Stefan,
I'll play around with it a bit more, but just an FYI, I have been
dropping/re-creating the table both using CTAS and manual definition and
getting the same results (but it could be me). No chance for maintenance jobs
to run when I'm doing this.
I have opened an SR with Oracle, so if they come back with anything related to
the NDV values, I'll update this thread.
Thanks!
Scott
-----Original Message-----
From: Stefan Koehler [mailto:contact@xxxxxxxx]
Sent: Tuesday, December 01, 2015 12:56 PM
To: Deas, Scott <Scott.Deas@xxxxxxx>; oracle-l@xxxxxxxxxxxxx
Subject: RE: Question on Stats for Indexes Created in Parallel
Hi Scott,
DB Version 11.2.0.4 - NDV is wrong when index is built in parallel,
correct when built serially. Stale_stats is set to NO (this is not
what I saw in my earlier tests, maybe a user error.....)
"Deas, Scott" <Scott.Deas@xxxxxxx> hat am 1. Dezember 2015 um 17:23
geschrieben:
Hi Stefan,
I guessed the same regarding the cause, but thought there may be something
I'm missing.
My concern about setting the stats is that for this low-cardinality
column, I know what the distinct values are, but I would assume that
any of our bitmap indexes will have bad cardinality estimates as a result of
this issue (and I won't know the NDV for each). Re-gathering stats is not a
deal-breaker, but it seems like it shouldn't have to be done.
I ran a few more tests in our environments:
DB Version 10.2.0.4 - NDV is correct, regardless of whether the index
is built in parallel or not. Stale_stats is set to NO DB Version
11.2.0.4 - NDV is wrong when index is built in parallel, correct when
built serially. Stale_stats is set to NO (this is not what I saw in my
earlier tests, maybe a user error.....) DB Version 12.1.0.2.0 - NDV is
wrong when index is built in parallel, correct when built serially.
Stale_stats is set to NO
Sounds like I need to open a ticket with Oracle.
Thanks,
Scott