TheR1sing3un opened a new pull request, #9843: URL: https://github.com/apache/paimon/pull/9843
### Purpose Keep each local multimodal search execution on one snapshot. A commit between candidate search and result lookup can currently drop a selected row or return its updated values even when they violate the original pre-filter. Hybrid routes can also plan against different commits, and full-text reads lose the planning snapshot. Capture an execution-local read view and carry it through vector, batch, full-text, hybrid, and result lookup paths. Reusing a query resolves a fresh view; explicit snapshot, tag, and timestamp selectors remain effective. Retain resolved snapshot metadata for tags and preserve an empty view across the first concurrent commit. Share the read-view helper with existing snapshot-aware readers without mutating reusable queries or builders. ### Tests 367 passed and 12 skipped across the targeted Python 3.9 suites (`paimon_vindex` installed): - New search snapshot regressions: concurrent delete/update before lookup, query reuse, hybrid route planning, empty-table append, snapshot/tag/timestamp selection, retained tag metadata, read-view copies, branch switching, full-text live-row/raw reads, and native indexed-vector lookup. - Existing vector filter/metric, batch raw/refinement, hybrid execution/ranking, multimodal table, primary-key index, scalar-index/build, table-copy/time-travel, scan-mode, native planner, and file-scanner schema suites. - Skips cover optional native planner, full-text, and Ray dependencies. Full-text snapshot reads are exercised with real table data and stub scoring. - Changed Python files pass repository-configured flake8. License-header checks and `git diff --check` pass. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
