LibrePortal/scripts/instance/instance_create.sh
librelad e25c69e2a1 Make multi-instance work without a domain
An instance's isolation never needed a domain — its own slug, dir, secrets,
IP and randomly-allocated host port already make two copies independent. But
the routing layer assumed one, so a LAN-only box got a broken instance rather
than a port-served one. Four fixes:

- instanceCreate now rewrites the parent-service column of the cloned config's
  PORT_ rows to match the service names it stamps into the compose. That value
  is stored as network_resources.parent_service and joined against the
  compose-derived service names, so an instance left carrying the TYPE's
  service name matched nothing: it rendered in the WebUI with no port, no URL
  and no login row despite being up and reachable.

- `instance create --local` (plus a LAN-only toggle in the modal) forces every
  port to access=private, traefik=false, for a second copy that should stay
  off the domain even when one is configured.

- initializeAppVariables forces the traefik column false when no CFG_DOMAIN_n
  is set. Previously a traefik=true port with an empty domain stamped
  Host(`app.`) — a trailing-dot host matching nothing — and dragged APP_URL to
  https://app. with it, breaking every app that builds its links from APP_URL.
  host_setup is blanked for the same reason. The published host port is
  untouched; access type, not the traefik flag, gates allocation.

- APP_URL's direct host-port branch now prefers a new $local_ip_v4 (the source
  IP for the default route) over $public_ip_v4, which is the WAN address from
  an external resolver. LibrePortal never forwards ports, so the WAN address
  was unreachable for exactly the LAN/VPN clients that branch serves.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-19 02:53:43 +01:00

368 lines
17 KiB
Bash

#!/bin/bash
# Multi-instance support.
#
# Some apps are worth running more than once on a single box — e.g. two
# WordPress/Bookstack sites, or a "family" and a "work" Nextcloud kept in
# separate trust/blast-radius/backup domains. Internal multi-tenancy answers
# logical separation; this answers instance-level isolation (independent data,
# version cadence, admin, restore granularity).
#
# The model: an instance is just another app. It gets its own slug
# (<type>_<id>), its own CFG_<SLUG>_* namespace, its own deployed dir, DB row,
# IP/port allocation, subdomain and backups — so the entire downstream pipeline
# (scan, install, services, routing, updater, backups) treats it like any other
# app with ZERO changes. Everything instance-specific happens here, on a cloned
# copy of the type's template, leaving the shipped template and the core engine
# untouched.
#
# Only apps that opt in via CFG_<TYPE>_MULTI_INSTANCE=true can be instanced;
# structurally-singleton apps (Traefik, DNS, VPN, the *arr stack, LibrePortal
# itself) never get the flag.
# Read a CFG_<TYPE>_<KEY> value straight from a type's template config, without
# relying on it already being sourced.
instanceTypeCfg() {
local type="$1" key="$2"
local cfg="${install_containers_dir%/}/$type/$type.config"
[[ -f "$cfg" ]] || return 1
local line
line=$(grep -E "^CFG_${type^^}_${key}=" "$cfg" | head -n1)
[[ -z "$line" ]] && return 1
line="${line#*=}"
line="${line//$'\r'/}"
line="${line#\"}"
line="${line%\"}"
printf '%s' "$line"
}
# Turn a user-supplied instance name into the <id> half of the slug. App configs
# are SOURCED, so the slug (uppercased) must be a valid shell identifier — that
# means [a-z0-9] only (underscores are fine, hyphens are not). Hostname-safety is
# handled separately by the subdomain, which may contain hyphens.
instanceIdPart() {
local raw="${1,,}"
raw="${raw//[^a-z0-9]/}"
printf '%s' "$raw"
