I've seen a couple of +1s on the spec PR:
https://github.com/apache/iceberg/pull/15746

Shall we move this to a vote?

On Sat, Mar 28, 2026 at 2:21 PM Kevin Liu <[email protected]> wrote:

> We can use `IcebergErrorResponse` to differentiate between route doesnt
> exists (404) and resource/warehouse doesnt exists (404). This is what the
> spec PR describes,
> https://github.com/apache/iceberg/pull/15746/files#diff-02549ca620d020dc9ead80088cc14e311e12a69651fa8d394cd41a4308debb2eR165-R173
>
> For example,
> `google.com/v1/config` <http://google.com/v1/config> doesnt exist, so
> ```
> python3 -c "import urllib.request; urllib.request.urlopen('
> http://google.com/v1/config')"
> ```
> returns `urllib.error.HTTPError: HTTP Error 404: Not Found`
>
> And I would expect an IRC endpoint to return `NoSuchWarehouseError`
> instead, with `"type": "NoSuchWarehouseException"`.
>
> Best,
> Kevin Liu
>
> On Sat, Mar 28, 2026 at 10:08 AM Steven Wu <[email protected]> wrote:
>
>> Chatted with Yufei offline. The `warehouse/catalog` is a hidden concept
>> in the REST spec. If we could redo it, including it in the path (instead of
>> as a query parameter) might make more sense. E.g., The endpoint could look
>> like "/v1/{prefix}/config," where a 404 status would be perfect.
>>
>> Since it is too late to change that, I agree 404 is fine here.
>>
>> On Fri, Mar 27, 2026 at 10:00 AM Ryan Blue <[email protected]> wrote:
>>
>>> I think the difference between those examples and the config route is
>>> that those examples identify resources that do not exist (namespace in both
>>> cases). We also have cases where you could get a 404 indicating a namespace
>>> or a table does not exist, but that indicates that the resource you're
>>> looking for either does not exist (table) or can't exist (namespace
>>> preventing table from being present).
>>>
>>> The config endpoint always exists, which is why this is odd. I think you
>>> could argue that this is okay because it isn't really a resource that has
>>> create/update/delete operations. I just don't know what the "correct" way
>>> to handle this is in REST APIs. But then I've never been one that's too
>>> strict about REST principles.
>>>
>>> On Thu, Mar 26, 2026 at 3:44 PM Yufei Gu <[email protected]> wrote:
>>>
>>>> I'd be cautious about 204 since it indicates a successful response. 404
>>>> seems fine to me. IRC spec uses it in multiple places, like [1] and [2] to
>>>> indicate that certain entities do not exist.
>>>>
>>>> 1.
>>>> https://github.com/apache/iceberg/blob/9534c9b3adc29d127ecc541ce131f49fd72f1980/open-api/rest-catalog-open-api.yaml#L539
>>>> 2.
>>>> https://github.com/apache/iceberg/blob/9534c9b3adc29d127ecc541ce131f49fd72f1980/open-api/rest-catalog-open-api.yaml#L490
>>>>
>>>> Yufei
>>>>
>>>>
>>>> On Thu, Mar 26, 2026 at 2:03 PM Steven Wu <[email protected]> wrote:
>>>>
>>>>>
>>>>> 404 may not be the best fit, as it generally indicates that the
>>>>> endpoint itself could not be found. The endpoint receiving the query
>>>>> parameters exists, and a lack of results is a valid outcome of the
>>>>> search/filter operation, not a client error in forming the request URI.
>>>>>
>>>>> Maybe return 204 No Content as the request itself was valid and
>>>>> successfully processed.
>>>>>
>>>>> On Thu, Mar 26, 2026 at 1:48 PM Ryan Blue <[email protected]> wrote:
>>>>>
>>>>>> This seems reasonable to me. I don't know if 404 is the right
