Re: [PoC] Federated Authn/z with OAUTHBEARER

Hi,

On 2025-03-20 17:08:54 -0400, Tom Lane wrote:
> Bruce Momjian <bruce(at)momjian(dot)us> writes:
> > On Thu, Mar 20, 2025 at 01:33:26PM -0700, Jacob Champion wrote:
> >> So one question for the collective is -- putting Curl itself aside --
> >> is having a basic-but-usable OAuth flow, out of the box, worth the
> >> costs of a generic HTTP client?
>
> > One observation is that security scanning tools are going to see the
> > curl dependency and look at any CSVs related to them and ask us, whether
> > they are using OAUTH or not.
>
> Yes. Also, none of this has addressed my complaint about the extent
> of the build and install dependencies. Yes, simply not selecting
> --with-libcurl removes the problem ... but most packagers are under
> very heavy pressure to enable all features of a package.

How about we provide the current libpq.so without linking to curl and also a
libpq-oauth.so that has curl support? If we do it right libpq-oauth.so would
itself link to libpq.so, making libpq-oauth.so a fairly small library.

That way packagers can split libpq-oauth.so into a separate package, while
still just building once.

That'd be a bit of work on the buildsystem side, but it seems doable.

> From what's been said here, only a small minority of users are likely
> to have any interest in this feature. So my answer to "is it worth
> the cost" is no, and would be no even if I had a lower estimate of
> the costs.

I think this is likely going to be rather widely used, way more widely than
e.g. kerberos or ldap support in libpq. My understanding is that there's a
fair bit of pressure in lots of companies to centralize authentication towards
centralized systems, even for server applications.

> I don't have any problem with making a solution available to those
> users who want it --- but I really do NOT want this to be part of
> stock libpq nor done as part of the core Postgres build. I do not
> think that the costs of that have been fully accounted for, especially
> not the fact that almost all of those costs fall on people other than
> us.

I am on board with not having it as part of stock libpq, but I don't see what
we gain by not building it as part of postgres (if the dependencies are
available, of course).

Greetings,

Andres Freund