Merge pull request #1522 from typhoonzero/k8s_cluster_rerun

Fix k8s cluster job rerunable

doc/howto/usage/k8s/src/k8s_train/start_paddle.py

@@ -132,6 +132,7 @@ def startPaddle(idMap={}, train_args_dict=None):
     logDir = JOB_PATH_OUTPUT + "/node_" + str(trainerId)
     if not os.path.exists(JOB_PATH_OUTPUT):
         os.makedirs(JOB_PATH_OUTPUT)
+    if not os.path.exists(logDir):
         os.mkdir(logDir)
     copyCommand = 'cp -rf ' + JOB_PATH + \
         "/" + str(trainerId) + "/data/*" + " ./data/"