Merge branch 'develop' of https://kamco.git.gs.dabeeo.com/MVPTeam/kamco-train-api into develop

2026-03-04 06:12:27 +09:00
parent df3bedfbda 1c7c213977
commit 43d0e55cb7
8 changed files with 365 additions and 105 deletions
--- a/src/main/java/com/kamco/cd/training/common/utils/FIleChecker.java
+++ b/src/main/java/com/kamco/cd/training/common/utils/FIleChecker.java
@@ -3,7 +3,6 @@ package com.kamco.cd.training.common.utils;
 import static java.lang.String.CASE_INSENSITIVE_ORDER;

 import com.jcraft.jsch.ChannelExec;
-import com.jcraft.jsch.ChannelSftp;
 import com.jcraft.jsch.JSch;
 import com.jcraft.jsch.Session;
 import com.kamco.cd.training.common.exception.CustomApiException;
--- a/src/main/java/com/kamco/cd/training/dataset/service/DatasetService.java
+++ b/src/main/java/com/kamco/cd/training/dataset/service/DatasetService.java
@@ -398,9 +398,8 @@ public class DatasetService {
    return datasetCoreService.getFilePathByUUIDPathType(uuid, pathType);
  }

-
-
  private String escape(String path) {
+    // 쉘 커맨드에서 안전하게 사용할 수 있도록 문자열을 작은따옴표로 감싸면서, 내부의 작은따옴표를 이스케이프 처리
    return "'" + path.replace("'", "'\"'\"'") + "'";
  }

--- a/src/main/java/com/kamco/cd/training/postgres/core/ModelTrainMngCoreService.java
+++ b/src/main/java/com/kamco/cd/training/postgres/core/ModelTrainMngCoreService.java
@@ -430,6 +430,44 @@ public class ModelTrainMngCoreService {
    master.setUpdatedDttm(ZonedDateTime.now());
  }

+  /**
+   * step1 정지 처리
+   *
+   * @param modelId
+   * @param errorMessage
+   */
+  public void markStep1Stop(Long modelId, String errorMessage) {
+    ModelMasterEntity master =
+        modelMngRepository
+            .findById(modelId)
+            .orElseThrow(() -> new IllegalArgumentException("Model not found: " + modelId));
+
+    master.setStatusCd(TrainStatusType.STOPPED.getId());
+    master.setStep1State(TrainStatusType.STOPPED.getId());
+    master.setLastError(errorMessage);
+    master.setUpdatedUid(userUtil.getId());
+    master.setUpdatedDttm(ZonedDateTime.now());
+  }
+
+  /**
+   * step2 정지 처리
+   *
+   * @param modelId
+   * @param errorMessage
+   */
+  public void markStep2Stop(Long modelId, String errorMessage) {
+    ModelMasterEntity master =
+        modelMngRepository
+            .findById(modelId)
+            .orElseThrow(() -> new IllegalArgumentException("Model not found: " + modelId));
+
+    master.setStatusCd(TrainStatusType.STOPPED.getId());
+    master.setStep2State(TrainStatusType.STOPPED.getId());
+    master.setLastError(errorMessage);
+    master.setUpdatedUid(userUtil.getId());
+    master.setUpdatedDttm(ZonedDateTime.now());
+  }
+
  @Transactional
  public void markSuccess(Long modelId) {
    ModelMasterEntity master =
--- a/src/main/java/com/kamco/cd/training/train/service/DataSetCountersService.java
+++ b/src/main/java/com/kamco/cd/training/train/service/DataSetCountersService.java
@@ -222,7 +222,9 @@ public class DataSetCountersService {
    log.info("missing               = {}", missing.size());
    log.info("extra                 = {}", extra.size());

+    log.info("[MISSING] total = {}", missing.size());
    missing.stream().sorted().limit(50).forEach(f -> log.warn("[MISSING] {}", f));
+    log.info("[EXTRA] total = {}", extra.size());
    extra.stream().sorted().limit(50).forEach(f -> log.warn("[EXTRA] {}", f));
  }
 }
--- a/src/main/java/com/kamco/cd/training/train/service/JobRecoveryOnStartupService.java
+++ b/src/main/java/com/kamco/cd/training/train/service/JobRecoveryOnStartupService.java
@@ -20,7 +20,9 @@ import java.util.stream.Stream;
 import lombok.RequiredArgsConstructor;
 import lombok.extern.log4j.Log4j2;
 import org.springframework.beans.factory.annotation.Value;
+import org.springframework.boot.context.event.ApplicationReadyEvent;
 import org.springframework.context.annotation.Profile;
+import org.springframework.context.event.EventListener;
 import org.springframework.stereotype.Component;
 import org.springframework.transaction.annotation.Transactional;

