From 7e602489509ed1e957eff48595bdbc367fabfcf4 Mon Sep 17 00:00:00 2001
From: Jerry-Xin <jerryxin0@gmail.com>
Date: Sat, 14 Mar 2026 22:05:45 +0800
Subject: [PATCH] fix(memory): do not close cached manager during doctor status
 probe
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

The doctor.memory.status handler was closing the manager in a finally
block, but MemoryIndexManager.get returns a cached instance that may
be the long-lived startup manager with an active file watcher. Closing
it tears down the watcher and disables external-change auto-reindexing.

Remove the close call — manager lifecycle is handled by
closeAllMemorySearchManagers at gateway shutdown.
---
 src/gateway/server-methods/doctor.ts | 8 +++-----
 1 file changed, 3 insertions(+), 5 deletions(-)

diff --git a/src/gateway/server-methods/doctor.ts b/src/gateway/server-methods/doctor.ts
index f842f40f4eb..a2db2841a79 100644
--- a/src/gateway/server-methods/doctor.ts
+++ b/src/gateway/server-methods/doctor.ts
@@ -56,11 +56,9 @@ export const doctorHandlers: GatewayRequestHandlers = {
       };
       respond(true, payload, undefined);
     } finally {
-      try {
-        await manager.close();
-      } catch {
-        // Ignore close errors - response already sent
-      }
+      // Do NOT close the manager here — it may be the long-lived cached
+      // instance from startup whose file watcher must stay active.
+      // Manager lifecycle is handled by closeAllMemorySearchManagers at shutdown.
     }
   },
 };