Re: Causes for high "Rollback per transaction %" in AWR report

  • From: Thomas Kellerer <thomas.kellerer@xxxxxxxxxx>
  • To: "oracle-l@xxxxxxxxxxxxx" <oracle-l@xxxxxxxxxxxxx>
  • Date: Tue, 20 Nov 2012 15:46:32 +0100

Jonathan Lewis, 20.11.2012 15:37:
> | Hello,
> |
> | I am analyzing some performance problems of an Oracle 10.2.0.5.0 2-node RAC 
> system and one figure in the
> |AWR report caught my attention:
> |
> | Rollback per transaction %: 98.15
> |
> | What exactly does this figure indicate, and what factors contribute to this?
> |
>
>
> The first things to check are the stats:
>      rollback changes - undo records applied
>      user commits
>      user rollbacks
>      transaction rollbacks
>
> If the first figure is very small it doesn't really matter how many
> rollbacks you are doing, they are "no work" rollbacks. It is a common
> feature of web app servers to issue a (redundant) rollback after every
> single statement sent to the database - you may be seeing the effects of
> such a WAS.
>
> If transaction rollbacks is much larger than user rollbacks then there may
> be an internal problem of some sort - e.g. a loader program is constantly
> trying to insert duplicate keys which are rolled back by Oracle, raising an
> error that should be trapped (but might be ignored) by the application.
>

Thanks for the quick respsonse

It indeed seems to be that "no work rollback" type of thing:


                                          Total   per Second   per Trans
rollback changes - undo records applied      6         0.00        0.00
transaction rollbacks                        3         0.00        0.00
user rollbacks                          23,511         6.52        0.98

I already requested to get the configuration of the connection pools to find out
if e.g. a "validation query" or something like that is configured to be a 
ROLLBACK
statement.

So, seeing the low number of "rollback changes", I assume the bottomline is 
that I don't need to worry and concentrate on the other bottlenecks that we 
have already identified.

Regards
Thomas





--
//www.freelists.org/webpage/oracle-l


Other related posts: