turboFei commented on a change in pull request #26339: [SPARK-27194][SPARK-29302][SQL] Fix the issue that for dynamic partition overwrite a task would conflict with its speculative task URL: https://github.com/apache/spark/pull/26339#discussion_r407837974
########## File path: core/src/main/scala/org/apache/spark/internal/io/HadoopMapReduceCommitProtocol.scala ########## @@ -236,13 +267,25 @@ class HadoopMapReduceCommitProtocol( committer.setupTask(taskContext) addedAbsPathFiles = mutable.Map[String, String]() partitionPaths = mutable.Set[String]() + dynamicStagingTaskFiles = mutable.Set[Path]() } override def commitTask(taskContext: TaskAttemptContext): TaskCommitMessage = { val attemptId = taskContext.getTaskAttemptID logTrace(s"Commit task ${attemptId}") SparkHadoopMapRedUtil.commitTask( committer, taskContext, attemptId.getJobID.getId, attemptId.getTaskID.getId) + if (dynamicPartitionOverwrite) { + val fs = stagingDir.getFileSystem(taskContext.getConfiguration) + dynamicStagingTaskFiles.foreach { stagingTaskFile => + val fileName = stagingTaskFile.getName + val partitionPath = getDynamicPartitionPath(stagingTaskFile, taskContext) + val finalFile = new Path(partitionPath, fileName) + if (!fs.exists(finalFile) && !fs.rename(stagingTaskFile, finalFile)) { Review comment: Thanks, I just check the code. It seems that the `SparkHadoopMapRedUtil.commitTask` has no meaning for dynamic partition overwrite, because dynamicPartitionOverwrite operation has a specific staging dir (`.spark-staging-jobId`) instead of using the working temporary of FileOutputCommitter(`_temporary/0`). So, the conflict issue is not related with the enable of outputCommitCoordination. > Another question is if it's possible the existing finalFile is corrupted due to another task crashing during commit. It is possible, because there may be several dynamicStagingTaskFiles,if the task aborted when commitTask, partial outputs are moved to the partition paths. ---------------------------------------------------------------- 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: us...@infra.apache.org With regards, Apache Git Services --------------------------------------------------------------------- To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org For additional commands, e-mail: reviews-h...@spark.apache.org