Hi Rory,

Do you have a specific catalog server implementation in mind, or a use
case? This could help decide what is the best option moving forward.

It occurs to me that Iceberg Java indeed has a BasicAuthManager, so
Basic auth is "unofficially" supported at least in Java already.

If we really want to mention this new scheme globally, let's at least
make sure it is accompanied by a clear written warning regarding its
security limitations.

Thanks,
Alex

On Wed, Apr 8, 2026 at 2:37 PM Steve Loughran <[email protected]> wrote:
>
>
> why?
>
> It's not "a security scheme" by the modern definition of authentication 
> mechanisms, whose weaknesses after 30 years are well documented.
>
> No defence against MITM
> Transmits credentials over the wire; if stolen are gone until you next roll 
> the secre.
> Doesn't validate message integrity
> etc
>
> About the only thing in its favour is the clients don't need to rely on an 
> library to do the auth as it's built into most http clients. But if you only 
> support basic auth then you can only talk to those catalogs that implemented 
> it, so are either restricted to a subset of catalogs or, if the app turns out 
> to be popular, force all implementations to support an auth mechanism which 
> we all agree is flawed.
>
> I would personally advocate for the # of auth mechanisms to be kept to an 
> absolute minimum, ideally: 1:
>
> Testing interoperability becomes a lot easier as everyone authenticates the 
> same way.
> If a client has an auth problem, it's likely to surface everywhere
>
> i know basic auth is easy to set up so you can deploy a dev system without 
> having to set up a more complex authentication protocol -but requiring those 
> dev systems to use a stricter auth protocol forces everyone to get that 
> implementation right as failures will happen during development, rather than 
> only surfacing in production.
>
>>

Reply via email to