web-app-launcher/src/lib/server/services/metricsService.ts

import { prisma } from '../prisma.js';
import { AppStatusValue } from '$lib/utils/constants.js';

/**
 * Tiny Prometheus-text metrics gatherer. Avoids the prom-client dependency
 * (~150KB + extra runtime memory) by emitting the exposition format directly.
 * If we later want histograms or counters with labels at high cardinality,
 * swap this out for prom-client.
 */

interface CounterSnapshot {
	readonly name: string;
	readonly help: string;
	readonly value: number;
	readonly labels?: Record<string, string>;
}

function escapeLabel(value: string): string {
	return value.replace(/\\/g, '\\\\').replace(/"/g, '\\"').replace(/\n/g, '\\n');
}

function renderLabels(labels?: Record<string, string>): string {
	if (!labels) return '';
	const parts = Object.entries(labels).map(([k, v]) => `${k}="${escapeLabel(v)}"`);
	return parts.length ? `{${parts.join(',')}}` : '';
}

/**
 * In-memory counter / gauge state. Process-local — Prometheus is expected to
 * scrape a single launcher instance (the app is SQLite-bound to one process
 * anyway). Reset on restart, like most lightweight setups.
 */
class MetricRegistry {
	private counters = new Map<string, number>();
	private gauges = new Map<string, number>();

	incCounter(name: string, by = 1): void {
		this.counters.set(name, (this.counters.get(name) ?? 0) + by);
	}

	setGauge(name: string, value: number): void {
		this.gauges.set(name, value);
	}

	getCounter(name: string): number {
		return this.counters.get(name) ?? 0;
	}

	snapshot(): { counters: Map<string, number>; gauges: Map<string, number> } {
		return { counters: new Map(this.counters), gauges: new Map(this.gauges) };
	}
}

export const metricRegistry = new MetricRegistry();

// Counter names — keep them ASCII identifiers (Prometheus naming rules).
export const Counters = {
	HEALTHCHECK_TOTAL: 'wal_healthcheck_total',
	HEALTHCHECK_FAILED: 'wal_healthcheck_failed_total',
	LOGIN_SUCCESS: 'wal_login_success_total',
	LOGIN_FAILED: 'wal_login_failed_total',
	NOTIFICATION_SENT: 'wal_notification_sent_total',
	NOTIFICATION_FAILED: 'wal_notification_failed_total',
	INTEGRATION_FETCH_TOTAL: 'wal_integration_fetch_total',
	INTEGRATION_FETCH_FAILED: 'wal_integration_fetch_failed_total'
} as const;

/**
 * Build the full exposition. Combines:
 *  - process-local counters (login attempts, healthcheck ticks, etc.)
 *  - DB-backed gauges (current online/offline app count, user count, etc.)
 */
export async function renderMetrics(): Promise<string> {
	const lines: string[] = [];

	// --- Static help/type lines + counter snapshots ---
	const COUNTER_HELP: Record<string, string> = {
		[Counters.HEALTHCHECK_TOTAL]: 'Total healthcheck ticks executed since process start',
		[Counters.HEALTHCHECK_FAILED]: 'Healthcheck ticks where any app returned offline',
		[Counters.LOGIN_SUCCESS]: 'Successful local logins since process start',
		[Counters.LOGIN_FAILED]: 'Failed local logins since process start',
		[Counters.NOTIFICATION_SENT]: 'Notification dispatch attempts',
		[Counters.NOTIFICATION_FAILED]: 'Notification dispatch failures',
		[Counters.INTEGRATION_FETCH_TOTAL]: 'Integration fetch attempts',
		[Counters.INTEGRATION_FETCH_FAILED]: 'Integration fetch failures'
	};

	const { counters } = metricRegistry.snapshot();
	for (const name of Object.values(Counters)) {
		const value = counters.get(name) ?? 0;
		lines.push(`# HELP ${name} ${COUNTER_HELP[name]}`);
		lines.push(`# TYPE ${name} counter`);
		lines.push(`${name} ${value}`);
	}

	// --- DB-backed gauges ---
	const gauges: CounterSnapshot[] = [];

	try {
		const [totalApps, healthchecked, totalUsers, totalBoards] = await Promise.all([
			prisma.app.count(),
			prisma.app.count({ where: { healthcheckEnabled: true } }),
			prisma.user.count(),
			prisma.board.count()
		]);

		gauges.push(
			{ name: 'wal_apps_total', help: 'Total apps registered', value: totalApps },
			{
				name: 'wal_apps_healthchecked_total',
				help: 'Apps with healthcheck enabled',
				value: healthchecked
			},
			{ name: 'wal_users_total', help: 'Total user accounts', value: totalUsers },
			{ name: 'wal_boards_total', help: 'Total boards', value: totalBoards }
		);

		// Latest status per app — broken down by status value.
		// Subquery: for each app, take the most recent AppStatus row.
		const latest = await prisma.$queryRaw<{ status: string; count: number }[]>`
			SELECT status, COUNT(*) AS count
			FROM (
				SELECT appId, status, ROW_NUMBER() OVER (PARTITION BY appId ORDER BY checkedAt DESC) AS rn
				FROM AppStatus
			)
			WHERE rn = 1
			GROUP BY status
		`;
		for (const status of Object.values(AppStatusValue)) {
			const row = latest.find((r) => r.status === status);
			gauges.push({
				name: 'wal_app_status',
				help: 'Current count of apps by latest status',
				value: Number(row?.count ?? 0),
				labels: { status }
			});
		}
	} catch (err) {
		// DB issue — emit an "up" gauge of 0 so scrapers can alert on it.
		// eslint-disable-next-line no-console
		console.warn('[metrics] failed to gather DB gauges:', err);
		lines.push(`# HELP wal_db_up 1 if the metrics endpoint could read from the DB`);
		lines.push(`# TYPE wal_db_up gauge`);
		lines.push(`wal_db_up 0`);
		lines.push('');
		return lines.join('\n');
	}

	// Group same-name gauges so we emit HELP/TYPE once.
	const grouped = new Map<string, CounterSnapshot[]>();
	for (const g of gauges) {
		const arr = grouped.get(g.name);
		if (arr) arr.push(g);
		else grouped.set(g.name, [g]);
	}
	for (const [name, samples] of grouped) {
		lines.push(`# HELP ${name} ${samples[0].help}`);
		lines.push(`# TYPE ${name} gauge`);
		for (const s of samples) {
			lines.push(`${name}${renderLabels(s.labels)} ${s.value}`);
		}
	}

	lines.push(`# HELP wal_db_up 1 if the metrics endpoint could read from the DB`);
	lines.push(`# TYPE wal_db_up gauge`);
	lines.push(`wal_db_up 1`);
	lines.push('');
	return lines.join('\n');
}