────────────────────────────
# ==============================================================================================
# unRAID native notification system — integrates with the bell icon in the WebGUI.
# normal = job completed successfully / warning = something failed or needs attention
NOTIFY_UNRAID=true
# Discord webhook URL — defined per host in host*.conf.
# HOST1_DISCORD_WEBHOOK / HOST2_DISCORD_WEBHOOK
# Allows different webhooks per server, or only one server notifying.
# Upgrade webhook — standalone PHP listener that receives OnUpgrade events from
# Sonarr/Radarr/Lidarr and immediately pushes the upgraded file to all mesh nodes.
# Bypasses Unraid nginx auth — the secret in the URL is the only gate.
# Run Tools/webhook_setup.sh once to register the connection in each arr.
# WEBHOOK_PORT 0 disables the listener.
WEBHOOK_PORT=7821
WEBHOOK_SECRET="" # auto-generated on first start if empty
# ==============================================================================================
# ── GIT / REPO ────────────────────────────────────────────────────────────────────────────────
# ==============================================================================================
# Gitea self-hosted repository — used by git_pull_execute.sh.
# Detects Gitea container location at runtime — works through failover automatically.
# Falls back to GITEA_DOMAIN if local and Tailscale both fail.
GITEA_CONTAINER="Gitea"
GITEA_REPO_PATH="" # e.g. YourUser/Varaverk.git
GITEA_DOMAIN="" # e.g. git.yourdomain.com — requires NPM + DNS
TARGET_DIR="/mnt/user/appdata/Varaverk"
GITEA_SSH_KEY="/root/.ssh/unraid_gitea"
SSH_PORT=221 # Gitea SSH port (default 22, Gitea often uses 221/222)
GITEA_HTTP_PORT=3000 # Gitea web/API port — used by gitea_ssh_setup.sh
# ==============================================================================================
# ── ORCHESTRATORS ─────────────────────────────────────────────────────────────────────────────
# ==============================================================================================
# All orchestrator job lists live here — edit arrays to add/remove scripts.
# No changes to orchestrator scripts needed when adding or removing jobs.
# Sections ordered by run frequency: array events first, then shortest interval to longest.
# ━━━ Array Start ━━━
# Scripts launched by array_started.sh when the array comes online.
# Launched in order — each as a background process.
# One-shot scripts (ramdisk, syslog, fpm, inotify, network) run and exit naturally.
# Continuous scripts (failover) run until array stops.
# Watchdogs (resource_watchdog, docker_watchdog, system_watchdog) are cronned via
# watchdog_orchestrator.sh — NOT launched here.
ARRAY_START_SCRIPTS=(
"Plugin/unraid/System_Essentials/unraid_api_key_renew.sh" # re-register Varaverk API key — registry is ephemeral
"System_Essentials/conf_sync.sh" # pull partner confs + push own conf into /tmp/.vv/ RAM cache
"System_Essentials/conf_cache_restore.sh" # load partner confs from persistent backup if conf_sync couldn't reach partner
"Transcodes/ramdisk_setup.sh" # creates ramdisk + symlink before Emby starts
"System_Essentials/docker_syslog_filter.sh" # suppress veth noise before logs fill
"Plugin/unraid/System_Essentials/php_fpm_max_children.sh" # WebGUI performance tuning
"System_Essentials/inotify_tuning.sh" # bump inotify limits — containers miss events if exhausted
"Docker_Essentials/docker_network_connect.sh" # ensure networks exist + connect containers
"Arrs_Stack/arr_cache_prefill.sh" # warm Lidarr/Sonarr/Radarr tracked-data caches before anything reads them cold
"Arrs_Stack/start_webhook_listener.sh" # arr upgrade webhook listener — continuous
"Fallback/fallback.sh" # mutual failover — continuous
)
# ━━━ Array Stop ━━━
# Scripts run by array_stopping.sh for a planned shutdown — stops everything cleanly in order.
# Run sequentially (foreground) — each must complete before the next starts.
# Order matters: user scripts first (prevents new ops), then data movement, then containers.
ARRAY_STOP_SCRIPTS=(
"System_Essentials/conf_cache_save.sh" # snapshot partner conf RAM cache → /boot before anything stops
"Plugin/unraid/System_Essentials/user_scripts_stop.sh" # stop background scripts before they start new ops
"Fallback/fallback.sh --stop" # gracefully stop fallback (not caught by user_scripts_stop)
"System_Essentials/rsync_stop.sh --rsync-only" # kill rsync; skip container recovery (handled below)
"Plugin/unraid/System_Essentials/mover_stop.sh" # stop mover after rsync (they conflict on same files)
"Docker_Essentials/docker_container_stop.sh" # stop all containers last
)
# ━━━ Transcode Management ━━━
# transcode_management.sh runs TRANSCODE_MANAGEMENT_SCRIPTS in order each cron cycle.
# Schedule: */7 * * * * (every 7 minutes)
# Order matters — cleanup first so the manager measures real current ramdisk usage,
# not usage inflated by stale segment files from ended sessions.
TRANSCODE_MANAGEMENT_SCRIPTS=(
"Transcodes/transcode_cleanup.sh" # remove aged segment files before usage is measured
"Transcodes/transcode_manager.sh" # flip ramdisk/SSD symlink, write daily log entry
)
# ━━━ Watchdog Orchestrator ━━━
# watchdog_orchestrator.sh runs WATCHDOG_ORCHESTRATOR_SCRIPTS in order each cron cycle.
# Schedule: */15 * * * * (every 15 minutes)
# NOT in ARRAY_START_SCRIPTS — has its own cron entry.
# Order matters — resource first (frees pressure), docker second (heals with freed resources),
# system third (storage + webgui component health), stability last (last line of defense).
WATCHDOG_ORCHESTRATOR_SCRIPTS=(
"Watchdogs/resource_watchdog.sh" # reduce system pressure before healing attempts
"Watchdogs/docker_watchdog.sh" # heal containers with freed resources
"Watchdogs/system_watchdog.sh" # system component health — storage + webgui
"Plugin/unraid/System_Essentials/unraid_api_key_renew.sh" # re-register API key if registry lost it
"Plugin/unraid/Tools/ai_repair_sweep.sh" # read what the last cycle logged; off unless AI_REPAIR_ENABLED
"Watchdogs/stability_watchdog.sh" # reboot if all else fails — last line of defense
)
# The repair sweep sits ahead of stability deliberately, which is the one exception to "stability
# last". It reads the previous cycle's logs and may correct the very misconfiguration stability
# would otherwise reboot for — a wrong port is not fixed by restarting the machine. It is bounded
# by AI_PROBE_TIMEOUT, exits 0 in every case, and does nothing at all unless AI_REPAIR_ENABLED.
# ━━━ System Watchdog ━━━
# system_watchdog.sh runs SYSTEM_WATCHDOG_SCRIPTS sequentially each cycle.
# Called by watchdog_orchestrator.sh — not scheduled directly.
SYSTEM_WATCHDOG_SCRIPTS=(
"Watchdogs/System/storage_watchdog.sh" # pool growth + runaway log detection
"Plugin/unraid/Watchdogs/System/webgui_watchdog.sh" # WebGUI availability — nginx → php-fpm → emhttp
"Watchdogs/System/network_watchdog.sh" # internet, DDNS, Tailscale, NPM proxy
"Watchdogs/System/conf_cache_watchdog.sh" # maintain persistent conf backup while partner is offline
)
# ━━━ Critical Sync Maintenance ━━━
# critical_sync_maintenance.sh runs every 30 minutes.
# Order: CRITICAL_MAINTENANCE_SCRIPTS (jobs) → CRITICAL_SYNC_SHARES (rsync) → partnership --check
# partnership --check always runs last regardless of rsync gate.
# Comment out entries to disable without removing.
CRITICAL_MAINTENANCE_SCRIPTS=(
"Docker_Essentials/downloaders_reset.sh" # clear stuck download states every 30min
"Media/play_state_sync.sh" # sync watched/resume state across Emby + Jellyfin
)
# Shares synced every 30 minutes — defined per host in host*.conf.
# HOST1_CRITICAL_SYNC_SHARES / HOST2_CRITICAL_SYNC_SHARES
# Format: "/path/to/share" or "/path/to/share|profile-name"
# Order matters — Critical-Data first (auth stack), then Emby dirty sync.
# ━━━ Intermediate Sync Maintenance ━━━
# intermediate_sync_maintenance.sh runs every 4 hours — arr library sync, artwork fetch,
# and optional mid-day rsync for any shares that need sub-daily propagation.
# Schedule: 0 */4 * * *
# INTERMEDIATE_SYNC_SHARES is host-specific — configure HOST*_INTERMEDIATE_SYNC_SHARES in host*.conf.
INTERMEDIATE_MAINTENANCE_SCRIPTS=(
# Moved here from CRITICAL (30min) 2026-07-17 — nothing in the critical tier actually
# reads this cache, and every real consumer (arr_sync/arrs_failed_stalled_recovery
# here, lidarr/sonarr/radarr_cleanup + the classification scans in DAILY) already
# tolerates staleness well past 4h (arr_get_tracked_data()'s own fallback tolerates up
# to 1 day). 30min was true "always fresh" but had no consumer that needed it that often.
"Arrs_Stack/arr_cache_prefill.sh ARR_PREFILL_WAIT_MINUTES=1" # keep the shared arr tracked-data cache fresh
"Arrs_Stack/arrs_failed_stalled_recovery.sh" # blocklist + re-search failed/stalled arr queue items
"AI/ai_token_sync.sh" # pull partner AI token ledgers into the tmpfs cache
)
# arr_sync.sh runs as a fixed first step in intermediate_sync_maintenance.sh — not listed here.
# It is controlled by ARR_SYNC_ENABLED (see Arr Sync section above).
# ━━━ Daily Sync Maintenance ━━━
# daily_sync_maintenance.sh runs media share sync first, then iterates
# DAILY_MAINTENANCE_SCRIPTS for all jobs.
# Schedule: 0 1 * * * (1am daily)
DAILY_MAINTENANCE_SCRIPTS=(
"git_pull_execute.sh" # pull latest scripts — always runs first
"Media/media_shares_permissions.sh" # apply permissions — runs before cleaners
"Media/media_cleaner.sh anime" # remove junk from anime shares
"Media/media_cleaner.sh media" # remove junk from media shares
"Arrs_Stack/lidarr_release_fixer.sh" # fix wrong release editions before cleanup runs
"Arrs_Stack/lidarr_duplicate_artist_cleanup.sh" # resolve duplicate MusicBrainz artist entries before cleanup runs
"Arrs_Stack/lidarr_cleanup.sh" # remove orphaned music files
"Arrs_Stack/sonarr_cleanup.sh" # remove orphaned TV files
"Arrs_Stack/radarr_cleanup.sh" # remove orphaned movie files
# Daily, not weekly — DOWNLOAD_ORPHAN_AGE gates each folder on its own mtime, so
# frequency costs nothing in safety, but a weekly run clears a whole week of
# newly-eligible orphans at once and can exceed DOWNLOAD_ORPHAN_MAX_DELETE_GB, which
# aborts the pass entirely and rolls the backlog into an even larger next run.
"Arrs_Stack/arr_download_orphan_cleaner.sh" # sweep orphaned completed downloads out of the SAB Completed folders — deletes junk + already-imported leftovers, triggers import scans for genuinely-missing content
# Runs after cleanup, not before — frees disk space from orphans/junk first, so the
# searches triggered below (for relocated hasFile=false/episodeFileCount=0 entries)
# have headroom for the new grabs they cause, rather than contending with clutter
# cleanup hasn't swept yet.
"Arrs_Stack/radarr_classification_scan.sh --remove-junk --move" # fix anime/kids misclassification + remove bad-metadata entries
"Arrs_Stack/sonarr_classification_scan.sh --move" # fix anime/kids misclassification
"Arrs_Stack/lidarr_missing_art.sh" # fetch missing album/artist artwork (HOST1 only — self-guards)
"Arrs_Stack/radarr_tmdb_removed.sh" # remove movies dropped from TMDb
"Arrs_Stack/sonarr_tvdb_removed.sh" # remove series dropped from TVDB
"Docker_Essentials/docker_update.sh" # pull container image updates before restart
"Docker_Essentials/docker_daily_restart.sh" # daily container restarts — runs last
)
