On Tue, Nov 8, 2011 at 1:07 PM, Greg Smith <greg@2ndquadrant.com> wrote:
> On 11/06/2011 11:55 PM, Mark Kirkwood wrote:
>
> I am guessing (at this point very much guessing) that pg_freespace map may
> return its data faster, as pageinspect is gonna have to grovel through all
> the pages for itself (whereas pg_freespacemap relies on using info from the
> ... free space map fork).
>
> I started with pageinspect because I wasn't sure if other methods would be
> as accurate. For example, I wasn't sure until just before submission that
> only the free space and size of the relation are needed to get a useful
> measure here; at one point I was considering some other things too. I've
> ruled them out as unnecessary as far as I can tell, but I can't claim my
> tests are exhaustive. Some review confirmation that this measure is useful
> for other people is one thing I was hoping for feedback on, as one thing to
> consider in addition to the actual implementation.
>
> If this measurement is the only one needed, than as I said at the start of
> the thread here it might easily be implemented to run just against the free
> space map instead. I'm thinking of what's been sent so far as a UI with
> matching output it should produce. If it's possible to get the same numbers
> faster, exactly how to implement the function under the hood is easy enough
> to change. Jaime already has a new version in development that adds a ring
> buffer for example.
It's already easy to get "good enough" numbers based on user space
tools with very little overhead, so I think it's more important that
the server side tool be accurate rather than fast. Of course, if we
can get both, that's a bonus, but I'd rather not go that route at the
expense of accuracy. Just my .02.
Robert Treat
conjecture: xzilla.net
consulting: omniti.com