On 2018-Jul-04, Andrey Borodin wrote:
> Thanks, Tom!
>
> So, I can create the script for pg_upgrade that will walk through each old enough[0] GiST index, scan for invalid
tuplesand repair them. This procedure seems quite trivial, but there will be more code that we have now. Does it sound
reasonable?
>
> [0] Actually, I do not know how to understand which index is old enough.
Requiring a scan of all indexes during pg_upgrade might increase the
upgrade time prohibitively for some sites, so I don't think that's a
good solution.
I think keeping track of which indexes might be old enough not to have
invalid tuples anymore is a good idea in the long run. If we start
doing it in pg12, then by the time pg17 comes about and we abandon pg11
(the one without the cataloguing) then we can retire the code to support
invalid tuples. Most people, come this point, say "naaah this too long,
this project is useless" so the cataloguing is never done :-)
*If* we make it to 2023, that is.
--
Álvaro Herrera https://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services