@@ -41,6 +43,7 @@ public class JobRecoveryOnStartupService {

  private final ModelTrainJobCoreService modelTrainJobCoreService;
  private final ModelTrainMngCoreService modelTrainMngCoreService;
+  private final ModelTrainMetricsJobService modelTrainMetricsJobService;

  /**
   * Docker 컨테이너가 쓰는 response(산출물) 디렉토리의 "호스트 측" 베이스 경로. 예) /data/train/response
@@ -56,7 +59,7 @@ public class JobRecoveryOnStartupService {
   * <p>@Transactional: - recover() 메서드 전체가 하나의 트랜잭션으로 감싸집니다. - Job 하나씩 처리하다가 예외가 발생하면 전체 롤백이 될 수
   * 있으므로 "잡 단위로 확실히 커밋"이 필요하면 (권장) 잡 단위로 분리 트랜잭션(REQUIRES_NEW) 고려하세요.
   */
-  //  @EventListener(ApplicationReadyEvent.class)
+  @EventListener(ApplicationReadyEvent.class)
  @Transactional
  public void recover() {

@@ -91,57 +94,92 @@ public class JobRecoveryOnStartupService {
          if (out.completed()) {
            log.info("[RECOVERY] outputs look completed. mark SUCCESS. jobId={}", job.getId());
            modelTrainJobCoreService.markSuccess(job.getId(), 0);
+            // model 상태 변경
            markStepSuccessByJobType(job);
+            // 결과 csv 파일 정보 등록
+            modelTrainMetricsJobService.findTrainValidMetricCsvFiles();

          } else {
-            // 3-3) 산출물이 부족하면 실패 처리(운영 정책에 따라 "유예"도 가능)
+
+            // 3-3) 산출물이 부족하면 중단처리
+            // 산출물이 부족하면 "중단/보류"로 처리
+            // 운영자가 재시작 할 수 있게 한다.
            log.warn(
-                "[RECOVERY] outputs incomplete. mark FAILED. jobId={} reason={}",
+                "[RECOVERY] outputs incomplete. mark PAUSED/STOP for restart. jobId={} reason={}",
                job.getId(),
                out.reason());

-            modelTrainJobCoreService.markFailed(
-                job.getId(), -1, "SERVER_RESTART_CONTAINER_MISSING_OUTPUT_INCOMPLETE");
+            Integer modelId = job.getModelId() == null ? null : Math.toIntExact(job.getModelId());

-            markStepErrorByJobType(job, out.reason());
+            //  PAUSED/STOP
+            modelTrainJobCoreService.markPaused(
+                job.getId(), modelId, "SERVER_RESTART_CONTAINER_MISSING_OUTPUT_INCOMPLETE");
+
+            // 모델도 에러가 아니라 STOP으로
+            markStepStopByJobType(
+                job, "SERVER_RESTART_CONTAINER_MISSING_OUTPUT_INCOMPLETE: " + out.reason());
          }
          continue;
        }

        // 4) 컨테이너는 존재하고, 아직 running=true
        //    - 서버만 재기동됐고 컨테이너는 그대로 살아있는 케이스
-        //    - 이 경우 DB를 건드리면 오히려 꼬일 수 있으니 RUNNING 유지
+        //    - 실행중 docker 를 kill 하고 이어하기를 한다,
        if (state.running()) {
-          log.info("[RECOVERY] container still running. container={}", containerName);
+          log.warn("[RECOVERY] container still running. force kill. container={}", containerName);
+
          try {
-            ProcessBuilder pb = new ProcessBuilder("docker", "stop", "-t", "20", containerName);
+            // ============================================================
+            // 1) docker kill (SIGKILL) 바로 전송
+            //    - kill은 즉시 종료
+            // ============================================================
+            ProcessBuilder pb = new ProcessBuilder("docker", "kill", containerName);
            pb.redirectErrorStream(true);

            Process p = pb.start();

