[Spark] Avoid non-deterministic UDF to filter deleted rows #2576
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
This was referenced Apr 23, 2024
vkorukanti
pushed a commit
that referenced
this pull request
Apr 26, 2024
## Description Currently, when Deletion Vectors are enabled we disable predicate pushdown and splitting in scans. This is because we rely on a custom row index column which is constructed in the executors and cannot not handle splits and predicates. These restrictions can now be lifted by relying instead on `metadata.row_index` which was exposed recently after relevant [work](https://issues.apache.org/jira/browse/SPARK-37980) was concluded. Overall, this PR adds predicate pushdown and splits support as follows: 1. Replaces `__delta_internal_is_row_deleted` with `_metadata.row_index`. 2. Adds a new implementation of `__delta_internal_is_row_deleted` that is based on `_metadata.row_index`. 3. `IsRowDeleted` filter is now non deterministic to allow predicate pushdown. Furthermore, it includes previous relevant [work](#2576) to remove the UDF from `IsRowDeleted` filter. ## How was this patch tested? Added new suites.
scottsand-db
pushed a commit
that referenced
this pull request
Apr 26, 2024
<!-- Thanks for sending a pull request! Here are some tips for you: 1. If this is your first time, please read our contributor guidelines: https://github.com/delta-io/delta/blob/master/CONTRIBUTING.md 2. If the PR is unfinished, add '[WIP]' in your PR title, e.g., '[WIP] Your PR title ...'. 3. Be sure to keep the PR description updated to reflect all changes. 4. Please write your PR title to summarize what this PR proposes. 5. If possible, provide a concise example to reproduce the issue for a faster review. 6. If applicable, include the corresponding issue number in the PR title and link it in the body. --> #### Which Delta project/connector is this regarding? <!-- Please add the component selected below to the beginning of the pull request title For example: [Spark] Title of my pull request --> - [x] Spark - [ ] Standalone - [ ] Flink - [ ] Kernel - [ ] Other (fill in here) ## Description <!-- - Describe what this PR changes. - Describe why we need the change. If this PR resolves an issue be sure to include "Resolves #XXX" to correctly link and close the issue upon merge. --> Currently, when Deletion Vectors are enabled we disable predicate pushdown and splitting in scans. This is because we rely on a custom row index column which is constructed in the executors and cannot not handle splits and predicates. These restrictions can now be lifted by relying instead on `metadata.row_index` which was exposed recently after relevant [work](https://issues.apache.org/jira/browse/SPARK-37980) was concluded. Overall, this PR adds predicate pushdown and splits support as follows: 1. Replaces `__delta_internal_is_row_deleted` with `_metadata.row_index`. 2. Adds a new implementation of `__delta_internal_is_row_deleted` that is based on `_metadata.row_index`. 3. `IsRowDeleted` filter is now non deterministic to allow predicate pushdown. Furthermore, it includes previous relevant [work](#2576) to remove the UDF from `IsRowDeleted` filter. ## How was this patch tested? <!-- If tests were added, say they were added here. Please make sure to test the changes thoroughly including negative and positive cases if possible. If the changes were tested in any way other than unit tests, please clarify how you tested step by step (ideally copy and paste-able, so that other reviewers can test and check, and descendants can verify in the future). If the changes were not tested, please explain why. --> Added new suites. ## Does this PR introduce _any_ user-facing changes? <!-- If yes, please clarify the previous behavior and the change this PR proposes - provide the console output, description and/or an example to show the behavior difference if possible. If possible, please also clarify if this is a user-facing change compared to the released Delta Lake versions or within the unreleased branches such as master. If no, write 'No'. --> No.
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.
Which Delta project/connector is this regarding?
Description
Currently, filtering of rows that are marked as deleted (DVs) is performed with a non-deterministic UDF that is added in the plan during
PrepareDeltaScanrule. The problem is that the non-deterministic UDF prevents any filters to be pushed down to the scan, resulting in bad performance. In addition, the non-deterministic UDF prevents a number of optimizations, e.g. reusing subqueries.To avoid the above issues, this commit replaces the non-deterministic UDF with a standard filter expressions that is injected by the new
PreprocessTableWithDVsStrategybefore converting the logical plan to a physical one. The DV filter will be the bottom-most filter in the logical plan and so will be placed at the beginning of the filters that are pushed to theFileSourceScanExecnode.Note that the DV filter will not be further pushed down to the Parquet reader because filter pushdown is disabled when DVs are enabled.
How was this patch tested?
Existing tests.
Does this PR introduce any user-facing changes?
No