Comments on: SQL Tuning Advisor Profiles

By: Blogroll Report 29/01/2009 – 05/02/2010 « Coskan’s Approach to Oracle

Blogroll Report 29/01/2009 – 05/02/2010 « Coskan’s Approach to Oracle — Fri, 05 Mar 2010 10:24:39 +0000

[...] SQL Profiles and manual sql profile (comments) Kerry Osborne-Single Hint SQL Profiles Kerry Osborne-SQL Tuning Advisor Profiles 22-How to do pivoting pre-11G? Marc Billette-Easy Pivot Query Result in pre-11g Oracle Comments [...]

By: osborne

osborne — Sat, 13 Feb 2010 18:51:36 +0000

Jonathan,

Well rats! I hate it when I am wrong! My script was combining hints for multiple Profiles (if there happened to be more than one Profile on the same statement, which is possible if you use multiple categories - a bit unusual but very possible). This was exactly the case in my test. Brilliant piece of deduction on your part by the way, without actually seeing the system - especially since I hacked up the hints to protect the table and column names. At any rate, I’m back to where I was before with a Profile of unknown origin that has a mixture of opt_estimate and other hints, but no proof that it was generated by the Tuning Advisor. It’s possible someone manually added hints or changed the name of the profile, although I can’t get anyone to fess up to that. So still a bit of a mystery to unravel there - another post I’m sure.

At any rate, I’ve updated my sql_profile_hints script to add the additional join (on category). Thanks for taking the time to look carefully at this post and comment on it.

Kerry

By: osborne

osborne — Sat, 13 Feb 2010 16:19:56 +0000

Excellent observation. I always just figured they didn’t do a very good job of programming, but now that I think about it a little, it seems unlikely that the programmers would not catch multiple all_rows hints in the same profile, much less multiple opt_estimates on the same operation (which I hadn’t noticed). I’ll have a look at that specific example and let you know what I find.

Kerry

By: Jonathan Lewis

Jonathan Lewis — Sat, 13 Feb 2010 15:17:33 +0000

Kerry,

Thanks for following up that topic.
Interesting result - but there’s an anomaly that I’d want to look into.

Note that your output shows the ALL_ROWS hint three times; that’s suspect. Then near the end of the output you’ve got:

OPT_ESTIMATE(@”SEL$5DA710D3″, TABLE, “D”@”SEL$1″, SCALE_ROWS=11.39782103)
OPT_ESTIMATE(@”SEL$5DA710D3″, TABLE, “D”@”SEL$1″, SCALE_ROWS=11.95241429)

It’s the same transformed query block referencing the same alias twice with different scale factors. Again, that’s suspect; and it happens at many points in the hint set.

I suspect that you’ve managed to report two profiles as one - which doesn’t preclude the possibility that there is a faked set that holds the ‘outline’ type hints and your generated set is just ’stats’ hints.

Regards
Jonathan Lewis

By: David Mann

David Mann — Fri, 05 Feb 2010 19:53:28 +0000

I agree with your points. I was recently horrified when a recent STA job had recommended a plan with about 20 cardinality hints. No way I would consider implementing that suggestion as it would quickly be wrong.

It did give me a direction to start looking and further analysis with GATHER_PLAN_STATISTICS showed me exactly where Estimated Rows and Actual Rows were getting hosed.

But sometimes a simple switch between ALL_ROWS and FIRST_ROWS will be something that is generic enough to feel safe implementing.