# Pull latest images for DAILY_RESTART_CONTAINERS before the daily restart.
# Containers keep running during pull — no extra downtime.
# Set false to skip updates while still running the daily restart.
DAILY_CONTAINER_UPDATES=true
# Media shares synced daily by daily_sync_maintenance.sh.
# Defined per-host in host*.conf — HOST1_DAILY_SYNC_SHARES and HOST2_DAILY_SYNC_SHARES.
# Mesh model: every node pushes every media share. rsync has no --delete so pushes are additive.
# arr_sync (union) ensures all arr libraries converge first. the arr cleanups remove true orphans.
# Any node can download content to any share — it propagates to all nodes on the next cycle.
# Adding HOST3: list every media share in HOST3_DAILY_SYNC_SHARES. No ownership to track.
# These shares use DEFAULT_RSYNC_OPTS — no profile entry needed.
# For shares needing custom options or container stops — create a profile in RSYNC section.
# Personal encrypted shares defined per-host in host*.conf.
# ZFS encrypted at dataset level — remote receives encrypted blocks, cannot read content.
# See README-Rsync_Setup.md for ZFS encryption setup before uncommenting.
# ━━━ Weekly Sync Maintenance ━━━
# weekly_sync_maintenance.sh stops containers both sides → pulls updates →
# syncs WEEKLY_SYNC_SHARES → restarts → then iterates WEEKLY_MAINTENANCE_SCRIPTS.
# Schedule: 30 2 * * 0 (Sunday 2:30am)
WEEKLY_MAINTENANCE_SCRIPTS=(
"Docker_Essentials/docker_update.sh --weekly" # pull latest images for WEEKLY_RESTART_CONTAINERS before restart
"Docker_Essentials/docker_weekly_restart.sh" # weekly container restarts after sync
"System_Essentials/clear_logs.sh" # purge aged logs — Sunday only, low priority
"Arrs_Stack/arr_full_rescan.sh" # full disk↔DB reconciliation for Lidarr/Sonarr/Radarr — keeps tracked stats honest, runs before discovery so it works off fresh data
"Arrs_Stack/arr_corruption_scan.sh --remediate" # ffprobe-based corruption sweep of Sonarr's tracked files — deletes+re-searches only after CORRUPTION_SCAN_STRIKE_LIMIT consecutive hits on the same file
"Arrs_Stack/playback_aware_lidarr_discovery.sh" # behavior-driven music discovery using weekly Emby playback history
"Arrs_Stack/playback_aware_radarr_discovery.sh" # behavior-driven movie discovery using TMDB recommendations
"Arrs_Stack/playback_aware_sonarr_discovery.sh" # behavior-driven TV discovery using TMDB recommendations
)
# Pull updates for WEEKLY_RESTART_CONTAINERS before docker_weekly_restart.sh runs.
# Set false to skip — docker_weekly_restart.sh still runs regardless.
WEEKLY_CONTAINER_UPDATES=true
# Pull updates for all running containers NOT in daily/weekly managed lists.
# Ensures every deployed container receives at least one image pull per month.
# Set false to skip.
MONTHLY_REMAINING_UPDATES=true
# docker_update.sh rebuilds (stop+recreate onto new image) any container whose image changed,
# in every mode. For daily/weekly, that rebuild is immediately followed by the restart script's
# own unconditional pass — a container that just got rebuilt would be stopped and started again
# right after for no reason. docker_update.sh records which containers it rebuilt to these files;
# docker_daily_restart.sh / docker_weekly_restart.sh read them and skip those containers rather
# than restarting them a second time. A file older than DOCKER_UPDATE_REBUILT_STALE_HOURS is
# treated as untrustworthy (docker_update.sh likely didn't run, or didn't run recently) — deleted,
# and every container in that tier restarts normally, same as if the file never existed.
DOCKER_UPDATE_REBUILT_DAILY_FILE="${DB_DIR}/docker_update_rebuilt_daily.list"
DOCKER_UPDATE_REBUILT_WEEKLY_FILE="${DB_DIR}/docker_update_rebuilt_weekly.list"
DOCKER_UPDATE_REBUILT_STALE_HOURS=12
# Shares synced during the weekly maintenance window — defined per host in host*.conf.
# HOST1_WEEKLY_SYNC_SHARES / HOST2_WEEKLY_SYNC_SHARES
# Containers stopped both sides before sync — full clean state guaranteed.
# Profiles drive container stops, excludes, and options — configure in RSYNC section.
# Order matters — Emby first (larger), then Critical-Data (auth stack).
# Container update toggles for the weekly sync window.
# Containers already stopped for sync — updates pull at no extra downtime.
# Both false → sync only, no updates.
WEEKLY_SYNC_UPDATES=true # pull container updates locally during weekly window
WEEKLY_SYNC_UPDATES_REMOTE=true # pull container updates on remote via SSH
# ━━━ Monthly Maintenance ━━━
# monthly_maintenance.sh fires only when BOTH gates pass:
# 1. Server uptime >= MONTHLY_UPTIME_THRESHOLD_DAYS days
# 2. Last run was >= MONTHLY_RUN_INTERVAL_DAYS days ago (or never run)
# Cron: 0 0 15 * * (15th of the month, midnight — script self-gates, so a spare run is safe)
# NOT in WATCHDOG_ORCHESTRATOR_SCRIPTS — has its own cron entry.
# Add scripts that require a long-stable settled system — scrubs, extended drive tests.
# State file on /boot/config — survives reboots (interval gate independent of uptime gate).
MONTHLY_MAINTENANCE_SCRIPTS=(
#"Tools/zfs_pool_scrub.sh" # ZFS pool integrity scrub — not yet built
#"Tools/smart_long_test.sh" # SMART extended drive health test — not yet built
)
MONTHLY_UPTIME_THRESHOLD_DAYS=30 # minimum uptime in days before maintenance fires
MONTHLY_RUN_INTERVAL_DAYS=30 # minimum days since last run before running again
MONTHLY_LAST_RUN_FILE="$STATE_DIR/monthly_maintenance_last_run.db"
# ━━━ Sunday Morning Coffee Report ━━━
# Orchestrator that runs all Sunday monitor scripts in sequence.
# Schedule: 0 7 * * 0 (Sunday 7am — after weekly_sync_maintenance.sh finishes at ~3am)
# Each script runs independently and notifies on its own findings.
COFFEE_REPORT_SCRIPTS=(
"Monitors/zfs_memory_snapshot.sh" # ZFS pool health + ARC + Docker memory snapshot
"Monitors/smart_health.sh" # drive SMART attributes — reallocated, pending, temp
"Monitors/cert_monitor.sh" # SSL certificate expiry for all configured domains
"Monitors/backup_verify.sh" # rsync mirror integrity via independent MD5 checksums
"Monitors/bandwidth_monitor.sh" # weekly rsync transfer totals and per-share breakdown
"Monitors/emby_session_report.sh" # Emby usage — streams, users, library, transcode ratio
"Monitors/weekly_health_digest.sh" # aggregated digest — watchdogs, fallback, skip list
)
# ==============================================================================================
# ── RSYNC ─────────────────────────────────────────────────────────────────────────────────────
# ==============================================================================================
# ━━━ Rsync Enable/Disable ━━━
# Two-tier toggle system — Tier 1 overrides Tier 2.
#
# Tier 1 — Global gate:
# RSYNC_ENABLED=false → ALL rsync stops everywhere, no exceptions
# Use when: remote completely offline, major maintenance, disaster recovery
#
# Tier 2 — Per-orchestrator (only applies when Tier 1 is true):
# Fine grained control — disable specific orchestrators while keeping others
# Use when: rebuilding secondary, testing, per-window bandwidth management
#
# Example — HOST2 data rebuild:
# RSYNC_ENABLED=true ← rsync works, individual scripts run fine
# CRITICAL_RSYNC_ENABLED=true ← 30min auth stack sync still runs
# INTERMEDIATE_RSYNC_ENABLED=false ← skip 4h arr/mid-day rsync during rebuild
# DAILY_RSYNC_ENABLED=false ← skip daily HDD syncs during rebuild
# WEEKLY_RSYNC_ENABLED=true ← Emby + Critical-Data still sync (NVMe)
# MONTHLY_RSYNC_ENABLED=true ← monthly_maintenance.sh rsync section
# FALLBACK_RSYNC_ENABLED=true ← handback writeback still works
# → Run individual: bash Rsync/rsync.sh /mnt/user/Movies
# → When ready: INTERMEDIATE_RSYNC_ENABLED=true DAILY_RSYNC_ENABLED=true
RSYNC_ENABLED=true # Tier 1 — global gate, overrides everything below
CRITICAL_RSYNC_ENABLED=true # Tier 2 — critical_sync_maintenance.sh rsync section
INTERMEDIATE_RSYNC_ENABLED=true # Tier 2 — intermediate_sync_maintenance.sh rsync section
DAILY_RSYNC_ENABLED=true # Tier 2 — daily_sync_maintenance.sh rsync section
WEEKLY_RSYNC_ENABLED=true # Tier 2 — weekly_sync_maintenance.sh rsync section
MONTHLY_RSYNC_ENABLED=true # Tier 2 — monthly_maintenance.sh rsync section
FALLBACK_RSYNC_ENABLED=true # Tier 2 — fallback.sh writeback jobs on handback
# ━━━ Download Webhook ━━━
# Immediate push to remote nodes on every Sonarr/Radarr/Lidarr Download event.
# Prevents remotes from searching for content that is already on disk locally.
DOWNLOAD_WEBHOOK_ENABLED=true
# ━━━ Rsync Merge Auto-Promote ━━━
# When enabled, rsync.sh pre-scans the remote before syncing and promotes to merge
# mode (pull-then-push) if ≥75% of remote top-level entries exist locally.
# Named profiles are always excluded from auto-promote regardless of this toggle.
RSYNC_MERGE_ENABLED=true
# ━━━ Rsync Defaults ━━━
# Global fallback values used when no profile match is found.
# Media shares in HOST*_DAILY_SYNC_SHARES always use these globals — no profile needed.
# Appdata shares match profiles by directory basename (lowercased).
BW_LIMIT=12500 # KB/s — 12500 ≈ 100Mbit
RETRY_COUNT=3 # retry attempts before giving up
SLEEP=300 # seconds between retry attempts
RSYNC_MAX_RUNTIME_HOURS=19 # cap per transfer attempt — pauses and resumes next scheduled run
# lowered from 23 on 2026-07-16 — daily maintenance jobs alone now
# take ~4h; 19h cap leaves room for them to still run same-day
# before the next 1am fire even if a share hits the cap
CRITICAL_CONTAINER_NAMES=() # containers stopped on REMOTE before rsync — profiles override
DELAYED_CONTAINERS=() # containers needing delay before starting — profiles override
CONTAINER_DELAY=5 # seconds before starting delayed containers
RESTART_VERIFY_WAIT=3 # seconds to wait after restart before checking container is running
