Re: Cursor With_Hold Performance Workarounds/Optimization

> When we declare a cursor for a select on the mentioned big table, it takes
> multiple seconds and a big temp file is created which to me seems like the
> materialization took place immediately.

Since you mentioned, Postgres already postponed materialization until commit
operations we checked again and you were right. When we manually checked, we
executed a "declare" statement without opening a transaction block first, which
causes instant materialization. When a transaction is opened, it is in fact
postponed.
Unfortunately we do not use cursors but portals so we can (re-)use prepared
statements in multiple instances and there is not "hold" feature for portals so
we cannot benefit from the lazy "with hold" of the cursor.