-            boolean finished = p.waitFor(30, TimeUnit.SECONDS);
+            boolean finished = p.waitFor(20, TimeUnit.SECONDS);
            if (!finished) {
              p.destroyForcibly();
-              throw new IOException("docker stop timeout");
+              throw new IOException("docker kill timeout");
            }

            int code = p.exitValue();
            if (code != 0) {
-              throw new IOException("docker stop failed. exit=" + code);
+              throw new IOException("docker kill failed. exit=" + code);
            }

-            log.info(
-                "[RECOVERY] container stopped (will be auto removed by --rm). container={}",
-                containerName);
+            // ============================================================
+            // 2) kill 후 실제로 죽었는지 확인
+            // ============================================================
+            DockerInspectState after = inspectContainer(containerName);
+            if (after.exists() && after.running()) {
+              throw new IOException("docker kill returned 0 but container still running");
+            }

-            // 여기서 상태를 PAUSED로 바꿔도 되고
-            modelTrainJobCoreService.markPaused(job.getId(), -1, "AUTO_STOP_FAILED_ON_RESTART");
+            log.info("[RECOVERY] container killed successfully. container={}", containerName);
+
+            // ============================================================
+            // 3) job 상태를 PAUSED로 변경 (서버 재기동으로 강제 중단)
+            // ============================================================
+            Integer modelId = job.getModelId() == null ? null : Math.toIntExact(job.getModelId());
+
+            modelTrainJobCoreService.markPaused(
+                job.getId(), modelId, "AUTO_KILLED_ON_SERVER_RESTART");
+
+            log.info("job = {}", job);
+            markStepStopByJobType(job, "AUTO_KILLED_ON_SERVER_RESTART");

          } catch (Exception e) {
-            log.error("[RECOVERY] docker stop failed. container={}", containerName, e);

-            modelTrainJobCoreService.markFailed(job.getId(), -1, "AUTO_STOP_FAILED_ON_RESTART");
+            log.error("[RECOVERY] docker kill failed. container={}", containerName, e);
+
+            modelTrainJobCoreService.markFailed(
+                job.getId(), -1, "AUTO_KILL_FAILED_ON_SERVER_RESTART");
+
+            markStepErrorByJobType(job, "AUTO_KILL_FAILED_ON_SERVER_RESTART");
          }
+
          continue;
        }

@@ -154,7 +192,10 @@ public class JobRecoveryOnStartupService {
        if (exitCode != null && exitCode == 0) {
          log.info("[RECOVERY] container exited(0). mark SUCCESS. container={}", containerName);
          modelTrainJobCoreService.markSuccess(job.getId(), 0);
+          // model 상태 변경
          markStepSuccessByJobType(job);
+          // 결과 csv 파일 정보 등록
+          modelTrainMetricsJobService.findTrainValidMetricCsvFiles();

        } else {
          // 5-2) exitCode != 0 이거나 null이면 실패로 간주 → FAILED 처리
@@ -166,7 +207,7 @@ public class JobRecoveryOnStartupService {

          modelTrainJobCoreService.markFailed(
              job.getId(), exitCode, "SERVER_RESTART_CONTAINER_EXIT_NONZERO");
-
+          // model 상태 변경
          markStepErrorByJobType(job, "exit=" + exitCode + " status=" + status);
        }

@@ -178,7 +219,7 @@ public class JobRecoveryOnStartupService {

        modelTrainJobCoreService.markFailed(
            job.getId(), -1, "SERVER_RESTART_CONTAINER_INSPECT_ERROR");
-
+        // model 상태 변경
        markStepErrorByJobType(job, "inspect-error");
      }
    }
@@ -204,6 +245,16 @@ public class JobRecoveryOnStartupService {
   *
   * <p>예: - jobType == "EVAL" → step2(평가 단계) 에러 - 그 외 → step1 혹은 전체 에러
   */
+  private void markStepStopByJobType(ModelTrainJobDto job, String msg) {
+    Map<String, Object> params = job.getParamsJson();
+    boolean isEval = params != null && "EVAL".equals(String.valueOf(params.get("jobType")));
+    if (isEval) {
+      modelTrainMngCoreService.markStep2Stop(job.getModelId(), msg);
+    } else {
+      modelTrainMngCoreService.markStep1Stop(job.getModelId(), msg);
+    }
+  }
+
  private void markStepErrorByJobType(ModelTrainJobDto job, String msg) {
    Map<String, Object> params = job.getParamsJson();
    boolean isEval = params != null && "EVAL".equals(String.valueOf(params.get("jobType")));
@@ -310,33 +361,66 @@ public class JobRecoveryOnStartupService {
   */
  private OutputResult probeOutputs(ModelTrainJobDto job) {
    try {
+
+      log.info(
+          "[RECOVERY] probeOutputs start. jobId={}, modelId={}", job.getId(), job.getModelId());
+
+      // 1) 출력 디렉토리 확인
      Path outDir = resolveOutputDir(job);
      if (outDir == null || !Files.isDirectory(outDir)) {
+        log.warn("[RECOVERY] output directory missing. jobId={}, path={}", job.getId(), outDir);
        return new OutputResult(false, "output-dir-missing");
      }

