On Wed, Nov 13, 2019 at 11:39:04AM +0100, Julien Rouhaud wrote:
> (moved to -hackers)
>
> On Tue, Nov 12, 2019 at 9:55 PM Andres Freund <andres@anarazel.de> wrote:
> >
> > This last point is more oriented towards other PG developers: I wonder
> > if we ought to display buffer statistics for plan time, for EXPLAIN
> > (BUFFERS). That'd surely make it easier to discern cases where we
> > e.g. access the index and scan a lot of the index from cases where we
> > hit some CPU time issue. We should easily be able to get that data, I
> > think, we already maintain it, we'd just need to compute the diff
> > between pgBufferUsage before / after planning.
>
> That would be quite interesting to have. I attach as a reference a
> quick POC patch to implement it:
+1
+ result.shared_blks_hit = stop->shared_blks_hit - start->shared_blks_hit;
+ result.shared_blks_read = stop->shared_blks_read - start->shared_blks_read;
+ result.shared_blks_dirtied = stop->shared_blks_dirtied -
+ start->shared_blks_dirtied;
[...]
I think it would be more readable and maintainable using a macro:
#define CALC_BUFF_DIFF(x) result.##x = stop->##x - start->##x
CALC_BUFF_DIFF(shared_blks_hit);
CALC_BUFF_DIFF(shared_blks_read);
CALC_BUFF_DIFF(shared_blks_dirtied);
...
#undefine CALC_BUFF_DIFF