Re: Configuration Parameter/GUC value validation hook

On Mon, May 2, 2022 at 10:54 AM Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
> Bharath Rupireddy <bharath(dot)rupireddyforpostgres(at)gmail(dot)com> writes:
> > I propose to add a simple new hook in set_config_option (void
> > set_config_option_hook(struct config_generic *record);) and the
> > vendors can implement their own platform-dependent extensions to
> > accept or reject certain parameters (based on platform/VM
> > configuration) that are of interest to them.
>
> This seems entirely useless. Vendors are unlikely to have any better
> idea than we do about what are "reasonable" values. Moreover, if they
> did, modifying the source code directly would be an easier route to
> introducing their code than making use of a hook (which'd require
> finding a way to ensure that some extension is loaded).

I don't think we should be in the business of encouraging vendors to
fork the source code, and I think it is quite likely that a vendor
will have better ideas than we do about what values they want to allow
their users to set. It's far easier to know what is reasonable in a
particular context than it is to make a statement about reasonableness
in general.

I have some desire here to see us solve this problem not just for
service providers, but for users in general. You don't have to be a
service provider to want to disallow SET work_mem = '1TB' -- you just
need to be a DBA on a system where such a setting will cause bad
things to happen. But, if you are a DBA on some random system, you
won't likely find a hook to be a particularly useful way of
controlling this sort of thing. I feel like Alice wants to do
something like GRANT work_mem BETWEEN '1MB' AND '2GB' to bob, not that
I'm proposing that particular syntax. I also don't have a clear idea
for what to do about GUCs where a range constraint isn't useful. One
could allow a list of permissible values, but that might not be
powerful enough. One could maybe allow a PL validation function for a
value, but that might be complicated to make work.

In the end, I don't think providing a hook here is particularly
unreasonable. I would be a little sad if it ended up being the only
thing we provided, but I'm also not a huge believer in the idea of
forcing people to write the patch that I want written as a condition
of writing the patch that they want written.

--
Robert Haas
EDB: http://www.enterprisedb.com