Shuffle coalesce read supports split-and-retry #11598
Open
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Signed-off-by: Firestarman <[email protected]>
Signed-off-by: Firestarman <[email protected]>
Signed-off-by: Firestarman <[email protected]>
|
build |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
close #11584
This PR adds in the split-and-retry support when reading buffers from Shuffle for GpuShuffleCoalesceExec and sized hash joins.
When OOM happens, the retry will reduce the target size by half, re-collect buffers from the cache with total size up to the new target batch size, then concatenate the collected buffers and move the concatenated result to GPU. Since the concatenated buffer can not be split, we have to re-collect the buffers and concatenate them again with this new smaller target size.
The new target size will be used by all the following read in this task. The more split-and-retry it gets, the smaller the target size will be. This is just a choice. We prefer to believe that the GPU is busy with high memory pressure when getting an OOM. And more OOMs will possibly occur if using the original target batch size.
This PR also covers some code refactor in the shuffle hash join for the coming Kudo support. See #11590. It tries to move all the code pieces handling serialized tables into GpuShuffleCoalesceExec for easier code maintenance.