Pavel Stehule escribió:
> But question?
>
> about MVCC?
> Is necessary to use MVCC on pg_statistic and some columns from pg_proc?
Right now, we use MVCC everywhere because we have no choice. But there
are some things that would be better off not using MVCC; and indeed we
use a hack to make things that way in certain, very limited cases (see
callers of heap_update_inplace).
Years ago I proposed a patch (search for pg_class_nt and/or pg_ntclass)
that created a catalog for non-MVCC attributes, with an eye on extending
it to more attributes. The use case for those other attributes had
fatal flaws so it went nowhere, but perhaps we can give the idea another
whirl now. For example I think it would be possible to implement
read-only partitions that don't need vacuuming using that idea.
--
Alvaro Herrera http://www.CommandPrompt.com/
The PostgreSQL Company - Command Prompt, Inc.