Skip to content

feat(table): equality-delete read support in merge-on-read scans#19

Open
abnobdoss wants to merge 5 commits into
mainfrom
v3/t5-eqdelete
Open

feat(table): equality-delete read support in merge-on-read scans#19
abnobdoss wants to merge 5 commits into
mainfrom
v3/t5-eqdelete

Conversation

@abnobdoss

Copy link
Copy Markdown
Owner

No description provided.

Abanoub Doss added 5 commits June 23, 2026 09:13
Implements the READ path for equality deletes (DataFileContent.EQUALITY_DELETES)
in PyArrow merge-on-read scans, building on the DeleteFileIndex matching/pruning
from PR apache#3285.

- DeleteFileIndex now indexes equality deletes at (seq-1) so they apply only to
  data files with a strictly smaller data sequence number, with null-count and
  bounds-based pruning ported from PR apache#3285.
- plan_files no longer raises on equality deletes; routes them into the index.
- _read_all_delete_files splits positional vs equality delete files; equality
  delete files are read into Arrow tables keyed by file path.
- _task_to_record_batches applies equality deletes after positional deletes and
  the row filter, anti-joining the data batch on the equality_ids columns. The
  common no-null case uses a fast Arrow left-anti join; delete rows with NULL keys
  use a null-aware pass implementing Iceberg IS NOT DISTINCT FROM semantics.
- Equality-id key columns are added to the projected field ids so they are read
  even when not in the user projection.

Tests: 4 end-to-end ArrowScan read tests (single-column, multi-column AND-match,
no-match, null-key) plus DeleteFileIndex seq-scoping/pruning unit tests.
Address adversarial review: make test_equality_delete_no_match load-bearing
(include a matching key so it fails if deletes are ignored) and add
test_equality_delete_multi_batch verifying eq-delete application across
record-batch boundaries.
…chema evolution and type promotion, support REST plan path
@abnobdoss abnobdoss closed this Jun 25, 2026
@abnobdoss abnobdoss reopened this Jun 25, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant