feat: Add Aggregate::addRawClusteredInput and streaming_aggregation_eager_flush #12975
Conversation
This file contains hidden or 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
facebook-github-bot
added
the
CLA Signed
|
This pull request was exported from Phabricator. Differential Revision: D72677410 |
✅ Deploy Preview for meta-velox canceled.
|
Yuhta
changed the title
|
@Yuhta Jimmy, thank you for the optimization. Would you update the PR description to share some findings about why this is a useful optimization? 3x improvements for a specific query shape common in data loading for AI training. |
mbasmanova
reviewed
Apr 9, 2025
mbasmanova
reviewed
Apr 9, 2025
mbasmanova
reviewed
Apr 9, 2025
mbasmanova
reviewed
Apr 9, 2025
mbasmanova
reviewed
Apr 9, 2025
mbasmanova
reviewed
Apr 10, 2025
mbasmanova
reviewed
Apr 10, 2025
mbasmanova
reviewed
Apr 10, 2025
mbasmanova
reviewed
Apr 10, 2025
mbasmanova
reviewed
Apr 10, 2025
mbasmanova
reviewed
Apr 10, 2025
Yuhta
added a commit
to Yuhta/velox
that referenced
this pull request
Apr 10, 2025
…ager_flush (facebookincubator#12975) Summary: Add fast path for streaming aggregation where we have input rows from same group located together. For certain functions, we can leverage this property to reduce the number of copy calls and create larger and fewer ranges for copy. This brings 3x improvements for a specific query shape common in data loading for AI training. We implement this optimization for `arbitrary` and `array_agg`. For `arbitrary`, if the input is clustered, we just keep a reference to the input vector and index that is selected; when we extract values from the container, we group all copies from same vector to one `copyRange` call so the cost is minimized. For `array_agg`, we do similar thing, only track the range (offset and size) where the input will be taken for each group, and do the copy in bulk when we extract value. There is another optimization to flush the streaming aggregation output whenever there is result available, via a new query config `streaming_aggregation_eager_flush`. This allows us to minimize the memory used by accumulators. Differential Revision: D72677410
|
This pull request was exported from Phabricator. Differential Revision: D72677410 |
Yuhta
added a commit
to Yuhta/velox
that referenced
this pull request
Apr 10, 2025
…ager_flush (facebookincubator#12975) Summary: Add fast path for streaming aggregation where we have input rows from same group located together. For certain functions, we can leverage this property to reduce the number of copy calls and create larger and fewer ranges for copy. This brings 3x improvements for a specific query shape common in data loading for AI training. We implement this optimization for `arbitrary` and `array_agg`. For `arbitrary`, if the input is clustered, we just keep a reference to the input vector and index that is selected; when we extract values from the container, we group all copies from same vector to one `copyRange` call so the cost is minimized. For `array_agg`, we do similar thing, only track the range (offset and size) where the input will be taken for each group, and do the copy in bulk when we extract value. There is another optimization to flush the streaming aggregation output whenever there is result available, via a new query config `streaming_aggregation_eager_flush`. This allows us to minimize the memory used by accumulators. Differential Revision: D72677410
Yuhta
added a commit
to Yuhta/velox
that referenced
this pull request
Apr 10, 2025
…ager_flush (facebookincubator#12975) Summary: Add fast path for streaming aggregation where we have input rows from same group located together. For certain functions, we can leverage this property to reduce the number of copy calls and create larger and fewer ranges for copy. This brings 3x improvements for a specific query shape common in data loading for AI training. We implement this optimization for `arbitrary` and `array_agg`. For `arbitrary`, if the input is clustered, we just keep a reference to the input vector and index that is selected; when we extract values from the container, we group all copies from same vector to one `copyRange` call so the cost is minimized. For `array_agg`, we do similar thing, only track the range (offset and size) where the input will be taken for each group, and do the copy in bulk when we extract value. There is another optimization to flush the streaming aggregation output whenever there is result available, via a new query config `streaming_aggregation_eager_flush`. This allows us to minimize the memory used by accumulators. Reviewed By: mbasmanova Differential Revision: D72677410
|
This pull request was exported from Phabricator. Differential Revision: D72677410 |
Yuhta
added a commit
to Yuhta/velox
that referenced
this pull request
Apr 10, 2025
…ager_flush (facebookincubator#12975) Summary: Add fast path for streaming aggregation where we have input rows from same group located together. For certain functions, we can leverage this property to reduce the number of copy calls and create larger and fewer ranges for copy. This brings 3x improvements for a specific query shape common in data loading for AI training. We implement this optimization for `arbitrary` and `array_agg`. For `arbitrary`, if the input is clustered, we just keep a reference to the input vector and index that is selected; when we extract values from the container, we group all copies from same vector to one `copyRange` call so the cost is minimized. For `array_agg`, we do similar thing, only track the range (offset and size) where the input will be taken for each group, and do the copy in bulk when we extract value. There is another optimization to flush the streaming aggregation output whenever there is result available, via a new query config `streaming_aggregation_eager_flush`. This allows us to minimize the memory used by accumulators. Reviewed By: mbasmanova Differential Revision: D72677410
