Link: https://github.com/SQLMesh/sqlmesh/actions/runs/32190188666/job/95882655785
AI Notes
Failing Test
tests/core/engine_adapter/integrationtest_integration_clickhouse.py::test_insert_overwrite_by_condition_by_key_partitioned[[query]clickhouse_cluster]
What’s Wrong
After a unique-key insert overwrite, id=4 should be 2024-05-01 but the table has 2024-05-02. The assert also looks totally broken because ClickHouse returns dates as unix seconds and the test expects pandas nanoseconds, so every row looks wrong even when most dates match.
Possible Fix
Make the unique-key path pick a deterministic row for duplicates (e.g. ordered), and/or normalize date comparison in compare_with_current so ClickHouse dates compare as real dates, not raw ints.
Link: https://github.com/SQLMesh/sqlmesh/actions/runs/32190188666/job/95882655785
AI Notes
Failing Test
tests/core/engine_adapter/integrationtest_integration_clickhouse.py::test_insert_overwrite_by_condition_by_key_partitioned[[query]clickhouse_cluster]What’s Wrong
After a unique-key insert overwrite, id=4 should be 2024-05-01 but the table has 2024-05-02. The assert also looks totally broken because ClickHouse returns dates as unix seconds and the test expects pandas nanoseconds, so every row looks wrong even when most dates match.
Possible Fix
Make the unique-key path pick a deterministic row for duplicates (e.g. ordered), and/or normalize date comparison in compare_with_current so ClickHouse dates compare as real dates, not raw ints.