Re: Displaying accumulated autovacuum cost

On Sat, Nov 26, 2011 at 10:10 AM, Robert Haas <robertmhaas(at)gmail(dot)com> wrote:
> On Fri, Nov 25, 2011 at 11:39 AM, Alvaro Herrera
> <alvherre(at)commandprompt(dot)com> wrote:
>> I'm going to push this now anyway, thanks.
>
> This patch adds a count of the number of buffers dirtied to VACUUM,
> but it strikes me that it would be useful to add similar tracking to
> pgBufferUsage.  Attached is a patch for that.  You can see the new
> counters through pg_stat_statements or with EXPLAIN (ANALYZE,
> BUFFERS).  This is useful because the number of buffers that a query
> *writes* doesn't necessarily have much to do with anything - it may
> end up writing buffers dirtied by other queries while being read-only
> itself, or conversely it may not write anything at all even though it
> dirties quite a bit.
>
> Thoughts?  Comments?  Objections?

Here are review comments:

The document about EXPLAIN needs to be updated.

You forgot to add the long-integer-valued property of shared/local_blks_dirtied.
So when I ran EXPLAIN and used json as a format, no information about
blks_dirtied
was reported.

Regards,

--
Fujii Masao
NIPPON TELEGRAPH AND TELEPHONE CORPORATION
NTT Open Source Software Center