Re: Add important info about ANALYZE after create Functional Index
От | Tomas Vondra |
---|---|
Тема | Re: Add important info about ANALYZE after create Functional Index |
Дата | |
Msg-id | 20201028185538.fzux2lhim472h46o@development обсуждение исходный текст |
Ответ на | Re: Add important info about ANALYZE after create Functional Index ("David G. Johnston" <david.g.johnston@gmail.com>) |
Ответы |
Re: Add important info about ANALYZE after create Functional Index
Re: Add important info about ANALYZE after create Functional Index |
Список | pgsql-hackers |
On Mon, Oct 26, 2020 at 03:46:10PM -0700, David G. Johnston wrote: >On Mon, Oct 26, 2020 at 3:08 PM Fabrízio de Royes Mello < >fabriziomello@gmail.com> wrote: > >> Hi all, >> >> As you all already know Postgres supports functions in index expressions >> (marked as immutable ofc) and for this special index the ANALYZE command >> creates some statistics (new pg_statistic entry) about it. >> >> The problem is just after creating a new index or rebuilding concurrently >> (using the new REINDEX .. CONCURRENTLY or the old manner creating new one >> and then swapping) we need to run ANALYZE to update statistics but we don't >> mention it in any part of our documentation. >> >> Last weekend Gitlab went down because the lack of an ANALYZE after >> rebuilding concurrently a functional index and they followed the >> recommendation we have into our documentation [1] about how to rebuild it >> concurrently, but we don't warn users about the ANALYZE after. >> >> Would be nice if add some information about it into our docs but not sure >> where. I'm thinking about: >> - doc/src/sgml/ref/create_index.sgml >> - doc/src/sgml/maintenance.sgml (routine-reindex) >> >> Thoughts? >> >> [1] >> https://gitlab.com/gitlab-com/gl-infra/production/-/issues/2885#note_436310499 >> > >It would seem preferable to call the lack of auto-analyzing after these >operations a bug and back-patch a fix that injects an analyze side-effect >just before their completion. It doesn't have to be smart either, >analyzing things even if the created (or newly validated) index doesn't >have statistics of its own isn't a problem in my book. > I agree the lack of stats may be quite annoying and cause issues, but my guess is the chances of backpatching such change are about 0.000001%. We have a usable 'workaround' for this - manual analyze. regards -- Tomas Vondra http://www.2ndQuadrant.com PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services
В списке pgsql-hackers по дате отправления: