[BUG][Client] get_dataset is ommitting part of the path in the URL

[carlosmagan]

Describe the bug
There is a behavioural difference on how to access the URLs in client.get_dataset, client._get_dataset_id_by_name and client.get_dataset_versions, specifically the step where the url is generated.

In get_dataset we have
urljoin(self._base_url, f"/v1/datasets/{quote(id)}/examples")

In _get_dataset_id_by_name we have
urljoin(self._base_url, "/v1/datasets")

In get_dataset_versions we have
url = urljoin(self._base_url, f"v1/datasets/{dataset_id}/versions")

The correct one is the one in get_dataset_versions.

The '/' before v1 implies that the relative url is absolute and must follow the domain at a root level. This is a problem when the base url is something on the line of https://base_domain.com/phoenix because the /phoenix is removed resulting in a final url like https://base_domain.com/v1/datasets/example_id/examples instead of the correct one https://base_domain.com/phoenix/v1/datasets/example_id/examples

To Reproduce
Steps to reproduce the behavior:

1. Deploy phoenix on an endpoint whose url has some kind of path after the domain
2. Upload a dataset
3. Instantiate a phoenix client in a notebook
4. query the dataset with the get_dataset method

Expected behavior
The path for get_dataset should include the complete base_path
Screenshots

[axiomofjoy]

@carlosmagan Thanks for reporting and good find, we'll take a look.

[axiomofjoy]

Great catch @carlosmagan, thanks for the detailed explanation.

[axiomofjoy]

This is resolved in arize-phoenix 4.20.0

[carlosmagan]

Thanks a lot for the quick fix