Excerpts from Tom Lane's message of mié sep 22 12:39:24 -0400 2010:
> OK ... so the problem is that the function uses the pg_settings view,
> which results in calling show_all_settings(), and in particular it
> runs show_session_authorization(), which then dumps core because
> session_authorization_string is NULL in an autovacuum process.
Ugh.
> As far as a fix for the crash goes, I'm not sure if it'd be better to
> try to make show_session_authorization() return some sort of default
> value in this scenario, or to try to ensure that the variable has been
> set to something valid before we start running user-supplied code.
> In either case the problem is potentially wider than this one function
> and variable. Thoughts anyone?
My first thought is that the weird encoding of
session_authorization_string should better be contained in as few places
as possible, so we shouldn't try to initialize it to something
valid-looking. Seems easier to have show_session_authorization() cope
with a NULL value. But what would this default value be?
--
Ãlvaro Herrera <alvherre@commandprompt.com>
The PostgreSQL Company - Command Prompt, Inc.
PostgreSQL Replication, Consulting, Custom Development, 24x7 support