-
[OPIK-6983] [BE][FE] Surface queue-scoped feedback scores in general views (#7209)
frostbyte_neo 发布于
2026-06-29 08:31:27 +00:00 - [OPIK-6983] [BE][FE] Add sourceQueueId to ValueEntry and update feedback score utilities
Add sourceQueueId field to ValueEntry and FeedbackScoreMapper parsing
(tuple index 7) so the backend can surface which annotation queue a
score originated from.On the frontend, extend FeedbackScoreValueByAuthorMap with
source_queue_id, update findValueByAuthor to accept an optional
sourceQueueId for queue-scoped lookups, and thread sourceQueueId
through generateUpdateMutation / generateDeleteMutation so optimistic
cache updates use composite keys (author_queueId).- [OPIK-6983] [BE][FE] Surface queue-scoped scores in trace detail panel
TraceDAO SELECT_BY_IDS: add source_queue_id to UNION ALL selects,
LIMIT 1 BY, entries tuple, and value_by_author map key/value for both
trace and span feedback score CTEs.Add author field to ValueEntry and value tuple (index 8) so the
frontend reads entry.author directly instead of parsing composite keys.FeedbackScoreTable (v2): replace extractAuthorName with getAuthorName
using entry.author, propagate source_queue_id and per-entry reason/source
to child rows, render SourceCell as ResourceLink to the annotation queue
(plain text fallback for SDK/deleted queue).- [OPIK-6983] [BE][FE] Update remaining TraceDAO CTEs, score cells, hover cards, and delete scoping
Backend:
- Add source_queue_id to all 8 remaining TraceDAO LIMIT 1 BY, UNION ALL,
entries tuples, and map key/value constructions. - Scope authored_feedback_scores DELETE by source_queue_id to prevent
deleting queue-scoped scores when removing a direct score.
Frontend:
- FeedbackScoreCellValue: multi-value display (categorical count format,
numeric avg), showTooltip prop for experiment context vs hover card
for logs context. - MultiValueFeedbackScoreHoverCard: AuthorLabel with queue name resolution
via useAnnotationQueueById, source label right-aligned in lighter color,
show card for single entries, Average row only for multi-entry. - findValueByAuthor: default to exact match (safe), prefixMatch opt-in
via hasAnyValueByAuthor helper. Prevents direct scoring from picking
up queue-scoped entries. - getCategoricFeedbackScoreValuesMap: return UserEntry with mapKey+entry
so AuthorLabel can resolve source in categorical hover cards.
- [OPIK-6983] [BE] Add source_queue_id to ThreadDAO CTEs
Same mechanical pattern: add source_queue_id to UNION ALL selects,
LIMIT 1 BY, entries tuples, and map key/value constructions in
ThreadDAO (4 instances). SpanDAO intentionally skipped — annotation
queues never target spans.- [OPIK-6983] [BE] Add source_queue_id to ProjectMetricsDAO and KpiCardDAO CTEs
Same mechanical pattern: add source_queue_id to UNION ALL selects
and LIMIT 1 BY in ProjectMetricsDAO (3 instances) and KpiCardDAO
(3 instances) so dashboard widgets and KPI cards include all
queue-scoped scores in their aggregations.- [OPIK-6983] [BE][FE] Scope queue export by source_queue_id
Backend: add source_queue_id to comments CTE inner SELECTs and
tuple in TraceDAO and ThreadDAO. Update CommentResultMapper to
parse source_queue_id from tuple index 6.Frontend: add source_queue_id to CommentItem type. Update
getCommentsByUser to filter by sourceQueueId when provided.
Pass annotationQueue.id to both getFeedbackScoresByUser and
getCommentsByUser in ExportAnnotatedDataButton so export only
includes scores and comments from that specific queue.- [OPIK-6983] [BE][FE] Fix experiment compare grid queue-scoped scores
Backend: add source_queue_id to LIMIT 1 BY, UNION ALL, entries tuples,
and map key/value in ExperimentItemDAO, DatasetItemDAO,
DatasetItemVersionDAO, and ExperimentAggregatesDAO. Fix count(*)
in ExperimentAggregatesDAO to count all entries (not just distinct
authors) so avg is used instead of any_value for multi-queue scores.Frontend: replace showTooltip with footer prop on FeedbackScoreCellValue.
