rommapp_romm/docker/init_scripts/init

#!/usr/bin/env bash

set -o errexit           # treat errors as fatal
set -o nounset           # treat unset variables as an error
set -o pipefail          # treat errors in pipes as fatal
shopt -s inherit_errexit # inherit errexit

LOGLEVEL="${LOGLEVEL:="INFO"}"

# make it possible to disable the inotify watcher process
ENABLE_RESCAN_ON_FILESYSTEM_CHANGE="${ENABLE_RESCAN_ON_FILESYSTEM_CHANGE:="false"}"
ENABLE_SCHEDULED_RESCAN="${ENABLE_SCHEDULED_RESCAN:="false"}"
ENABLE_SCHEDULED_UPDATE_LAUNCHBOX_METADATA="${ENABLE_SCHEDULED_UPDATE_LAUNCHBOX_METADATA:="false"}"
ENABLE_SCHEDULED_UPDATE_SWITCH_TITLEDB="${ENABLE_SCHEDULED_UPDATE_SWITCH_TITLEDB:="false"}"

# if REDIS_HOST is set, we assume that an external redis is used
REDIS_HOST="${REDIS_HOST:=""}"

# logger colors
RED='\033[0;31m'
LIGHTMAGENTA='\033[0;95m'
GREEN='\033[0;32m'
YELLOW='\033[0;33m'
BLUE='\033[0;34m'
CYAN='\033[0;36m'
RESET='\033[0;00m'

print_banner() {
	local version
	version=$(python3 -c "exec(open('__version__.py').read()); print(__version__)")
	info_log "               _____                 __  __ "
	info_log '              |  __ \               |  \/  |'
	info_log '              | |__) |___  _ __ ___ | \  / |'
	info_log "              |  _  // _ \\| '_ \` _ \\| |\\/| |"
	info_log '              | | \ \ (_) | | | | | | |  | |'
	info_log '              |_|  \_\___/|_| |_| |_|_|  |_|'
	info_log ""
	info_log "The beautiful, powerful, self-hosted Rom manager and player"
	info_log ""
	info_log "Version: ${version}"
	info_log ""
}

debug_log() {
	# print debug log output if enabled
	if [[ ${LOGLEVEL} == "DEBUG" ]]; then
		echo -e "${LIGHTMAGENTA}DEBUG:    ${BLUE}[RomM]${LIGHTMAGENTA}[init]${CYAN}[$(date +"%Y-%m-%d %T")]${RESET}" "${@}" || true
	fi
}

info_log() {
	echo -e "${GREEN}INFO:     ${BLUE}[RomM]${LIGHTMAGENTA}[init]${CYAN}[$(date +"%Y-%m-%d %T")]${RESET}" "${@}" || true
}

warn_log() {
	echo -e "${YELLOW}WARNING:  ${BLUE}[RomM]${LIGHTMAGENTA}[init]${CYAN}[$(date +"%Y-%m-%d %T")]${RESET}" "${@}" || true
}

error_log() {
	echo -e "${RED}ERROR:    ${BLUE}[RomM]${LIGHTMAGENTA}[init]${CYAN}[$(date +"%Y-%m-%d %T")]${RESET}" "${@}" || true
	exit 1
}

# Commands to run initial startup tasks
run_startup() {
	if ! PYTHONPATH="/backend:${PYTHONPATH-}" opentelemetry-instrument \
		--service_name "${OTEL_SERVICE_NAME_PREFIX-}startup" \
		python3 /backend/startup.py; then
		error_log "Startup script failed, exiting"
	fi
}

wait_for_gunicorn_socket() {
	debug_log "Waiting for gunicorn socket file..."
	local retries=60
	while [[ ! -S /tmp/gunicorn.sock && retries -gt 0 ]]; do
		sleep 0.5
		((retries--))
	done

	if [[ -S /tmp/gunicorn.sock ]]; then
		debug_log "Gunicorn socket file found"
	else
		warn_log "Gunicorn socket file not found after waiting 30s!"
	fi
}

# function that runs or main process and creates a corresponding PID file,
start_bin_gunicorn() {
	# cleanup potentially leftover socket
	rm /tmp/gunicorn.sock -f

