From 9ed574050f58bf75226095df67579cdea5b8ffeb Mon Sep 17 00:00:00 2001
From: tian <11429339@qq.com>
Date: Wed, 29 Jul 2026 23:03:11 +0800
Subject: [PATCH] fix: save last_good.json on successful reload, no rollback on
 failure

---
 agent/internal/httpapi/server.go | 30 +++++++++++++++++++++++++-----
 1 file changed, 25 insertions(+), 5 deletions(-)

diff --git a/agent/internal/httpapi/server.go b/agent/internal/httpapi/server.go
index e068f70..28a8031 100644
--- a/agent/internal/httpapi/server.go
+++ b/agent/internal/httpapi/server.go
@@ -416,13 +416,33 @@ func (s *Server) writeConfigAndReload(ctx context.Context, body []byte, restoreB
 		return fmt.Errorf("write config failed: %w", err)
 	}
 
-	// Signal media-server to reload. Don't wait for health check —
-	// graph rebuild can take 10-30s and the 1s health check timeout
-	// causes false failures. Edge-server will load the new config on
-	// next restart if hot-reload fails.
+	// Signal media-server to reload.
 	reloadCtx, reloadCancel := context.WithTimeout(ctx, 5*time.Second)
 	defer reloadCancel()
-	return s.ms.Reload(reloadCtx)
+	if err := s.ms.Reload(reloadCtx); err != nil {
+		return fmt.Errorf("reload failed: %w", err)
+	}
+
+	// Wait for media-server to recover, then save last_good as known-safe backup.
+	deadline := time.After(60 * time.Second)
+	ticker := time.NewTicker(3 * time.Second)
+	defer ticker.Stop()
+	for {
+		select {
+		case <-deadline:
+			return fmt.Errorf("media-server did not recover within 60s; config saved, will apply on restart")
+		case <-ticker.C:
+			if s.mediaServerHealthy() {
+				// Config works — persist as known-good backup.
+				if data, err := os.ReadFile(s.agentCfg.ConfigPath); err == nil {
+					_ = files.WriteFileAtomic(s.agentCfg.ConfigPath+".last_good.json", data, 0o644)
+				}
+				return nil
+			}
+		case <-ctx.Done():
+			return ctx.Err()
+		}
+	}
 }
 
 // mediaServerHealthy checks whether the media-server process is running