-
[OPIK-7247] [BE] Fix duplicate experiment items in test suite detail view (#7388)
frostbyte_neo 发布于
2026-07-08 16:01:39 +00:00 - [OPIK-7247] [BE] Fix duplicate experiment items in test suite detail view
Add LIMIT 1 BY dedup to aggregated-branch lookup JOINs that resolve
dataset_item_id to stable IDs via dataset_item_versions.The LEFT JOIN dataset_item_versions FINAL matches multiple rows per
experiment item when the same physical id appears across multiple
dataset versions (which happens whenever items are edited, since each
version row keeps id == dataset_item_id). Without dedup, the
groupArray that builds experiment_items_array captures N identical
copies per item (N = number of dataset versions), causing the UI to
render the same trial result multiple times.The raw branch was already protected by LIMIT 1 BY ei.id; this commit
adds the same protection to the 3 aggregated-branch locations in
DatasetItemVersionDAO and ExperimentAggregatesDAO.Includes a regression test that edits dataset items across multiple
versions and verifies experiment items are not duplicated after
aggregation.- Revision: Add parameterized regression test for OPIK-7247
Parameterized test covering the full matrix of dataset vs test suite
and single-trial vs multi-trial (runsPerItem=3) on multi-version
datasets. Verifies experiment item count is preserved exactly after
aggregation — no duplication from the lookup JOIN, and multi-trial
experiments retain the correct number of trials per item.- Revision 2: Address review feedback
- Add SQL comment above each LIMIT 1 BY eia.id explaining why arbitrary
pick is safe (all dupes resolve to the same stable_dataset_item_id) - Replace .collect(Collectors.toList()) with .toList() in test
- Revision 3: Split tests into branch-free parameterized methods
Replace single parameterized test with if-branching into two clean
parameterized tests (dataset vs test suite), each parameterized by
runsPerItem only. No conditional logic, no unused variables.下载附件