On 11 January 2018 at 04:37, Alvaro Herrera <alvherre@alvh.no-ip.org> wrote:
> In prior incarnations of the patch, I had an if-test to prevent
> attaching invalid indexes, but I decided to remove it at some point --
> mainly thinking of attaching a partition for which a CREATE INDEX
> CONCURRENTLY was running which already had the index as invalid and was
> later expected to become valid. I suppose that doesn't really work
> anyway because of locking considerations (you can't attach a partition
> in which CIC is concurrently running, can you). I'll think some more
> about this case and post an updated version later.
I guess CIC will need to check if the index has a parent index when
setting indisvalid = true, and do likewise to the parent index if all
other siblings are valid.
--
David Rowley http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Training & Services