st 22. 11. 2023 v 7:20 odesílatel Julien Rouhaud <rjuju123@gmail.com> napsal:
Hi,
On Tue, Oct 17, 2023 at 08:52:13AM +0200, Pavel Stehule wrote: > > When I thought about global temporary tables, I got one maybe interesting > idea. The one significant problem of global temporary tables is place for > storing info about size or column statistics. > > I think so these data can be stored simply in session variables. Any global > temporary table can get assigned one session variable, that can hold these > data.
I don't know how realistic this would be. For instance it will require to properly link the global temporary table life cycle with the session variable and I'm afraid it would require to add some hacks to make it work as needed.
But this still raises the question of whether this feature could be used internally for the need of another feature. If we think it's likely, should we try to act right now and reserve the "pg_" prefix for internal use rather than do that a few years down the line and probably break some user code as it was done recently for the role names?
I don't think it is necessary. Session variables (in this design) are joined with schemas. If we use some session variables for system purposes, we can use some dedicated schema. But when I think about it in detail, probably my own dedicated storage (hash table in session memory) can be much better than session variables. What can be shared (maybe) is probably sinval message processing.