Re: New feature request: FlashBack Query - Mailing list pgsql-hackers

From Csaba Nagy
Subject Re: New feature request: FlashBack Query
Date
Msg-id 1172049456.3101.335.camel@coppola.muc.ecircle.de
Whole thread Raw
In response to Re: New feature request: FlashBack Query  (Rod Taylor <rod.taylor@gmail.com>)
List pgsql-hackers
> Do 97% of transactions commit because Oracle has slow rollbacks and  
> developers are working around that performance issue, or because they  
> really commit?
> 
> I have watched several developers that would prefer to issue numerous  
> selects to verify things like foreign keys in the application in  
> order to avoid a rollback.

Most of the code we have will not afford a rollback because it can be
part of a much bigger transaction which would have much higher
performance penalty if retried than a simple rollback. And you know that
in postgres you can't roll back just the last insert, you will crash the
whole transaction with it... and it's simply a performance bottleneck to
retry in a high contention scenario (which is usually so in our case).

So I would say we don't avoid rollbacks because of the cost of the
rollback, but because of the cost of the retry...

Cheers,
Csaba.




pgsql-hackers by date:

Previous
From: Magnus Hagander
Date:
Subject: Re: msvc failure in largeobject regression test
Next
From: Markus Schiltknecht
Date:
Subject: Re: tsearch in core patch, for inclusion