waitinfuture commented on code in PR #2064:
URL:
https://github.com/apache/incubator-celeborn/pull/2064#discussion_r1389113307
##########
worker/src/main/scala/org/apache/celeborn/service/deploy/worker/PushDataHandler.scala:
##########
@@ -1246,6 +1204,62 @@ class PushDataHandler(val workerSource: WorkerSource)
extends BaseMessageHandler
}
}
+ private def writeLocalData(
+ fileWriters: Seq[FileWriter],
+ body: ByteBuf,
+ shuffleKey: String,
+ isPrimary: Boolean,
+ batchOffsets: Option[Array[Int]],
+ writePromise: Promise[Unit]): Unit = {
+ def writeData(fileWriter: FileWriter, body: ByteBuf, shuffleKey: String):
Unit = {
+ try {
+ fileWriter.write(body)
+ } catch {
+ case e: Exception =>
+ if (e.isInstanceOf[AlreadyClosedException]) {
+ val (mapId, attemptId) = getMapAttempt(body)
+ val endedAttempt =
+ if (shuffleMapperAttempts.containsKey(shuffleKey)) {
+ shuffleMapperAttempts.get(shuffleKey).get(mapId)
+ } else -1
+ // TODO just info log for ended attempt
+ logWarning(s"Append data failed for task(shuffle $shuffleKey, map
$mapId, attempt" +
+ s" $attemptId), caused by AlreadyClosedException, endedAttempt
$endedAttempt, error message: ${e.getMessage}")
+ } else {
+ logError("Exception encountered when write.", e)
+ }
+ if (!writePromise.isCompleted) {
+ val cause =
+ if (isPrimary) {
+ StatusCode.PUSH_DATA_WRITE_FAIL_PRIMARY
+ } else {
+ StatusCode.PUSH_DATA_WRITE_FAIL_REPLICA
+ }
+ writePromise.failure(new CelebornIOException(cause))
+ }
+ fileWriter.decrementPendingWrites()
+ }
+ }
+ batchOffsets match {
+ case Some(batchOffsets) =>
+ batchOffsets.zip(fileWriters).foreach { case (offset, writer) =>
+ val length =
+ if (offset == batchOffsets.last) {
+ body.readableBytes() - offset
+ } else {
+ batchOffsets(batchOffsets.indexOf(offset) + 1) - offset
Review Comment:
I think the original `while` has better performance because it avoids
calling `batchOffsets.indexOf (which is O(n) complexity)`, see
https://github.com/apache/incubator-celeborn/pull/2086#issuecomment-1804944666
##########
worker/src/main/scala/org/apache/celeborn/service/deploy/worker/PushDataHandler.scala:
##########
@@ -1246,6 +1204,62 @@ class PushDataHandler(val workerSource: WorkerSource)
extends BaseMessageHandler
}
}
+ private def writeLocalData(
+ fileWriters: Seq[FileWriter],
+ body: ByteBuf,
+ shuffleKey: String,
+ isPrimary: Boolean,
+ batchOffsets: Option[Array[Int]],
+ writePromise: Promise[Unit]): Unit = {
+ def writeData(fileWriter: FileWriter, body: ByteBuf, shuffleKey: String):
Unit = {
+ try {
+ fileWriter.write(body)
+ } catch {
+ case e: Exception =>
+ if (e.isInstanceOf[AlreadyClosedException]) {
+ val (mapId, attemptId) = getMapAttempt(body)
+ val endedAttempt =
+ if (shuffleMapperAttempts.containsKey(shuffleKey)) {
+ shuffleMapperAttempts.get(shuffleKey).get(mapId)
+ } else -1
+ // TODO just info log for ended attempt
+ logWarning(s"Append data failed for task(shuffle $shuffleKey, map
$mapId, attempt" +
+ s" $attemptId), caused by AlreadyClosedException, endedAttempt
$endedAttempt, error message: ${e.getMessage}")
+ } else {
+ logError("Exception encountered when write.", e)
+ }
+ if (!writePromise.isCompleted) {
Review Comment:
I think we need to move this check inside the loop over batchOffsets to
avoid calling `writeData` for successive partitions if
`writePromise.isCompleted`
--
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.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
