Re: Measuring relation free space

On 12/15/2011 04:11 PM, Noah Misch wrote:
> Is one of those index figures simply wrong, or do they measure two senses of
> free space, both of which are interesting to DBAs?
>

I think the bigger one--the one I was aiming to measure--also includes
fill-factor space. It should be possible to isolate whether that's true
by running the function against a fresh index, or by trying tests with a
table where there's no useful fill. I need to add some of those to the
test example suite.

While in theory both measures of free space might be interesting to
DBAs, I'd prefer to have the one that reflects the lost space to
fill-factor if I'm checking an index. But as Robert Treat was pointing
out, even the very rough estimates being made with existing user-space
tools not using the contrib module features are helpful enough for a lot
of users. So long as it's easy and accuracy is good enough to find
bloated indexes, either implementation is probably good enough.

Shaking out the alternate implementation ideas was really my goal for
this CF here. The major goal of the next revision is to present the
options with a measure of their respective accuracy and runtime. If I
have to give up just a of bit of accuracy and make it much faster,
that's probably what most people want as an option. When Jaime and I
come back with an update, it really needs to have benchmarks and
accuracy numbers for each option. That may be complicated a bit
depending on how much of the table or index is cached, so isolating that
out will be a pain.

--
Greg Smith 2ndQuadrant US greg(at)2ndQuadrant(dot)com Baltimore, MD
PostgreSQL Training, Services, and 24x7 Support www.2ndQuadrant.us