Re: performance drop on 8.2.4, reverting to 8.1.4

Tom Lane escribió:

> I was hoping that the auto plan invalidation code in CVS HEAD would get
> it out of this problem, but it seems not to for the problem-as-given.
> The trouble is that it won't change plans until autovacuum analyzes the
> tables, and that won't happen until the transaction commits and sends
> off its I-inserted-lotsa-rows report to the stats collector.  So any
> given large transaction is stuck with the plans it first forms.  There's
> probably nothing we can do about that in time for 8.3, but it's
> something to think about for future releases ...

I think there is something we can do about this -- drop the default
value for analyze threshold.  We even discussed way back that we could
drop the concept of thresholds altogether, and nobody came up with an
argument for defending them.

> it won't change plans until autovacuum analyzes the
> tables, and that won't happen until the transaction commits and sends
> off its I-inserted-lotsa-rows report to the stats collector.  So any
> given large transaction is stuck with the plans it first forms.  There's
> probably nothing we can do about that in time for 8.3, but it's
> something to think about for future releases ...

Ah, *within* a single large transaction :-(  Yeah that's probably not
very solvable for the moment.

-- 
Alvaro Herrera       Valdivia, Chile   ICBM: S 39º 49' 18.1", W 73º 13' 56.4"
"Ninguna manada de bestias tiene una voz tan horrible como la humana" (Orual)