Skip to Main Content
  • Questions
  • speeding up rollback of current transactions

Breadcrumb

Question and Answer

Chris Saxon

Thanks for the question, gary.

Asked: March 04, 2008 - 9:59 am UTC

Last updated: March 04, 2008 - 12:38 pm UTC

Version: 10.2.0.1

Viewed 10K+ times! This question is

You Asked

An update job against a huge table ran for 20 hours, was accidentally killed, then started rolling back for a few hours. Someone suggested we set fast_start_parallel_rollback=High, to speed-up the rollback that was currently taking place.

Is there any chance that changing fast_start_parallel_rollback could hasten a rollback that's already taking place?

and we said...

typically it takes *at least* as long if not *longer* to rollback.

A rollback is a logical operation, if you just did 1,000,000 row inserts and roll them back - we now have 1,000,000 deletes to perform (including the index maintenance and everything)

rollback insert = do a logical delete
rollback delete = do a logical insert
rollback update = do a logical update to put it back the way it was

row by row pretty much.


Now, you can use fast_start_parallel_rollback to try to have the rollback done in parallel - yes - but you'll want to make sure that you have sufficient free resources for it or else doing something in parallel like that will appear to have the qualities of a denial of service attack.


Rating

  (3 ratings)

Is this answer out of date? If it is, please let us know via a Comment

Comments

How about current transactions

gary smith, March 04, 2008 - 11:32 am UTC

The crux of my question regards a transaction that has already started rolling back.

Can changing fast_start_parallel_rollback to 'HIGH' (it's dynamic) hasten a rollback that's already in progress?


Tom Kyte
March 04, 2008 - 12:38 pm UTC

if you crash the system and restart it - it is for "fast start", so if you shutdown abort/startup or startup force, it would.

it is for SMON, if your session is doing the rollback - not SMON - it won't affect it.

fast_start_parallel_rollback

Marcus, October 17, 2013 - 1:38 pm UTC

Maybe this could help. Do at you own risk.

alter system set fast_start_parallel_rollback=HIGH sid='*' ;

kill -9 <pid>

It will make smon do the rollback using all the resources available.

Parallel Rollback ? Could be...

J. Laurindo Chiappa, September 29, 2016 - 6:50 pm UTC

Yes, in some scenarios this could be useful, BUT in others not - for example, see metalink/MOS note "Database Hangs Because SMON Is Taking 100% CPU Doing Transaction Recovery" (Doc ID 414242.1) referencing one case where the parallel recover slaves could interfere/create concurrency one with another and the solution is to DISABLE parallel rollback/recover...
And just adding :

=> the same note have links about refs regarding SMON being blocked and/or suffering from CPU starvation

=> the ROLLBACK operation basically implies in an awful lot of I/Os to READ the data and another mountain of operations (in the Oracle dictionary and in memory) to PROCESS the read data - for sure, things like undo data being in the fastest disk/device (to help the I/O), internal/dictionay objects┬┤ statistics and/or rescheduling of the data load and things like that COULD be in order, too...

Regards,

J. Laurindo Chiappa