[SPARK-55817][SQL] Enable Parquet row-group skipping for shredded variant - #54598
Closed
qlong wants to merge 1 commit into
Closed
[SPARK-55817][SQL] Enable Parquet row-group skipping for shredded variant#54598qlong wants to merge 1 commit into
qlong wants to merge 1 commit into
Conversation
…iant When PushVariantIntoScan rewrites variant_get() calls into struct field accesses, the rewritten predicates reference logical paths like "v.`0`" that ParquetFilters cannot resolve to any physical column, so they are dropped and row-group skipping is disabled for all shredded variant queries. This change adds variantExtractionSchema to ParquetFilters, and resolves the logical path to the corresponding typed_value leaf in the physical Parquet schema.The resolved entries allow predicates on shredded variant to participate in row-group skipping. Array-index paths and fields absent from a file's physical schema are skipped.
qlong
force-pushed
the
SPARK-55817-row-group-skipping
branch
from
March 3, 2026 21:41
079b01f to
ef8bc2f
Compare
Contributor
Author
|
@chenhao-db Can you take a look at this? It is a follow up to your PR #49235 |
This was referenced Apr 28, 2026
|
We're closing this PR because it hasn't been updated in a while. This isn't a judgement on the merit of the PR in any way. It's just a way of keeping the PR queue manageable. |
Contributor
Author
|
I think this PR is still needed to improve variant read performance |
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
What changes were proposed in this pull request?
When PushVariantIntoScan rewrites variant_get() calls into struct field accesses, the rewritten predicates reference logical paths like "v.
0" that ParquetFilters cannot resolve to any physical column, so they are dropped and row-group skipping is disabled for all shredded variant queries.This change adds variantExtractionSchema to ParquetFilters, and resolves the logical path to the corresponding typed_value leaf in the physical Parquet schema.The resolved entries allow predicates on shredded variant to participate in row-group skipping.
Array-index paths and fields absent from a file's physical schema are skipped.
Jira: https://issues.apache.org/jira/browse/SPARK-55817
Why are the changes needed?
Performance improvement. The shreded variant predicates are pushed down to participate row group filtering.
Does this PR introduce any user-facing change?
No.
How was this patch tested?
Was this patch authored or co-authored using generative AI tooling?
co-authorized with Claude 4.6 Sonnet.