EXCLUDE_DIRS=() # directories excluded from transfer — profiles override
# --inplace writes directly to destination — delta against existing file, better for large media
# --partial keep partial file on interrupted transfer so next run resumes, not re-transfers
# --timeout kill stalled transfers instead of hanging indefinitely
# --numeric-ids use UIDs/GIDs numerically — prevents ownership mismatches between servers
# --delete intentionally omitted — the per-arr cleanups (lidarr/sonarr/radarr_cleanup.sh)
# enforce media truth post-rsync.
# Media shares use this default; rsync spreads files only, never removes them.
# Note: --no-whole-file removed — redundant over SSH (delta transfer is already the default).
# Note: the arr cleanups only catch true orphans under the union model — intentional removals
# require arr_sync.sh --blocklist-add first, then manual file deletion.
DEFAULT_RSYNC_OPTS=(-av --info=progress2 --human-readable --bwlimit="$BW_LIMIT" --inplace --partial --timeout=60 --numeric-ids)
# ━━━ Remote Health Checks ━━━
# Pre-flight — aborts if remote rootfs (/) usage is at or above this percentage.
# When remote array is down, rsync writes land on rootfs and fill it rapidly.
ROOTFS_WARN=75
# ━━━ Rsync Profile System ━━━
# Profiles allow per-share rsync behaviour without touching script logic.
# Profile key matched by basename of directory passed to rsync.sh (lowercased).
# Override with --profile=name flag.
#
# IMPORTANT: PROFILE_RSYNC_OPTS does NOT inherit DEFAULT_RSYNC_OPTS.
# List ALL desired options explicitly when defining a profile.
#
# Current profiles:
# arrs_stack — arr databases — lower bandwidth, containers stopped for consistency
# critical-data — auth stack — full stop both sides, Authelia delayed start
# called by weekly_sync_maintenance.sh — full clean sync weekly
# critical-fallback — dirty sync — auth stays running both sides, WAL excluded
# called by critical_sync_maintenance.sh every 30min
# host1-appdata — HOST1 server-specific appdata — defined in host1.conf
# host2-appdata — HOST2 server-specific appdata — defined in host2.conf
# important-data — NextCloud + Postgres — NextCloud delayed start after Postgres
# emby — weekly clean sync — both Emby stopped, full mirror
# called by weekly_sync_maintenance.sh only — do NOT schedule separately
#
# Every profile lookup falls back to the global default when a key is absent
# (${PROFILE_X[name]:-$DEFAULT}), so the value/delay/retry/sleep arrays below only list
# profiles that differ from the default — anything not listed inherits it.
declare -A PROFILE_RSYNC_OPTS=(
[arrs_stack]="-av --info=progress2 --human-readable --bwlimit=$BW_LIMIT --delete --inplace"
[critical-data]="-av --human-readable --bwlimit=$BW_LIMIT --delete"
[critical-fallback]="-av --human-readable --bwlimit=$BW_LIMIT --delete --inplace --no-whole-file"
[important-data]="-av --human-readable --bwlimit=$BW_LIMIT"
[emby]="-av --human-readable --bwlimit=$BW_LIMIT --delete --inplace --no-whole-file"
)
# Per-profile bandwidth limits in KB/s
declare -A PROFILE_BW_LIMIT=(
[arrs_stack]=5000 # lower — runs alongside other syncs
[critical-data]=9500 # high — small dataset, sync fast
[critical-fallback]=9500 # high — small dataset, sync fast
[important-data]=9500 # high — database sync
[emby]=8000 # medium — large full mirror
)
# Seconds between retry attempts — only non-default profiles listed (default SLEEP applies otherwise)
declare -A PROFILE_SLEEP=(
[critical-fallback]=120 # shorter — frequent dirty sync, retry faster
)
# Containers stopped on BOTH LOCAL and REMOTE before rsync.
# Local stops first — flushes databases cleanly. Remote stops next — prevents writes.
# Only running containers get restarted — stopped containers stay stopped.
# SPACE-SEPARATED STRINGS — converted to array at runtime
declare -A PROFILE_CRITICAL_CONTAINER_NAMES=(
[arrs_stack]="Sonarr Lidarr Readarr Radarr Prowlarr Bazarr Pinchflat"
[critical-data]="" # e.g. "Mariadb Authelia Redis NginxProxyManager"
[critical-fallback]="" # dirty sync — auth stays running both sides
[important-data]="" # e.g. "Postgres NextCloud"
[emby]="Emby"
)
# Containers needing a delay after rsync before starting.
# SPACE-SEPARATED STRINGS — converted to array at runtime
declare -A PROFILE_DELAYED_CONTAINERS=(
[arrs_stack]=""
[critical-data]="" # e.g. "Authelia" — containers that need a delay after db starts
[critical-fallback]=""
[important-data]="NextCloud" # wait for Postgres
[emby]=""
)
# Seconds before starting delayed containers — only non-default profiles listed (default CONTAINER_DELAY applies otherwise)
declare -A PROFILE_CONTAINER_DELAY=(
[critical-data]=15 # Mariadb + Redis need time to accept connections
[important-data]=10 # Postgres needs time before NextCloud
)
# Directories excluded from rsync per profile.
# SPACE-SEPARATED STRINGS — converted to array at runtime
declare -A PROFILE_EXCLUDE_DIRS=(
[arrs_stack]="logs *.tmp"
[critical-data]="logs *.tmp *.log nginx/temp nginx/cache __pycache__ notification.txt"
[critical-fallback]="logs *.tmp *.log nginx/temp nginx/cache __pycache__ notification.txt *.db-wal *.db-shm"
[important-data]="logs *.tmp"
[emby]="logs transcodes cache crash*"
)
# Remote restart after dirty sync — restart these on remote IF they were running before sync.
# Same logic as stop/start — was stopped = stays stopped, was running = gets restarted.
# Used by dirty sync profiles (critical-fallback) so remote picks up changes.
# SPACE-SEPARATED STRINGS — converted to array at runtime
declare -A PROFILE_REMOTE_RESTART_CONTAINERS=(
[critical-fallback]="" # containers to restart on remote after dirty sync
)
# Note: disk check is auto-detected from disks.ini — no PROFILE_SKIP_DISK_CHECK needed.
# check_remote_disks() reads fsType per disk and handles XFS, ZFS, and cache pools automatically.
# ==============================================================================================
# ── FALLBACK ──────────────────────────────────────────────────────────────────────────────────
# ==============================================================================================
# Mutual container failover between two unRAID servers.
# Each server runs Fallback/fallback.sh independently via array_started.sh.
# All decisions based on two pings: remote reachable + internet reachable.
#
# States: NORMAL | FALLBACK | NO_INTERNET | DARK
#
# DDNS rules — absolute:
# Internet loss → stop own DDNS immediately
# Failover → start remote DDNS first (Tier 1)
# Handback → stop remote DDNS → rsync → start containers → start local DDNS last
#
# Per-host container lists and tier delays live in host*.conf.
# Shared settings (intervals, state file, thresholds) live here.
EXTERNAL_IP="8.8.8.8"
FALLBACK_CHECK_INTERVAL=30 # seconds between fallback state checks
FALLBACK_HANDBACK_STRIKES=3 # consecutive healthy checks before initiating handback (3×30s = 90s)
FALLBACK_STATE_FILE="$STATE_DIR/fallback_state.db"
FALLBACK_ENABLED=false # set true once both servers are configured and paired
# false = suppresses "not running" warnings in status scripts
FALLBACK_PARTNERSHIP_SUSPEND_AFTER=120 # minutes without active partnership before suspending
# 0 = suspend immediately when partnership goes inactive
# ━━━ Play State Sync ━━━
PLAY_SYNC_ENABLED=true
PLAY_SYNC_REMOTE=true # sync across hosts via Tailscale
PLAY_SYNC_TYPES="Movie,Episode" # Audio excluded — music library too large; favorites handled separately
PLAY_SYNC_FAV_TYPES="MusicArtist,MusicAlbum,Movie,Series" # union sync — never unmarks; Audio tracks future
PLAY_SYNC_PROBE=true # hash fetched state, skip per-item comparison when nothing changed
PLAY_SYNC_PROBE_MAX_AGE_HOURS=24 # force a full comparison when the fingerprint is older than this
# ━━━ Play State Sync — Handback ━━━
# play_state_sync runs before DNS cutover so users land on current watch state.
# Retried until success — DNS is held until sync completes or retries are exhausted.
PLAY_SYNC_HANDBACK_RETRIES=5 # attempts before giving up and proceeding to DNS cutover
PLAY_SYNC_HANDBACK_RETRY_DELAY=60 # seconds between retry attempts
# ━━━ Failover Test ━━━
# Controlled simulation of a failover event — run manually via fallback_test.sh.
FALLBACK_TEST_BLOCK_WAIT=150 # seconds to wait after blocking connectivity
FALLBACK_TEST_HANDBACK_WAIT=360 # seconds to wait before initiating handback
# ==============================================================================================
# ── DOCKER ESSENTIALS ─────────────────────────────────────────────────────────────────────────
# ==============================================================================================
# ━━━ Downloaders Reset ━━━
# Runs every 30 minutes via CRITICAL_MAINTENANCE_SCRIPTS.
# Clears stuck states, purges old history, prepares each download client for a clean cycle.
# Per-host URLs and API keys live in host*.conf.
DOWNLOADER_RETENTION_DAYS=7 # days — purge history older than this
# qBittorrent failsafe — removes torrents older than threshold regardless of ratio
# deleteFiles=false — removes from qBit but leaves files on disk (arr manages files)
QBIT_FAILSAFE_MIN_DAYS=180 # days — minimum age before failsafe deletion
QBIT_FAILSAFE_MIN_RATIO=0 # 0 = age only, no ratio requirement
# ━━━ Docker Daily Restart ━━━
# Containers restarted every day via DAILY_MAINTENANCE_SCRIPTS.
# Per-host lists live in host*.conf:
# HOST1_DAILY_RESTART_CONTAINERS
# HOST2_DAILY_RESTART_CONTAINERS
# detect_hosts() sets DAILY_RESTART_CONTAINERS to the correct host array at runtime.
# ━━━ Docker Weekly Restart ━━━
# Less critical services restarted weekly via WEEKLY_MAINTENANCE_SCRIPTS (Sunday 2:30am).
# Containers already stopped for weekly sync — restart adds zero extra downtime.
# Per-host lists live in host*.conf:
# HOST1_WEEKLY_RESTART_CONTAINERS
# HOST2_WEEKLY_RESTART_CONTAINERS
# ━━━ Docker Watchdog ━━━
# Continuous two-tier self-healing container monitoring.
# Started by array_started.sh — runs until array stops.
# Re-sources all three conf files each cycle — add/remove containers without restarting watchdog.
#
# Tier 1 — strict monitoring of explicitly configured containers:
# Memory hard limits — immediate restart if exceeded
# CPU thresholds — strike system, restart after CPU_FAIL_LIMIT sustained strikes
# HTTP responsiveness — strike system, restart after RESP_FAIL_LIMIT failed checks
# Required containers — must always be running, strike + skip list with auto-clear
#
# Tier 2 — global health scan of ALL running containers:
# Unhealthy status — Docker HEALTHCHECK unhealthy → restart
# OOM killed — kernel killed → restart + notify
# Crash loop detection — RestartCount climbing → notify, critical above limit
# Dead containers — remove and restart
# Unexpected exits — non-zero exit code → restart
#
# All per-host container lists live in host*.conf:
# HOST*_WATCHDOG_CONTAINERS — memory hard limits per container
# HOST*_WATCHDOG_CONTAINER_URLS — HTTP health check URLs
# HOST*_WATCHDOG_REQUIRED_CONTAINERS — must always be running
# HOST*_WATCHDOG_SCAN_IGNORE — skip in Tier 2 scan
# HOST*_WATCHDOG_DEPENDENCIES — dependency ordering for restart decisions
# Strike state file — /tmp resets on reboot (correct — no stale strikes after reboot)
WATCHDOG_STATE_FILE="$STATE_DIR/container_watchdog_state.db"
# CPU thresholds — normalised against total core count at runtime
SOFT_CPU_THRESHOLD=80 # warn at this % of total system CPU
HARD_CPU_THRESHOLD=85 # strike at this % of total system CPU
CPU_FAIL_LIMIT=2 # consecutive hard CPU strikes before container restart
# Memory soft threshold — warn when container reaches this % of its hard limit
SOFT_MEM_THRESHOLD=80
# HTTP responsiveness
RESP_FAIL_LIMIT=2 # consecutive failed checks before restart
CURL_TIMEOUT=5 # seconds per check before timeout
# Tier 2 master toggle
WATCHDOG_SCAN_ALL=true # false = only WATCHDOG_CONTAINERS + required containers
# Individual Tier 2 check toggles
WATCHDOG_RESTART_UNHEALTHY=true
WATCHDOG_RESTART_DEAD=true
WATCHDOG_RESTART_CRASHED=true
WATCHDOG_NOTIFY_OOM=true
WATCHDOG_NOTIFY_CRASHLOOP=true
# Crash loop threshold
WATCHDOG_CRASH_LIMIT=5 # RestartCount above this = critical crash loop
# Required-container strikes — consecutive down-checks before docker_watchdog.sh attempts a restart
WATCHDOG_REQUIRED_STRIKE_LIMIT=2
# Startup grace period — skip restarts while system is still booting
WATCHDOG_STARTUP_GRACE=600 # seconds after boot before watchdog acts on failures
# Restart loop protection — prevents watchdog from endlessly restarting a broken container
WATCHDOG_CONTAINER_RESTART_LIMIT=3
WATCHDOG_CONTAINER_RESTART_WINDOW=1 # rolling window in hours
WATCHDOG_CONTAINER_RESTART_LOG="${DB_DIR}/container_restart_history.db"
# Notification batching — one summary per cycle instead of one ping per event
WATCHDOG_BATCH_NOTIFY=true
# Docker daemon health check — hung daemon detection at the start of each cycle
WATCHDOG_DAEMON_TIMEOUT=20 # seconds — timeout for all docker commands
WATCHDOG_DAEMON_STRIKE_LIMIT=3 # consecutive failed checks before restart attempt
WATCHDOG_DAEMON_RESTART_WAIT=900 # seconds to wait after restart before verifying