}
# Upsert a single CFG line in a config file (append if absent, else update).
_instanceSetCfg() {
local key="$1" val="$2" file="$3"
if grep -qE "^${key}=" "$file"; then
updateConfigOption "$key" "$val" "$file" >/dev/null
else
echo "${key}=\"${val}\"" >> "$file"
fi
}
# Rewrite field 10 (the subdomain column) of the instance's primary webui port so
# the instance routes to its own host instead of inheriting the type's. Empty
# subdomain would otherwise resolve to <slug>.<domain> — and the slug carries an
# underscore, which isn't valid in a hostname.
_instanceSetSubdomain() {
local slug_u="$1" subdomain="$2" file="$3"
local key="CFG_${slug_u}_PORT_1"
local line
line=$(grep -E "^${key}=" "$file" | head -n1)
[[ -z "$line" ]] && return 0
local val="${line#*=}"
val="${val//$'\r'/}"
val="${val#\"}"
val="${val%\"}"
local IFS='|'
local -a f=($val)
while [[ ${#f[@]} -lt 11 ]]; do f+=(""); done
f[10]="$subdomain"
local newval="${f[*]}"
updateConfigOption "$key" "$newval" "$file" >/dev/null
}
# Rewrite the parent-service column (field 1) of every PORT_<n> row in the cloned
# config so it names the service _instanceRewriteCompose actually stamped into the
# compose.
#
# portAllocate stores this value as the port row's `parent_service`, and the WebUI
# joins ports to an app's Docker services on it (webui_services.sh,
# db_list_installed_app.sh) — matching against service names read back from the
# compose's SERVICE_TAG_<n>. The config re-namespace only rewrites KEYS, so an
# instance kept the TYPE's service name in the value while its compose moved on:
# the join found nothing and the instance rendered in the WebUI with no port, no
# URL and no login row, despite being up and reachable.
#
# MUST mirror rule 1 of _instanceRewriteCompose — same two token forms, so the
# config and the compose can't drift apart. Scoped to PORT_ lines so a bare app
# name elsewhere in the config (titles, descriptions, the subdomain column) is
# left alone.
_instanceRewriteConfigPorts() {
local type="$1" slug="$2" file="$3"
[[ -f "$file" ]] || return 0
sed -i -E "/^CFG_[A-Z0-9_]+_PORT_[0-9]+=/{
s/\b${type}-service\b/${slug}-service/g
s/\b${type}_db\b/${slug}_db/g
}" "$file"
}
# Force every one of the instance's ports to LAN-only: access `private`, Traefik
# off. Used by `instance create --local`, for a second copy that should be reached
# at http://<ip>:<port> and never given a router or a certificate.
#
# `disabled` ports are left as-is — the user opted them out of host exposure
# entirely, and flipping them to private would start publishing them.
_instanceSetLocalOnly() {
local slug_u="$1" file="$2"
local key line val newval
local -a f
local n
for n in {1..20}; do
key="CFG_${slug_u}_PORT_${n}"
line=$(grep -E "^${key}=" "$file" | head -n1)
[[ -z "$line" ]] && continue
val="${line#*=}"
val="${val//$'\r'/}"
val="${val#\"}"
val="${val%\"}"
local IFS='|'
f=($val)
unset IFS
while [[ ${#f[@]} -lt 11 ]]; do f+=(""); done
[[ "${f[3]}" == "disabled" ]] && continue
f[3]="private" # access
f[6]="false" # traefik
local IFS='|'
newval="${f[*]}"
