There's no PARTITION stats on KEY_T and no re-collection for three weeks.
The UPI on T_CUST does not provide a better distribution than a NUPI (FIRST_DT ,VISITOR_ID), but prevents a fast AMP-local join.
And the number of rows is probbaly underestimated.
Btw, you'll probably not run a select like that returning millions of rows, is this part of an Insert/Select or Create Table As?
There's no PARTITION stats on KEY_T and no re-collection for three weeks.
The UPI on T_CUST does not provide a better distribution than a NUPI (FIRST_DT ,VISITOR_ID), but prevents a fast AMP-local join.
And the number of rows is probbaly underestimated.
Btw, you'll probably not run a select like that returning millions of rows, is this part of an Insert/Select or Create Table As?