containerName 생성 변경

src/main/java/com/kamco/cd/training/postgres/core/ModelTrainJobCoreService.java

@@ -47,7 +47,8 @@ public class ModelTrainJobCoreService {
 
   /** 실행 시작 처리 */
   @Transactional
-  public void markRunning(Long jobId, String containerName, String logPath, String lockedBy) {
+  public void markRunning(
+      Long jobId, String containerName, String logPath, String lockedBy, Integer totalEpoch) {
     ModelTrainJobEntity job =
         modelTrainJobRepository
             .findById(jobId)
@@ -59,6 +60,10 @@ public class ModelTrainJobCoreService {
     job.setStartedDttm(ZonedDateTime.now());
     job.setLockedDttm(ZonedDateTime.now());
     job.setLockedBy(lockedBy);
+
+    if (totalEpoch != null) {
+      job.setTotalEpoch(totalEpoch);
+    }
   }
 
   /** 성공 처리 */

src/main/java/com/kamco/cd/training/postgres/entity/ModelTrainJobEntity.java

@@ -78,6 +78,12 @@ public class ModelTrainJobEntity {
   @Column(name = "locked_by", length = 100)
   private String lockedBy;
 
+  @Column(name = "total_epoch")
+  private Integer totalEpoch;
+
+  @Column(name = "current_epoch")
+  private Integer currentEpoch;
+
   public ModelTrainJobDto toDto() {
     return new ModelTrainJobDto(
         this.id,
@@ -90,6 +96,8 @@ public class ModelTrainJobEntity {
         this.paramsJson,
         this.queuedDttm,
         this.startedDttm,
-        this.finishedDttm);
+        this.finishedDttm,
+        this.totalEpoch,
+        this.currentEpoch);
   }
 }

src/main/java/com/kamco/cd/training/train/dto/ModelTrainJobDto.java

@@ -20,4 +20,6 @@ public class ModelTrainJobDto {
   private ZonedDateTime queuedDttm;
   private ZonedDateTime startedDttm;
   private ZonedDateTime finishedDttm;
+  private Integer totalEpoch;
+  private Integer currentEpoch;
 }

src/main/java/com/kamco/cd/training/train/service/DockerTrainService.java

@@ -87,7 +87,7 @@ public class DockerTrainService {
                     log.info("[EPOCH] container={} {}/{}", containerName, currentEpoch, totalEpoch);
 
                     // TODO 실행중인 에폭 저장 필요하면 만들어야함
-                    // TODO 완료여부를 여기다가?
+                    // TODO 하지만 여기서 트랜젝션 걸리는 db 작업하면 안좋다고하는데..?
                     //                modelTrainMngCoreService.updateCurrentEpoch(modelId,
                     // currentEpoch, totalEpoch);
                   }

src/main/java/com/kamco/cd/training/train/service/TrainJobService.java

@@ -58,6 +58,7 @@ public class TrainJobService {
     Map<String, Object> paramsMap = objectMapper.convertValue(trainRunRequest, Map.class);
     paramsMap.put("jobType", "TRAIN");
     paramsMap.put("uuid", trainRunRequest.getUuid());
+    paramsMap.put("totalEpoch", trainRunRequest.getEpochs());
 
     Long jobId =
         modelTrainJobCoreService.createQueuedJob(

src/main/java/com/kamco/cd/training/train/service/TrainJobWorker.java

@@ -47,9 +47,17 @@ public class TrainJobWorker {
 
     boolean isEval = "EVAL".equals(jobType);
 
-    String containerName = (isEval ? "eval-" : "train-") + jobId + "-" + params.get("uuid");
+    String containerName =
+        (isEval ? "eval-" : "train-") + jobId + "-" + params.get("uuid").toString().substring(0, 8);
 
-    modelTrainJobCoreService.markRunning(jobId, containerName, null, "TRAIN_WORKER");
+    Integer totalEpoch = null;
+    if (params.containsKey("totalEpoch")) {
+      if (params.get("totalEpoch") != null) {
+        totalEpoch = Integer.parseInt(params.get("totalEpoch").toString());
+      }
+    }
+
+    modelTrainJobCoreService.markRunning(jobId, containerName, null, "TRAIN_WORKER", totalEpoch);
 
     try {
       TrainRunResult result;