Privithvi, good catch on NAMESPACE_FULL_METADATA or CATALOG_FULL_METADATA! Fixed in a new commit. Would you mind also taking a look at the PR?
Yufei On Sun, Sep 13, 2026 at 5:14 AM Prithvi S <[email protected]> wrote: > Hi Yufei, > > Thanks for sending this to the list. Dedicated SEMANTIC_MODEL_* privileges > look right to me, and the names match the policy analog well (READ/WRITE > rather than READ_PROPERTIES/WRITE_PROPERTIES). > > Two things I'd keep explicit: > > - Model privileges should stay independent of table/view privileges in both > directions. Deferring source-access checks is fine. > - CATALOG_MANAGE_CONTENT and CATALOG_MANAGE_METADATA covering semantic > models is the right compatibility story. I wouldn't also treat > NAMESPACE_FULL_METADATA or CATALOG_FULL_METADATA as umbrellas here, those > mean "full metadata of the namespace/catalog entity" today, and > tables/views/policies don't inherit that way. > > The grant type and the LIST/CREATE split onto namespace/catalog scope look > good. > > WDYT? > > Thanks, > Prithvi S > > On Sat, Sep 12, 2026 at 5:41 AM Yufei Gu <[email protected]> wrote: > > > Hi everyone, > > > > I'd like to get feedback on the API changes in PR #5492 ( > > https://github.com/apache/polaris/pull/5492), as suggested during > review. > > > > Today, all semantic model operations require a single privilege, > > CATALOG_MANAGE_CONTENT. This PR adds dedicated privileges for list, > create, > > read, write, drop, full metadata access, and grant management. This > enables > > finer grant access control, similar to what we implemented for Iceberg > > tables/views and Polaris policies. > > > > Any thoughts on the API shape or privilege names? > > > > Yufei > > >
