fix: DIA-1061: Start Training should send START_TRAINING event as web…

…hook instead of PROJECT_UPDATE (#5761)

#### Change has impacts in these area(s)
_(check all that apply)_
- [ ] Product design
- [ ] Backend (Database)
- [x] Backend (API)
- [ ] Frontend



### Describe the reason for change
Start Training button on Model settings page sends PROJECT_UPDATE event
and it's not possible to distinguish when Start Training really was
called.



#### What does this fix?
ML backend model method fit(..., event, ...) will be able to distinguish
the right training event.


#### What is the current behavior?
ML backend model method fit(..., event, ...) receives PROJECT_UPDATE as
event and this can be triggered on any project change event.

docs/source/guide/webhook_reference.md

@@ -712,3 +712,24 @@ You must [enable organization-level webhooks](webhooks.html#Enable-organization-
 ```
 
 
+### Start Training 
+
+This webhook is triggered when a user clicks `Start Training` button on the ML Model card in the Project Settings page.
+This event will be sent to the ML Backend and can be caught in the model.fit(event, ...) method:
+
+```
+class MyModel(LabelStudioMLBase):
+  def fit(self, event, *args, **kwargs):
+    if event == 'START_TRAINING': 
+      ...
+```
+
+### Webhook payload details
+
+| Key | Type | Description                           |
+| --- | --- |---------------------------------------| 
+| action | string | Name of the action: `START_TRAINING`. | 
+| id | integer | ID of the project where training is started. |
+| project | JSON dictionary | All fields related to the project that was updated. See the [API documentation for updating a project](/api#operation/api_projects_partial_update). |
+
+

label_studio/ml/api_connector.py

@@ -191,7 +191,7 @@ def train(self, project, use_ground_truth=False):
         # Identify if feature flag is turned on
         if flag_set('ff_back_dev_1417_start_training_mlbackend_webhooks_250122_long', user):
             request = {
-                'action': 'PROJECT_UPDATED',
+                'action': 'START_TRAINING',
                 'project': load_func(settings.WEBHOOK_SERIALIZERS['project'])(instance=project).data,
             }
             return self._request('webhook', request, verbose=False, timeout=TIMEOUT_PREDICT)

label_studio/tests/webhooks/test_webhooks.py

@@ -32,6 +32,17 @@ def project_webhook(configured_project):
     )
 
 
+@pytest.fixture
+def ml_start_training_webhook(configured_project):
+    organization = configured_project.organization
+    uri = 'http://0.0.0.0:9090/webhook'
+    return Webhook.objects.create(
+        organization=organization,
+        project=configured_project,
+        url=uri,
+    )
+
+
 @pytest.mark.django_db
 def test_run_webhook(setup_project_dialog, organization_webhook):
     webhook = organization_webhook
@@ -390,3 +401,42 @@ def test_webhooks_for_tasks_from_storages(configured_project, business_client, o
     assert r.json()['action'] == WebhookAction.TASKS_CREATED
     assert 'tasks' in r.json()
     assert 'project' in r.json()
+
+
+@pytest.mark.django_db
+def test_start_training_webhook(setup_project_dialog, ml_start_training_webhook, business_client):
+    """
+    1. Setup: The test uses the project_webhook fixture, which assumes that a webhook
+    is already configured for the project.
+    2. Mocking the POST Request: The requests_mock.Mocker is used to mock
+    the POST request to the webhook URL. This is where you expect the START_TRAINING action to be sent.
+    3. Making the Request: The test makes a POST request to the /api/ml/{id}/train endpoint.
+
+    Assertions:
+        - The response status code is checked to ensure the request was successful.
+        - It verifies that exactly one request was made to the webhook URL.
+        - It checks that the request method was POST.
+        - The request URL and the JSON payload are validated against expected values.
+    """
+    from ml.models import MLBackend
+
+    webhook = ml_start_training_webhook
+    project = webhook.project
+    ml = MLBackend.objects.create(project=project, url='http://0.0.0.0:9090')
+
+    # Mock the POST request to the ML backend train endpoint
+    with requests_mock.Mocker(real_http=True) as m:
+        m.register_uri('POST', webhook.url)
+        response = business_client.post(
+            f'/api/ml/{ml.id}/train',
+            data=json.dumps({'action': 'START_TRAINING'}),
+            content_type='application/json',
+        )
+
+    assert response.status_code == 200
+    request_history = m.request_history
+    assert len(request_history) == 1
+    assert request_history[0].method == 'POST'
+    assert request_history[0].url == webhook.url
+    assert 'project' in request_history[0].json()
+    assert request_history[0].json()['action'] == 'START_TRAINING'