[pgpool-hackers: 192] Re: pgpool stucks in pgpool_regclass() call

[Tatsuo Ishii]

> Recently we have multiple reports complaining that pgpool-II stucks
> and PostgreSQL is left in parsing SELECT pgpool_regclass.
> 
> Today in bug report http://www.pgpool.net/mantisbt/view.php?id=53 ,
> the reporter identified that that was a problem with user side.  They
> do have pgpool_regclass() but they do *not* have execute privilege on
> the function. I know
> 
> You can reproduce the problem by revoking execute privilege from a
> user something like this:
> 
> createuser foo
> revoke revoke all on function pgpool_regclass(cstring) from public;
> grant execute on function pgpool_regclass(cstring) to postgres; 
> 
> The apparent solution is before calling pgpool_regclass(), make sure
> that the user has a privilege to execute the function. This is easy
> but will add lots of overhead. Checking the existence is managed by
> relcache module, and it is not a problem. However checking the
> privilege would be a BIG performance problem.
> 
> So I lean to the idea that if calling pgpool_regclass() fails, just
> report it and exit the session.
> 
> Thoughts?

After thinking more, I decided to employ this solution because we can
avoid the overhead by caching has_function_privilege() for each user.

> The apparent solution is before calling pgpool_regclass(), make sure
> that the user has a privilege to execute the function. This is easy
> but will add lots of overhead. Checking the existence is managed by
> relcache module, and it is not a problem. However checking the
> privilege would be a BIG performance problem.

Fix committed to master and all V3 branches.
--
Tatsuo Ishii
SRA OSS, Inc. Japan
English: http://www.sraoss.co.jp/index_en.php
Japanese: http://www.sraoss.co.jp