You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Describe the bug GpuBringBackToHost creates the ColumnarBatch without numRows, which will be set to 0 in this case. This can confuse any logic relying on the numRows metric during the query processing, such as adaptive query optimizer. The ColumnarBatch returned would be always regarded as if it were an empty batch.
Expected behavior
The right number of rows should be retrieved from the hostColumns and then set for the ColumnarBatch before it is returned.
The text was updated successfully, but these errors were encountered:
@jihoonson great catch. Could you also check other places that we create a ColumnarBatch to be sure there are no other issues like this? The API is kind of bad because you can create a ColumnarBatch and then set the number of rows afterwards.
@revans2 I have checked other places, and did not see any. I agree that this API does not have the best design. I wonder if we should do something about it, such as adding safe wrappers, e.g., rapids.ColumnarBatches.empty() and rapids.ColumnarBatches.create(ColumnVector[] columns, int numRows), and forbidding unsafe APIs.
Describe the bug
GpuBringBackToHost
creates theColumnarBatch
withoutnumRows
, which will be set to 0 in this case. This can confuse any logic relying on thenumRows
metric during the query processing, such as adaptive query optimizer. TheColumnarBatch
returned would be always regarded as if it were an empty batch.Expected behavior
The right number of rows should be retrieved from the
hostColumns
and then set for theColumnarBatch
before it is returned.The text was updated successfully, but these errors were encountered: