otterc commented on a change in pull request #30433: URL: https://github.com/apache/spark/pull/30433#discussion_r529823126
########## File path: common/network-shuffle/src/main/java/org/apache/spark/network/shuffle/RemoteBlockPushResolver.java ########## @@ -827,13 +833,16 @@ void resetChunkTracker() { 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); + indexFile.write(Longs.toByteArray(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); Review comment: Had a discussion with @Victsm @mridulm yesterday. This is the approach currently we are thinking of: 1. Verify if `seek` is not recoverable. If it is not then let the clients know to stop pushing and stop merging this partition. 2. Have a threshold on number of `IOExceptions` from writes and when this threshold is reached for a single partition, inform the client to stop pushing and stop merging this partition. 3. When the update to metadata fails, not propagate this exception back to client so that they push the block again. The size of the current chunk may grow but with (2) in place it will still be of a manageable size. These changes will not be that complex. ---------------------------------------------------------------- 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: us...@infra.apache.org --------------------------------------------------------------------- To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org For additional commands, e-mail: reviews-h...@spark.apache.org