|
This pull request was exported from Phabricator. Differential Revision: D72677410 |
Yuhta
added a commit
to Yuhta/velox
that referenced
this pull request
Apr 10, 2025
…ager_flush (facebookincubator#12975) Summary: Add fast path for streaming aggregation where we have input rows from same group located together. For certain functions, we can leverage this property to reduce the number of copy calls and create larger and fewer ranges for copy. This brings 3x improvements for a specific query shape common in data loading for AI training. We implement this optimization for `arbitrary` and `array_agg`. For `arbitrary`, if the input is clustered, we just keep a reference to the input vector and index that is selected; when we extract values from the container, we group all copies from same vector to one `copyRange` call so the cost is minimized. For `array_agg`, we do similar thing, only track the range (offset and size) where the input will be taken for each group, and do the copy in bulk when we extract value. There is another optimization to flush the streaming aggregation output whenever there is result available, via a new query config `streaming_aggregation_eager_flush`. This allows us to minimize the memory used by accumulators. Reviewed By: mbasmanova Differential Revision: D72677410
|
This pull request was exported from Phabricator. Differential Revision: D72677410 |
…ager_flush (facebookincubator#12975) Summary: Add fast path for streaming aggregation where we have input rows from same group located together. For certain functions, we can leverage this property to reduce the number of copy calls and create larger and fewer ranges for copy. This brings 3x improvements for a specific query shape common in data loading for AI training. We implement this optimization for `arbitrary` and `array_agg`. For `arbitrary`, if the input is clustered, we just keep a reference to the input vector and index that is selected; when we extract values from the container, we group all copies from same vector to one `copyRange` call so the cost is minimized. For `array_agg`, we do similar thing, only track the range (offset and size) where the input will be taken for each group, and do the copy in bulk when we extract value. There is another optimization to flush the streaming aggregation output whenever there is result available, via a new query config `streaming_aggregation_eager_flush`. This allows us to minimize the memory used by accumulators. Reviewed By: mbasmanova Differential Revision: D72677410
|
This pull request was exported from Phabricator. Differential Revision: D72677410 |
|
This pull request has been merged in afb236a. |
|
Rebase triggered for oap-project/velox. |
1 task
zhanglistar
pushed a commit
to bigo-sg/velox
that referenced
this pull request
Apr 22, 2025
…ager_flush (facebookincubator#12975) Summary: Pull Request resolved: facebookincubator#12975 Add fast path for streaming aggregation where we have input rows from same group located together. For certain functions, we can leverage this property to reduce the number of copy calls and create larger and fewer ranges for copy. This brings 3x improvements for a specific query shape common in data loading for AI training. We implement this optimization for `arbitrary` and `array_agg`. For `arbitrary`, if the input is clustered, we just keep a reference to the input vector and index that is selected; when we extract values from the container, we group all copies from same vector to one `copyRange` call so the cost is minimized. For `array_agg`, we do similar thing, only track the range (offset and size) where the input will be taken for each group, and do the copy in bulk when we extract value. There is another optimization to flush the streaming aggregation output whenever there is result available, via a new query config `streaming_aggregation_eager_flush`. This allows us to minimize the memory used by accumulators. Reviewed By: mbasmanova Differential Revision: D72677410 fbshipit-source-id: eedd664174b13784b47325eb2ab8274445470235
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:
Add fast path for streaming aggregation where we have input rows from same group located together. For certain functions, we can leverage this property to reduce the number of copy calls and create larger and fewer ranges for copy. This brings 3x improvements for a specific query shape common in data loading for AI training.
We implement this optimization for
arbitraryandarray_agg. Forarbitrary, if the input is clustered, we just keep a reference to the input vector and index that is selected; when we extract values from the container, we group all copies from same vector to onecopyRangecall so the cost is minimized. Forarray_agg, we do similar thing, only track the range (offset and size) where the input will be taken for each group, and do the copy in bulk when we extract value.There is another optimization to flush the streaming aggregation output whenever there is result available, via a new query config
streaming_aggregation_eager_flush. This allows us to minimize the memory used by accumulators.Differential Revision: D72677410