> > > > On the other hand, it looks appealing to make index_set_state_flags() > transactional. This would solve the consistency problem, and looking > at the code scanning pg_index, I don't see a reason why we could not > do that. What do you think? >
TBH, I am not sure. I think it is a reasonable change. It is even indicated in the comment above index_set_state_flags() that it can be made transactional. At the same time, probably we can just go ahead with current inconsistent update of relisreplident and indisvalid flags. Can't see what will break with that. Thank you, -- Rahila Syed Performance Engineer 2ndQuadrant http://www.2ndQuadrant.com <http://www.2ndquadrant.com/> PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services
