Re: how much disk space does a VACUUM FULL take?

On Dec 3, 2013, at 3:53 PM, Andreas Brandl <ml(at)3(dot)141592654(dot)de> wrote:

> John,
>
>> ...
>> For a variety of reasons I would prefer disk usage to be as low as
>> possible, thus I would like to run a VACUUM FULL during some
>> maintenance cycle (since it exclusively locks the table). However,
>
> you might want to consider setting fillfactor to 100 [1] to completely compact the table (before doing a VACUUM FULL).
>
> Though I'm not 100% sure but I assume that VACUUM FULL considers the fillfactor when rewriting the table (maybe someone can comment on this?).

In the 9.0 documentation (which we're on) it suggests fillfactor is 100 by default, and I don't believe the table creation overrode these. This likely means that incoming data will attempt to re-use pages with deleted rows? It does seem that our database growth has slowed (as reflected with on-disk space), but it never decreases, which is why I was hoping to VACUUM FULL.

>> given the details of VACUUM FULL:
>>
>>> VACUUM FULL actively compacts tables by writing a complete new
>>> version of the table file with no dead space. This minimizes the
>>> size of the table, but can take a long time. It also requires
>>> extra disk space for the new copy of the table, until the
>>> operation completes.
>>
>> Does this suggest that VACUUM FULL needs free disk space on the order
>> of the full size of the table that it's vacuuming to be able to
>> complete? Or does it / can it write the filesystem files in the 1GB
>> chunks stored in /base while removing the new "unused" files at the
>> same time, thus requiring only a few GB of free space?
>
> AFAIK a VACUUM FULL frees the old data after having completely written the new version. So the size of the original table is an upper bound for the space requirement and it can be much less (in case the original table is bloated a lot).

That's what I'm afraid of. Thanks for the anecdote,

~ john