	# commands to start our main application and store its PID to check for crashes
	info_log "Starting backend"

	export PYTHONUNBUFFERED=1
	export PYTHONDONTWRITEBYTECODE=1

	opentelemetry-instrument \
		--service_name "${OTEL_SERVICE_NAME_PREFIX-}api" \
		gunicorn \
		--bind=0.0.0.0:"${DEV_PORT:-5000}" \
		--bind=unix:/tmp/gunicorn.sock \
		--pid=/tmp/gunicorn.pid \
		--forwarded-allow-ips="*" \
		--worker-class uvicorn_worker.UvicornWorker \
		--workers "${WEB_SERVER_CONCURRENCY:-1}" \
		--timeout "${WEB_SERVER_TIMEOUT:-300}" \
		--keep-alive "${WEB_SERVER_KEEPALIVE:-2}" \
		--max-requests "${WEB_SERVER_MAX_REQUESTS:-1000}" \
		--max-requests-jitter "${WEB_SERVER_MAX_REQUESTS_JITTER:-100}" \
		--worker-connections "${WEB_SERVER_WORKER_CONNECTIONS:-1000}" \
		--error-logfile - \
		--log-config /etc/gunicorn/logging.conf \
		main:app &
}

# Commands to start nginx (handling PID creation internally)
start_bin_nginx() {
	wait_for_gunicorn_socket

	info_log "Starting nginx"
	if [[ ${EUID} -ne 0 ]]; then
		nginx
	else
		# if container runs as root, drop permissions
		nginx -g 'user romm;'
	fi

	: "${ROMM_BASE_URL:=http://0.0.0.0:8080}"
	info_log "🚀 RomM is now available at ${ROMM_BASE_URL}"
}

# Commands to start valkey-server (handling PID creation internally)
start_bin_valkey-server() {
	info_log "Starting internal valkey"

	if [[ -f /usr/local/etc/valkey/valkey.conf ]]; then
		if [[ ${LOGLEVEL} == "DEBUG" ]]; then
			valkey-server /usr/local/etc/valkey/valkey.conf &
		else
			valkey-server /usr/local/etc/valkey/valkey.conf >/dev/null 2>&1 &
		fi
	else
		if [[ ${LOGLEVEL} == "DEBUG" ]]; then
			valkey-server --dir /redis-data &
		else
			valkey-server --dir /redis-data >/dev/null 2>&1 &
		fi
	fi

	VALKEY_PID=$!
	echo "${VALKEY_PID}" >/tmp/valkey-server.pid

	local host="127.0.0.1"
	local port="6379"
	local max_retries=120
	local retry=0

	debug_log "Waiting for internal valkey to be ready..."

	# Temporarily disable errexit for this part of the script
	set +o errexit

	while ((retry < max_retries)); do
		# Attempt to check if valkey TCP port is open
		if (echo >/dev/tcp/"${host}"/"${port}") 2>/dev/null; then
			debug_log "Internal valkey is ready and accepting connections"
			set -o errexit # Re-enable errexit after success
			return 0
		fi

		sleep 0.5
		((retry++))
	done

	error_log "Internal valkey did not become ready after $((max_retries * 500))ms"
}

# Commands to start RQ scheduler
start_bin_rq_scheduler() {
	info_log "Starting RQ scheduler"

	RQ_REDIS_HOST=${REDIS_HOST:-127.0.0.1} \
		RQ_REDIS_PORT=${REDIS_PORT:-6379} \
		RQ_REDIS_USERNAME=${REDIS_USERNAME:-""} \
		RQ_REDIS_PASSWORD=${REDIS_PASSWORD:-""} \
		RQ_REDIS_DB=${REDIS_DB:-0} \
		RQ_REDIS_SSL=${REDIS_SSL:-0} \
		rqscheduler \
		--path /backend \
		--pid /tmp/rq_scheduler.pid &
}

# Commands to start RQ worker
start_bin_rq_worker() {
	info_log "Starting RQ worker"

	# Build Redis URL properly
	local redis_url
	if [[ -n ${REDIS_PASSWORD-} ]]; then
		redis_url="redis${REDIS_SSL:+s}://${REDIS_USERNAME-}:${REDIS_PASSWORD}@${REDIS_HOST:-127.0.0.1}:${REDIS_PORT:-6379}/${REDIS_DB:-0}"
	elif [[ -n ${REDIS_USERNAME-} ]]; then
		redis_url="redis${REDIS_SSL:+s}://${REDIS_USERNAME}@${REDIS_HOST:-127.0.0.1}:${REDIS_PORT:-6379}/${REDIS_DB:-0}"
	else
		redis_url="redis${REDIS_SSL:+s}://${REDIS_HOST:-127.0.0.1}:${REDIS_PORT:-6379}/${REDIS_DB:-0}"
	fi

