On Mon, Dec 3, 2012 at 7:08 AM, Philip Martin <philip.martin_at_wandisco.com>wrote:
> Prompted by a question on users I wondered how SQLite's vacuum
> (http://sqlite.org/lang_vacuum.html) would affect wc.db size. On a
> Subversion trunk working copy I have been using for months the size was
> reduced from 2.3MB to 1.3MB which isn't really a significant change.
>
> For a further test I checked-out a ^/subversion/branches working copy
> for a wc.db of 93MB with 121738 rows, I made it sparse with 66046 rows
> and it was still 93MB, then I ran vacuum and it was reduced to 51MB. I
> have a gcc working copy with some subtrees switched to an empty
> directory. There vacuum reduced wc.db from 47MB to 8.1MB.
> So it appears that vacuum is interesting if the number of rows decreases
> dramatically.
>
> SQLite has auto_vacuum but it comes with a warning that it may make
> fragmentation worse (http://sqlite.org/pragma.html#pragma_auto_vacuum)
> so it's not clear whether we should enable it. Perhaps we should add a
> "vacuum" to cleanup? A full vacuum rewrites all the tables so it's not
> a trivial operation but it is reasonably fast for the working copies on
> local disk that I tried.
>
I think adding "vacuum" to cleanup is a reasonable first step. "cleanup"
is an explicit operation that a user could reasonably expect to take some
non-trivial amount of time. We already remove unneeded pristines during
cleanup, might as well do the same thing with wc.db space.
-Hyrum
Received on 2012-12-03 13:39:34 CET