Hi Prithvi,

Thanks for starting this thread!

Your proposal and use case look very reasonable to me.

I left some non-blocking enhancement comments in GH. I hope you find them
useful.

Cheers,
Dmitri.

On Fri, Nov 13, 2026 at 6:10 PM Prithvi S <[email protected]>
wrote:

> Hi Yufei,
>
> Thanks for taking a look.
>
> The gap this is meant to close is Iceberg REST catalogs that speak
> SigV4 but have no IAM role and no STS endpoint. Typical cases are an
> on-prem or private-cloud REST catalog, or anything behind a SigV4
> front door that only accepts a static key pair. Today the only SigV4
> path is AssumeRole, so those catalogs cannot be federated.
>
> The same field also works for AWS services (Glue, S3 Tables,
> execute-api). storageName means "sign with these named keys, skip
> STS". Glue continues to use roleArn as it does today; storageName is
> there if an operator already has a dedicated IAM user in server
> config and would rather not assume a role.
>
> I've updated the spec and docs on the PR so they describe that
> signing mode. Server config and the named-credentials flag are off
> the REST spec. The flag is still checked at catalog create, now
> through CatalogConfigValidator, as Dimitri asked after #5788.
>
> Setting both storageName and roleArn is still rejected in this change.
> Dimitri suggested that combination could later mean "use the named keys
> as the STS client credentials". I'd like to keep that as a follow-up
> so this PR stays the no-STS path.
>
> Thanks,
> Prithvi
>
> On Sat, Nov 14, 2026 at 1:21 AM Yufei Gu <[email protected]> wrote:
>
> > Thanks for raising this, Prithvi. Sorry for the late reply. The
> > approach looks good to me. Would you mind elaborating the use cases?
> > Is this for a non-AWS catalog with sigV4 capabilities?
> >
> > Yufei
> >
> > On Sat, Oct 31, 2026 at 1:32 PM Prithvi S <[email protected]>
> > wrote:
> > >
> > > Hi all,
> > >
> > > I'd like to gather feedback on the Management API change in #5664 [1]
> > > before it lands.
> > >
> > > The PR adds an optional storageName to SigV4AuthenticationParameters.
> > > It references an AWS key pair already configured on the server via
> > > polaris.storage.aws.<name>.access-key and
> > > polaris.storage.aws.<name>.secret-key, following the same naming
> > > convention as RESOLVE_CREDENTIALS_BY_STORAGE_NAME.
> > >
> > > The catalog stores only the storageName reference. The access andsecret
> > > keys are never passed through the API or persisted in thesecrets
> manager.
> > > When storageName is set, roleArn is not required. The server
> > signsrequests
> > > directly with the configured keys, without going through STS.
> > > Setting both storageName and roleArn is rejected for now.
> > I'veintentionally
> > > left that combination out of the spec so it can latersupport using the
> > > configured keys as base credentials for assumingthe roleArn.
> > > The feature flag is checked both at catalog creation and atcredential
> > > vending time.
> > >
> > > Does this approach sound right? Feedback on the API shape and the
> > > validation behaviour would be especially helpful.
> > >
> > > Thanks,
> > > Prithvi
> > >
> > > [1] https://github.com/apache/polaris/pull/5664
> >
>

Reply via email to