Thread: Possible interesting extra information for explain analyze?

Possible interesting extra information for explain analyze?

From
Jeff
Date:
Given some recent posts / irc issues with dead tuple bloat..

And given that a lot of these people have at least been smart enough to
explain analyze would it be A. possible B. useful C. None of the above
to have various "scan" nodes of explain analyze also report how many
invisible / dead tuples they had to disqualify (Just to clarify, they
matched the search criteria, but were invisible due to MVCC rules).
Some thing like:

  Seq Scan on boards  (cost=0.00..686.30 rows=25430 width=0) (actual
time=8.866..5407.693 rows=18636 loops=1 invisiblerows=8934983098294)

This may help us to point out tuple bloat issues quicker... or it may
give the developer enough of a clue to search around and find out he
needs to vacuum... hmm.. but once we have an integrated autovacuum it
will be a moot point.....

Also another thing I started working on back in the day and hope to
finish when I get time (that is a funny idea) is having explain analyze
report when a step required the use of temp files.
--
Jeff Trout <jeff@jefftrout.com>
http://www.jefftrout.com/
http://www.stuarthamm.net/


Re: Possible interesting extra information for explain analyze?

From
Tom Lane
Date:
Jeff <threshar@torgo.978.org> writes:
> Given some recent posts / irc issues with dead tuple bloat..
> And given that a lot of these people have at least been smart enough to
> explain analyze would it be A. possible B. useful C. None of the above
> to have various "scan" nodes of explain analyze also report how many
> invisible / dead tuples they had to disqualify (Just to clarify, they
> matched the search criteria, but were invisible due to MVCC rules).

I think this would not help a whole lot because (particularly on
indexscans) you won't get a very accurate picture of the true extent
of bloat.  The contrib/pgstattuple utility is more useful for measuring
that sort of thing.

            regards, tom lane

Re: Possible interesting extra information for explain

From
Simon Riggs
Date:
On Fri, 2005-02-25 at 08:49 -0500, Jeff wrote:
> Also another thing I started working on back in the day and hope to
> finish when I get time (that is a funny idea) is having explain analyze
> report when a step required the use of temp files.

Sounds useful. Please work on it...

Best Regards, Simon Riggs