Re: Reporting planning time with EXPLAIN

* Ashutosh Bapat (ashutosh(dot)bapat(at)enterprisedb(dot)com) wrote:
> >> One can use this option as
> >> postgres=# explain (summary on) select * from pg_class c, pg_type t
> >> where c.reltype = t.oid;
> >> QUERY PLAN
> >> --------------------------------------------------------------------------
> >> Hash Join (cost=17.12..35.70 rows=319 width=511)
> >> Hash Cond: (c.reltype = t.oid)
> >> -> Seq Scan on pg_class c (cost=0.00..14.19 rows=319 width=259)
> >> -> Hash (cost=12.61..12.61 rows=361 width=256)
> >> -> Seq Scan on pg_type t (cost=0.00..12.61 rows=361 width=256)
> >> Planning time: 48.823 ms
> >> (6 rows)
> >>
> >> When analyze is specified, summary is also set to ON. By default this
> >> flag is OFF.
> >>
> >
> > I am not sure whether using *summary* to print just planning time is a
> > good idea. Another option could be SUMMARY_PLAN_TIME.
>
> I have just used the same name as the boolean which controls the
> printing of planning time. Suggestions are welcome though. We haven't
> used words with "_" for EXPLAIN options, so I am not sure about
> SUMMARY_PLAN_TIME.

Using 'summary' seems entirely reasonable to me, I don't think we need
to complicate it by saying 'summary_plan_time'- I know that I'd had to
have to write that out.

> > + /* Execution time matters only when analyze is requested */
> > + if (es->summary && es->analyze)
> >
> > Do you really need es->summary in above check?

I'm pretty sure we do.

EXPLAIN (ANALYZE, SUMMARY OFF)

Should not print the summary (which means it shouldn't print either the
planning or execution time).

> I think es->summary controls printing overall timing, planning as well
> as execution (hence probably the name "summary").

I am imagining it controlling if we print the summary or not, where the
summary today is the last two lines:

Planning time: 14.020 ms
Execution time: 79.555 ms

Thanks!

Stephen