Re: Need to run CLUSTER to keep performance

[Rafael Martinez <r(dot)m(dot)guerrero(at)usit(dot)uio(dot)no>]

Heikki Linnakangas wrote:
> Rafael Martinez wrote:
>> DETAIL:  83623 dead row versions cannot be removed yet.
> 
> Looks like you have a long-running transaction in the background, so
> VACUUM can't remove all dead tuples. I didn't see that in the vacuum
> verbose outputs you sent earlier. Is there any backends in "Idle in
> transaction" state, if you run ps?
> 

I don't see any long transaction in progress (<IDLE> in transaction) and
if we run the vacuum jobb manual just after checking this, it still
cannot remove the dead rows.

Any suggestions cause vacuum cannot remove these dead rows?

> In 8.1, CLUSTER will remove those tuples anyway, but it's actually not
> correct. 

With other words, .... we have to be very carefull to not run CLUSTER on
a table been modified inside a transaction if we do not want to lose
data? ...

Does this mean that if we run a transaction which update/delete many
rows, run cluster before the transaction is finnish, and then rollback
the transaction after cluster has been executed, all dead rows
updated/deleted  by the transaction can not be rollbacked back because
they are not there anymore?


-- 
 Rafael Martinez, <r(dot)m(dot)guerrero(at)usit(dot)uio(dot)no>
 Center for Information Technology Services
 University of Oslo, Norway

 PGP Public Key: http://folk.uio.no/rafael/