>>>>>> response since the endpoint always exists, but it's fine with me.
>>>>>>
>>>>>> On Wed, Mar 25, 2026 at 6:04 PM Yufei Gu <[email protected]>
>>>>>> wrote:
>>>>>>
>>>>>>> It seems reasonable to add the 404 response. I noticed that the
>>>>>>> warehouse parameter is optional. I assume this is meant for catalog
>>>>>>> implementations that support exactly one catalog or warehouse here so 
>>>>>>> that
>>>>>>> client is OK to skip it, though please correct me if I am mistaken. In 
>>>>>>> that
>>>>>>> case, a 404 would still make sense when that single warehouse is not yet
>>>>>>> ready.
>>>>>>>
>>>>>>> parameters:
>>>>>>>   - name: warehouse
>>>>>>>     in: query
>>>>>>>     required: false
>>>>>>>
>>>>>>>
>>>>>>> Yufei
>>>>>>>
>>>>>>>
>>>>>>> On Tue, Mar 24, 2026 at 8:33 AM Kevin Liu <[email protected]>
>>>>>>> wrote:
>>>>>>>
>>>>>>>> Thanks for raising this proposal! I think it makes sense to add
>>>>>>>> this to the spec and be explicit about the error case. I found the 
>>>>>>>> place
>>>>>>>> where Apache Polaris throws `NotFoundException` for the `/v1/config`
>>>>>>>> endpoint. The specific error `type` field can be used to disambiguate a
>>>>>>>> route 404 (URL doesn't exist) from a resource 404 (URL is valid, but 
>>>>>>>> the
>>>>>>>> server cannot find the warehouse).
>>>>>>>>
>>>>>>>> Best,
>>>>>>>> Kevin Liu
>>>>>>>>
>>>>>>>> [1]
>>>>>>>> https://github.com/apache/polaris/blob/67daa9bb479eaa0ee6c4428984e253afc01b6efd/runtime/service/src/main/java/org/apache/polaris/service/catalog/iceberg/IcebergCatalogHandler.java#L1360
>>>>>>>>
>>>>>>>> On Tue, Mar 24, 2026 at 4:34 AM Oğuzhan Ünlü <
>>>>>>>> [email protected]> wrote:
>>>>>>>>
>>>>>>>>> Hi everyone,
>>>>>>>>>
>>>>>>>>> I'd like to propose a small addition to the REST catalog spec:
>>>>>>>>> documenting HTTP 404 as a valid response for the /v1/config endpoint 
>>>>>>>>> when a
>>>>>>>>> requested warehouse does not exist.
>>>>>>>>>
>>>>>>>>> The Rationale
>>>>>>>>>
>>>>>>>>> The /v1/config endpoint allows an optional query parameter for a
>>>>>>>>> warehouse identifier, e.g. /v1/config?warehouse=mywarehouse.  But the
>>>>>>>>> openapi spec does not specify what should happen if the requested 
>>>>>>>>> warehouse
>>>>>>>>> does not exist.
>>>>>>>>>
>>>>>>>>> Snowflake Open Catalog already returns a 404 for non-existent
>>>>>>>>> warehouses:
>>>>>>>>>
>>>>>>>>> ```
>>>>>>>>> {
>>>>>>>>>   "error": {
>>>>>>>>>     "message": "Unable to find warehouse
>>>>>>>>> NONEXISTENT_WAREHOUSE_12345",
>>>>>>>>>     "type": "NotFoundException",
>>>>>>>>>     "code": 404
>>>>>>>>>   }
>>>>>>>>> }
>>>>>>>>> ```
>>>>>>>>>
>>>>>>>>> This proposal therefore formalizes what Snowflake Open Catalog is
>>>>>>>>> already doing in production. It seems sensible to formalize the 404
>>>>>>>>> response code, because this is consistent with other Iceberg REST 
>>>>>>>>> endpoints
>>>>>>>>> which allow a 404 response code for missing resources (tables, 
>>>>>>>>> namespaces,
>>>>>>>>> views).
>>>>>>>>>
>>>>>>>>> The Proposed Solution (PR-15746)
>>>>>>>>>
>>>>>>>>> Add a NoSuchWarehouseResponse to the OpenAPI spec for the
>>>>>>>>> /v1/config endpoint, formalizing 404 as the response when a warehouse 
>>>>>>>>> does
>>>>>>>>> not exist. You can view the PR here:
>>>>>>>>> https://github.com/apache/iceberg/pull/15746 .
>>>>>>>>>
>>>>>>>>> Looking forward to your thoughts.
>>>>>>>>>
>>>>>>>>> Best,
>>>>>>>>> Oguzhan
>>>>>>>>>
>>>>>>>>

Reply via email to