On Thu, Feb 19, 2015 at 3:44 PM, Kevin Grittner <kgrittn@ymail.com> wrote:
>> I'm also interested in handling the case Stephen Frost described, where
>> a tuple is effectively dead but we don't currently have the means of
>> discovering the fact, because there is an older long running transaction
>> which is never in fact going to be able to see the tuple.
>
> Absolutely. That's one of several other issues that I've been
> looking at over the last few weeks. It sounds like people are
> already working on that one, which is great. My personal priority
> list included that, but after the two I submitted here and a patch
> to allow combining near-empty btree pages so that btree bloat is
> constrained without having to reindex periodically for the cases
> where index tuples are added in index order (at one or a few
> insertion points) and most-but-not-all are deleted. You can
> currently wind up with a worst-case of one index tuple per page
> with no action to reduce that bloat by vacuum processes.
I'd be willing to test that patch.
I have a big database that does that, and a script that fills the disk
with said bloat. That's forced me to do periodic reindexing, which
sucks.
