-
[NA] [BE] Performant ClickHouse bulk update queries to ensure latest row selection (#4206)
frostbyte_neo 发布于
2025-11-24 15:03:29 +00:00 - [BE] Fix ClickHouse TraceThreadDAO bulk update query to use latest row
- Add ORDER BY tt.last_updated_at DESC LIMIT 1 BY tt.id to BULK_UPDATE query
- Prevents bug where bulk updates could select old historical rows
- Aligns with existing fixes in SpanDAO and TraceDAO which already have this protection
- Add regression test batchUpdate__whenMultiplePartialUpdates__thenLatestValuesPreserved
- Test verifies that multiple sequential bulk updates preserve the latest values
Note: SpanDAO and TraceDAO already have ORDER BY + LIMIT 1 BY protection and did not need fixes
- Optimize ORDER BY clauses in bulk update queries to follow primary key order
- Update TraceDAO BULK_UPDATE to order by (workspace_id, project_id, id) DESC, last_updated_at DESC
- Update TraceThreadDAO BULK_UPDATE to order by (workspace_id, project_id, thread_id, id) DESC, last_updated_at DESC
- Aligns with ClickHouse best practice: ORDER BY primary key tuple, then last_updated_at for deduplication
- Improves query performance by following table's natural sort order
- ExperimentDAO and SpanDAO already follow this pattern correctly
-
Revision 2: Remove 'final' from TraceThreadDAO bulk update for consistency and performance
-
Format fix
下载附件