[Gluten-1.2] Port #10652 to Branch-1.2 for Let HashProbe keep track of memory consumption when listing join results (#10652) #495
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
…lts (facebookincubator#10652) Summary: Hash probe currently has limited memory control when extracting results from the hash table. When a small number of large sized rows from the build side is frequently joined with the left side, the total extracted size will explode, making HashProbe using a large amount of memory. And the process of filling output is not in spillable state, and will often cause OOM. This PR computes the total size when listing join results in hash probe if there are any variable size columns from the build side that is going to be extracted. It stops listing further when it reaches the maximum size. This can help to control hash probe side memory usage to a confined limit. Pull Request resolved: facebookincubator#10652 Reviewed By: xiaoxmeng Differential Revision: D60771773 Pulled By: tanjialiang fbshipit-source-id: 2cb8c58ba795a0aa1df0485b58e4f6d0100be8f8 (cherry picked from commit 82e5492)
|
@zhouyuan Can you help review this? |
|
The patch causes performance regression in newest Gluten. Are you already aware of that? @kecookier |
zhztheplayer
approved these changes
Aug 27, 2024
|
@zhztheplayer Thanks, I'm not clear about this situation yet. Let us hold on for now and consider merging after we have resolve it. cc @zsmj2017 |
|
The PR has run pass 333 velox tests with fail 1 test related to ParquetTableScanTest.timestampFilter test report empty parquet file. |
weiting-chen
approved these changes
Sep 18, 2024
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Outdated suggestions cannot be applied.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.
Summary:
Hash probe currently has limited memory control when extracting results from the hash table. When a small number of large sized rows from the build side is frequently joined with the left side, the total extracted size will explode, making HashProbe using a large amount of memory. And the process of filling output is not in spillable state, and will often cause OOM. This PR computes the total size when listing join results in hash probe if there are any variable size columns from the build side that is going to be extracted. It stops listing further when it reaches the maximum size. This can help to control hash probe side memory usage to a confined limit.
Pull Request resolved: facebookincubator#10652
Reviewed By: xiaoxmeng
Differential Revision: D60771773
Pulled By: tanjialiang
fbshipit-source-id: 2cb8c58ba795a0aa1df0485b58e4f6d0100be8f8 (cherry picked from commit 82e5492)