+      log.info("[RECOVERY] output directory found. jobId={}, path={}", job.getId(), outDir);
+
+      // 2) totalEpoch 확인
      Integer totalEpoch = extractTotalEpoch(job).orElse(null);
      if (totalEpoch == null || totalEpoch <= 0) {
+        log.warn(
+            "[RECOVERY] totalEpoch missing or invalid. jobId={}, totalEpoch={}",
+            job.getId(),
+            totalEpoch);
        return new OutputResult(false, "total-epoch-missing");
      }

+      log.info("[RECOVERY] totalEpoch={}. jobId={}", totalEpoch, job.getId());
+
+      // 3) val.csv 존재 확인
      Path valCsv = outDir.resolve("val.csv");
      if (!Files.exists(valCsv)) {
+        log.warn("[RECOVERY] val.csv missing. jobId={}, path={}", job.getId(), valCsv);
        return new OutputResult(false, "val.csv-missing");
      }

+      // 4) val.csv 라인 수 확인
      long lines = countNonHeaderLines(valCsv);

-      // “같아야 완료” 정책
+      log.info(
+          "[RECOVERY] val.csv lines counted. jobId={}, lines={}, expected={}",
+          job.getId(),
+          lines,
+          totalEpoch);
+
+      // 5) 완료 판정
      if (lines == totalEpoch) {
+        log.info("[RECOVERY] outputs look COMPLETE. jobId={}", job.getId());
        return new OutputResult(true, "ok");
      }

+      log.warn(
+          "[RECOVERY] val.csv line mismatch. jobId={}, lines={}, expected={}",
+          job.getId(),
+          lines,
+          totalEpoch);
+
      return new OutputResult(
          false, "val.csv-lines-mismatch lines=" + lines + " expected=" + totalEpoch);

    } catch (Exception e) {
+
      log.error("[RECOVERY] probeOutputs error. jobId={}", job.getId(), e);
+
      return new OutputResult(false, "probe-error");
    }
  }
--- a/src/main/java/com/kamco/cd/training/train/service/TmpDatasetService.java
+++ b/src/main/java/com/kamco/cd/training/train/service/TmpDatasetService.java
@@ -20,71 +20,187 @@ public class TmpDatasetService {
  @Value("${train.docker.basePath}")
  private String trainBaseDir;

-  private final DataSetCountersService dataSetCountersService;
-
  /**
-   * 다른 데이터셋 파일과 이름이 겹치면 그 파일은 skip함
+   * train, val, test 폴더별로 link
   *
-   * @param uid
-   * @param type
-   * @param links
+   * @param uid 임시폴더 uuid
+   * @param type train, val, test
+   * @param links tif pull path
   * @return
   * @throws IOException
   */
-  public String buildTmpDatasetSymlink(String uid, String type, List<ModelTrainLinkDto> links)
+  public void buildTmpDatasetHardlink(String uid, String type, List<ModelTrainLinkDto> links)
      throws IOException {

-    if (uid == null || uid.isBlank()) throw new IOException("uid is empty");
-    if (type == null || type.isBlank()) throw new IOException("type is empty");
-    if (links == null || links.isEmpty()) throw new IOException("links is empty");
+    if (links == null || links.isEmpty()) {
+      throw new IOException("links is empty");
+    }

