Re: Why is vacuum_freeze_min_age 100m?

On Thu, Aug 13, 2009 at 5:15 PM, Josh Berkus<josh@agliodbs.com> wrote:
> Robert,
>
>> Ah.  Yeah, I agree with Tom: how would it help to make this smaller?
>> It seems like that could possibly increase I/O, if the old data is
>> changing at all, but even if it doesn't it I don't see that it saves
>> you anything to freeze it sooner.
>
> Before 8.4, it actually does on tables which are purely cumulative
> (WORM).  Within a short time, say, 10,000 transactions, the rows to be
> frozen are still in the cache.  By 100m transactions, they are in an
> archive partition which will need to be dragged from disk.  So if I know
> they won't be altered, then freezing them sooner would be better.
>
> However, I can easily manage this through the autovacuum settings.  I
> just wanted confirmation of what I was thinking.

Interesting.  Thanks for the explanation.

...Robert