# Appdata size monitoring — two-part catch-all for runaway growth and oversized log files.
#
# Part 1 — Growth rate (zero-config):
# Reads per-container dir totals each cycle via du, compares to previous cycle.
# Any container growing more than WATCHDOG_APPDATA_GROWTH_GB triggers a focused *.log scan
# inside that container. No per-container config required — new containers are covered
# automatically. Baseline built on first run after boot; growth detection starts cycle 2.
#
# Part 2 — Absolute log size:
# Finds *.log / *.log.* files over WATCHDOG_APPDATA_LOG_MAX_GB across all appdata paths.
# Catches logs that have already stabilised at a large size and are no longer actively growing.
#
# Strike system (reuses existing watchdog infrastructure):
# Strike 1 — warn + notify: condition first detected
# Strike 2 — warn + escalated notify: still present next cycle
# Strike 3 (WATCHDOG_APPDATA_STRIKE_LIMIT) — action cycle:
# If WATCHDOG_APPDATA_TRUNCATE_LOGS=true: truncate *.log files in-place, clear strikes
# If false: critical notify only, strikes held until condition resolves
# Condition resolves (growth stops / log drops below threshold) → strikes auto-clear
#
# HOST*_WATCHDOG_APPDATA_SIZES (in host*.conf) suppresses growth warnings for a
# container until its dir exceeds the configured ceiling. Only needed when a container
# legitimately has large stable data and you want to guarantee it never triggers a false alarm.
WATCHDOG_CHECK_APPDATA=true
WATCHDOG_APPDATA_PATHS=("/mnt/docker-unraid/appdata")
WATCHDOG_APPDATA_GROWTH_GB=2 # flag containers growing more than this per cycle
WATCHDOG_APPDATA_LOG_MAX_GB=2 # flag *.log files exceeding this size (absolute)
WATCHDOG_APPDATA_TRUNCATE_LOGS=false # set true to auto-truncate oversized *.log files on action cycle
WATCHDOG_APPDATA_STRIKE_LIMIT=3 # cycles before action fires (matches existing watchdog pattern)
WATCHDOG_APPDATA_GROWTH_FILE="$STATE_DIR/watchdog_appdata_growth.db"
STORAGE_WATCHDOG_STATE_FILE="$STATE_DIR/storage_watchdog_state.db"
# ━━━ Docker Network Connect ━━━
# Ensures custom networks exist and connects containers at array start.
# Runs once via ARRAY_START_SCRIPTS — idempotent, safe to re-run.
# Container and network lists are host-specific — defined in host*.conf:
# HOST1_NETWORK_CONNECT_CONTAINERS / HOST2_NETWORK_CONNECT_CONTAINERS
# HOST1_NETWORK_CONNECT_NETWORKS / HOST2_NETWORK_CONNECT_NETWORKS
# ==============================================================================================
# ── UNRAID ESSENTIALS ─────────────────────────────────────────────────────────────────────────
# ==============================================================================================
# ━━━ inotify Tuning ━━━
# Linux inotify limits — applied at every array start by inotify_tuning.sh.
# Default unRAID values are very low — with many Docker containers watching files
# you can silently exhaust the limit causing containers to miss file events.
# Symptoms: containers miss file events, downloads not detected, library not updated.
# Live TV stutter is a known symptom of inotify exhaustion.
# These settings are lost on reboot — reapplied automatically at array start.
INOTIFY_MAX_INSTANCES=1024 # default: 128 — max inotify instances per user
# 1024 handles ~20-30 containers watching files simultaneously
INOTIFY_MAX_WATCHES=1048576 # default: 8192 — SHARED budget across ALL users/containers
# 524288 (512K) was previous value — raised to 1M (1048576)
# VSCode/Code-Server alone needs ~50K-200K for large workspaces
# with node_modules. All arr containers + Emby + VSCode share
# this budget. 1M is safe on 128GB RAM (~128MB kernel memory)
# If VSCode shows "unable to watch for file changes" → too low
INOTIFY_MAX_QUEUED_EVENTS=32768 # default: 16384 — max events queued before dropping
# ━━━ System Tuning Monitor ━━━
# Tracks inotify and php-fpm usage over time.
# Snapshot written every 6 hours by system_tuning_monitor.sh.
# Read by sunday_morning_coffee_report.sh for weekly peak/avg/warning summary.
INOTIFY_WARN_PCT=80 # warn if inotify instances exceed this % of limit
PHP_FPM_WARN_PCT=80 # warn if php-fpm workers exceed this % of max_children
TUNING_MONITOR_LOG="${DB_DIR}/system_tuning_history.db"
TUNING_LOG_RETENTION=30 # days before old entries are purged
# ━━━ Reboot ━━━
# Seconds of warning broadcast to logged-in users before server_reboot.sh reboots.
# Gives users time to save work — 300s = 5 minutes.
REBOOT_SLEEP=300
# Seconds to wait for graceful VM shutdown (ACPI signal via virsh) before libvirt
# stops it anyway — reboot takes priority.
REBOOT_VM_WAIT=30
# ━━━ Mover ━━━
# Seconds to wait before mover_stop.sh sends SIGTERM to the mover process.
# Gives mover time to finish current file transfer before being interrupted.
MOVER_STOP_TIMEOUT=300
# ━━━ Syslog Filter ━━━
# Path for the rsyslog filter file that suppresses Docker veth interface noise.
# Docker creates a new veth interface for each container — generates hundreds of
# log lines per hour that have no diagnostic value. Filter removes them at source.
FILTER_FILE="/etc/rsyslog.d/ignore-docker-veth.conf"
# ━━━ PHP-FPM ━━━
# Higher max_children allows more concurrent PHP requests to the unRAID WebGUI.
# Default is very low — increasing it prevents WebGUI slowdowns under load.
# 250 is safe for servers with 32GB+ RAM.
PHP_CONF="/etc/php-fpm.d/www.conf"
PHP_MAX_CHILDREN=250
# ━━━ Clear Logs ━━━
# System log files cleared weekly to prevent rootfs fill over time.
# These grow continuously — without clearing they eventually consume all rootfs space.
# clear_logs.sh runs in WEEKLY_MAINTENANCE_SCRIPTS — Sunday 2:30am.
#
# Size threshold approach — only clear if log exceeds threshold.
# Avoids destroying useful recent diagnostic context when logs are small.
# LOG_MIN_SIZE_MB: skip clearing if log is under this size (not worth clearing)
# LOG_DOCKER_MAX_MB: clear a container log only if it exceeds this size
# Docker logs grow fastest on active containers (Emby, SABnzbd, Sonarr)
# 100MB per container × 30 containers = 3GB before clearing kicks in
LOG_FILES=(/var/log/syslog /var/log/messages /var/log/dmesg)
LOG_MIN_SIZE_MB=10 # skip system log if under this size (already small)
LOG_DOCKER_MAX_MB=100 # clear Docker container log only if over this size (MB)
# ━━━ Network Watchdog ━━━
# Services-layer connectivity — internet reachability, DDNS sync, Tailscale, NPM proxy.
# Host-specific values (domain, container, NPM URL) live in host*.conf.
NETWORK_WATCHDOG_ENABLED=true
NETWORK_WATCHDOG_INTERNET_URL="https://1.1.1.1"
NETWORK_WATCHDOG_INTERNET_TIMEOUT=5
NETWORK_WATCHDOG_CHECK_TAILSCALE=true
NETWORK_WATCHDOG_NPM_TIMEOUT=10
NETWORK_WATCHDOG_NPM_STRIKE_LIMIT=2
NETWORK_WATCHDOG_NPM_STATE_FILE="$STATE_DIR/network_watchdog_state.db"
# ━━━ WebGUI Watchdog ━━━
# Monitors unRAID WebGUI responsiveness — escalates through nginx restart → emhttp restart.
# Separate from docker_watchdog — this monitors the unRAID UI itself, not containers.
WEBGUI_URL="http://localhost"
WEBGUI_TIMEOUT=5 # seconds before curl gives up on WebGUI check
WEBGUI_NGINX_WAIT=15 # seconds after nginx restart before rechecking
WEBGUI_PHP_WAIT=10 # seconds after php-fpm restart before rechecking
WEBGUI_EMHTTP_WAIT=30 # seconds after emhttp restart before rechecking
# ==============================================================================================
# ── MEDIA ─────────────────────────────────────────────────────────────────────────────────────
# ==============================================================================================
# ━━━ Media Permissions ━━━
# Applied recursively by media_shares_permissions.sh.
# Runs first in DAILY_MAINTENANCE_SCRIPTS — arr cleanup depends on correct ownership.
#
# Why split directory vs file permissions:
# Directories need execute bit to enter — 755 allows owner+group+others to traverse
# Files should NOT be executable — 664 allows owner+group read/write, others read
# This is POSIX best practice for media servers, not a blanket 777 band-aid
#
# Why nobody:users ownership:
# All ecosystem containers run as PUID=99 (nobody) PGID=100 (users) on unRAID
# Files owned by nobody:users are accessible to all containers without world-write
# If this script fixes many files each run → a container has wrong PUID/PGID set
# Fix: add PUID=99 PGID=100 to the container's environment variables
#
# This script runs daily as a failsafe — even with correct container config:
# rsync may bring files with source ownership if not run with --chown
# Manual admin copies create root:root files
# New containers may not have PUID/PGID set yet
PERMISSIONS_DIR_MODE="755" # directories — traverse + list, no world-write
PERMISSIONS_FILE_MODE="664" # files — group read/write, no execute
PERMISSIONS_OWNER="nobody:users"
