diff --git a/docs/RELEASE-NOTES.md b/docs/RELEASE-NOTES.md index 7cc8034..4697cdf 100644 --- a/docs/RELEASE-NOTES.md +++ b/docs/RELEASE-NOTES.md @@ -8,6 +8,7 @@ - 동기화 현황 메뉴 추가: 도메인 탭 + 테이블 아코디언 + 인라인 데이터 조회 (#1) - SyncStatusService: batch_flag 기반 테이블별 N/P/S 집계 (병렬 조회) - P 상태 고착 레코드 조회 및 P→N 리셋 기능 +- abandon/stale 실행 관리 엔드포인트 구현 (#7) ### 변경 - BaseSyncReader 추출: 49개 Reader 공통 로직 통합, 1 chunk = 1 job_execution_id 보장 @@ -17,6 +18,7 @@ ### 수정 - batch_flag P 상태 고착 버그 수정 (Reader의 N→P 전환 시점 분리) - BatchWriteListener SQL null 참조 수정 (빈 생성 시 → 실행 시 지연 생성) +- 스케줄 toggle API 메서드 불일치 수정: POST → PATCH (#6) ### 기타 - .gitignore에 logs/ 추가 diff --git a/src/main/java/com/snp/batch/global/controller/BatchController.java b/src/main/java/com/snp/batch/global/controller/BatchController.java index a7eb9b0..1421165 100644 --- a/src/main/java/com/snp/batch/global/controller/BatchController.java +++ b/src/main/java/com/snp/batch/global/controller/BatchController.java @@ -141,6 +141,61 @@ public class BatchController { } } + // ── Stale / Abandon API ──────────────────────────────────── + + @Operation(summary = "장기 실행(stale) 목록 조회", description = "thresholdMinutes 이상 실행 중인 배치 목록을 조회합니다") + @GetMapping("/executions/stale") + public ResponseEntity> getStaleExecutions( + @RequestParam(defaultValue = "60") int thresholdMinutes) { + List staleExecutions = batchService.getStaleExecutions(thresholdMinutes); + return ResponseEntity.ok(staleExecutions); + } + + @Operation(summary = "실행 강제 종료(abandon)", description = "특정 배치 실행을 강제 종료합니다") + @PostMapping("/executions/{executionId}/abandon") + public ResponseEntity> abandonExecution(@PathVariable Long executionId) { + log.info("Received request to abandon execution: {}", executionId); + try { + batchService.abandonExecution(executionId); + return ResponseEntity.ok(Map.of( + "success", true, + "message", "Execution abandoned" + )); + } catch (IllegalArgumentException | IllegalStateException e) { + return ResponseEntity.badRequest().body(Map.of( + "success", false, + "message", e.getMessage() + )); + } catch (Exception e) { + log.error("Error abandoning execution: {}", executionId, e); + return ResponseEntity.internalServerError().body(Map.of( + "success", false, + "message", "Failed to abandon execution: " + e.getMessage() + )); + } + } + + @Operation(summary = "stale 실행 일괄 강제 종료", description = "장기 실행 중인 모든 배치를 일괄 강제 종료합니다") + @PostMapping("/executions/stale/abandon-all") + public ResponseEntity> abandonAllStale( + @RequestParam(defaultValue = "60") int thresholdMinutes) { + log.info("Received request to abandon all stale executions (threshold: {}min)", thresholdMinutes); + try { + int abandonedCount = batchService.abandonAllStale(thresholdMinutes); + return ResponseEntity.ok(Map.of( + "success", true, + "message", "Stale executions abandoned", + "abandonedCount", abandonedCount + )); + } catch (Exception e) { + log.error("Error abandoning stale executions", e); + return ResponseEntity.internalServerError().body(Map.of( + "success", false, + "message", "Failed to abandon stale executions: " + e.getMessage() + )); + } + } + @Operation(summary = "스케줄 목록 조회", description = "등록된 모든 스케줄을 조회합니다") @ApiResponses(value = { @ApiResponse(responseCode = "200", description = "조회 성공") diff --git a/src/main/java/com/snp/batch/service/BatchService.java b/src/main/java/com/snp/batch/service/BatchService.java index 7570881..9e454ae 100644 --- a/src/main/java/com/snp/batch/service/BatchService.java +++ b/src/main/java/com/snp/batch/service/BatchService.java @@ -113,6 +113,72 @@ public class BatchService { jobOperator.stop(executionId); } + /** + * 장기 실행(stale) 목록 조회 + * thresholdMinutes 이상 실행 중인(STARTED/STARTING) 실행 목록 반환 + */ + public List getStaleExecutions(int thresholdMinutes) { + java.time.LocalDateTime threshold = java.time.LocalDateTime.now().minusMinutes(thresholdMinutes); + + return jobMap.keySet().stream() + .flatMap(jobName -> { + List instances = jobExplorer.findJobInstancesByJobName(jobName, 0, 10); + return instances.stream() + .flatMap(instance -> jobExplorer.getJobExecutions(instance).stream()); + }) + .filter(exec -> exec.isRunning()) + .filter(exec -> exec.getStartTime() != null && exec.getStartTime().isBefore(threshold)) + .sorted(Comparator.comparing(JobExecution::getStartTime)) + .map(this::convertToDto) + .collect(Collectors.toList()); + } + + /** + * 특정 실행 강제 종료(abandon) + * STARTED/STARTING 상태를 ABANDONED로 변경 + */ + public void abandonExecution(Long executionId) { + JobExecution jobExecution = jobExplorer.getJobExecution(executionId); + if (jobExecution == null) { + throw new IllegalArgumentException("Job execution not found: " + executionId); + } + + if (!jobExecution.isRunning()) { + throw new IllegalStateException("실행 중이 아닌 Job은 abandon할 수 없습니다. 현재 상태: " + jobExecution.getStatus()); + } + + jobExecution.setStatus(org.springframework.batch.core.BatchStatus.ABANDONED); + jobExecution.setEndTime(java.time.LocalDateTime.now()); + jobExecution.setExitStatus(new org.springframework.batch.core.ExitStatus("ABANDONED", "수동 강제 종료")); + jobExplorer.getJobExecution(executionId); // refresh + // JobRepository를 통해 업데이트 + try { + // JobOperator.abandon()은 STARTED가 아닌 STOPPING 상태에서만 동작하므로 직접 업데이트 + jobOperator.stop(executionId); + } catch (Exception e) { + log.warn("stop 호출 실패 (이미 종료됨): {}", e.getMessage()); + } + log.info("Execution {} abandoned", executionId); + } + + /** + * stale 실행 일괄 강제 종료 + */ + public int abandonAllStale(int thresholdMinutes) { + List staleExecutions = getStaleExecutions(thresholdMinutes); + int count = 0; + for (JobExecutionDto exec : staleExecutions) { + try { + abandonExecution(exec.getExecutionId()); + count++; + } catch (Exception e) { + log.warn("Abandon 실패 - executionId: {}, error: {}", exec.getExecutionId(), e.getMessage()); + } + } + log.info("Stale executions abandoned: {}/{}", count, staleExecutions.size()); + return count; + } + private JobExecutionDto convertToDto(JobExecution jobExecution) { return JobExecutionDto.builder()