Re: Allow ssl_renegotiation_limit in PG 9.5

On 17 October 2015 at 13:27, Andres Freund <andres(at)anarazel(dot)de> wrote:

> On 2015-10-17 12:49:17 +0100, Simon Riggs wrote:
> > Agreed, but I don't like the idea of hardcoding something so horribly
> > specific into the server.
>
> What's that specific about accepting the value for 'disabled' for a
> feature that's not supported anymore?
>

Because we don't do that in any other non-supported feature.

Do I really need to explain why a specific, hardcoded kluge is a bad idea?

> > I'd rather the driver added "driver=npgsql" as an additional parameter in
> > the StartupMessage. We can then make the server run some driver specific
> > logic, rather than hardcoding something that could cause breakage
> > elsewhere. This mechanism would then be extensible to all drivers.
>
> How could this cause breakage alsewhere?
>

Because we are adding code for use by one specific driver, but doing
nothing to ensure it runs only for that driver. We'll forget why we did
this and it could cause breakage elsewhere.

> Having to backpatch a new parameter to all supported versions seems far
> more invasive than adding a guc that can only be set to one value.
>

I doubt it, since as I pointed out the protocol already supports it. The
suggested method is principled and extensible.

--
Simon Riggs http://www.2ndQuadrant.com/
<http://www.2ndquadrant.com/>
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services