fix(inference_endpoints): use GET healthRoute instead of GET / to check status (#3165)

* fix(inference_endpoints): use GET /health instead of GET / to check status

* misc(quality): format

* feat(inference_endpoints): use healthRoute from the API to retrieve the path where to query endpoint status

* feat(inference_endpoints): missing health_url in classdef

* feat(inference_endpoints): wrong variable name ... 🤦🏻‍♂️

* feat(inference_endpoints): need coffee at this stage.

* feat(inference_endpoints): fix remaining tests

* feat(inference_endpoints): address comments use health_route and compute _health_url

* feat(inference_endpoints): move _health_url computation when endpoint is running and we have an actual url

* Apply suggestions from code review

* add test

* Update src/huggingface_hub/_inference_endpoints.py

---------

Co-authored-by: Lucain <lucain@huggingface.co>
Co-authored-by: Lucain Pouget <lucainp@gmail.com>

Author: mfuntowicz

src/huggingface_hub/_inference_endpoints.py

@@ -100,6 +100,7 @@ class InferenceEndpoint:
     namespace: str
     repository: str = field(init=False)
     status: InferenceEndpointStatus = field(init=False)
+    health_route: str = field(init=False)
     url: Optional[str] = field(init=False)
 
     # Other fields
@@ -220,7 +221,8 @@ def wait(self, timeout: Optional[int] = None, refresh_every: int = 5) -> "Infere
                 )
             if self.status == InferenceEndpointStatus.RUNNING and self.url is not None:
                 # Verify the endpoint is actually reachable
-                response = get_session().get(self.url, headers=self._api._build_hf_headers(token=self._token))
+                _health_url = f"{self.url.rstrip('/')}/{self.health_route.lstrip('/')}"
+                response = get_session().get(_health_url, headers=self._api._build_hf_headers(token=self._token))
                 if response.status_code == 200:
                     logger.info("Inference Endpoint is ready to be used.")
                     return self
@@ -400,6 +402,7 @@ def _populate_from_raw(self) -> None:
         self.repository = self.raw["model"]["repository"]
         self.status = self.raw["status"]["state"]
         self.url = self.raw["status"].get("url")
+        self.health_route = self.raw["healthRoute"]
 
         # Other fields
         self.framework = self.raw["model"]["framework"]

tests/test_hf_api.py

@@ -4521,6 +4521,7 @@ def test_create_inference_endpoint_from_catalog(self, mock_get_session: Mock) ->
                 },
                 "name": "llama-3-2-3b-instruct-eey",
                 "provider": {"region": "us-east-1", "vendor": "aws"},
+                "healthRoute": "/health",
                 "status": {
                     "createdAt": "2025-03-07T15:30:13.949Z",
                     "createdBy": {"id": "6273f303f6d63a28483fde12", "name": "Wauplin"},
@@ -4650,6 +4651,7 @@ def test_create_inference_endpoint_custom_image_payload(
         },
         "name": "llama-3-2-3b-instruct-eey",
         "provider": {"region": "us-east-1", "vendor": "aws"},
+        "healthRoute": "/health",
         "status": {
             "createdAt": "2025-03-07T15:30:13.949Z",
             "createdBy": {"id": "6273f303f6d63a28483fde12", "name": "Wauplin"},

tests/test_inference_endpoints.py

@@ -19,6 +19,7 @@
     "type": "protected",
     "accountId": None,
     "provider": {"vendor": "aws", "region": "us-east-1"},
+    "healthRoute": "/health",
     "compute": {
         "accelerator": "cpu",
         "instanceType": "intel-icl",
@@ -51,6 +52,7 @@
     "type": "protected",
     "accountId": None,
     "provider": {"vendor": "aws", "region": "us-east-1"},
+    "healthRoute": "/health",
     "compute": {
         "accelerator": "cpu",
         "instanceType": "intel-icl",
@@ -84,6 +86,7 @@
     "type": "protected",
     "accountId": None,
     "provider": {"vendor": "aws", "region": "us-east-1"},
+    "healthRoute": "/health",
     "compute": {
         "accelerator": "cpu",
         "instanceType": "intel-icl",
@@ -116,6 +119,7 @@
     "type": "protected",
     "accountId": None,
     "provider": {"vendor": "aws", "region": "us-east-1"},
+    "healthRoute": "/health",
     "compute": {
         "accelerator": "cpu",
         "instanceType": "intel-icl",
@@ -158,6 +162,7 @@ def test_from_raw_initialization():
     assert endpoint.revision == "11c5a3d5811f50298f278a704980280950aedb10"
     assert endpoint.task == "text-generation"
     assert endpoint.type == "protected"
+    assert endpoint.health_route == "/health"
 
     # Datetime parsed correctly
     assert endpoint.created_at == datetime(2023, 10, 26, 12, 41, 53, 263078, tzinfo=timezone.utc)
@@ -197,6 +202,7 @@ def test_get_client_ready():
     # Endpoint is ready
     assert endpoint.status == "running"
     assert endpoint.url == "https://vksrvs8pc1xnifhq.us-east-1.aws.endpoints.huggingface.cloud"
+    assert endpoint.health_route == "/health"
 
     # => Client available
     client = endpoint.client
@@ -218,6 +224,7 @@ def test_fetch(mock_get: Mock):
 
     assert endpoint.status == "running"
     assert endpoint.url == "https://vksrvs8pc1xnifhq.us-east-1.aws.endpoints.huggingface.cloud"
+    assert endpoint.health_route == "/health"
 
 
 @patch("huggingface_hub._inference_endpoints.get_session")
@@ -245,6 +252,11 @@ def test_wait_until_running(mock_get: Mock, mock_session: Mock):
     assert endpoint.status == "running"
     assert len(mock_get.call_args_list) == 6
 
+    # Ensure the health route has been called
+    assert mock_session.return_value.get.call_count == 2
+    for call in mock_session.return_value.get.call_args_list:
+        assert call[0][0] == "https://vksrvs8pc1xnifhq.us-east-1.aws.endpoints.huggingface.cloud/health"
+
 
 @patch("huggingface_hub.hf_api.HfApi.get_inference_endpoint")
 def test_wait_timeout(mock_get: Mock):