Thread

Re: postgres_fdw: use_scram_passthrough on user mapping is ignored when also set on server

Fujii Masao <masao.fujii@gmail.com> — 2026-05-23T01:58:40Z

On Mon, May 18, 2026 at 11:42 PM Matheus Alcantara
<matheusssilv97@gmail.com> wrote:
> I think that the test is worth to have to avoid such issues in the
> future again, but I agree that adding as a TAP test is overkill. I've
> moved to sql/dblink.sql on the new attached version.
>
> Do you think that we need to add such test for postgres_fdw too?

No, I don't think so.

In dblink, use_scram_passthrough is handled by a dblink-specific validation
path, and patch 0003 fixes that specific code path. So it might be worth
adding such test for dblink.

In postgres_fdw, however, this option is handled by the generic option/context
validation machinery, not by any special-case code. We already have tests for
context-sensitive option validation there. So adding such test for postgres_fdw
would be mostly redundant, I think. Thought?


> Yeah, we don't have any documentation about this on 18, but we do have
> for sslkey and sslcert where in postgres-fdw.sgml we have the following:
>     sslkey and sslcert - these may appear in either or both a connection
>     and a user mapping. If both are present, the user mapping setting
>     overrides the connection setting.
>
> So I think that is desirable to have the same behavior for
> use_scram_passthrough.

Agreed. I'm therefore thinking of backpatching patches 0001 and 0002 to v18.

Even in v18, this change is very narrow. It only affects cases where
use_scram_passthrough is specified at both the server and user-mapping levels
with conflicting values. In such cases, the natural interpretation is that
the user-mapping setting is intended to override the server-level setting;
otherwise, there would be little reason to specify conflicting values.

Given the limited impact, leaving v18 as the only branch with different
semantics for a feature introduced in v18 seems more undesirable and
potentially confusing. So I think backpatching to v18 makes sense.

Regards,

-- 
Fujii Masao