Hey,
I totally agree with Andy here. However you can make it a little bit more easy
for us (and you).
Just create a SQLd360 report (https://github.com/sqldb360/sqldb360) and send it
to us.
Best Regards
Stefan Koehler
Independent Oracle performance consultant and researcher
Website: http://www.soocs.de
Twitter: @OracleSK
Andy Sayer <andysayer@xxxxxxxxx> hat am 19. Juli 2018 um 09:27 geschrieben:--
You’ve had a lot of suggestions, but personally I believe they’re either over
kill or guess work they’ll just waste plenty of your time.
I would just do it very simply:
Compare the two plans (by eye)
Find the pivotol line of the plan that “changes everything” - does like 13
have a tiny cardinality in a But not b? (Etc)
Identify what went into that decision
Find out how that input changed between a and b (different peeked bind,
different stat etc)
I don’t see any need to dig into traces, look into system stats, check for
Witch doctors nearby etc.
If you shared the two complete plans, then I’m sure it wouldn’t take long to
just point out where the issue is likely coming from and make some
suggestions. Make sure you include the predicate section, peeked binds and
notes - they are all packed full of important information
Hope this helps,
Andrew