ajithme edited a comment on issue #24142: [SPARK-27194][core] Job failures when task attempts do not clean up spark-staging parquet files URL: https://github.com/apache/spark/pull/24142#issuecomment-474866759 So as we can see from stacktrace, when we have ``spark.sql.sources.partitionOverwriteMode=DYNAMIC`` and we have overwrite enabled @ ``InsertIntoHadoopFsRelationCommand``, ``` val enableDynamicOverwrite = sparkSession.sessionState.conf.partitionOverwriteMode == PartitionOverwriteMode.DYNAMIC // This config only makes sense when we are overwriting a partitioned dataset with dynamic // partition columns. val dynamicPartitionOverwrite = enableDynamicOverwrite && mode == SaveMode.Overwrite && staticPartitions.size < partitionColumns.length ``` we directly try to write to final location via ``DynamicPartitionWriteTask`` which fails as it tries to create the file because of left over file from previous task. i.e because ``org.apache.spark.sql.execution.datasources.parquet.ParquetOutputWriter#recordWriter`` always calls ``org.apache.parquet.hadoop.ParquetOutputFormat#getRecordWriter(Configuration, Path, org.apache.parquet.hadoop.metadata.CompressionCodecName)`` which always has ``org.apache.parquet.hadoop.ParquetFileWriter`` with ``org.apache.parquet.hadoop.ParquetFileWriter.Mode#CREATE`` So for the retry task, ``getRecordWriter`` will never succeed
---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected] With regards, Apache Git Services --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