# Share list defined per host in host*.conf:
# HOST1_MEDIA_PERMISSION_SHARES / HOST2_MEDIA_PERMISSION_SHARES
# ━━━ Media Cleaner ━━━
# Removes junk files from media shares — two profiles: anime and media.
# Called via DAILY_MAINTENANCE_SCRIPTS. Run manually: Media/media_cleaner.sh anime|media
# Folder lists defined per host in host*.conf:
# HOST1_ANIME_CLEAN_FOLDERS / HOST2_ANIME_CLEAN_FOLDERS
# HOST1_MEDIA_CLEAN_FOLDERS / HOST2_MEDIA_CLEAN_FOLDERS
ANIME_FILE_PATTERNS=(
# ── Checksums ────────────────────────────────────────────────────────────────────────
'*.sfv' '*.md5' '*.sha1' # verification files — useless post-download
# ── Scene / download metadata ─────────────────────────────────────────────────────────
'*.url' '*.lnk' # scene links
'*.info' '*.diz' # scene description files
'*.nzb' # usenet download files
'*.torrent' # torrent files left by download clients
# ── Archives and segments ─────────────────────────────────────────────────────────────
'*.rar' '*.zip' '*.7z' '*.ace' # archives — source files not needed after extract
'*.r00' '*.r01' '*.r02' '*.r03' # multi-part rar segments
'*.r04' '*.r05' '*.r06' '*.r07'
'*.r08' '*.r09' '*.srr' # scene repair files
'*.001' '*.002' '*.003' # split archive parts
'*.gz' '*.tar' '*.bz2' # linux archives
# ── Scene samples and proofs ──────────────────────────────────────────────────────────
'*.sample*' '*.proof*' # scene samples — never needed in library
# ── Executables and scripts ───────────────────────────────────────────────────────────
'*.exe' '*.scr' '*.com' # Windows executables
'*.bat' '*.cmd' '*.vbs' '*.ps1' # Windows scripts
'*.msi' '*.dll' '*.sys' # Windows installers and system files
'*.sh' # shell scripts in media folders = suspicious
# ── Incomplete downloads ──────────────────────────────────────────────────────────────
'*.!ut' '*.!qB' # uTorrent / qBittorrent incomplete markers
'*.crdownload' '*.opdownload' # Chrome / Opera incomplete downloads
'*.part' # partial download files
# ── Sync conflicts ────────────────────────────────────────────────────────────────────
'*sync-conflict*' # Syncthing conflict copies
)
# Media = the anime set plus two media-specific extras. Defined by composition so the
# shared patterns can never drift between the two lists.
MEDIA_FILE_PATTERNS=(
"${ANIME_FILE_PATTERNS[@]}"
# ── Media-specific extras ─────────────────────────────────────────────────────────────
'*.iso' # disc images — extracted content is in library
'*.lrc' # lyrics files — handled by music apps not Emby
)
# ━━━ Arr Cleanup ━━━
# Orphan file cleanup via Lidarr, Sonarr, and Radarr APIs.
# Per-host URLs, API keys, and path maps live in host*.conf.
# detect_hosts() selects correct host vars at runtime.
#
# API versions — update MAJOR version here when script is updated for a new arr version:
# Sonarr v4 → /api/v3/series → /api/v3/episodefile?seriesId=X
# Radarr v6 → /api/v3/movie → /api/v3/moviefile?movieId=X
# Lidarr v3 → /api/v1/artist → /api/v1/trackFile?artistId=X
SONARR_VERSION_MAJOR=4
RADARR_VERSION_MAJOR=6
LIDARR_VERSION_MAJOR=3
# arr_profile_enforcer.sh — quality profile names by root folder type
# Root folder paths containing "kids" or "anime" → ARR_KIDS_PROFILE_NAME
# All other root folders → ARR_*_DEFAULT_PROFILE
ARR_KIDS_PROFILE_NAME="Kids shows"
ARR_SONARR_DEFAULT_PROFILE="Any"
ARR_RADARR_DEFAULT_PROFILE="Any (mine)"
# Lidarr shared settings
LIDARR_RELEASE_FIXER_ENABLED=true # set false to disable without removing from job list
LIDARR_LOCK_WARN_AGE=3600 # 1hr — large libraries take time, not stuck
LIDARR_ORPHAN_AGE=7 # days — files must be older than this before eligible for deletion
LIDARR_MAX_DELETE_GB=5 # require --i-know-what-im-doing if deletion exceeds this
LIDARR_MIN_TRACKED_PCT=80 # abort if tracked count drops below this % of last run
# protects against API returning partial data on a bad day
LIDARR_TRACKED_COUNT_FILE="${DB_DIR}/lidarr_tracked.count"
LIDARR_IMPORT_SCAN_TIMEOUT=600 # seconds to wait for pre-flight import scan
# Lidarr tracked-data cache — shared by lidarr_cleanup.sh, lidarr_duplicate_artist_cleanup.sh,
# lidarr_missing_art.sh, lidarr_release_fixer.sh, and arr_cache_prefill.sh. See
# lidarr_get_tracked_data() in common.sh for the fresh/stale/rescan-active branching logic.
LIDARR_CACHE_FILE="${ARR_CACHE_BACKUP_DIR}/lidarr_tracked_cache.json"
LIDARR_RESCAN_DURATION_DB="${DB_DIR}/lidarr_rescan_duration.db"
LIDARR_CACHE_MAX_AGE_DAYS=1 # force a live refresh (or rescan-aware wait) past this age
ARR_PREFILL_WAIT_MINUTES=10 # array-start prefill: how long to retry reaching each arr
LIDARR_EXTENSIONS=("flac" "mp3" "m4a" "wav" "aac" "ogg" "opus" "wma")
LIDARR_PROTECTED_PATTERNS=(
# Metadata
"*.nfo" "*.tbn"
# Images — album art, artist images, Emby artwork
"*.jpg" "*.jpeg" "*.png" "*.webp" "*.svg"
"poster.*" "fanart.*" "backdrop.*" "clearlogo.*"
"banner.*" "thumb.*" "landscape.*"
"folder.*" "cover.*" "album.*" "artist.*" "disc.*"
# Lyrics
"*.lrc"
)
# Lidarr artwork fetcher settings
LIDARR_ART_MIN_SIZE=10000 # bytes — reject downloads smaller than this
LIDARR_ART_MAX_PARALLEL=4 # concurrent background download jobs
LIDARR_ART_RETRIES=2 # download retry attempts per image
LIDARR_ART_SLEEP_BETWEEN=0.2 # seconds between fanart.tv API calls
LIDARR_ART_RECHECK_DAYS=30 # days before re-querying art that upstream didn't have
LIDARR_ART_MISS_CACHE="${DB_DIR}/lidarr_art_miss_cache.tsv" # negative cache — art upstream has never had
# HOST*_FANART_API_KEY / HOST*_LASTFM_API_KEY — set in host*.conf
# Lidarr discovery settings (playback_aware_lidarr_discovery.sh)
LIDARR_DISCOVERY_THRESHOLD=70 # score to accept candidate (0-100)
LIDARR_DISCOVERY_LOOKBACK_DAYS=7 # Emby play history window in days
LIDARR_DISCOVERY_MIN_PLAYS=3 # min plays in window before evaluating an artist
LIDARR_DISCOVERY_USER_CAP_PCT=35 # max % any single user can contribute to play score (prevents one listener dominating)
LIDARR_DISCOVERY_MAX_ADDS=5 # max artists to add per run — quality over bulk
LIDARR_DISCOVERY_REJECT_COOLDOWN=30 # days before re-evaluating a rejected artist
LIDARR_DISCOVERY_HISTORY="${DB_DIR}/lidarr_discovery_history.db"
# Sonarr discovery settings (playback_aware_sonarr_discovery.sh)
SONARR_DISCOVERY_THRESHOLD=52 # score to accept candidate (0-100)
SONARR_DISCOVERY_LOOKBACK_DAYS=14 # Emby episode play history window in days (shorter than movies — TV watched more frequently)
SONARR_DISCOVERY_MAX_SEEDS=5 # max seed series from Stage 1
SONARR_DISCOVERY_MAX_ADDS=3 # max shows to add per run — TV is a larger commitment than movies
SONARR_DISCOVERY_MIN_VOTE_COUNT=50 # min TMDB votes (TV has fewer votes than movies at same popularity)
SONARR_DISCOVERY_MIN_RATING=65 # min TMDB vote_average × 10 (65 = 6.5/10)
SONARR_DISCOVERY_REJECT_COOLDOWN=60 # days before re-evaluating a rejected show
SONARR_DISCOVERY_USER_EPISODE_CAP=8 # max episodes any one user contributes to seed volume score
SONARR_DISCOVERY_MONITOR_MODE="all" # Sonarr monitor mode on add: all | future | first | latest | none
SONARR_DISCOVERY_HISTORY="${DB_DIR}/sonarr_discovery_history.db"
# Radarr discovery shared settings
RADARR_DISCOVERY_THRESHOLD=52 # score to accept candidate (0-100) — lower than Lidarr since diverse seeds rarely overlap
RADARR_DISCOVERY_LOOKBACK_DAYS=30 # Emby watch history window in days (movies rewatched less often)
RADARR_DISCOVERY_MAX_SEEDS=5 # max seed movies from Stage 1
RADARR_DISCOVERY_MAX_ADDS=5 # max movies to add per run
RADARR_DISCOVERY_MIN_VOTE_COUNT=100 # min TMDB votes to be considered a candidate
RADARR_DISCOVERY_MIN_RATING=60 # min TMDB vote_average × 10 (60 = 6.0/10)
RADARR_DISCOVERY_REJECT_COOLDOWN=60 # days before re-evaluating a rejected movie
RADARR_DISCOVERY_SEED_LIBRARIES=("Movies") # Emby libraries to draw seed movies from
RADARR_DISCOVERY_HISTORY="${DB_DIR}/radarr_discovery_history.db"
# Emby → arr sync library allowlists
# Only these Emby library names will be considered by the sync tools.
# Names must match exactly as shown in Emby > Dashboard > Libraries.
SONARR_EMBY_LIBRARIES=("Anime" "Kids Shows" "Stand-Up Comedy" "TV shows")
RADARR_EMBY_LIBRARIES=("Movies" "Kid's Movies")
# Sonarr shared settings
SONARR_ORPHAN_AGE=7 # days — files must be older than this before eligible for deletion
SONARR_MAX_DELETE_GB=10 # require --i-know-what-im-doing if deletion exceeds this
SONARR_MIN_TRACKED_PCT=80 # abort if tracked count drops below this % of last run
# protects against API returning partial data on a bad day
SONARR_TRACKED_COUNT_FILE="${DB_DIR}/sonarr_tracked.count"
SONARR_IMPORT_SCAN_TIMEOUT=600 # seconds to wait for pre-flight import scan
SONARR_MOVE_POLL_TIMEOUT=3600 # seconds to wait for a single async MoveSeries command to
# reach "completed" — generous because a large series can sit
# queued behind other moves already in progress, not just its
# own copy time
CORRUPTION_SCAN_STATE_FILE="${DB_DIR}/corruption_scan_state.tsv" # clean-file skip-cache
CORRUPTION_SCAN_STRIKES_FILE="${DB_DIR}/corruption_scan_strikes.tsv" # consecutive corrupt-detection counts, keyed by host path
CORRUPTION_SCAN_STRIKE_LIMIT=2 # consecutive corrupt detections (across separate scan runs)
# required before --remediate deletes+re-searches — guards
# against a one-off ffprobe hiccup (mid-write file, NFS blip)
# triggering an unnecessary delete. Resets to 0 the moment a
# file probes clean again.
SONARR_EXTENSIONS=("mkv" "mp4" "avi" "m4v" "ts" "wmv" "mov")
SONARR_PROTECTED_PATTERNS=(
# Subtitles
"*.srt" "*.sub" "*.ass" "*.ssa" "*.idx" "*.vtt"
# Metadata
"*.nfo" "*.tbn"
# Images
"*.jpg" "*.jpeg" "*.png" "*.webp" "*.svg"
"poster.*" "fanart.*" "backdrop.*" "clearlogo.*"
"banner.*" "thumb.*" "landscape.*"
# Kodi/Emby extras — not tracked by Sonarr API
"*-trailer.*" "*-featurette.*" "*-behindthescenes.*"
"*-interview.*" "*-scene.*" "*-short.*" "*-deleted.*"
"*-clip.*" "*-other.*"
# Theme songs — stored in show folder, not tracked by arr
"theme.mp3" "theme.flac" "theme.wav" "theme.m4a" "theme.mka"
"theme.mkv" "theme.mp4"
"trailer.*" "trailer *.*" "trailer-*.*"
)
# Radarr shared settings
RADARR_ORPHAN_AGE=7 # days — files must be older than this before eligible for deletion
RADARR_MAX_DELETE_GB=30 # require --i-know-what-im-doing if deletion exceeds this
RADARR_MIN_TRACKED_PCT=80 # abort if tracked count drops below this % of last run
# protects against API returning partial data on a bad day
RADARR_TRACKED_COUNT_FILE="${DB_DIR}/radarr_tracked.count"
RADARR_IMPORT_SCAN_TIMEOUT=600 # seconds to wait for pre-flight import scan
RADARR_MOVE_POLL_TIMEOUT=3600 # seconds to wait for a single async MoveMovie command to
# reach "completed" — mirrors SONARR_MOVE_POLL_TIMEOUT
RADARR_EXTENSIONS=("mkv" "mp4" "avi" "m4v" "wmv" "mov")
