On Monday, September 26, 2016 9:44 AM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>> Paul Jones <pbj@cmicdo.com> writes:
>> For a freshly pg_restore'd 9.2 database, would VACUUM ANALYZE update
>> statistics any better than just an ANALYZE?
>
> VACUUM would have caused the page-all-visible flags to get set for all
> pages of unchanging tables. I don't recall whether ANALYZE has any side
> effects on those flags at all, but it certainly would not have set them
> for pages it didn't even visit, which would be most.
>
> Net result is that the pg_class.relallvisible fractions didn't get high
> enough to persuade the planner that index-only scans would be effective.
> I guess you could call that a statistic, but it's really about the
> contents of the tables' free space maps.
>
> regards, tom lane
This is good to know. I think we will be running VACUUM ANALYZE from
now on after restore instead of just ANALYZE.
I do note that sect. 49.11 claims that ANALYZE updates
pg_class.relallvisible. I don't know if this is a documentation problem
in light of what you explained.
PJ