test: cover ai platform demo workflow

backend/apps/modelhub/tests/test_services.py

@@ -1,4 +1,6 @@
-from openai import APITimeoutError
+import httpx
+import pytest
+from openai import APITimeoutError, AuthenticationError, InternalServerError, RateLimitError
 
 from apps.modelhub.crypto import encrypt_secret
 from apps.modelhub.models import ModelDeployment, ModelInvocation, ModelProvider
@@ -57,3 +59,43 @@ def test_non_transient_error_fails_without_exposing_raw_error(db, monkeypatch):
     assert invocation.status == "failed"
     assert invocation.error_code == "unexpected_error"
     assert "secret provider detail" not in invocation.error_message
+
+
+def provider_error(error_type, status_code):
+    request = httpx.Request("POST", "https://models.example.com/v1/chat/completions")
+    response = httpx.Response(status_code, request=request)
+    return error_type("provider raw error", response=response, body={})
+
+
+@pytest.mark.parametrize(
+    ("error", "expected_code", "expected_calls"),
+    [
+        (provider_error(AuthenticationError, 401), "authentication_failed", 1),
+        (provider_error(RateLimitError, 429), "rate_limited", 2),
+        (provider_error(InternalServerError, 500), "provider_unavailable", 2),
+    ],
+)
+def test_provider_error_classification_and_retry(
+    db,
+    monkeypatch,
+    error,
+    expected_code,
+    expected_calls,
+):
+    invocation = create_invocation()
+    calls = []
+
+    def request(*args, **kwargs):
+        calls.append(1)
+        raise error
+
+    monkeypatch.setattr("apps.modelhub.services._request_completion", request)
+    monkeypatch.setattr("apps.modelhub.services.sleep", lambda *_: None)
+
+    execute_invocation(invocation, invocation.input_snapshot["messages"])
+
+    invocation.refresh_from_db()
+    assert invocation.status == "failed"
+    assert invocation.error_code == expected_code
+    assert len(calls) == expected_calls
+    assert "provider raw error" not in invocation.error_message