Thanks, Youngrae, for working on it. +1 on the direction. Left some
comments in the PR.
Yufei


On Mon, Nov 16, 2026 at 8:30 PM youngrae kim <[email protected]> wrote:

> Hi all,
>
> As requested by Dmitri in the review, I would like to bring PR #5999 [1] to
> the list, since it changes how the Management API treats
> `connectionConfigInfo.uri`.
>
> The Hadoop federated catalog is initialized from `warehouse` and never uses
> `uri`. Today the field is still required and validated as a `java.net.URL`,
> so users have to put a placeholder there, and values such as
> `hdfs://namenode:8020/warehouse` or `s3://bucket/warehouse` are rejected
> with "400 Invalid remote URI". Omitting it fails with a 500 (#5905).
>
> The PR proposes:
>   - `uri` becomes optional for HADOOP connections and is not validated
> (after Dmitri's comment that not every storage location is a valid URI).
>   - ICEBERG_REST and HIVE still require `uri`, but a missing value now
> returns a 400 with "uri is required for <TYPE> connections" instead of a
> 500. BIGQUERY keeps defaulting to its public endpoint.
>   - The spec descriptions of `uri` and `HadoopConnectionConfigInfo` say
> which connection types require it. The schema itself does not change: `uri`
> was never in a `required` list.
>
> Compatibility: existing catalogs are unaffected. A HADOOP catalog created
> without `uri` cannot be loaded by an older Polaris version, which only
> matters when downgrading.
>
> Feedback welcome, here or on the PR.
> [1] https://github.com/apache/polaris/pull/5999
>
> Regards,
> Youngrae Kim
>

Reply via email to