Using the latest community build (3.1.1-201901281837360699-30c9d74), I’m running into an issue where the query is failing without an error message (just “400 - Bad Request”).
The set up is as follows:
- Create a dataset using a query from Table A, and activate a “raw reflection” on the dataset (I’ll call Reflection A)
- Create a view (in postgres, the main driver was to expose attributes of a postgres jsonb column, but contents of the view end up not being important) of Table B (I’ll call View B)
- Join View B to Reflection A, “Group By” a column in View B and have a measure from Reflection A
Running will fail, with 10 profiles in the profiles tab of the job. Drilling into the profile, on the Error tab it complains:
INVALID_DATASET_METADATA ERROR: Unexpected columns redacted_column_a,redacted_column_b Expected skipped columns SqlOperatorImpl JDBC_SUB_SCAN Location 2:0:6 SqlOperatorImpl JDBC_SUB_SCAN Location 2:0:6 Fragment 2:0
“redacted_column_a” and “redacted_column_b” are the other columns that are in View B that were not being used for the “Group By”. It appears that this only happens when the job is being accelerated by a reflection, because other test cases that are not accelerated seem to work properly.
I thought maybe it was happening when the full query couldn’t be pushed down into the database, but I tested a similar query with a postgres view and a dataset from an uploaded spreadsheet and it worked as expected.
Additionally, it appears to only happen with a reflection from the same database? Having a reflection on the uploaded spreadsheet that is used for a similar query does not exhibit the same error.