Re: Documentation of underlying functions for statistics views

Josh Kupershmidt <schmiddy@gmail.com> writes:
> On Sat, Apr 28, 2012 at 2:24 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>> Currently, some of the per-column entries in monitoring.sgml mention the
>> underlying stats function; but the coverage is incomplete, and it seems
>> very cluttering anyway.  I propose that we remove all that and instead
>> just suggest that people examine the definition of the view if they want
>> to know the names of the underlying functions.

> I did find it marginally useful to see clarifications of milliseconds
> vs. microseconds  (e.g. "block_write_time" column). Yeah, you could
> figure out that based on the " / 1000" in the view description, but
> it'd be a bit less obvious. So maybe keep just those?

There's an existing paragraph that explains that the underlying
functions take backend ID or table/index/function OID.  I was thinking
of expanding that to add the recommendation to look at the view
definitions.  We could also add a sentence there that says that for
timing columns, the underlying function reports integer microseconds
which are converted to milliseconds by the view.  I think that's better
than repeating the info for each such column.

Alternatively, we could bite the bullet and change all those functions
to report float8 msec now, instead of putting that change off ;-).
I'm actually not sure what the point of waiting on such a change is;
it's not going to become any less painful, especially not if we are
encouraging people to develop custom views that use the functions
directly.  But this is the wrong list to debate such a proposal.

            regards, tom lane