Re: Clarify "allow_system_table_mods"

David, Melvin,

* David G. Johnston (david(dot)g(dot)johnston(at)gmail(dot)com) wrote:
> On Monday, April 25, 2016, Melvin Davidson <melvin6925(at)gmail(dot)com> wrote:
> > I need clarification on allow_system_table_mods parameter
> > Per the documentation:
> > *Allows modification of the structure of system tables.* This is used by
> > initdb. This parameter can only be set at server start.
> >
> > However, attempting to modify pg_class to add another column fails with
> > "STATUS_ACCESS_VIOLATION" as below.
> > So either only certain system catalogs may be changed, or only certain
> > types of structure changes are allowed.

There might be some things about system catalogs you can modify, but
generally speaking, you can't add a column or otherwise change the
structure. System catalogs are represented in memory by C structures
(and more), so it's not too surprising that adding a column causes a
crash.

> If it's going to prevent the action it should be able to do so without
> crashing the server..and I doubt it's intended to be opinionated aside from
> covering all DDL. What little there is on Google seems to support this.

Sorry, but as a superuser, there's a lot of things you can do to crash
the server, this is just one way. The effort required to prevent
anything bad from happening when a user is running as a superuser is far
from trivial.

In short, no, you can't just add a column to pg_class via SQL, and I
don't think we're going to be very interested in trying to "fix" such
cases.

Thanks!

Stephen