Add metrics

ansible/roles/tldraw-server/templates/configmap.yml.j2

@@ -9,4 +9,5 @@ data:
   API_HOST: "http://api-svc:3030"
   WS_PATH_PREFIX: "/tldraw-server"
   LOG: "^(yjs|@y|server)"
+  FEATURE_PROMETHEUS_METRICS_ENABLED: true
   REDIS_SENTINEL_SERVICE_NAME: valkey-headless.{{ NAMESPACE }}.svc.cluster.local

ansible/roles/tldraw-server/templates/deployment.yml.j2

@@ -49,6 +49,9 @@ spec:
         - containerPort: 3345
           name: tldraw-ws
           protocol: TCP
+        - containerPort: 9090
+          name: tldraw-server-metrics
+          protocol: TCP
         envFrom:
         - configMapRef:
             name: tldraw-server-configmap

ansible/roles/tldraw-server/templates/pod-monitor.yml.j2

@@ -0,0 +1,14 @@
+apiVersion: monitoring.coreos.com/v1
+kind: PodMonitor
+metadata:
+  name: tldraw-server-pod-monitor
+  namespace: {{ NAMESPACE }}
+  labels:
+    app: tldraw-server
+spec:
+  selector:
+    matchLabels:
+      app: tldraw-server
+  endpoints:
+  - path: /metrics
+    port: tldraw-server-metrics

package.json

@@ -14,9 +14,10 @@
 		"test": "echo \"Error: no test specified\" && exit 1"
 	},
 	"dependencies": {
-		"@y/redis": "github:hpi-schul-cloud/y-redis#e7a2965d85a668f4f97040f18c129af6eb21287a",
+		"@y/redis": "github:hpi-schul-cloud/y-redis#ee6d920f77c3d30521f49506797e3d71dd026903",
 		"ioredis": "^5.4.1",
 		"lib0": "^0.2.93",
+		"prom-client": "^15.1.3",
 		"uws": "github:uNetworking/uWebSockets.js#v20.40.0",
 		"y-protocols": "^1.0.6"
 	},

src/metrics.js

@@ -0,0 +1,33 @@
+import { number } from 'lib0';
+import * as env from 'lib0/environment';
+import { Gauge, register } from 'prom-client';
+import * as uws from 'uws';
+
+const openConnectionsGauge = new Gauge({
+	name: 'tldraw_open_connections',
+	help: 'Number of open WebSocket connections on tldraw-server.',
+});
+
+export const incOpenConnectionsGauge = () => {
+	openConnectionsGauge.inc();
+};
+
+export const decOpenConnectionsGauge = () => {
+	openConnectionsGauge.dec();
+};
+
+export const exposeMetricsToPrometheus = () => {
+	const route = env.getConf('prometheus-metrics-route') ?? '/metrics';
+	const port = number.parseInt(env.getConf('prometheus-metrics-port') ?? '9090');
+
+	const app = uws.App({});
+
+	app.get(route, async (res) => {
+		const metrics = await register.metrics();
+		res.end(metrics);
+	});
+
+	app.listen(port, () => {
+		console.log('Prometheus metrics exposed on port 9090');
+	});
+};

src/server.js

@@ -5,6 +5,7 @@ import * as logging from 'lib0/logging';
 import * as number from 'lib0/number';
 import * as promise from 'lib0/promise';
 import * as uws from 'uws';
+import { decOpenConnectionsGauge, exposeMetricsToPrometheus, incOpenConnectionsGauge } from './metrics.js';
 import { redis } from './redis.js';
 import { initStorage } from './storage.js';
 
@@ -34,7 +35,18 @@ export const createYWebsocketServer = async ({ redisPrefix = 'y', port, store })
 
 	const app = uws.App({});
 
-	await registerYWebsocketServer(app, `${wsPathPrefix}/:room`, store, checkAuthz, { redisPrefix }, redis);
+	await registerYWebsocketServer(
+		app,
+		`${wsPathPrefix}/:room`,
+		store,
+		checkAuthz,
+		{
+			redisPrefix,
+			openWsCallback,
+			closeWsCallback,
+		},
+		redis,
+	);
 
 	await promise.create((resolve, reject) => {
 		app.listen(port, (token) => {
@@ -94,8 +106,20 @@ const createAuthzRequestOptions = (room, token) => {
 	return requestOptions;
 };
 
+const openWsCallback = () => {
+	incOpenConnectionsGauge();
+};
+
+const closeWsCallback = () => {
+	decOpenConnectionsGauge();
+};
+
 const port = number.parseInt(env.getConf('port') || '3345');
 const redisPrefix = env.getConf('redis-prefix') || 'y';
 const store = await initStorage();
 
+if (env.getConf('feature-prometheus-metrics-enabled') === 'true') {
+	exposeMetricsToPrometheus();
+}
+
 createYWebsocketServer({ port, store, redisPrefix });