Plan Stability: SQLs switching to Bad Plans (HPV) and Load Very High (12c+)

Hi all,

That’s not the first post about this topic. Please read this one first.

Thing is: I access this post soooo often, as I face this sort of issue sooo frequently I decided to make a new one with some updated info. Even because, 3 years ago I wrote this article for 12.1 databases, now we have some relevant changes to 12.2 onwards.

Ok, so, for a little bit of context from the previous post:

  • After some critical issues with Plan Stability (new HSP being created out of nowhere very often) I mapped this to the usage of Adaptive Query Features.
  • At that point, I had a match to the latchfree events and mapped it specifically to latch misses on “Result Cache: RC Latch”.
  • As per MOS at the time, when Automatic Dynamic Sampling is used for the SQL statements, it can decide, based upon these statistics, that a better response time could be achieved by using the result cache for those queries. This can cause heavy usage of the result cache leading to the contention on latch free for “Result Cache: RC Latch”
  • So the villain was the Automatic Dynamic Sampling, as part to the SQL Plan Directives.
  • On the following months and migrations I started seeing more and more issues related to the SQL Plan Directives as part of the Adaptive Query Optimization features. So, I have been following the referred steps more and more.
  • Happens this means setting a few underscore parameters, which we should avoid unless approved/recommended by Oracle, so I’ve been trying to be less intrusive and simply disabling the whole Adaptive Features.
  • How to do it? That’s the tricky part and the thing to keep in mind, as we had changes from 12.1 to 12.2

Some words about the parameters settings:

  • The parameter “optimizer_adaptive_features” is a valid parameter for 12.1 and it’s enabled by default. I recommend leaving it as FALSE if you are facing this sort of issues.
  • This parameter is obsolete con 12.2 onwards though. It was replaced by both optimizer_adaptive_statistics and optimizer_adaptive_plans.
    • 12.1:

adap1

    • 12.2+:

adap2

(images from https://blogs.oracle.com/optimizer/optimizer-adaptive-features-in-oracle-database-12c-release-2)

  • Also, keep one eye on all %optim% parameters, as specific mechanisms can be enabled (for some reason) as I recently faced in a client ( _optimizer_adaptive_plan = TRUE).

When Adaptive Queries are good?

Ok Matheus, but if we should disable it, why did Oracle created in the first place?

Well, if something would be always bad, it would be disabled by default… There are some good usage cases: Dynamic Sampling is typically recommended when you know you are getting a bad execution plan due to complex predicates.

oh, noice, so I’ll change on my database I have a lot of complex queries” . Don’t, I suggest you to try and use an alter session statement to set the value for optimizer_dynamic_sampling parameter as it can be extremely difficult to come up with a system-wide setting. And you can mess up with other things.

When Adaptive Queries are bad?

If the queries compile times need to be as fast as possible, for example, unrepeated OLTP queries where you can’t amortize the additional cost of compilation over many executions. Serial statements are typically short running and any Dynamic Sampling overhead at compile time can have a large impact on overall system performance (if statements are frequently hard parsed).

For systems that match this profile, setting OAF=FALSE is recommended. For Oracle Database 12c Release 2 onwards, using the default OAS=FALSE is recommended.

I hope this helps you!