unset IFS
updateConfigOption "$key" "$newval" "$file" >/dev/null
done
}
# Rewrite identity-bearing tokens in the cloned compose so the instance's
# containers, Traefik routers and backup labels are unique. image: lines are
# deliberately left untouched (rule 2/3 anchor on their line prefix; the
# *-service / *_db tokens never appear in an image path).
_instanceRewriteCompose() {
local type="$1" slug="$2" dir="$3"
local f="$dir/docker-compose.yml"
[[ -f "$f" ]] || return 0
# 1. Traefik router/service names (<type>-service) and the db container/host
# (<type>_db) — these tokens are unambiguous, rewrite everywhere.
sed -i -E "s/\b${type}-service\b/${slug}-service/g; s/\b${type}_db\b/${slug}_db/g" "$f"
# 2. The standalone app container (container_name: <type>) — anchored so the
# image: line ending in <type> is never touched.
sed -i -E "s/(container_name:[[:space:]]*)${type}\b/\1${slug}/g" "$f"
# 3. The files-backup label's container ref (libreportal.backup.files: "<type>:/...").
sed -i -E "s/(libreportal\.backup\.files:[[:space:]]*\")${type}\b/\1${slug}/g" "$f"
# 4. The per-app tag namespace. tagsProcessorAppConfigValues derives tag names
# mechanically from the config keys (CFG_<APP>_APP_KEY_1 -> the tag
# <APP>_APP_KEY_1_TAG), so a clone still carrying the TYPE's tag names has
# nothing to match its own CFG_<SLUG>_* vars: the placeholders survive and
# the pre-start guard refuses to launch the instance.
# Deliberately narrow — the tag name right after the #LIBREPORTAL| marker,
# and the *_DATA placeholder tokens. A blanket <TYPE>_ rewrite would also
# hit an app whose compose sets a real container env var named after itself
# (- <TYPE>_SECRET=...), renaming the variable the image reads.
local type_u="${type^^}" slug_u="${slug^^}"
type_u="${type_u//-/_}"; slug_u="${slug_u//-/_}"
sed -i -E "s/(#LIBREPORTAL\|)${type_u}_/\1${slug_u}_/g" "$f"
sed -i -E "s/\b${type_u}_([A-Z0-9_]*)_DATA\b/${slug_u}_\1_DATA/g" "$f"
}
# Clone + prefix-rename the per-app tools/scripts so an instance's helpers target
# its own container and don't collide (by function name) with the type's. This is
# best-effort: it keeps the tool tree internally consistent, but apps with unusual
# tool wiring may need review before their flag is flipped.
_instanceRewriteTools() {
local type="$1" slug="$2" dir="$3"
local d f base
for d in "$dir/tools" "$dir/scripts"; do
[[ -d "$d" ]] || continue
for f in "$d/${type}_"*.sh "$d/${type}.tools.json"; do
[[ -e "$f" ]] || continue
base="$(basename "$f")"
mv "$f" "$d/${base/#${type}/${slug}}"
done
for f in "$d"/*.sh "$d"/*.json; do
[[ -e "$f" ]] || continue
# Uniform lowercase-prefix rename keeps file names, function defs and
# tools.json ids consistent; then fix container-exec + config refs.
sed -i -E "s/\b${type}_/${slug}_/g" "$f"
sed -i -E "s/(docker[[:space:]]+(exec|logs|restart|stop|start|inspect)[[:space:]]+)${type}\b/\1${slug}/g" "$f"
sed -i -E "s/\b${type}\.config\b/${slug}.config/g" "$f"
# Config reads are uppercase and so escape the lowercase rename above:
# an instance hook left reading CFG_<TYPE>_ADMIN_EMAIL would provision
# itself from the type's config (its own value silently ignored).
sed -i -E "s/\bCFG_${type^^}_/CFG_${slug^^}_/g" "$f"
# container="<type>" / container_name="<type>" holds the docker target
# for exec-based helpers (auth adapters, tools). The bare literal has
# no trailing underscore, so the rename above misses it and the clone
# would operate on the BASE app's container. Kept to these two
# assignment forms — a blanket bare-<type> rewrite would hit image
# names and prose.
sed -i -E "s/(\b(container|container_name)=\")${type}(\")/\1${slug}\3/g" "$f"
done
done
}
# Provision and install a new instance of a multi-instance-capable app.
# instanceCreate <type> <name> [domain_index] [subdomain] [local_only]
#
# local_only=true skips Traefik entirely — the instance is served on its own
# published host port and nothing else. That is also what you get implicitly when
# no CFG_DOMAIN_<n> is configured (initializeAppVariables forces the same thing),
# so the flag exists for the case where a domain IS set but this particular
# instance should stay off it.
instanceCreate() {
local type="$1" rawname="$2" domain_idx="$3" subdomain="$4" local_only="$5"
local type_dir="${install_containers_dir%/}/$type"
if [[ -z "$type" || ! -d "$type_dir" || ! -f "$type_dir/$type.config" ]]; then
isError "Instance create: unknown app type '$type'."
