HeartSaVioR commented on a change in pull request #22952: [SPARK-20568][SS]
Provide option to clean up completed files in streaming query
URL: https://github.com/apache/spark/pull/22952#discussion_r258234889
##########
File path:
sql/core/src/main/scala/org/apache/spark/sql/execution/streaming/FileStreamSource.scala
##########
@@ -257,16 +261,33 @@ class FileStreamSource(
* equal to `end` and will only request offsets greater than `end` in the
future.
*/
override def commit(end: Offset): Unit = {
- // No-op for now; FileStreamSource currently garbage-collects files based
on timestamp
- // and the value of the maxFileAge parameter.
+ val logOffset = FileStreamSourceOffset(end).logOffset
+
+ if (sourceOptions.cleanSource != CleanSourceMode.NO_OP) {
+ val files = metadataLog.get(Some(logOffset),
Some(logOffset)).flatMap(_._2)
+ val validFileEntities = files.filter(_.batchId == logOffset)
+ logDebug(s"completed file entries: ${validFileEntities.mkString(",")}")
+ sourceOptions.cleanSource match {
+ case CleanSourceMode.ARCHIVE =>
+ validFileEntities.foreach(sourceCleaner.archive)
+
+ case CleanSourceMode.DELETE =>
+ validFileEntities.foreach(sourceCleaner.delete)
+
+ case _ =>
+ }
+ } else {
+ // No-op for now; FileStreamSource currently garbage-collects files
based on timestamp
+ // and the value of the maxFileAge parameter.
+ }
+
Review comment:
Maybe we can also purge metadata here to prevent growing infinitely given
that Spark guarantees it will not happen to request offset equal or less than
`logOffset` (I guess it would also apply to `getBatch`).
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on 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]