# Identical protected set to Sonarr's (sidecars/artwork/extras the arr doesn't track are
# the same for movies and shows). Reused by reference so the two can never drift.
RADARR_PROTECTED_PATTERNS=("${SONARR_PROTECTED_PATTERNS[@]}")
# Radarr/Sonarr TMDb/TVDB removed entry cleanup
RADARR_DROPPED_ADD_EXCLUSION=true # add removed movies to import exclusion list
SONARR_DROPPED_ADD_EXCLUSION=true # add removed series to import exclusion list
# ━━━ Download Orphan Cleaner (arr_download_orphan_cleaner.sh) ━━━
# Weekly sweep of the SABnzbd Completed folders Sonarr/Radarr import from — deletes junk and
# parse-verified already-in-library leftovers the arr queue no longer references, triggers
# import scans for anything the library is actually missing. Built 2026-07-26 after 755G of
# orphaned completed downloads (accumulating since 2022) filled HOST1's cache pool to 89%.
# Per-host dirs: HOST*_SONARR_DOWNLOAD_DIR / HOST*_RADARR_DOWNLOAD_DIR (+ _CONTAINER_DIR).
DOWNLOAD_ORPHAN_CLEANER_ENABLED=true
DOWNLOAD_ORPHAN_AGE=7 # days — entries younger than this may be mid-import, never touched
DOWNLOAD_ORPHAN_MIN_VIDEO_MB=50 # no video file above this = junk (par2 debris, samples, dead archives)
DOWNLOAD_ORPHAN_MAX_DELETE_GB=100 # abort delete pass over this — a partial queue fetch would classify
# live downloads as orphans, and a huge total is that failure's symptom;
# --i-know-what-im-doing overrides for known backlogs
# ━━━ Arr Content Classification (radarr/sonarr_classification_scan.sh) ━━━
#
# Curated lists validated against real library data 2026-07-17 — every entry here was
# checked against the actual signal ratio (kids%/anime% of titles under that studio/network)
# before being added, specifically to avoid the false-positive traps found this session:
# general umbrella studios (Walt Disney Pictures, Paramount Pictures), general streaming
# platforms (Netflix, Disney+, Hulu), and networks that carry mixed content under one name
# (The WB, UPN) are deliberately excluded even though they produce plenty of kids/anime
# content — they also produce just as much non-kids/non-anime content under the same name.
#
# Anime always takes priority over kids when both signals could apply (explicit user rule).
#
RADARR_ANIME_STUDIOS=(
"Studio Ghibli" "ufotable" "Toei Company" "Toei Animation" "MAPPA" "Madhouse"
"Production I.G" "Wit Studio" "Bones" "Trigger" "Kyoto Animation" "CloverWorks"
"A-1 Pictures" "Sunrise" "J.C.Staff" "Shaft" "Studio Pierrot" "P.A. Works"
"David Production" "OLM Inc."
)
RADARR_KIDS_STUDIOS=(
"DreamWorks Animation" "Pixar" "Disney Television Animation"
"Walt Disney Animation Studios" "DisneyToon Studios" "Walt Disney Productions"
"Walt Disney Feature Animation" "Warner Bros. Animation" "Illumination"
"Sony Pictures Animation" "Blue Sky Studios" "Mattel" "Rankin/Bass Productions"
"Mainframe Entertainment" "DreamWorks Animation Television" "ZAG Entertainment"
"Paramount Animation" "Pacific Data Images" "Cinesite Animation"
"Cartoon Network Studios" "Nickelodeon Animation Studio" "Nelvana"
"Aardman Animations"
)
# Radarr's TMDb-vote/imdbId combo reliably flags bad metadata matches (proven live —
# caught 5 genuine junk entries: 2 fake X-Men "Wolverine" titles, Silent Hill: No Escape,
# Silent Hill 4: The Room, The Purge: Behind the Series — all hasFile=false, no imdbId,
# under this vote count). Do NOT reuse this threshold/approach for Sonarr — TheTVDB's
# ratings data is far sparser than TMDb's even for completely legitimate shows (confirmed
# live: "The Pussycat Dolls Present", a real 2007 MTV show, has ratings.votes=0), so the
# same heuristic there would flag real content for removal.
RADARR_JUNK_MIN_VOTES=15
SONARR_ANIME_NETWORKS=(
"Tokyo MX" "AT-X" "TV Tokyo" "Fuji TV" "Nippon TV" "TBS (JP)" "MBS" "TV Asahi"
"WOWOW" "NHK" "Animax" "tvk" "Teletama" "Kansai TV" "ABC (JP)"
)
SONARR_KIDS_NETWORKS=(
"Cartoon Network" "Cartoon Network (CA)" "Cartoon Network (Canada)" "Nickelodeon"
"Nick Jr." "Nicktoons" "Disney Channel" "Disney Jr." "Disney XD" "Toon Disney"
"PBS" "PBS KIDS" "CBeebies" "CBBC" "YTV" "TVOkids" "Discovery Kids"
"Discovery Family" "Boomerang" "ABC Kids" "Teletoon" "Télétoon" "Milkshake!"
"Tiny Pop" "Sprout" "Universal Kids" "BabyTV"
)
# ━━━ Arr Failed/Stalled Recovery ━━━
# Auto blocklist + re-search failed imports and stalled downloads.
# Runs every 4 hours via INTERMEDIATE_MAINTENANCE_SCRIPTS — no cron entry of its own.
#
# Targets five problem types:
# importFailed — downloaded but arr couldn't import
# importPending — downloaded, stuck waiting to import (won't self-resolve)
# importBlocked — arr matched the release by grab-history ID, not title, and
# refuses to auto-import — gets a smart-import check first,
# see ARR_SMART_IMPORT_ENABLED below
# error status — serious failure not covered above
# stalled — download stuck with no connections or progress
#
# Items newer than ARR_IMPORT_RECOVERY_AGE are skipped — gives arr time to retry first.
# Per-host recovery toggles (HOST1_SONARR_RECOVERY etc.) live in host*.conf.
ARR_IMPORT_RECOVERY_AGE=6 # hours — skip items newer than this
# matches cron interval — items eligible after one missed cycle
ARR_RECOVERY_MAX_ATTEMPTS=3 # consecutive failures before an item is flagged chronic
# and auto re-search stops (still blocklisted/cleaned up)
ARR_SMART_IMPORT_ENABLED=true # importBlocked items: try importing a clean same-language
# upgrade/gap-fill directly before falling back to blocklist
# + re-search. Sonarr/Radarr only — see try_smart_import()
ARR_SMART_IMPORT_PREFERRED_LANGUAGE="English" # only treated as a match/upgrade if the
# candidate is this language; an existing file in a
# different language is always considered upgradeable
# ━━━ Arr Full Library Rescan ━━━
# arr_full_rescan.sh — weekly disk↔DB reconciliation for Lidarr/Sonarr/Radarr. See script
# header for why this exists (2026-07-16 stale-stats incident).
ARR_FULL_RESCAN_TIMEOUT=3600 # seconds per arr — a full-library walk, not a single release
# ==============================================================================================
# ── TRANSCODES ────────────────────────────────────────────────────────────────────────────────
# ==============================================================================================
# Session-based storage allocator using filesystem symlink indirection.
# ffmpeg resolves the symlink ONCE at session start — existing sessions never affected.
# Symlink flips between ramdisk and SSD are transparent to active streams.
#
# ⚠️ Docker mount — must use shared propagation so symlink flips work inside container:
# --mount type=bind,source=/mnt/ram-transcode,target=/ext-ram-transcode,bind-propagation=shared
# Standard path mappings use rprivate — symlink changes are NOT visible inside container.
# ━━━ Transcode Manager ━━━
# tmpfs mount point — created at array start by ramdisk_setup.sh.
# Must exist before Emby starts so the symlink resolves correctly.
RAMDISK_PATH="/mnt/ramdisk_transcodes"
# Ramdisk size and flip thresholds — defined per host in host*.conf.
# All three are coupled — if size changes, thresholds must change with it.
# HOST1_RAMDISK_SIZE / HOST2_RAMDISK_SIZE
# HOST1_RAMDISK_WARN_GB / HOST2_RAMDISK_WARN_GB ← flip to SSD at this usage
# HOST1_RAMDISK_LOW_GB / HOST2_RAMDISK_LOW_GB ← flip back to ramdisk at this usage
# Symlink that Emby points at — this path NEVER changes regardless of ramdisk/SSD state.
# Emby resolves the symlink once per session at start — symlink flips are transparent.
# Must match the container path configured in Emby's Extra Parameters.
TRANSCODE_LINK="/mnt/ram-transcode"
# SSD fallback location — defined per host in host*.conf (cache path differs per server):
# HOST1_TRANSCODE_SSD / HOST2_TRANSCODE_SSD
# Minimum free GB on SSD before allowing flip from ramdisk to SSD.
RAMDISK_SSD_MIN_GB=20
TRANSCODE_MAX_AGE=20 # minutes — HLS segment age before cleanup eligibility
TRANSCODE_ORPHAN_AGE=30 # minutes — files with no matching active session
TRANSCODE_FLIP_WARN=3 # notify if symlink flips this many times in one hour
TRANSCODE_OWNER="nobody:users"
TRANSCODE_CHMOD="755"
# Operating mode — controls symlink direction behaviour.
# smart — auto-flips between ramdisk and SSD based on thresholds (default)
# ramdisk — always uses ramdisk, warns if RAMDISK_WARN_GB exceeded but holds
# ssd — always uses SSD, never flips to ramdisk
TRANSCODE_MANAGER_MODE="smart"
# Daily statistics log — read by weekly_health_digest.sh for transcode summary.
TRANSCODE_STATE_FILE="$STATE_DIR/transcode_state.db"
TRANSCODE_DAILY_LOG="${DB_DIR}/transcode_daily.db"