-    log.info("========== buildTmpDatasetHardlink MERGE START ==========");
-    log.info("uid={}, type={}, links.size={}", uid, type, links.size());
+    Path tmp = Path.of(trainBaseDir, "tmp", uid);
+
+    long hardlinksMade = 0;
+
+    for (ModelTrainLinkDto dto : links) {
+
+      if (type == null) {
+        log.warn("SKIP - trainType null: {}", dto);
+        continue;
+      }
+
+      // type별 디렉토리 생성
+      Files.createDirectories(tmp.resolve(type).resolve("input1"));
+      Files.createDirectories(tmp.resolve(type).resolve("input2"));
+      Files.createDirectories(tmp.resolve(type).resolve("label"));
+      Files.createDirectories(tmp.resolve(type).resolve("label-json"));
+
+      // comparePath → input1
+      hardlinksMade += link(tmp, type, "input1", dto.getComparePath());
+
+      // targetPath → input2
+      hardlinksMade += link(tmp, type, "input2", dto.getTargetPath());
+
+      // labelPath → label
+      hardlinksMade += link(tmp, type, "label", dto.getLabelPath());
+
+      // geoJsonPath -> label-json
+      hardlinksMade += link(tmp, type, "label-json", dto.getGeoJsonPath());
+    }
+
+    if (hardlinksMade == 0) {
+      throw new IOException("No hardlinks created.");
+    }
+
+    log.info("tmp dataset created: {}, hardlinksMade={}", tmp, hardlinksMade);
+  }
+
+  private long link(Path tmp, String type, String part, String fullPath) throws IOException {
+
+    if (fullPath == null || fullPath.isBlank()) return 0;
+
+    Path src = Path.of(fullPath);
+
+    if (!Files.isRegularFile(src)) {
+      log.warn("SKIP (not file): {}", src);
+      return 0;
+    }
+
+    String fileName = src.getFileName().toString();
+    Path dst = tmp.resolve(type).resolve(part).resolve(fileName);
+
+    // 충돌 시 덮어쓰기
+    if (Files.exists(dst)) {
+      Files.delete(dst);
+    }
+
+    Files.createLink(dst, src);
+
+    return 1;
+  }
+
+  private String safe(String s) {
+    return (s == null || s.isBlank()) ? null : s.trim();
+  }
+
+  /**
+   * request 전체 폴더 link
+   *
+   * @param uid
+   * @param datasetUids
+   * @return
+   * @throws IOException
+   */
+  public String buildTmpDatasetSymlink(String uid, List<String> datasetUids) throws IOException {
+
+    log.info("========== buildTmpDatasetHardlink START ==========");
+    log.info("uid={}", uid);
+    log.info("datasetUids={}", datasetUids);
+    log.info("requestDir(raw)={}", requestDir);

    Path BASE = toPath(requestDir);
    Path tmp = Path.of(trainBaseDir, "tmp", uid);

-    long hardlinksMade = 0;
-    long skippedCollision = 0;
-    long noDir = 0;
+    log.info("BASE={}", BASE);
+    log.info("BASE exists? {}", Files.isDirectory(BASE));
+    log.info("tmp={}", tmp);

-    // tmp/<type>/<part> 준비
+    long noDir = 0, scannedDirs = 0, regularFiles = 0, hardlinksMade = 0;
+
+    // tmp 디렉토리 준비
+    for (String type : List.of("train", "val", "test")) {
      for (String part : List.of("input1", "input2", "label", "label-json")) {
-      Files.createDirectories(tmp.resolve(type).resolve(part));
+        Path dir = tmp.resolve(type).resolve(part);
+        Files.createDirectories(dir);
+        log.info("createDirectories: {}", dir);
+      }
    }

-    for (ModelTrainLinkDto dto : links) {
-      String datasetUid = safe(dto.getDatasetUid());
-      if (datasetUid == null) {
-        log.warn("SKIP dto (datasetUid null): {}", dto);
-        continue;
+    // 하드링크는 "같은 파일시스템"에서만 가능하므로 BASE/tmp가 같은 FS인지 미리 확인(권장)
+    try {
+      var baseStore = Files.getFileStore(BASE);
+      var tmpStore = Files.getFileStore(tmp.getParent()); // BASE/tmp
+      if (!baseStore.name().equals(tmpStore.name()) || !baseStore.type().equals(tmpStore.type())) {
+        throw new IOException(
+            "Hardlink requires same filesystem. baseStore="
+                + baseStore.name()
+                + "("
+                + baseStore.type()
+                + "), tmpStore="
+                + tmpStore.name()
+                + "("
+                + tmpStore.type()
+                + ")");
+      }
+    } catch (Exception e) {
+      // FileStore 비교가 환경마다 애매할 수 있어서, 여기서는 경고만 주고 실제 createLink에서 최종 판단하게 둘 수도 있음.
+      log.warn("FileStore check skipped/failed (will rely on createLink): {}", e.toString());
    }