	# Set PYTHONPATH so RQ can find the tasks module
	PYTHONPATH="/backend:${PYTHONPATH-}" rq worker \
		--path /backend \
		--pid /tmp/rq_worker.pid \
		--url "${redis_url}" \
		high default low &
}

start_bin_watcher() {
	info_log "Starting watcher"
	watchfiles \
		--target-type command \
		"opentelemetry-instrument --service_name '${OTEL_SERVICE_NAME_PREFIX-}watcher' python3 watcher.py" \
		/romm/library &
	WATCHER_PID=$!
	echo "${WATCHER_PID}" >/tmp/watcher.pid
}

watchdog_process_pid() {
	PROCESS=$1
	if [[ -f "/tmp/${PROCESS}.pid" ]]; then
		# Check if the pid we last wrote to our state file is actually active
		PID=$(cat "/tmp/${PROCESS}.pid") || true
		if [[ ! -d "/proc/${PID}" ]]; then
			start_bin_"${PROCESS}"
		fi
	else
		# Start process if we dont have a corresponding PID file
		start_bin_"${PROCESS}"
	fi
}

stop_process_pid() {
	PROCESS=$1
	if [[ -f "/tmp/${PROCESS}.pid" ]]; then
		PID=$(cat "/tmp/${PROCESS}.pid") || true
		if [[ -d "/proc/${PID}" ]]; then
			info_log "Stopping ${PROCESS}"
			kill "${PID}" || true
			# wait for process exit
			while [[ -e "/proc/${PID}" ]]; do sleep 0.1; done
		fi
	fi
}

shutdown() {
	# shutdown in reverse order
	stop_process_pid rq_worker
	stop_process_pid rq_scheduler
	stop_process_pid watcher
	stop_process_pid nginx
	stop_process_pid gunicorn
	stop_process_pid valkey-server
}

# switch to backend directory
cd /backend || { error_log "/backend directory doesn't seem to exist"; }

print_banner

# setup trap handler
exited=0
trap 'exited=1 && shutdown' SIGINT SIGTERM EXIT

# clear any leftover PID files
rm /tmp/*.pid -f

# Disable OpenTelemetry if no OTEL_ prefixed environment variables are set
if ! printenv | grep -q '^OTEL_'; then
	info_log "No OpenTelemetry environment variables found, disabling OpenTelemetry SDK"
	export OTEL_SDK_DISABLED=true
fi

# Set ROMM_AUTH_SECRET_KEY if not already set
if [[ -z ${ROMM_AUTH_SECRET_KEY} ]]; then
	ROMM_AUTH_SECRET_KEY=$(python3 -c "import secrets; print(secrets.token_hex(32))")
	info_log "ROMM_AUTH_SECRET_KEY not set, generating random secret key"
	export ROMM_AUTH_SECRET_KEY
fi

# Start Valkey server if REDIS_HOST is not set (which would mean user is using an external Redis/Valkey)
if [[ -z ${REDIS_HOST} ]]; then
	watchdog_process_pid valkey-server
else
	info_log "REDIS_HOST is set, not starting internal valkey-server"
fi

# Run needed database migrations once at startup
info_log "Running database migrations"
if alembic upgrade head; then
	info_log "Database migrations succeeded"
else
	error_log "Failed to run database migrations"
fi

# Startup process requires database and cache to be already available
run_startup

# main loop
while ! ((exited)); do
	watchdog_process_pid gunicorn

	# only start the scheduler if enabled
	if [[ ${ENABLE_SCHEDULED_RESCAN} == "true" || ${ENABLE_SCHEDULED_UPDATE_SWITCH_TITLEDB} == "true" || ${ENABLE_SCHEDULED_UPDATE_LAUNCHBOX_METADATA} == "true" ]]; then
		watchdog_process_pid rq_scheduler
	fi

	watchdog_process_pid rq_worker

	# only start the watcher if enabled
	if [[ ${ENABLE_RESCAN_ON_FILESYSTEM_CHANGE} == "true" ]]; then
		watchdog_process_pid watcher
	fi

	watchdog_process_pid nginx

	# check for died processes every 5 seconds
	sleep 5
done