TRANSCODE_LOG_RETENTION=90 # days before old entries purged
TRANSCODE_CHECK_EMBY=true
# ━━━ Transcode Server Array ━━━
# All media servers sharing the ramdisk transcode space.
# Format: "ContainerName|URL|APIKey|Type" — Type: emby | jellyfin | plex
# Entries with placeholder API keys are skipped automatically.
# ⚠️ Tdarr does NOT belong here — keep Tdarr on SSD, not ramdisk.
# Defined per host in host*.conf — Emby container names and keys differ per server:
# HOST1_TRANSCODE_SERVERS / HOST2_TRANSCODE_SERVERS
# ==============================================================================================
# ── MONITORS ──────────────────────────────────────────────────────────────────────────────────
# ==============================================================================================
# ━━━ Certificate Monitor ━━━
# Checks SSL certificate expiry via direct openssl connection — no NPM dependency.
# Checks the actual certificate served by each domain, not what NPM thinks it has.
# Domains defined per host in host*.conf — each server monitors its own domains:
# HOST1_CERT_MONITOR_DOMAINS / HOST2_CERT_MONITOR_DOMAINS
CERT_WARN_DAYS=30 # warn when cert expires within this many days
CERT_CRIT_DAYS=7 # critical alert within this many days
CERT_TIMEOUT=10 # seconds per domain before giving up
# ━━━ Backup Verify ━━━
# Verifies rsync mirror health by comparing random file checksums between servers.
# Catches silent corruption or incomplete syncs that rsync itself wouldn't detect.
# Defined per host in host*.conf — leave empty to use HOST*_DAILY_SYNC_SHARES automatically:
# HOST1_BACKUP_VERIFY_SHARES / HOST2_BACKUP_VERIFY_SHARES
BACKUP_VERIFY_SAMPLE=10 # random files to check per share
BACKUP_VERIFY_MIN_SIZE=1M # minimum file size to include in sample
# ━━━ SMART Health ━━━
# Monitors drive SMART attributes — discovers all drives via /dev/sd* and /dev/nvme*.
# Thresholds read from /boot/config/plugins/dynamix/dynamix.cfg at runtime
# (hot/max/hotssd/maxssd) — these vars are fallback only if dynamix.cfg not found.
SMART_TEMP_WARN=45 # fallback — Celsius warn threshold
SMART_TEMP_CRIT=55 # fallback — Celsius critical threshold
# Drives to ignore defined per host in host*.conf — hardware is server-specific:
# HOST1_SMART_IGNORE_DRIVES / HOST2_SMART_IGNORE_DRIVES
# ━━━ ZFS Memory Snapshot ━━━
# Weekly ZFS pool health and memory diagnostic report — informational only.
ZFS_REPORT_LOG="${LOG_ARCHIVE_DIR}/zfs-weekly-health.log"
ZFS_REPORT_ARC_WARN_PCT=90 # warn if ARC using more than this % of its max
ZFS_REPORT_ARC_FREE_WARN_GB=10 # warn if ARC headroom (max - current) drops below this GB
ZFS_REPORT_AVAIL_WARN_GB=20 # warn if less than this GB available on ZFS pool
ZFS_REPORT_DOCKER_TOP=10 # how many top Docker containers to show by memory
# Pool ignore list defined per host in host*.conf — pool names are server-specific:
# HOST1_ZFS_REPORT_IGNORE_POOLS / HOST2_ZFS_REPORT_IGNORE_POOLS
# ━━━ Bandwidth Monitor ━━━
# Called automatically by rsync.sh after each sync — one bounded write per run.
# Tracks transfer size, duration and profile per sync for weekly summary reporting.
BANDWIDTH_LOG="${DB_DIR}/bandwidth_history.db"
BANDWIDTH_LOG_RETENTION=90 # days before old entries purged
BANDWIDTH_WARN_GB=50 # flag syncs larger than this in weekly report
# Stats files — written by cleanup and recovery scripts, read by coffee report.
# All in DATA_DIR — array always running when these are written.
ARR_CLEANUP_STATS="${DB_DIR}/arr_cleanup_stats.db" # lidarr/sonarr/radarr orphan stats
ARR_RECOVERY_STATS="${DB_DIR}/arr_recovery_stats.db" # blocklist + re-search stats
ARR_RECOVERY_FAILURE_COUNTS="${DB_DIR}/arr_recovery_failure_counts.db" # per-item chronic-failure tracking
# ━━━ Health Digest ━━━
# Aggregated system health summary — reads existing state files, no new writes.
# Three profiles control when the digest is sent:
# always — sends every run regardless of findings
# smart — sends only when DIGEST_SMART_ON_* conditions are found
# weekly — sends once per week on DIGEST_DAY only
DIGEST_PROFILE="weekly" # always | smart | weekly
DIGEST_DAY="Sunday"
DIGEST_SMART_ON_WATCHDOG=true # send if any watchdog strikes are active
DIGEST_SMART_ON_FALLBACK=true # send if fallback state is not NORMAL
DIGEST_SMART_ON_CERT_WARN=true # send if any cert is under CERT_WARN_DAYS
DIGEST_SMART_ON_BANDWIDTH=true # send if any transfer exceeded BANDWIDTH_WARN_GB
# ━━━ Emby Session Report ━━━
# Weekly Emby usage statistics via API — no persistent writes, queries fresh each run.
# URL and API key pulled from HOST*_EMBY_URL and HOST*_EMBY_API_KEY in host*.conf.
EMBY_REPORT_DAYS=7 # days to include in the report period
EMBY_REPORT_TOP_N=10 # number of top content items to show
# ==============================================================================================
# ── RESOURCE MANAGER ──────────────────────────────────────────────────────────────────────────
# ==============================================================================================
# Pressure reduction layer — keeps the system comfortable before things break.
# Called by watchdog_orchestrator.sh. Single-pass, not a continuous loop.
#
# ── PRESSURE LEVELS ───────────────────────────────────────────────────────────────────────────
# Level 1 (soft) — throttle SABnzbd + qBit download speeds
# Level 2 (medium) — further throttle + docker pause background containers
# Level 3 (hard) — docker stop optional containers, signal docker_watchdog to defer
#
# ── PER-HOST CONTAINER LISTS ──────────────────────────────────────────────────────────────────
# HOST*_RW_PAUSE_CONTAINERS — docker pause at medium pressure (in host*.conf)
# HOST*_RW_STOP_CONTAINERS — docker stop at hard pressure (in host*.conf)
RW_ENABLED=true
RW_STATE_FILE="$STATE_DIR/resource_watchdog_state.db"
# ━━━ Pressure Thresholds ━━━
# Graduated RAM response — resource_watchdog acts before system_watchdog reboots.
# RW_RAM_SOFT_GB > RW_RAM_MEDIUM_GB > RW_RAM_HARD_GB > SYS_WATCHDOG_MEM_GB always
RW_RAM_SOFT_GB=12 # throttle start — reduce background load
RW_RAM_MEDIUM_GB=8 # pause background containers
RW_RAM_HARD_GB=6 # stop optional containers (was SYS_WATCHDOG_MEM_SHUTDOWN_GB)
RW_RAM_RECOVER_GB=20 # RAM must reach this before restoring hard-stopped containers
# Load average thresholds — multiplier × core count
RW_LOAD_SOFT_MULTIPLIER=2.0 # soft pressure: 2× cores sustained
RW_LOAD_MEDIUM_MULTIPLIER=3.0 # medium pressure: 3× cores sustained
# Consecutive runs at lower pressure before de-escalating
RW_RECOVER_CYCLES=3
# ━━━ SABnzbd Throttle ━━━
# Speed values: "50M" = 50 MB/s, "0" = unlimited
RW_SABNZBD_ENABLED=true
RW_SABNZBD_SPEED_SOFT="50M"
RW_SABNZBD_SPEED_MEDIUM="10M"
# ━━━ qBittorrent Throttle ━━━
# KB/s — 0 = unlimited
RW_QBIT_ENABLED=true
RW_QBIT_DL_SOFT=51200 # 50 MB/s
RW_QBIT_DL_MEDIUM=10240 # 10 MB/s
# ━━━ Critical Containers ━━━
# Never paused or stopped regardless of pressure level.
# Keep DNS, auth, media serving, and live TV always running.
RW_CRITICAL_CONTAINERS=(
"NginxProxyManager" # reverse proxy — internet access
"Authelia" # auth — nothing accessible without it
"Authelia-Secondary"
"Mariadb-Authelia" # Authelia dependency
"Mariadb-Authelia-Secondary"
"Redis-Authelia" # Authelia dependency
"Redis-Authelia-Secondary"
"AdGuard-Home" # DNS — all LAN resolution
"Emby" # media server — Live TV buffering
"Dispatcharr" # Live TV scheduler — loses state if stopped
"Dispatcharr-Basic"
"Dispatcharr-Iptv-Users"
)
# ==============================================================================================
# ── SYSTEM WATCHDOG ───────────────────────────────────────────────────────────────────────────
# ==============================================================================================
# Single-pass system health check — last line of defense before a crash.
# Called by watchdog_orchestrator.sh every 15 minutes. NOT started by array_started.sh.
# Re-sources config at each orchestrator run — config changes take effect immediately.
#
# ── THREE-TIER RESPONSE SYSTEM ────────────────────────────────────────────────────────────────
# CRITICAL — bypass ALL strikes, reboot immediately
# Docker daemon down, rootfs 100%, kernel oops, FD exhaustion, /boot read-only
#
# URGENT — bypass strikes only when OOM confirms active crisis
# RAM < MEM_GB AND OOM kills >= OOM_LIMIT in this cycle → reboot NOW
# Without OOM confirmation → normal strike system
#
# STANDARD — strike system (N consecutive failures → reboot)
# RAM tiers, high load, CPU temp, zombies, /var/log, /tmp, containers
#
# ── RAM ───────────────────────────────────────────────────────────────────────────────────────
# SYS_WATCHDOG_MEM_GB — strike system → reboot (or OOM bypass)
# Warn/shutdown/recover RAM tiers are handled by resource_watchdog.sh
# ━━━ State Files ━━━
SYS_WATCHDOG_STATE_FILE="$STATE_DIR/system_watchdog_state.db"
DOCKER_WATCHDOG_FAILED_FILE="$STATE_DIR/docker_watchdog_failed.db"
DOCKER_WATCHDOG_INTENTIONAL_FILE="$STATE_DIR/docker_intentional_stops.db"
SYS_WATCHDOG_REBOOT_LOG="$STATE_DIR/system_watchdog_reboots.db"
SYS_WATCHDOG_OOM_FILE="$STATE_DIR/system_watchdog_oom.db"
# ━━━ Strike and Reboot Loop Settings ━━━
# Strike system: a check must fail this many consecutive cycles before action is taken.
# Single spikes (one bad reading) are ignored — sustained problems trigger reboot.
SYS_WATCHDOG_STRIKE_LIMIT=2 # consecutive failures before reboot trigger
# Reboot loop protection — if system keeps rebooting something is seriously wrong.
# After REBOOT_LIMIT reboots in REBOOT_WINDOW_HRS → shutdown instead of reboot.
SYS_WATCHDOG_REBOOT_LIMIT=3
SYS_WATCHDOG_REBOOT_WINDOW_HRS=12
# ━━━ RAM Reboot Threshold ━━━
# Reboot trigger only — warn/shutdown/recover handled by resource_watchdog.sh
# RW_RAM_HARD_GB > SYS_WATCHDOG_MEM_GB always (RM acts before watchdog reboots)
SYS_WATCHDOG_MEM_GB=4 # strike system → reboot
# ━━━ OOM Bypass Settings ━━━
# OOM bypass: if RAM is critically low AND kernel OOM kills exceed this threshold
# in a single cycle → bypass strike system and reboot immediately.
# Rate-based: kills per 5-minute cycle, not absolute count.
# Rationale: 1-2 kills = docker_watchdog handles it ✅
# 3+ kills while RAM critical = system dying faster than watchdogs can heal ✅
SYS_WATCHDOG_OOM_LIMIT=3 # OOM kills in one cycle to trigger bypass
# ━━━ Thresholds ━━━
# Set at "about to become unstable" — not "things are a bit high".
# rootfs (/) usage — two levels: strike at 95%, critical bypass at 99%
SYS_WATCHDOG_ROOTFS_PCT=95 # standard strike threshold
SYS_WATCHDOG_ROOTFS_CRITICAL_PCT=99 # bypass strikes — truly full, writes failing