return 1
fi
local capable
capable=$(instanceTypeCfg "$type" "MULTI_INSTANCE")
if [[ "$capable" != "true" ]]; then
isError "App type '$type' is not multi-instance-capable. Set CFG_${type^^}_MULTI_INSTANCE=true on a reviewed app to allow it."
return 1
fi
local id
id=$(instanceIdPart "$rawname")
if [[ -z "$id" ]]; then
isError "Instance create: '$rawname' has no usable letters/digits for an instance name."
return 1
fi
local slug="${type}_${id}"
local slug_u="${slug^^}"
if [[ -d "${install_containers_dir%/}/$slug" || -d "${containers_dir%/}/$slug" ]]; then
isError "An app or instance named '$slug' already exists. Pick a different name."
return 1
fi
# Default the host to a hyphen-safe form of the slug; let the caller override.
[[ -z "$subdomain" ]] && subdomain="${slug//_/-}"
if [[ "$local_only" == "true" ]]; then
isNotice "Creating new '$type' instance '$id' (slug: $slug, LAN-only on a published port)"
else
isNotice "Creating new '$type' instance '$id' (slug: $slug, host: ${subdomain}.<domain>)"
fi
# 1. Clone the type's template tree into a new instance template.
local inst_dir="${install_containers_dir%/}/$slug"
cp -r "$type_dir" "$inst_dir"
if [[ ! -d "$inst_dir" ]]; then
isError "Instance create: failed to clone template for '$slug'."
return 1
fi
# 2. Rename the files that are keyed by the type slug.
[[ -f "$inst_dir/$type.config" ]] && mv "$inst_dir/$type.config" "$inst_dir/$slug.config"
[[ -f "$inst_dir/$type.svg" ]] && cp "$inst_dir/$type.svg" "$inst_dir/$slug.svg"
[[ -f "$inst_dir/$type.png" ]] && cp "$inst_dir/$type.png" "$inst_dir/$slug.png"
local cfg="$inst_dir/$slug.config"
# 3. Re-namespace the config (CFG_<TYPE>_* -> CFG_<SLUG>_*) then stamp the
# instance metadata. Secrets keep their RANDOMIZED* placeholders so the
# install-time scanner mints fresh ones — instances never share secrets.
sed -i -E "s/CFG_${type^^}_/CFG_${slug_u}_/g" "$cfg"
local type_title
type_title=$(instanceTypeCfg "$type" "TITLE")
[[ -z "$type_title" ]] && type_title="$type"
