Re: Database size growing over time and leads to performance impact

On Mar 31, 2010, at 1:47 PM, Robert Haas wrote:

> On Wed, Mar 31, 2010 at 4:37 PM, Scott Carey <scott(at)richrelevance(dot)com> wrote:
>> On Mar 27, 2010, at 6:35 AM, Andy Colson wrote:
>>>
>>> Dont "VACUUM FULL", its not helping you, and is being removed in newer versions.
>>>
>>
>> Off topic: How is that going to work? CLUSTER doesn't work on tables without an index. I would love to be able to CLUSTER on some column set that doesn't necessarily have an index.
>
> I believe the new VF implementation just rewrites the data in the same
> physical order as it was in previously, but without the dead space.
> So it's sort of like cluster-by-no-index-at-all.
>

Still off topic:

Will CLUSTER/VF respect FILLFACTOR in 9.0?

As far as I can tell in 8.4, it does not. CLUSTER on a table with FILLFACTOR=100, then alter the table to FILLFACTOR=90, cluster again -- the file size reported by \dt+ is the same. This is a fairly big performance issue since it means that HOT doesn't function well on a table just CLUSTERed.

> ...Robert