시스템 사용율 모니터링 기능 로그 수정

docker-compose-dev.yml

@@ -5,7 +5,6 @@ services:
       dockerfile: Dockerfile-dev
     image: kamco-cd-training-api:${IMAGE_TAG:-latest}
     container_name: kamco-cd-training-api
-    runtime: nvidia
     deploy:
       resources:
         reservations:

docker-compose-prod.yml

@@ -5,6 +5,13 @@ services:
       dockerfile: Dockerfile
     image: kamco-train-api:${IMAGE_TAG:-latest}
     container_name: kamco-train-api
+    deploy:
+      resources:
+        reservations:
+          devices:
+            - driver: nvidia
+              count: all
+              capabilities: [gpu]
     expose:
       - "8080"
     environment:

src/main/java/com/kamco/cd/training/common/service/GpuDmonReader.java

@@ -60,7 +60,7 @@ public class GpuDmonReader {
         runDmon(); // GPU 사용률 수집 시작
       } catch (Exception e) {
         // dmon 프로세스 종료되면 여기로 들어옴
-        log.warn("dmon restart...");
+        log.warn("dmon restart: {}", e.getMessage());
       }
 
       // 5초 대기 후 재시작