feat(core): Implement task timeouts and heartbeats for runners (no-changelog) (#11690)

Co-authored-by: कारतोफ्फेलस्क्रिप्ट™ <aditya@netroy.in>

packages/@n8n/task-runner/src/config/base-runner-config.ts

@@ -1,4 +1,16 @@
-import { Config, Env } from '@n8n/config';
+import { Config, Env, Nested } from '@n8n/config';
+
+@Config
+class HealthcheckServerConfig {
+	@Env('N8N_RUNNERS_SERVER_ENABLED')
+	enabled: boolean = false;
+
+	@Env('N8N_RUNNERS_SERVER_HOST')
+	host: string = '127.0.0.1';
+
+	@Env('N8N_RUNNERS_SERVER_PORT')
+	port: number = 5680;
+}
 
 @Config
 export class BaseRunnerConfig {
@@ -13,4 +25,7 @@ export class BaseRunnerConfig {
 
 	@Env('N8N_RUNNERS_MAX_CONCURRENCY')
 	maxConcurrency: number = 5;
+
+	@Nested
+	healthcheckServer!: HealthcheckServerConfig;
 }

packages/@n8n/task-runner/src/healthcheck-server.ts

@@ -0,0 +1,38 @@
+import { ApplicationError } from 'n8n-workflow';
+import { createServer } from 'node:http';
+
+export class HealthcheckServer {
+	private server = createServer((_, res) => {
+		res.writeHead(200);
+		res.end('OK');
+	});
+
+	async start(host: string, port: number) {
+		return await new Promise<void>((resolve, reject) => {
+			const portInUseErrorHandler = (error: NodeJS.ErrnoException) => {
+				if (error.code === 'EADDRINUSE') {
+					reject(new ApplicationError(`Port ${port} is already in use`));
+				} else {
+					reject(error);
+				}
+			};
+
+			this.server.on('error', portInUseErrorHandler);
+
+			this.server.listen(port, host, () => {
+				this.server.removeListener('error', portInUseErrorHandler);
+				console.log(`Healthcheck server listening on ${host}, port ${port}`);
+				resolve();
+			});
+		});
+	}
+
+	async stop() {
+		return await new Promise<void>((resolve, reject) => {
+			this.server.close((error) => {
+				if (error) reject(error);
+				else resolve();
+			});
+		});
+	}
+}

packages/@n8n/task-runner/src/start.ts

@@ -3,8 +3,10 @@ import Container from 'typedi';
 
 import { MainConfig } from './config/main-config';
 import type { ErrorReporter } from './error-reporter';
+import type { HealthcheckServer } from './healthcheck-server';
 import { JsTaskRunner } from './js-task-runner/js-task-runner';
 
+let healthcheckServer: HealthcheckServer | undefined;
 let runner: JsTaskRunner | undefined;
 let isShuttingDown = false;
 let errorReporter: ErrorReporter | undefined;
@@ -22,6 +24,7 @@ function createSignalHandler(signal: string) {
 			if (runner) {
 				await runner.stop();
 				runner = undefined;
+				void healthcheckServer?.stop();
 			}
 
 			if (errorReporter) {
@@ -49,6 +52,14 @@ void (async function start() {
 
 	runner = new JsTaskRunner(config);
 
+	const { enabled, host, port } = config.baseRunnerConfig.healthcheckServer;
+
+	if (enabled) {
+		const { HealthcheckServer } = await import('./healthcheck-server');
+		healthcheckServer = new HealthcheckServer();
+		await healthcheckServer.start(host, port);
+	}
+
 	process.on('SIGINT', createSignalHandler('SIGINT'));
 	process.on('SIGTERM', createSignalHandler('SIGTERM'));
 })().catch((e) => {