Add more refactored openai test & in CI

python/sglang/srt/entrypoints/openai/api_server.py

@@ -36,7 +36,7 @@
 from fastapi.responses import Response
 
 from sglang.srt.disaggregation.utils import (
-    FakeBootstrapHost,
+    FAKE_BOOTSTRAP_HOST,
     register_disaggregation_server,
 )
 from sglang.srt.entrypoints.engine import Engine, _launch_subprocesses
@@ -265,7 +265,7 @@ def _wait_and_warmup(
                     "max_new_tokens": 8,
                     "ignore_eos": True,
                 },
-                "bootstrap_host": [FakeBootstrapHost] * server_args.dp_size,
+                "bootstrap_host": [FAKE_BOOTSTRAP_HOST] * server_args.dp_size,
                 # This is a hack to ensure fake transfer is enabled during prefill warmup
                 # ensure each dp rank has a unique bootstrap_room during prefill warmup
                 "bootstrap_room": [

test/srt/openai/conftest.py

@@ -12,9 +12,10 @@
 import requests
 
 from sglang.srt.utils import kill_process_tree  # reuse SGLang helper
+from sglang.test.test_utils import DEFAULT_SMALL_MODEL_NAME_FOR_TEST
 
 SERVER_MODULE = "sglang.srt.entrypoints.openai.api_server"
-DEFAULT_MODEL = "dummy-model"
+DEFAULT_MODEL = DEFAULT_SMALL_MODEL_NAME_FOR_TEST
 STARTUP_TIMEOUT = float(os.getenv("SGLANG_OPENAI_STARTUP_TIMEOUT", 120))
 
 
@@ -39,7 +40,7 @@ def _wait_until_healthy(proc: subprocess.Popen, base: str, timeout: float) -> No
 
 
 def launch_openai_server(model: str = DEFAULT_MODEL, **kw):
-    """Spawn the draft OpenAI-compatible server and wait until it’s ready."""
+    """Spawn the draft OpenAI-compatible server and wait until it's ready."""
     port = _pick_free_port()
     cmd = [
         sys.executable,
@@ -79,7 +80,7 @@ def launch_openai_server(model: str = DEFAULT_MODEL, **kw):
 
 @pytest.fixture(scope="session")
 def openai_server() -> Generator[str, None, None]:
-    """PyTest fixture that provides the server’s base URL and cleans up."""
+    """PyTest fixture that provides the server's base URL and cleans up."""
     proc, base, log_file = launch_openai_server()
     yield base
     kill_process_tree(proc.pid)

test/srt/openai/test_server.py

@@ -1,16 +1,52 @@
 # sglang/test/srt/openai/test_server.py
-import pytest
 import requests
 
+from sglang.test.test_utils import DEFAULT_SMALL_MODEL_NAME_FOR_TEST as MODEL_ID
+
 
 def test_health(openai_server: str):
     r = requests.get(f"{openai_server}/health")
-    assert r.status_code == 200, r.text
+    assert r.status_code == 200
+    # FastAPI returns an empty body → r.text == ""
     assert r.text == ""
 
 
-@pytest.mark.xfail(reason="Endpoint skeleton not implemented yet")
 def test_models_endpoint(openai_server: str):
     r = requests.get(f"{openai_server}/v1/models")
-    # once implemented this should be 200
-    assert r.status_code == 200
+    assert r.status_code == 200, r.text
+    payload = r.json()
+
+    # Basic contract
+    assert "data" in payload and isinstance(payload["data"], list) and payload["data"]
+
+    # Validate fields of the first model card
+    first = payload["data"][0]
+    for key in ("id", "root", "max_model_len"):
+        assert key in first, f"missing {key} in {first}"
+
+    # max_model_len must be positive
+    assert isinstance(first["max_model_len"], int) and first["max_model_len"] > 0
+
+    # The server should report the same model id we launched it with
+    ids = {m["id"] for m in payload["data"]}
+    assert MODEL_ID in ids
+
+
+def test_get_model_info(openai_server: str):
+    r = requests.get(f"{openai_server}/get_model_info")
+    assert r.status_code == 200, r.text
+    info = r.json()
+
+    expected_keys = {"model_path", "tokenizer_path", "is_generation"}
+    assert expected_keys.issubset(info.keys())
+
+    # model_path must end with the one we passed on the CLI
+    assert info["model_path"].endswith(MODEL_ID)
+
+    # is_generation is documented as a boolean
+    assert isinstance(info["is_generation"], bool)
+
+
+def test_unknown_route_returns_404(openai_server: str):
+    r = requests.get(f"{openai_server}/definitely-not-a-real-route")
+    assert r.status_code == 404

test/srt/run_suite.py

@@ -62,6 +62,7 @@ class TestFile:
         TestFile("test_openai_adapter.py", 1),
         TestFile("test_openai_function_calling.py", 60),
         TestFile("test_openai_server.py", 149),
+        TestFile("openai/test_server.py", 120),
         TestFile("test_openai_server_hidden_states.py", 240),
         TestFile("test_penalty.py", 41),
         TestFile("test_page_size.py", 60),