# /var/log usage — log spam indicates something broken
SYS_WATCHDOG_LOG_PCT=95
# /tmp usage — tmpfs fills from downloads, lock files fail at 100%
SYS_WATCHDOG_TMP_PCT=90 # warn + attempt clear
SYS_WATCHDOG_TMP_CRITICAL_PCT=98 # bypass strikes if clear failed
# ZFS ARC pinned percentage — ARC not releasing after reclaim = memory stuck
SYS_WATCHDOG_ARC_PINNED_PCT=98
SYS_WATCHDOG_ARC_RELEASE_PCT=95
# Load average multiplier — threshold = MULTIPLIER × CPU core count
# MULTIPLIER=3 on 16-core = load of 48 before triggering
SYS_WATCHDOG_LOAD_MULTIPLIER=3
# Zombie process count — large numbers = serious process management failure
SYS_WATCHDOG_ZOMBIE_LIMIT=50
# CPU temperature — sustained high temp causes throttling or kernel panic
SYS_WATCHDOG_CPU_TEMP_MAX=95
# File descriptor exhaustion — system-wide FD limit near exhaustion
# New connections fail silently, Docker can't spawn processes, SSH fails
SYS_WATCHDOG_FD_CRITICAL_PCT=95 # bypass strikes — critical tier
# Runaway process — single non-container process consuming excessive CPU
# Multiple strikes before action — single spikes are normal
SYS_WATCHDOG_RUNAWAY_CPU_PCT=90 # % single process must sustain
SYS_WATCHDOG_RUNAWAY_STRIKES=3 # consecutive cycles before warning
# Array disk errors — accumulating mdstat errors = disk failing NOW
SYS_WATCHDOG_MDSTAT_ERROR_LIMIT=5 # new errors in one cycle before acting
# ━━━ Check Toggles ━━━
# Per-host — moved to host*.conf
# Different servers may have different hardware, NICs, and check requirements
# See HOST*_SYS_WATCHDOG_CHECK_* in host*.conf
# ━━━ Abort Toggles ━━━
# Conditions that prevent reboot even when a threshold is hit.
# CRITICAL tier bypasses these — a truly critical condition reboots regardless.
# true = abort standard reboot if this condition is active
# false = reboot anyway
SYS_WATCHDOG_ABORT_ON_ZFS_UNHEALTHY=true # rebooting with bad pool risks data loss
SYS_WATCHDOG_ABORT_ON_PARITY=false # aborting parity better than crashing mid-check
SYS_WATCHDOG_ABORT_ON_MOVER=false # aborting move better than crashing mid-move
# ==============================================================================================
# ── AI / RAG ──────────────────────────────────────────────────────────────────────────────────
# ━━━ Conf Backups ━━━
# Every write through the plugin copies the conf aside first, to CONF_BACKUP_DIR with an
# ISO-8601 stamp. The confs are gitignored, so that directory is the whole recovery path — there
# is no history to revert to. A backup that cannot be taken cancels the write.
# Oldest are pruned past this count, per conf file.
CONF_BACKUP_RETAIN=30
# ==============================================================================================
# Varaverk works exactly as well with AI off as with it on. Nothing below is required for any
# script to function — every feature that can lean on AI has a complete non-AI path, and the
# AI path is an enhancement layered on top. If Ollama is unreachable, callers proceed without it.
#
# AI_ENABLED is necessary but not sufficient. Every feature stays individually off until it has
# earned it — narration for months before anything is allowed near a decision.
# ━━━ AI Master Switch ━━━
# Fail-closed: anything other than the literal "true" means off.
AI_ENABLED=false
AI_CONNECT_TIMEOUT=5 # seconds — probe when resolving which node has Ollama
AI_REQUEST_TIMEOUT=240 # seconds — must clear a cold model load
AI_RESOLVE_CACHE_TTL=300 # seconds — don't re-probe the mesh every invocation
AI_MAX_RETRIES=1 # AI is enhancement; do not retry hard
# ━━━ AI Retrieval Index ━━━
# The RAG index over this repo's own headers and documentation. Regenerable in minutes and
# gitignored — it is derived data, never a source of truth.
#
# Only git-tracked files are ever indexed. Configurations/, State_Files/ and data/ are
# gitignored, which is what makes it structurally impossible for a credential to reach the
# index: the files holding them were never in the repo. Do not "improve" this to a filesystem
# walk — an embedded secret cannot be rotated out of a vector.
AI_INDEX_DB="${AI_DATA_DIR}/ai_index.db"
AI_INDEX_BATCH=32 # chunks per embed request
# A pull is the only thing that changes tracked files on a server, so it is the only moment the
# index can go stale — and staleness is invisible in the answers, which keep citing the old
# text with full confidence. Incremental: unchanged files are skipped, a no-op run is ~66ms.
# Also gated on AI_ENABLED, and ai_index.sh refuses on its own unless that is true.
AI_INDEX_ON_PULL=true # re-index after a git pull that changed tracked files
AI_SEARCH_K=8 # chunks retrieved per query
AI_SEARCH_PER_FILE=3 # cap per file so one document cannot fill the context
# ━━━ AI Memory ━━━
# A small operator-maintained file the assistant is given at the start of every conversation:
# who you are, how this install is set up, decisions already made, things it should stop asking.
#
# Injected into the prompt, never indexed. It lives under DATA_DIR, which is gitignored — that
# is deliberate and load-bearing. Indexing it would embed a file that changes constantly, and
# vector similarity is the wrong way to retrieve "things I was told to remember"; it also keeps
# personal notes out of a repository that gets pushed.
#
# The character cap is a context budget, not a style guide. At 16384 the retrieved passages,
# the model's reasoning and the conversation history are already competing; memory takes its
# share off the top of every single turn, so keep it short and factual.
AI_MEMORY_FILE="${AI_DATA_DIR}/ai_memory.md"
AI_MEMORY_MAX_CHARS=4000 # ~1000 tokens — truncated with a notice if exceeded
# ━━━ AI Stored Conversations ━━━
# How many past conversations the AI tab and the Monitor tab's AI row keep. One JSON file per
# conversation under DATA_DIR/ai_chats, saved automatically when a turn completes; the oldest
# drop off once the count is exceeded.
#
# A cap rather than a retention age. These are read by picking one out of a short list, and a
# list you have to scroll is a list you stop using — the useful window is the handful of things
# you were recently working on, which is a count, not a date.
#
# Not indexed and never retrieved into a prompt on their own. A stored chat only re-enters the
# model's context when the operator explicitly reopens it, and it is re-validated per message on
# the way in, exactly as live history is.
AI_CHAT_HISTORY_MAX=10 # conversations kept — clamped to 1-50
# ━━━ AI Token Accounting ━━━
# One row per completed turn, appended by whichever path ran it — the WebGUI worker and the
# ai_query.sh CLI both write here, so the totals are not silently the tab's alone. Format is
# pipe-delimited to match the other data/*.db files:
#
# date|time|host|profile|source|prompt_tokens|completion_tokens|tok_s
#
# The host column records where the turn ran, not where it is read. Each host keeps its own
# data/ and nothing syncs it, so a host only ever sees its own rows — the column is there so
# the file is already shaped right if the partner payload fetch is ever extended to carry it.
# A column added later cannot be backfilled.
#
# Retention is by row count rather than age: pruning is considered only when the file passes a
# size threshold, so an ordinary turn costs one stat() and an append.
AI_TOKEN_DB="${AI_DATA_DIR}/ai_token_history.db"
AI_TOKEN_RETAIN_ROWS=20000 # oldest rows dropped past this — years of ordinary use
# AI/ai_token_sync.sh pulls each partner's ledger into the tmpfs cache the tab reads, so the
# fleet total is a fleet total. Same trick conf_sync.sh uses for partner confs, minus the push:
# nothing here is needed by anyone else, so the reader fetches its own data and controls its
# own freshness. An unreachable partner is a quiet skip, not a warning — a partner is expected
# to be down for long stretches, and a four-hourly warn trains you to ignore the script.
AI_TOKEN_SYNC_ENABLED=true
# ━━━ AI Feature Toggles ━━━
# Tier 1 is narration — it cannot change a decision. Tier 2 adds context to a decision a script
# already made. Tier 3 assists a human. Enable in that order, and give each one weeks.
AI_ASSIST_REPORTS=false # tier 1 — digest / coffee report narration
AI_ASSIST_WATCHDOG=false # tier 2 — context on a flagged condition
AI_ASSIST_DISCOVERY=false # tier 2 — discovery / classification judgement calls
AI_ASSIST_CLEANUP=false # tier 2 — orphan and stuck-import triage
AI_ASSIST_ONBOARD=false # tier 3 — onboarding / settings assistance
# ━━━ AI Repair ━━━
# Reads the log of a job that has finished, turns known error shapes into findings, and offers
# or applies a repair. Two switches, because detecting and repairing are separate things to
# trust.
#
# AI_REPAIR_ENABLED on its own reads, files findings and offers fixes, and writes nothing. Run
# it there first — long enough to read what it finds and disagree with some of it.
#
# AI_REPAIR_AUTOFIX_ENABLED is what allows a value to be written unattended, and only ever a
# value a probe has answered on. Never a toggle: whether something should be switched on is a
# decision about intent, and a probe cannot prove intent the way it can prove a port answers.
AI_REPAIR_ENABLED=false
AI_REPAIR_AUTOFIX_ENABLED=false
# ━━━ AI Repair Findings ━━━
# Misconfigurations found in this installation, as opposed to defects in Varaverk — those go to
# ai_bugs. A finding is open until the configuration is right, and closes itself when the probe
# that proved the fault starts passing. Closed ones are kept for a while, because "this happened
# before and here is what fixed it" is worth more than the disk. Open findings are never pruned:
# an unresolved problem does not stop mattering because it is old.
AI_FINDING_RETAIN_DAYS=90
# Seconds a single probe may take. Nothing is written to conf that has not answered a probe, so
# this is the budget for proving a candidate — kept short because a sweep may try several, and
# an address worth switching to answers quickly or is not worth switching to.
AI_PROBE_TIMEOUT=4
# Tell the operator when a finding needs them. Only findings the sweep cannot resolve itself are
# announced, one notification for all of them rather than one each, and each is announced once —
# it stays quiet through every later pass until the fault changes or gets worse. Delivery is
# whatever notify() is set up to use: NOTIFY_UNRAID and the host's Discord webhook.
#
# On by default, unlike the two switches above. Those gate reading and writing, which are things
# to be trusted first. This gates telling someone, which is the reason for having looked.
AI_REPAIR_NOTIFY_ENABLED=true
# ━━━ AI Repair: what it reads ━━━
# Beyond Varaverk's own logs and the arrs' health endpoints.
#
# The system log catches what Varaverk cannot see about itself — a disk throwing I/O errors, a
# filesystem going read-only, a PCIe link retraining every two minutes. Container restarts and
# OOM kills are deliberately left to docker_watchdog, which already notifies on them.
#
# Container logs catch the opposite blind spot: docker_watchdog watches whether a container is
# up and answering, which a container that has been unable to write to its database all day
# passes perfectly. Only environment faults are matched — disk full, read-only filesystem,
# corrupt database, fd limits, expired certificates — because those strings come from libc, the
# kernel and SQLite and mean the same thing in all fifty containers. Anything app-specific
# belongs in that app's own health endpoint.
#
# Both are bounded by time (since the last pass) and by a line cap, so a flood costs one pass.
# Tools/ai_log_check.sh replays this host's real logs against the patterns — run it after
# changing any of them.
# ━━━ AI Web Search ━━━
# General Chat only. Every other profile either reads this installation or changes it; chat holds
# no capability at all, and that is exactly why searching is the one thing it may do — a read that
# leaves the house is safe on the profile that cannot act on what it finds. The Varaverk assistant
# deliberately does not get it: its contract is that answers come from this installation's own
# documents, and a web result there is an answer that looks sourced and is not.
#
# Off by default, and not because it is dangerous. Searching sends the operator's question to
# something outside this house, and that is their decision to make rather than a default to
# inherit. Nothing turns it on.
#
# Asked for per turn as well — there is a checkbox on the AI tab, and a question is only searched
# when it is ticked. A question about this machine hands off to the assistant before the search
# would run, so it never reaches the internet even with the box ticked.
#
# Provider: searxng | brave | tavily
# searxng self-hosted, no key, no third party — set HOST*_SEARXNG_URL, and enable format: [json]
# in its own settings.yml, which the default image ships with off
# brave HOST*_WEB_SEARCH_API_KEY, free tier available
# tavily HOST*_WEB_SEARCH_API_KEY
AI_WEB_SEARCH_ENABLED=false
AI_WEB_SEARCH_PROVIDER=searxng
AI_WEB_SEARCH_RESULTS=4
AI_WEB_SEARCH_TIMEOUT=6
AI_REPAIR_SYSLOG_ENABLED=true
AI_REPAIR_SYSLOG_MAX_LINES=4000
AI_REPAIR_CONTAINER_LOGS_ENABLED=true
AI_REPAIR_CONTAINER_LOG_LINES=400
# ━━━ AI Conf Write Access ━━━
# Separate switch from AI_ENABLED, off by default, and an explicit key whitelist. Never paths,
# never credentials, never a container name. An empty whitelist means no writes regardless of
# the toggle.
AI_CONF_WRITE_ENABLED=false
AI_CONF_WRITE_KEYS=()
# ==============================================================================================
# ──────────────────────── End Of User Variables ───────────────────────────────────────────────
# ==============================================================================================
Reference in New Issue View Git Blame Copy Permalink