Tomas Vondra <tomas(dot)vondra(at)2ndquadrant(dot)com> writes:
> Yeah, I agree #1 seems like the cleanest/best option. Are you worried
> about the overhead due to the extra complexity, or overhead due to
> cache getting invalidated for this particular reason?
The overhead is basically a hash_seq_search traversal over the typcache
each time we get a pg_type inval event, which there could be a lot of.
On the other hand we have a lot of inval overhead already, so this might
not amount to anything noticeable.
regards, tom lane