fix: force contiguous copy for sliced list arrays in embed_array_storage #7896
+131
−0
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
|
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. |
4 tasks
When ds.shard() or ds.select() creates a sliced view, array.values returns values with internal offset references that can cause PyArrow's C++ layer to crash with SIGKILL when processing nested types like Sequence(Nifti()). The fix forces a contiguous copy via pa.concat_arrays([array]) when the array has a non-zero offset, breaking the internal references before further processing. Fixes: #6
1. Remove fork-specific issue URL placeholders (upstream-ready) 2. Add consistency assertions to LargeList test: - offset == 0 check - content verification (bytes embedded) 3. Add offset == 0 check to Nifti test for consistency
The-Obstacle-Is-The-Way
force-pushed
the
fix/embed-storage-crash
branch
from
b7839ff to
12117f2
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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Outdated suggestions cannot be applied.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
Fixes SIGKILL crash in
embed_array_storagewhen processing sliced/sharded datasets with nested types likeSequence(Nifti())orSequence(Image()).Root cause: When
ds.shard()ords.select()creates a sliced view,array.valueson a slicedListArrayreturns values with internal offset references. For nested types, PyArrow's C++ layer can crash (SIGKILL, exit code 137) when materializing these sliced nested structs.Fix: Force a contiguous copy via
pa.concat_arrays([array])when the array has a non-zero offset before processing list/large_list arrays.Changes
embed_array_storagefor list/large_list arraysarray.offset > 0to break internal referencesTest plan
tests/features/test_embed_storage_sliced.pywith 3 tests:test_embed_array_storage_sliced_list_imagetest_embed_array_storage_sliced_list_niftitest_embed_array_storage_sliced_large_listembedded.offset == 0(contiguous result)Context
This was discovered while uploading a 270GB neuroimaging dataset (ARC) with
Sequence(Nifti())columns. The process crashed with SIGKILL (no Python traceback) whenembed_table_storagewas called on sharded data.Workaround that confirmed the fix: pandas round-trip (
shard.to_pandas()→Dataset.from_pandas()) which forces a contiguous copy.Fixes #7894
Related: #6686, #7852, #6790