Skip to content

[SPARK-20065][SS][WIP] Avoid to output empty parquet files #17395

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

Closed
uncleGen wants to merge 3 commits into from
+14 −8
Closed

[SPARK-20065][SS][WIP] Avoid to output empty parquet files #17395

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
86a7d2f
avoid to output empty parquet files
uncleGen Mar 23, 2017
599218b
Merge branch 'master' into SPARK-20065
uncleGen Mar 23, 2017
42da5af
fix the unit test failure
uncleGen Mar 23, 2017
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
5 changes: 4 additions & 1 deletion 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 @@ -292,7 +292,10 @@ object FileFormatWriter extends Logging {
override def execute(iter: Iterator[InternalRow]): Set[String] = {
var fileCounter = 0
var recordsInFile: Long = 0L
newOutputWriter(fileCounter)
// Skip the empty partition to avoid creating a mass of 'empty' files.
if (iter.hasNext) {
newOutputWriter(fileCounter)

@HyukjinKwon HyukjinKwon Mar 23, 2017 •
edited
Loading

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

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

I proposed the similar PR before (about a year ago?) but got reverted. In this case, Parquet would not write out the footer and schema information. Namely, this will break the case below:

spark.range(100).filter("id > 100").write.parquet("/tmp/abc")
spark.read.parquet("/tmp/abc").show()

Up to my knowledge, we don't have test cases for them if I haven't missed related PRs it seems now there is.

@uncleGen uncleGen Mar 23, 2017

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

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

@HyukjinKwon IIUC, this case should fail as expected, as there is no output. Am i missing something?

spark.range(100).filter("id > 100").write.parquet("/tmp/abc")
spark.read.parquet("/tmp/abc").show()

@HyukjinKwon HyukjinKwon Mar 23, 2017 •
edited
Loading

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

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

Reading empty data should be fine too. It should preserve the schema. I am pretty sure that we want this case because mine was reverted due to the case above.

@HyukjinKwon HyukjinKwon Mar 23, 2017

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

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

See #12855, https://issues.apache.org/jira/browse/SPARK-10216 and https://issues.apache.org/jira/browse/SPARK-15393

@uncleGen uncleGen Mar 24, 2017

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

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

Thanks for your prompt. How about just left one empty file containing the metadata when df has empty partition? Furthmore, we may just left one metadata file?

@HyukjinKwon HyukjinKwon Mar 24, 2017 •
edited
Loading

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

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

Yes, I was thinking in that way. I remember I did several tries at that time but failed to make a confident fix, and could not have some time to work on that further.

Another problem is, it might be related with a datasource-specific issue because, for example, ORC does not write out empty df. For example,

scala> spark.range(100).filter("id > 100").write.orc("/tmp/abc1")

scala> spark.read.orc("/tmp/abc1").show()
org.apache.spark.sql.AnalysisException: Unable to infer schema for ORC. It must be specified manually.;
  at org.apache.spark.sql.execution.datasources.DataSource$$anonfun$8.apply(DataSource.scala:182)
  at org.apache.spark.sql.execution.datasources.DataSource$$anonfun$8.apply(DataSource.scala:182)

This issue is described in https://issues.apache.org/jira/browse/SPARK-15474.

FWIW, I happened to see https://issues.apache.org/jira/browse/SPARK-15693 around that time and I kind of felt we may be able to consolidate this issue with it although it is a rough idea.

}
while (iter.hasNext) {
if (description.maxRecordsPerFile > 0 && recordsInFile >= description.maxRecordsPerFile) {
fileCounter += 1
Expand Down
8 changes: 4 additions & 4 deletions ...src/main/scala/org/apache/spark/sql/execution/datasources/parquet/ParquetFileFormat.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -140,10 +140,10 @@ class ParquetFileFormat
// initialized.
private val parquetLogRedirector = ParquetLogRedirector.INSTANCE

override def newInstance(
path: String,
dataSchema: StructType,
context: TaskAttemptContext): OutputWriter = {
override def newInstance(
path: String,
dataSchema: StructType,
context: TaskAttemptContext): OutputWriter = {
new ParquetOutputWriter(path, context)
}

Expand Down