Always use rich hover card; experiment compare cell passes trial info
as footer. Re-add footer rendering to MultiValueFeedbackScoreHoverCard.- [OPIK-6983] [BE] Add source_queue_id to OptimizationDAO CTE
Aggregate-only: add source_queue_id to UNION ALL selects and
LIMIT 1 BY so optimization views include all queue-scoped scores
in their aggregations.- [OPIK-6983] [BE] Add E2E test for same-author multi-queue score averaging
Verify that scoring the same trace+name as the same author from two
different annotation queues produces the correct average and two
distinct composite-key entries in value_by_author.- Revision: Fix assertion_results LIMIT 1 BY, delete scoping, tuple cast, and PR review fixes
- Revert source_queue_id from assertion_results LIMIT 1 BY in ExperimentDAO
- Pass sourceQueueId through delete handlers in FeedbackScoreTable,
DeleteFeedbackScoreValueDialog, TraceDataViewer, AnnotatePanel,
ThreadAnnotations, ThreadAnnotatePanel, ExperimentFeedbackScoresViewer - Add FeedbackScoreTypeManufacturer to exclude sourceQueueId from PODAM
for FeedbackScore, FeedbackScoreBatchItem, FeedbackScoreBatchItemThread - Fix DatasetItemVersionDAO feedback_scores_array tuple cast to include
all 9 ValueEntry fields (spanType, spanId, sourceQueueId, author) - Extract buildValueByAuthorKey helper for composite key construction
- Add annotationQueue.id to useCallback deps in ExportAnnotatedDataButton
and SMEFlowContext - Fix comment typo in FeedbackScoreMapper (index 8 is author)
- Fix prettier formatting in annotation-queues.ts
- [OPIK-6983] [BE] Address PR review comments
Addresses andrescrz's and baz-reviewer's comments on PR #7209:
- ValueEntry: drop the dead overloaded constructors (the builder is used everywhere; no call sites).
- CommentResultMapper / FeedbackScoreMapper: use StringUtils.isNotBlank for source_queue_id checks (more resilient than isNotEmpty).
- TraceDAO: convert the SELECT_BY_IDS value_by_author note from a plain comment to a javadoc.
- SpanDAO: carry the author field in the value_by_author tuples for parity with traces/threads, so the frontend reads entry.author instead of parsing it from the map key. Queues never target spans, so source_queue_id stays empty.
- SpanDAO: fix span comments returning the span's own id as source_queue_id. Both comment subqueries (SELECT_BY_IDS and comments_final) now carry source_queue_id at slot 6, matching the CommentResultMapper contract; the SELECT_BY_IDS comment tuple is explicit instead of tuple(c.*) to avoid slot drift.
- FeedbackScoreMapper: in parseValueByAuthorFromJson, default ValueEntry.author to the value_by_author map key so legacy snapshots without an explicit author element don't return null authors (the explicit element still overrides for current data).
- MultiValueFeedbackScoresE2ETest: assert full ValueEntry entities via recursive comparison instead of field-by-field; extract the repeated feedback-score ignored-fields list to a local constant; assert ValueEntry.sourceQueueId explicitly in the queue-scoped delete tests; rename the added test to drop the forbidden test* prefix (sameAuthorTwoQueuesProducesAverage).
- [OPIK-6983] [FE] Address PR review comments from andriidudar and baz
- FeedbackScoreTable utils: lift source_queue_id from the entry in the single-value branch so a single-reviewer queue score shows the queue link in the Source column (and builds the author_queueId delete key), matching multi-reviewer rows.
- Extract a shared useFeedbackScoreSourceLabel hook (used by SourceCell and the hover card AuthorLabel) resolving the source display in one place: a spinner while the queue loads, the queue link/name when resolved, "" on 404, and "" on other errors (403/500/network) so transient failures are not mislabeled as deletion.
下载附件