+1 on the new endpoint. I agree that concurrent writes are the sticky part
where we cannot safely use the drop endpoint.

Yufei


On Tue, May 19, 2026 at 5:57 AM Jean-Baptiste Onofré <[email protected]>
wrote:

> +1
>
> It sounds like "logic" to me. Today the catalogs are dealing with that
> "specifically". So something in the spec makes sense.
>
> Regards
> JB
>
> On Mon, May 18, 2026 at 10:50 PM Ryan Blue <[email protected]> wrote:
>
>> Hi everyone,
>>
>> I just opened a PR to add an `unregister` endpoint to the REST spec,
>> #16400 <https://github.com/apache/iceberg/pull/16400>. Unregister is the
>> opposite of `register` and allows you to remove a table from a catalog
>> without deleting its underlying data and metadata files. The purpose is to
>> allow moving from one catalog to another.
>>
>> I think it was just an oversight that we didn't already have this. I know
>> that we've talked about adding a way to unregister a table in the past, but
>> I didn't see a previous email thread so I thought I'd start this one. We
>> originally assumed that the drop table endpoint would work, but we need a
>> different one for two reasons. First, most REST catalogs take care of table
>> cleanup (rather than putting the responsibility on the user or engine at
>> the time DROP is run) and we need a way to prevent files from being
>> removed. Second, DROP doesn't work for concurrent operations. We need to
>> return the current table state and metadata location from the `unregister`
>> operation and have the assurance that it is the latest state from the
>> catalog and that any other operations against the table will fail.
>>
>> Please take a look and reply. If there aren't many objections, I'll try
>> to start a vote thread sometime this week.
>>
>> Thanks,
>>
>> Ryan
>>
>

Reply via email to