-      Path srcRoot = BASE.resolve(datasetUid);
+    for (String id : datasetUids) {
+      Path srcRoot = BASE.resolve(id);
+      log.info("---- dataset id={} srcRoot={} exists? {}", id, srcRoot, Files.isDirectory(srcRoot));

+      for (String type : List.of("train", "val", "test")) {
        for (String part : List.of("input1", "input2", "label", "label-json")) {

          Path srcDir = srcRoot.resolve(type).resolve(part);
          if (!Files.isDirectory(srcDir)) {
+            log.warn("SKIP (not directory): {}", srcDir);
            noDir++;
            continue;
          }

-        // 하위폴더까지 전부
-        try (var walk = Files.walk(srcDir)) {
-          for (Path f : walk.filter(Files::isRegularFile).toList()) {
+          scannedDirs++;
+          log.info("SCAN dir={}", srcDir);

-            String fileName = f.getFileName().toString();
-            Path dst = tmp.resolve(type).resolve(part).resolve(fileName);
-
-            // 이름 유지 + 충돌은 skip
-            if (Files.exists(dst)) {
-              skippedCollision++;
+          try (DirectoryStream<Path> stream = Files.newDirectoryStream(srcDir)) {
+            for (Path f : stream) {
+              if (!Files.isRegularFile(f)) {
+                log.debug("skip non-regular file: {}", f);
                continue;
              }

+              regularFiles++;
+
+              String dstName = id + "__" + f.getFileName();
+              Path dst = tmp.resolve(type).resolve(part).resolve(dstName);
+
+              // dst가 남아있으면 삭제(심볼릭링크든 파일이든)
+              if (Files.exists(dst) || Files.isSymbolicLink(dst)) {
+                Files.delete(dst);
+                log.debug("deleted existing: {}", dst);
+              }
+
+              try {
+                // 하드링크 생성 (dst가 새 파일로 생기지만 inode는 f와 동일)
                Files.createLink(dst, f);
                hardlinksMade++;
+                log.debug("created hardlink: {} => {}", dst, f);
+              } catch (IOException e) {
+                // 여기서 바로 실패시키면 “tmp는 만들었는데 내용은 0개” 같은 상태를 방지할 수 있음
+                log.error("FAILED create hardlink: {} => {}", dst, f, e);
+                throw e;
+              }
+            }
          }
        }
      }
@@ -92,35 +208,29 @@ public class TmpDatasetService {

    if (hardlinksMade == 0) {
      throw new IOException(
-          "No hardlinks created. noDir=" + noDir + ", skippedCollision=" + skippedCollision);
+          "No hardlinks created. regularFiles="
+              + regularFiles
+              + ", scannedDirs="
+              + scannedDirs
+              + ", noDir="
+              + noDir);
    }

+    log.info("tmp dataset created: {}", tmp);
    log.info(
-        "tmp dataset merged: {} (type={}), hardlinksMade={}, skippedCollision={}, noDir={}",
-        tmp,
-        type,
-        hardlinksMade,
-        skippedCollision,
-        noDir);
+        "summary: scannedDirs={}, noDir={}, regularFiles={}, hardlinksMade={}",
+        scannedDirs,
+        noDir,
+        regularFiles,
+        hardlinksMade);

