fix: run boot recovery under the CLI entry point too

Boot recovery (trace-poll resume, orphan benchmark/evaluation-run
finalization) lived only in server/index.ts, which the CLI never
executes — 'agent-health serve' imports app.js and listens itself, so
recovery was dead code on the primary distribution path. Extract a
shared runBootRecoverySafely() and call it post-listen from both
entries (after AH_PORT reflects the bound port, since the trace poller
makes HTTP self-calls).

Signed-off-by: ashwin pc <ashwinpc@amazon.com>

Author: ashwin-pc

cli/utils/startServer.ts

@@ -49,7 +49,7 @@ export async function startServer(options: StartOptions): Promise<number> {
 
   const packageRoot = findPackageRoot();
   const serverPath = join(packageRoot, 'server', 'dist', 'app.js');
-  const { createApp } = await import(serverPath);
+  const { createApp, runBootRecoverySafely } = await import(serverPath);
 
   const app = await createApp();
 
@@ -83,5 +83,11 @@ export async function startServer(options: StartOptions): Promise<number> {
   if (actualPort !== options.port) {
     process.env.VITE_BACKEND_PORT = String(actualPort);
   }
+
+  // Post-listen boot recovery (orphan run finalization, trace-poll resume).
+  // Must run after AH_PORT reflects the bound port because the trace poller
+  // makes HTTP self-calls. Guarded for older compiled bundles without it.
+  runBootRecoverySafely?.();
+
   return actualPort;
 }

server/app.ts

@@ -128,3 +128,7 @@ export async function createApp(): Promise<Express> {
 }
 
 export default createApp;
+
+// Re-exported so the CLI entry (cli/utils/startServer.ts), which imports the
+// compiled app bundle, can run the same post-listen recovery as server/index.ts.
+export { runBootRecoverySafely } from './services/bootRecovery.js';

server/index.ts

@@ -15,10 +15,7 @@ import { createApp } from './app.js';
 import { getStorageConfigFromFile, getObservabilityConfigFromFile, getStorageConfigFromTs, getObservabilityConfigFromTs } from './services/configService.js';
 import { findObservioRoot, spawnObservioAgent, OBSERVIO_DEFAULT_PORT, resetObservioPort, isPortFree, setObservioPort, waitForObservioReady, killObservioAgent } from './services/observioAgent.js';
 import { validateAwsCredentials } from './services/tracesService.js';
-import { resumePendingTracePollsSafely } from './services/traceRecoveryOnBoot.js';
-import { recoverOrphanBenchmarkRunsSafely } from './services/benchmarkRunRecoveryOnBoot.js';
-import { recoverOrphanEvaluationRunsSafely } from './services/evaluationRunRecoveryOnBoot.js';
-import { getStorageModule } from './adapters/index.js';
+import { runBootRecoverySafely } from './services/bootRecovery.js';
 
 // Register server-side connectors (subprocess, claude-code)
 // This import has side effects that register connectors with the registry
@@ -159,25 +156,11 @@ async function startServer() {
         }
         console.log('');
 
-        // Resume orphan trace-mode polling that was lost during a restart.
-        // Fire-and-forget — must never block server startup or crash on failure.
-        // Runs after listen() so the poller's HTTP self-calls (asyncRunStorage)
-        // can reach the local API.
-        try {
-          const storage = getStorageModule();
-          if (storage) {
-            resumePendingTracePollsSafely(storage);
-            // Also fail out orphan BenchmarkRuns (status: 'running' for too long
-            // with the runner long dead). Different bug class — see
-            // server/services/benchmarkRunRecoveryOnBoot.ts.
-            recoverOrphanBenchmarkRunsSafely(storage);
-            // Top-level EvaluationRuns use a separate process-local registry;
-            // finalize any persisted `running` docs left by the prior process.
-            recoverOrphanEvaluationRunsSafely(storage);
-          }
-        } catch (err: any) {
-          console.warn(`[bootRecovery] Could not start: ${err?.message || err}`);
-        }
+        // Resume orphan trace polling / finalize orphan runs lost in a restart.
+        // Shared with the CLI entry point (cli/utils/startServer.ts) — see
+        // server/services/bootRecovery.ts. Runs after listen() so the trace
+        // poller's HTTP self-calls (asyncRunStorage) can reach the local API.
+        runBootRecoverySafely();
 
         // Graceful shutdown — stop background timers, kill child processes, drain connections
         const shutdown = (signal: string) => {

server/services/bootRecovery.ts

@@ -0,0 +1,35 @@
+/*
+ * Copyright OpenSearch Contributors
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+/**
+ * Boot Recovery — shared post-listen recovery hooks.
+ *
+ * Historically these hooks lived only in server/index.ts, so they ran under
+ * `node server/dist/index.js` (dev) but NEVER under the CLI (`agent-health
+ * serve`), which imports app.js and listens itself. Every entry point that
+ * listens should call this exactly once, after listen(), because the trace
+ * poller makes HTTP self-calls to the local API.
+ */
+
+import { getStorageModule } from '../adapters/index.js';
+import { resumePendingTracePollsSafely } from './traceRecoveryOnBoot.js';
+import { recoverOrphanBenchmarkRunsSafely } from './benchmarkRunRecoveryOnBoot.js';
+import { recoverOrphanEvaluationRunsSafely } from './evaluationRunRecoveryOnBoot.js';
+
+/**
+ * Fire-and-forget: must never block startup or throw.
+ * Call once per process, after the HTTP server is listening.
+ */
+export function runBootRecoverySafely(): void {
+  try {
+    const storage = getStorageModule();
+    if (!storage) return;
+    resumePendingTracePollsSafely(storage);
+    recoverOrphanBenchmarkRunsSafely(storage);
+    recoverOrphanEvaluationRunsSafely(storage);
+  } catch (err: any) {
+    console.warn(`[bootRecovery] Could not start: ${err?.message || err}`);
+  }
+}