Re: pg_upgrade problem with invalid indexes

On 2012-12-07 13:54:44 -0500, Tom Lane wrote:
> Andres Freund <andres(at)2ndquadrant(dot)com> writes:
> > !indislive indexes are created during DROP INDEX CONCURRENTLY. Thats a
> > different case than CREATE INDEX CONCURRENTLY. Accessing their
> > definition is actually problematic because i can vanish while youre
> > examing it which could cause errors both in the backend and in pg_dump.
>
> That's true of any index, not just !indislive ones. If you're doing DDL
> during a pg_dump, and that makes it fail, you get to keep both pieces.

Uhm. pg_dump is going to lock the tables at start, so I don't really see
anything dangerous happening otherwise? Once it has done so newly
started concurrent CREATE/DROPs won't finish and normal ones can't even
start.

What I am thinking about is parts of the dead indexes pg_attribute
entries vanishing while pg_get_indexdef() is running. That would
probably result in some scary error messages.

Greetings,

Andres Freund

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