    return uid;
  }

  private static Path toPath(String p) {
-    if (p == null || p.isBlank()) {
-      throw new IllegalArgumentException("path is null or blank");
+    if (p.startsWith("~/")) {
+      return Paths.get(System.getProperty("user.home")).resolve(p.substring(2)).normalize();
    }
-    String trimmed = p.trim();
-    if (trimmed.startsWith("~/")) {
-      return Paths.get(System.getProperty("user.home"))
-          .resolve(trimmed.substring(2))
-          .toAbsolutePath()
-          .normalize();
-    }
-    return Paths.get(trimmed).toAbsolutePath().normalize();
-  }
-
-  private static String safe(String s) {
-    return (s == null || s.isBlank()) ? null : s.trim();
+    return Paths.get(p).toAbsolutePath().normalize();
  }
 }
--- a/src/main/java/com/kamco/cd/training/train/service/TrainJobService.java
+++ b/src/main/java/com/kamco/cd/training/train/service/TrainJobService.java
@@ -266,21 +266,28 @@ public class TrainJobService {
    List<String> uids = modelTrainMngCoreService.findDatasetUid(datasetIds);

    try {
-      // train path
+      // 데이터셋 심볼링크 생성
+      //      String pathUid = tmpDatasetService.buildTmpDatasetSymlink(raw, uids);
+      // train path 모델 클래스별 조회
      List<ModelTrainLinkDto> trainList = modelTrainMngCoreService.findDatasetTrainPath(modelId);
-      // validation path
+      // validation path 모델 클래스별 조회
      List<ModelTrainLinkDto> valList = modelTrainMngCoreService.findDatasetValPath(modelId);
-      // test path
+      // test path 모델 클래스별 조회
      List<ModelTrainLinkDto> testList = modelTrainMngCoreService.findDatasetTestPath(modelId);

+      log.info(
+          "createTmpFile class list trainList = {} valList = {} testList = {}",
+          trainList.size(),
+          valList.size(),
+          testList.size());
+
      // train 데이터셋 심볼링크 생성
-      tmpDatasetService.buildTmpDatasetSymlink(raw, "train", trainList);
+      tmpDatasetService.buildTmpDatasetHardlink(raw, "train", trainList);
      // val 데이터셋 심볼링크 생성
-      tmpDatasetService.buildTmpDatasetSymlink(raw, "val", valList);
+      tmpDatasetService.buildTmpDatasetHardlink(raw, "val", valList);
      // test 데이터셋 심볼링크 생성
-      tmpDatasetService.buildTmpDatasetSymlink(raw, "test", testList);
-      // 카운트 로그
-      dataSetCounters.getCount(modelId);
+      tmpDatasetService.buildTmpDatasetHardlink(raw, "test", testList);
+
      ModelTrainMngDto.UpdateReq updateReq = new ModelTrainMngDto.UpdateReq();
      updateReq.setRequestPath(raw);

--- a/src/main/java/com/kamco/cd/training/train/service/TrainJobWorker.java
+++ b/src/main/java/com/kamco/cd/training/train/service/TrainJobWorker.java
@@ -108,6 +108,10 @@ public class TrainJobWorker {
        return;
      }

+      /**
+       * 0 정상 종료 SUCCESS 1~125 학습 코드 에러 FAILED 137 OOMKill FAILED 143 SIGTERM (stop) STOP -1 우리 내부
+       * 강제 중단 STOP
+       */
      if (result.getExitCode() == 0) {
        // 성공 처리
        modelTrainJobCoreService.markSuccess(jobId, result.getExitCode());
@@ -124,7 +128,23 @@ public class TrainJobWorker {
        }

      } else {
+
        String failMsg = result.getStatus() + "\n" + result.getLogs();
+        log.info("training fail exitCode={} Msg ={}", result.getExitCode(), failMsg);
+
+        if (result.getExitCode() == -1 || result.getExitCode() == 143) {
+          // 실패 처리
+          modelTrainJobCoreService.markPaused(
+              jobId, result.getExitCode(), result.getStatus() + "\n" + result.getLogs());
+
+          if (isEval) {
+            // 오류 정보 등록
+            modelTrainMngCoreService.markStep2Stop(modelId, "exit=" + result.getExitCode());
+          } else {
+            // 오류 정보 등록
+            modelTrainMngCoreService.markStep1Stop(modelId, "exit=" + result.getExitCode());
+          }
+        } else {
          // 실패 처리
          modelTrainJobCoreService.markFailed(
              jobId, result.getExitCode(), result.getStatus() + "\n" + result.getLogs());
@@ -137,6 +157,7 @@ public class TrainJobWorker {
            modelTrainMngCoreService.markError(modelId, "exit=" + result.getExitCode());
          }
        }
+      }

    } catch (Exception e) {
      modelTrainJobCoreService.markFailed(jobId, null, e.getMessage());