Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

[SPARK-41448] Make consistent MR job IDs in FileBatchWriter and FileFormatWriter #38980

Closed
wants to merge 1 commit into from
Closed

[SPARK-41448] Make consistent MR job IDs in FileBatchWriter and FileFormatWriter #38980

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
15 changes: 13 additions & 2 deletions core/src/main/scala/org/apache/spark/internal/io/SparkHadoopWriterUtils.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -47,11 +47,22 @@ object SparkHadoopWriterUtils {
* @return a job ID
*/
def createJobID(time: Date, id: Int): JobID = {
val jobTrackerID = createJobTrackerID(time)
createJobID(jobTrackerID, id)
}

/**
* Create a job ID.
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

how about extending the comment here by noting that the job id needs to be unique across all jobs (linking to SPARK-33402) and consistently across places used (SPARK-26873 + SPARK-41448). That way, whoever next goes near the code knows what is needed

*
* @param jobTrackerID unique job track id
* @param id job number
* @return a job ID
*/
def createJobID(jobTrackerID: String, id: Int): JobID = {
if (id < 0) {
throw new IllegalArgumentException("Job number is negative")
}
val jobtrackerID = createJobTrackerID(time)
new JobID(jobtrackerID, id)
new JobID(jobTrackerID, id)
}

/**
Expand Down
8 changes: 4 additions & 4 deletions sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/FileFormatWriter.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -201,14 +201,14 @@ object FileFormatWriter extends Logging {
rdd
}

val jobIdInstant = new Date().getTime
val jobTrackerID = SparkHadoopWriterUtils.createJobTrackerID(new Date())
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

i must have missed this -but also we've not had any reports of the timstamp clash surfacing.

that's probably because the s3a and abfs/gcs committers all pick up the uuid in "spark.sql.sources.writeJobUUID" in preference to anything else, and they are being generated uniquely

val ret = new Array[WriteTaskResult](rddWithNonEmptyPartitions.partitions.length)
sparkSession.sparkContext.runJob(
rddWithNonEmptyPartitions,
(taskContext: TaskContext, iter: Iterator[InternalRow]) => {
executeTask(
description = description,
jobIdInstant = jobIdInstant,
jobTrackerID = jobTrackerID,
sparkStageId = taskContext.stageId(),
sparkPartitionId = taskContext.partitionId(),
sparkAttemptNumber = taskContext.taskAttemptId().toInt & Integer.MAX_VALUE,
Expand Down Expand Up @@ -244,15 +244,15 @@ object FileFormatWriter extends Logging {
/** Writes data out in a single Spark task. */
private def executeTask(
description: WriteJobDescription,
jobIdInstant: Long,
jobTrackerID: String,
sparkStageId: Int,
sparkPartitionId: Int,
sparkAttemptNumber: Int,
committer: FileCommitProtocol,
iterator: Iterator[InternalRow],
concurrentOutputWriterSpec: Option[ConcurrentOutputWriterSpec]): WriteTaskResult = {

val jobId = SparkHadoopWriterUtils.createJobID(new Date(jobIdInstant), sparkStageId)
val jobId = SparkHadoopWriterUtils.createJobID(jobTrackerID, sparkStageId)
val taskId = new TaskID(jobId, TaskType.MAP, sparkPartitionId)
val taskAttemptId = new TaskAttemptID(taskId, sparkAttemptNumber)

Expand Down
4 changes: 3 additions & 1 deletion ...core/src/main/scala/org/apache/spark/sql/execution/datasources/v2/FileWriterFactory.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -29,6 +29,9 @@ import org.apache.spark.sql.execution.datasources.{DynamicPartitionDataSingleWri
case class FileWriterFactory (
description: WriteJobDescription,
committer: FileCommitProtocol) extends DataWriterFactory {

private val jobId = SparkHadoopWriterUtils.createJobID(new Date, 0)
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@boneanxs It looks like tha jobId is not serializable?

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

it does implement WritableComparable; you could use WritableConverter to do the marshalling if you don't just want to do via a string


Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I wonder if a uuid should be created here which is then passed down in that "spark.sql.sources.writeJobUUID" option in createTaskAttemptContext(). It would be consistent with the rest and the mapreduce manifest committer would pick it up. (so would the s3a one, but as you can't do dynamic partitioning there it's less relevant)

override def createWriter(partitionId: Int, realTaskId: Long): DataWriter[InternalRow] = {
val taskAttemptContext = createTaskAttemptContext(partitionId)
committer.setupTask(taskAttemptContext)
Expand All @@ -40,7 +43,6 @@ case class FileWriterFactory (
}

private def createTaskAttemptContext(partitionId: Int): TaskAttemptContextImpl = {
val jobId = SparkHadoopWriterUtils.createJobID(new Date, 0)
val taskId = new TaskID(jobId, TaskType.MAP, partitionId)
val taskAttemptId = new TaskAttemptID(taskId, 0)
// Set up the configuration object
Expand Down