otterc commented on a change in pull request #30433:
URL: https://github.com/apache/spark/pull/30433#discussion_r535507691
##########
File path:
common/network-shuffle/src/main/java/org/apache/spark/network/shuffle/RemoteBlockPushResolver.java
##########
@@ -825,45 +881,79 @@ void resetChunkTracker() {
* @param mapIndex the map index to be added to chunk tracker.
*/
void updateChunkInfo(long chunkOffset, int mapIndex) throws IOException {
- long idxStartPos = -1;
try {
- // update the chunk tracker to meta file before index file
- writeChunkTracker(mapIndex);
- idxStartPos = indexFile.getFilePointer();
logger.trace("{} shuffleId {} reduceId {} updated index current {}
updated {}",
- appShuffleId.appId, appShuffleId.shuffleId, reduceId,
this.lastChunkOffset,
- chunkOffset);
- indexFile.writeLong(chunkOffset);
- } catch (IOException ioe) {
- if (idxStartPos != -1) {
- // reset the position to avoid corrupting index files during
exception.
- logger.warn("{} shuffleId {} reduceId {} reset index to position {}",
- appShuffleId.appId, appShuffleId.shuffleId, reduceId, idxStartPos);
- indexFile.seek(idxStartPos);
+ appShuffleId.appId, appShuffleId.shuffleId, reduceId,
this.lastChunkOffset, chunkOffset);
+ if (indexMetaUpdateFailed) {
+ indexFile.getChannel().position(indexFile.getPos());
}
+ indexFile.getDos().writeLong(chunkOffset);
+ // Chunk bitmap should be written to the meta file after the index
file because if there are
+ // any exceptions during writing the offset to the index file, meta
file should not be
+ // updated. If the update to the index file is successful but the
update to meta file isn't
+ // then the index file position is reset in the catch clause.
+ writeChunkTracker(mapIndex);
+ indexFile.updatePos(8);
+ this.lastChunkOffset = chunkOffset;
+ indexMetaUpdateFailed = false;
+ } catch (IOException ioe) {
+ logger.warn("{} shuffleId {} reduceId {} update to index/meta failed",
appShuffleId.appId,
+ appShuffleId.shuffleId, reduceId);
+ indexMetaUpdateFailed = true;
+ // Any exception here is propagated to the caller and the caller can
decide whether to
+ // abort or not.
throw ioe;
}
- this.lastChunkOffset = chunkOffset;
}
private void writeChunkTracker(int mapIndex) throws IOException {
if (mapIndex == -1) {
return;
}
chunkTracker.add(mapIndex);
- long metaStartPos = metaFile.getFilePointer();
try {
logger.trace("{} shuffleId {} reduceId {} mapIndex {} write chunk to
meta file",
appShuffleId.appId, appShuffleId.shuffleId, reduceId, mapIndex);
- chunkTracker.serialize(metaFile);
+ if (indexMetaUpdateFailed) {
+ metaFile.getChannel().position(metaFile.getPos());
+ }
+ int chunkTrackerSize = chunkTracker.serializedSizeInBytes();
+ chunkTracker.serialize(metaFile.getDos());
+ metaFile.updatePos(chunkTrackerSize);
} catch (IOException ioe) {
- logger.warn("{} shuffleId {} reduceId {} mapIndex {} reset position of
meta file to {}",
- appShuffleId.appId, appShuffleId.shuffleId, reduceId, mapIndex,
metaStartPos);
- metaFile.seek(metaStartPos);
+ logger.warn("{} shuffleId {} reduceId {} mapIndex {} update to meta
file failed",
+ appShuffleId.appId, appShuffleId.shuffleId, reduceId, mapIndex);
+ // Any exceptions while writing to meta file is propagated to
updateChunkInfo so that
+ // the index file position can be reset as well.
throw ioe;
}
}
+ private void incrementIOExceptions() {
+ numIOExceptions++;
+ }
+
+ private boolean shouldAbort(int ioExceptionsThresholdDuringMerge) {
+ return numIOExceptions > ioExceptionsThresholdDuringMerge;
+ }
+
+ private void finalizePartition() throws IOException {
+ if (dataFilePos != lastChunkOffset) {
+ try {
+ updateChunkInfo(dataFilePos, lastMergedMapIndex);
+ } catch (IOException ioe) {
+ // Any exceptions here while updating the meta files can be ignored.
If the files
+ // aren't successfully updated they will be truncated.
+ }
+ }
+ // Get rid of any partial block data at the end of the file. This could
either
+ // be due to failure, or a request still being processed when the shuffle
+ // merge gets finalized, or any exceptions while updating index/meta
files.
+ dataChannel.truncate(lastChunkOffset);
+ indexFile.getChannel().truncate(indexFile.getPos());
+ metaFile.getChannel().truncate(metaFile.getPos());
Review comment:
We are not ignoring the exceptions for truncate. If any of these
truncate fail, the finalize of this shuffle partition fails and the shuffle
partition is considered not to be merged. So clients will fetch original
un-merged data.
If we want, we can ignore exceptions during truncation of data file.
However, I didn't do that because the existing code doesn't ignore that.
----------------------------------------------------------------
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:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
