Thanks.

Just for the record, there were "merge phase" counters with nonzero values for the DataTarget operator, and the COPY did end up functioning correctly (i.e., the data did finally appear in the correct table).

There's no rows at all in EE Profiles for operator Sort (for any counter).

We have a long-running COPY operation going. I can see that rows are coming in via LOAD_STREAMS, but when i look in projection_storage, I don't see anything in ROS or WOS (no rows, no bytes). (LOAD_STREAMS sees billions of rows.) This operation has been running for tens of hours, and I'm a little wo...

Yes - it would for these two queries.

And there's no funny edge cases that are treated differently (e.g. NULLs in the grouping columns)?

Understood. But I'm not referring to the differences in output between aggregate and analytic functions in general, but rather the differences in output between the two queries I posted above.

sKwa,

Thanks for the reply (as usual!).

So there might be performance differences, but no differences in result?

Are the following generally equivalent? SELECT DISTINCT a1, a2, COUNT(*) OVER w AS cnt, SUM(b) OVER w AS sum_b FROM ... WHERE ... WINDOW w AS (PARTITION BY a1, a2); and SELECT a1, a2, COUNT(*) AS cnt, SUM(b) AS sum_b FROM ... WHERE ... GROUP BY a1, a2; Or are there edge cases where they might not be?