Implement health check

agents/src/http_server.ts

@@ -4,10 +4,10 @@
 import { type IncomingMessage, type Server, type ServerResponse, createServer } from 'node:http';
 import { log } from './log.js';
 
-const healthCheck = async (res: ServerResponse) => {
-  res.writeHead(200);
-  res.end('OK');
-};
+export interface HealthCheckResult {
+  healthy: boolean;
+  message: string;
+}
 
 interface WorkerResponse {
   agent_name: string;
@@ -23,13 +23,25 @@ export class HTTPServer {
   app: Server;
   #logger = log();
 
-  constructor(host: string, port: number, workerListener: () => WorkerResponse) {
+  constructor(
+    host: string,
+    port: number,
+    healthCheckListener: () => HealthCheckResult,
+    workerListener: () => WorkerResponse,
+  ) {
     this.host = host;
     this.port = port;
 
     this.app = createServer((req: IncomingMessage, res: ServerResponse) => {
       if (req.url === '/') {
-        healthCheck(res);
+        const result = healthCheckListener();
+        if (result.healthy) {
+          res.writeHead(200);
+          res.end(result.message);
+        } else {
+          res.writeHead(503);
+          res.end(result.message);
+        }
       } else if (req.url === '/worker') {
         res.writeHead(200, { 'Content-Type': 'application/json' });
         res.end(JSON.stringify(workerListener()));

agents/src/ipc/supervised_proc.ts

@@ -59,6 +59,10 @@ export abstract class SupervisedProc {
     return this.#started;
   }
 
+  get isAlive(): boolean {
+    return this.#started && !this.#closing && !!this.proc?.connected;
+  }
+
   get runningJob(): RunningJobInfo | undefined {
     return this.#runningJob;
   }

agents/src/worker.ts

@@ -339,13 +339,30 @@ export class AgentServer {
     );
 
     this.#opts = opts;
-    this.#httpServer = new HTTPServer(opts.host, opts.port, () => ({
-      agent_name: opts.agentName,
-      worker_type: JobType[opts.serverType],
-      active_jobs: this.activeJobs.length,
-      sdk_version: version,
-      project_type: PROJECT_TYPE,
-    }));
+    this.#httpServer = new HTTPServer(
+      opts.host,
+      opts.port,
+      () => {
+        // Check if inference executor exists and is not alive
+        if (this.#inferenceExecutor && !this.#inferenceExecutor.isAlive) {
+          return { healthy: false, message: 'inference process not running' };
+        }
+
+        // Check if we've lost connection to LiveKit (session closed while not intentionally closing)
+        if (!this.#closed && !this.#connecting && !this.#session) {
+          return { healthy: false, message: 'not connected to livekit' };
+        }
+
+        return { healthy: true, message: 'OK' };
+      },
+      () => ({
+        agent_name: opts.agentName,
+        worker_type: JobType[opts.serverType],
+        active_jobs: this.activeJobs.length,
+        sdk_version: version,
+        project_type: PROJECT_TYPE,
+      }),
+    );
   }
 
   /** @throws {@link WorkerError} if worker failed to connect or already running */