fix worker timeout (#368)

miotech · Jan 13, 2021 · 9c28e15 · 9c28e15
1 parent 27b1dd4
commit 9c28e15
Showing 1 changed file with 3 additions and 3 deletions.
diff --git a/...orkflow-executor/src/main/java/com/miotech/kun/workflow/executor/local/LocalExecutor.java b/...orkflow-executor/src/main/java/com/miotech/kun/workflow/executor/local/LocalExecutor.java
@@ -384,9 +384,8 @@ public void run() {
     }
 
     private void handleTimeoutAttempt(Long taskAttemptId) {
-        workerToken.release();
-        logger.debug("taskAttemptId = {} release worker token, current size = {}", taskAttemptId, workerToken.availablePermits());
-        workerPool.remove(taskAttemptId);
+        //kill worker when worker time out
+        killTaskAttempt(taskAttemptId);
         miscService.changeTaskAttemptStatus(taskAttemptId, TaskRunStatus.ERROR);
         TaskAttempt taskAttempt = taskRunDao.fetchAttemptById(taskAttemptId).get();
         submit(taskAttempt, true);
@@ -414,6 +413,7 @@ public void run() {
                 if (worker.shutdown()) {
                     workerPool.remove(taskAttemptId);
                     workerToken.release();
+                    logger.debug("taskAttemptId = {} release worker token, current size = {}", taskAttemptId, workerToken.availablePermits());
                     notifyFinished(taskAttemptId, TaskRunStatus.ABORTED, OperatorReport.BLANK);
                     miscService.changeTaskAttemptStatus(taskAttemptId,
                             TaskRunStatus.ABORTED, null, DateTimeUtils.now());