feat/training_260202 #97
@@ -57,8 +57,7 @@ public class ModelTrainJobEntity {
|
|||||||
@Column(name = "exit_code")
|
@Column(name = "exit_code")
|
||||||
private Integer exitCode;
|
private Integer exitCode;
|
||||||
|
|
||||||
@Size(max = 2000)
|
@Column(name = "error_message", columnDefinition = "TEXT")
|
||||||
@Column(name = "error_message", length = 2000)
|
|
||||||
private String errorMessage;
|
private String errorMessage;
|
||||||
|
|
||||||
@ColumnDefault("now()")
|
@ColumnDefault("now()")
|
||||||
|
|||||||
@@ -60,9 +60,9 @@ public class TrainJobWorker {
|
|||||||
totalEpoch = Integer.parseInt(params.get("totalEpoch").toString());
|
totalEpoch = Integer.parseInt(params.get("totalEpoch").toString());
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
log.info("[JOB] markRunning start jobId={}, containerName={}", jobId, containerName);
|
||||||
modelTrainJobCoreService.markRunning(jobId, containerName, null, "TRAIN_WORKER", totalEpoch);
|
modelTrainJobCoreService.markRunning(jobId, containerName, null, "TRAIN_WORKER", totalEpoch);
|
||||||
|
log.info("[JOB] markRunning done jobId={}", jobId);
|
||||||
try {
|
try {
|
||||||
TrainRunResult result;
|
TrainRunResult result;
|
||||||
|
|
||||||
@@ -109,6 +109,7 @@ public class TrainJobWorker {
|
|||||||
}
|
}
|
||||||
|
|
||||||
} else {
|
} else {
|
||||||
|
String failMsg = result.getStatus() + "\n" + result.getLogs();
|
||||||
modelTrainJobCoreService.markFailed(
|
modelTrainJobCoreService.markFailed(
|
||||||
jobId, result.getExitCode(), result.getStatus() + "\n" + result.getLogs());
|
jobId, result.getExitCode(), result.getStatus() + "\n" + result.getLogs());
|
||||||
|
|
||||||
|
|||||||
Reference in New Issue
Block a user