在spark2.3的hdfs中编写分区表时遇到了一些问题。我认为问题来自partitionoverwritemode配置。有可能吗？保持覆盖模式如何解决？

spark.conf.set("spark.sql.sources.partitionOverwriteMode", "dynamic")
 df \
            .repartition(3) \
            .write \
            .format('orc') \
            .partitionBy(['data_date_part']) \
            .mode("overwrite") \
            .option("compression",'zlib') \
            .option("path", table_path+ '/' + table_name) \
            .saveAsTable(scheme + '.' + table_name)

错误：

Caused by: org.apache.spark.SparkException: Job aborted due to stage failure: Task 8 in stage 3.0 failed 4 times, most recent failure: Lost task 8.3 in stage 3.0 Task failed while writing rows.
Caused by: org.apache.hadoop.ipc.RemoteException(org.apache.hadoop.fs.FileAlreadyExistsException): file/ for client already exists