# APP_NAME must follow the slug. The sed above re-namespaces the KEY but
# leaves the VALUE at the type ("bookstack"), and installApp resolves the app
# it operates on from CFG_<SLUG>_APP_NAME — so an instance install ran the
# whole pipeline against the BASE app instead: same deployed dir, same ports,
# compose down/up on the already-running base container, base DB row updated,
# and the instance itself never installed. Every base app ships APP_NAME ==
# its own slug; this keeps instances to that invariant.
_instanceSetCfg "CFG_${slug_u}_APP_NAME" "$slug" "$cfg"
_instanceSetCfg "CFG_${slug_u}_INSTANCE_OF" "$type" "$cfg"
_instanceSetCfg "CFG_${slug_u}_MULTI_INSTANCE" "false" "$cfg"
_instanceSetCfg "CFG_${slug_u}_TITLE" "${type_title} · ${id}" "$cfg"
[[ -n "$domain_idx" ]] && _instanceSetCfg "CFG_${slug_u}_DOMAIN" "$domain_idx" "$cfg"
_instanceSetSubdomain "$slug_u" "$subdomain" "$cfg"
# 4. Make the cloned compose + config ports + tools target the instance's own
# identity. The config ports must follow the compose or the WebUI can't
# join the instance's ports to its services.
_instanceRewriteCompose "$type" "$slug" "$inst_dir"
_instanceRewriteConfigPorts "$type" "$slug" "$cfg"
_instanceRewriteTools "$type" "$slug" "$inst_dir"
if [[ "$local_only" == "true" ]]; then
_instanceSetLocalOnly "$slug_u" "$cfg"
isNotice "Instance '$slug' is LAN-only — no Traefik router, reachable on its published host port."
fi
isSuccessful "Instance template ready: $slug (instance of $type)"
# 5. Make the instance's freshly-cloned installers/hooks callable in THIS
# process. Both loaders ran at startup, before this dir existed: the eager
# scan (sourceScanFiles "containers") never saw it, and the lazy manifest
# has no stub for it. Without this, _appCallHook's `declare -F` finds
# nothing and every <slug>_install_* hook silently no-ops — for bookstack
# that is the readiness probe and the admin-account bootstrap, so the
# instance installs "successfully" with no usable login.
local _inst_f
while IFS= read -r -d '' _inst_f; do
source "$_inst_f"
done < <(find "$inst_dir" -maxdepth 2 -type d -name resources -prune -o -type f -name '*.sh' -print0 2>/dev/null)
# Persist that for later runs (and the WebUI): regenerate the file arrays +
# function manifest now that a new app dir exists. Best-effort — a stale
# manifest only affects lazy mode, and the in-process sourcing above already
# covers this install.
if declare -F lpRegenArrays >/dev/null 2>&1; then
lpRegenArrays force >/dev/null 2>&1 || true
fi
# 6. Hand off to the standard installer — from here it's just another app.
if ! declare -F dockerInstallApp >/dev/null 2>&1; then
isError "dockerInstallApp unavailable; instance template created but not installed."
return 1
fi
dockerInstallApp "$slug" "" "false"
}
# List instances, optionally filtered to one type. An instance is any app whose
# config declares CFG_<SLUG>_INSTANCE_OF.
instanceList() {
local want_type="$1"
local dir folder slug instance_of
for dir in "${install_containers_dir%/}"/*/; do
folder="$(basename "$dir")"
[[ -f "$dir/$folder.config" ]] || continue
instance_of=$(grep -E "^CFG_${folder^^}_INSTANCE_OF=" "$dir/$folder.config" 2>/dev/null | head -n1)
[[ -z "$instance_of" ]] && continue
instance_of="${instance_of#*=}"; instance_of="${instance_of//\"/}"; instance_of="${instance_of//$'\r'/}"
[[ -n "$want_type" && "$instance_of" != "$want_type" ]] && continue
echo "$folder (instance of $instance_of)"
done
}
# Remove an instance: standard uninstall (deployed dir + DB + compose down) then
# drop the instance's template clone. Refuses to touch a non-instance app.
instanceRemove() {
local slug="$1"
local cfg="${install_containers_dir%/}/$slug/$slug.config"
if [[ ! -f "$cfg" ]]; then
isError "Instance remove: no such instance '$slug'."
return 1
fi
if ! grep -qE "^CFG_${slug^^}_INSTANCE_OF=" "$cfg"; then
isError "'$slug' is a base app, not an instance — uninstall it via 'libreportal app uninstall $slug'."
return 1
fi
if declare -F dockerUninstallApp >/dev/null 2>&1; then
dockerUninstallApp "$slug" "false" "false"
fi
rm -rf "${install_containers_dir%/}/$slug"
isSuccessful "Removed instance '$slug'."
}