Re: Ignore invalid indexes in pg_dump.

On 2013-03-28 17:35:08 -0400, Bruce Momjian wrote:
> On Thu, Mar 28, 2013 at 10:31:51PM +0100, anarazel@anarazel.de wrote:
> > 
> > 
> > Tom Lane <tgl@sss.pgh.pa.us> schrieb:
> > 
> > >Bruce Momjian <bruce@momjian.us> writes:
> > >> Should I just patch pg_upgrade to remove the "indisvalid", skip
> > >> "indisvalid" indexes, and backpatch it?  Users should be using the
> > >> version of pg_upgrade to match new pg_dump.  Is there any case where
> > >> they don't match?  Do I still need to check for "indisready"?
> > >
> > >Yeah, if you can just ignore !indisvalid indexes that should work fine.
> > >I see no need to look at indisready if you're doing that.
> > 
> > You need to look at inisready in 9.2 since thats used for about to be dropped indexes. No?
> 
> Well, if it is dropped, pg_dump will not dump it.  At this point though,
> pg_upgrade is either running in check mode, or it is the only user.  I
> think we are OK.

Its about DROP INDEX CONCURRENTLY which can leave indexes in a partial state
visible to the outside. Either just transiently while a DROP CONCURRENLTY is
going on or even permanently if the server crashed during such a drop or
similar. c.f. index.c:index_drop

Greetings,

Andres Freund

-- Andres Freund                       http://www.2ndQuadrant.com/PostgreSQL Development, 24x7 Support, Training &
Services