We have discussed this yesterday in the catalog sync from *44:00 to 1:03:36*
https://drive.google.com/file/d/1SMOlxus5C92FvW9T4LOXy8Ue_9cmPLyX/view?usp=sharing&t=2643

IIRC, the conclusion was that each catalog may handle namespaces
differently, with some supporting multi-level namespaces and others not.
Iceberg does not intend to standardize this behavior, so it is up to each
catalog to update their documentation regarding their specific behavior.

- Ajantha


On Tue, Jul 16, 2024 at 3:12 PM Jean-Baptiste Onofré <j...@nanthrax.net>
wrote:

> Hi
>
> As namespace is not part of the spec, each REST server implementation
> can use a different logic from another one.
> The TCK will be able to verify the most abstract operations on
> namespace (common denominator on namespace behavior).
>
> Regards
> JB
>
> On Tue, Jul 16, 2024 at 11:11 AM Yong Zhang <zhangyong1025...@gmail.com>
> wrote:
> >
> > It seems TCK will make sure the logic of the REST is consistent. So in
> > the future iceberg will make all things align.
> >
> > We can merge the current PR to avoid a burden on the user. WDYT?
> >
> > Thanks!
> > Yong
> >
> > On Thu, 11 Jul 2024 at 01:04, Ajantha Bhat <ajanthab...@gmail.com>
> wrote:
> >>
> >> Thanks for initiating this discussion.
> >>
> >> I suggested moving it to the mailing list because the javadoc of
> SupportsNamespaces and the REST API spec don't clearly define how to handle
> missing parent namespaces.
> >>
> >> I'm generally +1 of the idea to automatically create missing namespaces
> now, as it reduces the burden on the user. However, I believe this approach
> should be standardized and adopted by all catalogs that support multi-level
> namespaces to avoid confusion.
> >> That is we need to update the java doc and REST API spec descriptions
> for this behavior once we have consensus.
> >>
> >> - Ajantha
> >>
> >> On Wed, Jul 10, 2024 at 7:44 PM Renjie Liu <liurenjie2...@gmail.com>
> wrote:
> >>>
> >>> Hi, Yong:
> >>>
> >>> Thanks for reporting this.
> >>>
> >>>> Do you think iceberg needs to standardize the REST spec to make sure
> the
> >>>> logic is the same for using the SupportsNamespace interface?
> >>>
> >>>
> >>> Yes, I think it's valuable. I remember @Jean-Baptiste Onofré had a
> proposal for TCK of rest catalog, and this case shows that TCK has a lot of
> merit.
> >>>
> >>> On Wed, Jul 10, 2024 at 5:47 PM Yong Zhang <zhangyong1025...@gmail.com>
> wrote:
> >>>>
> >>>> Hi all,
> >>>>
> >>>> When I was using the SupportsNamespaces to do the namespace
> >>>> create and exists check with other catalog services that implement
> >>>> the Iceberg REST API, I found it has different results with different
> >>>> catalog services. Iceberg has provided a standard API to do the
> >>>> namespace operations, but seems iceberg does not have a standard
> >>>> spec on the REST API implementation. That makes the other catalog
> >>>> services implement differently and then breaks the SupportsNamespace
> >>>> interface usage.
> >>>>
> >>>> Do you think iceberg needs to standardize the REST spec to make sure
> the
> >>>> logic is the same for using the SupportsNamespace interface?
> >>>> Otherwise, the SupportsNamespace interface is meaningless because the
> client
> >>>> needs to handle different logic to create a namespace or other
> operations
> >>>> when using different catalog.
> >>>>
> >>>> For example, Nessie doesn't support auto-creating parent namespace:
> >>>> https://github.com/apache/iceberg/pull/10630
> >>>> But the JDBC catalog supports it.
> >>>>
> >>>> Thanks.
> >>>> Yong
> >>>>
>

Reply via email to