[V0 Deprecation] Remove VLLM_USE_V1 from tests (#26341)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>

tests/entrypoints/openai/test_metrics.py

@@ -22,24 +22,6 @@ MODEL_NAME = "TinyLlama/TinyLlama-1.1B-Chat-v1.0"
 PREV_MINOR_VERSION = version._prev_minor_version()
 
 
-@pytest.fixture(scope="module", params=[True])
-def use_v1(request):
-    # Module-scoped variant of run_with_both_engines
-    #
-    # Use this fixture to run a test with both v0 and v1, and
-    # also to conditionalize the test logic e.g.
-    #
-    # def test_metrics_exist(use_v1, server, client):
-    #     ...
-    #     expected = EXPECTED_V1_METRICS if use_v1 else EXPECTED_METRICS
-    #     for metric in expected:
-    #         assert metric in response.text
-    #
-    # @skip_v1 wouldn't work here because this is a module-level
-    # fixture - per-function decorators would have no effect
-    yield request.param
-
-
 @pytest.fixture(scope="module")
 def default_server_args():
     return [
@@ -63,13 +45,11 @@ def default_server_args():
         f"--show-hidden-metrics-for-version={PREV_MINOR_VERSION}",
     ],
 )
-def server(use_v1, default_server_args, request):
+def server(default_server_args, request):
     if request.param:
         default_server_args.append(request.param)
-    env_dict = dict(VLLM_USE_V1="1" if use_v1 else "0")
-    with RemoteOpenAIServer(
-        MODEL_NAME, default_server_args, env_dict=env_dict
-    ) as remote_server:
+
+    with RemoteOpenAIServer(MODEL_NAME, default_server_args) as remote_server:
         yield remote_server
 
 
@@ -129,7 +109,8 @@ EXPECTED_VALUES = {
 
 @pytest.mark.asyncio
 async def test_metrics_counts(
-    server: RemoteOpenAIServer, client: openai.AsyncClient, use_v1: bool
+    server: RemoteOpenAIServer,
+    client: openai.AsyncClient,
 ):
     for _ in range(_NUM_REQUESTS):
         # sending a request triggers the metrics to be logged.
@@ -145,7 +126,7 @@ async def test_metrics_counts(
 
     # Loop over all expected metric_families
     for metric_family, suffix_values_list in EXPECTED_VALUES.items():
-        if (use_v1 and metric_family not in EXPECTED_METRICS_V1) or (
+        if (metric_family not in EXPECTED_METRICS_V1) or (
             not server.show_hidden_metrics
             and metric_family in HIDDEN_DEPRECATED_METRICS
         ):
@@ -183,62 +164,6 @@ async def test_metrics_counts(
         assert found_metric, f"Did not find {metric_family} in prom endpoint"
 
 
-EXPECTED_METRICS = [
-    "vllm:num_requests_running",
-    "vllm:num_requests_waiting",
-    "vllm:gpu_cache_usage_perc",
-    "vllm:time_to_first_token_seconds_sum",
-    "vllm:time_to_first_token_seconds_bucket",
-    "vllm:time_to_first_token_seconds_count",
-    "vllm:time_per_output_token_seconds_sum",
-    "vllm:time_per_output_token_seconds_bucket",
-    "vllm:time_per_output_token_seconds_count",
-    "vllm:e2e_request_latency_seconds_sum",
-    "vllm:e2e_request_latency_seconds_bucket",
-    "vllm:e2e_request_latency_seconds_count",
-    "vllm:request_queue_time_seconds_sum",
-    "vllm:request_queue_time_seconds_bucket",
-    "vllm:request_queue_time_seconds_count",
-    "vllm:request_inference_time_seconds_sum",
-    "vllm:request_inference_time_seconds_bucket",
-    "vllm:request_inference_time_seconds_count",
-    "vllm:request_prefill_time_seconds_sum",
-    "vllm:request_prefill_time_seconds_bucket",
-    "vllm:request_prefill_time_seconds_count",
-    "vllm:request_decode_time_seconds_sum",
-    "vllm:request_decode_time_seconds_bucket",
-    "vllm:request_decode_time_seconds_count",
-    "vllm:request_prompt_tokens_sum",
-    "vllm:request_prompt_tokens_bucket",
-    "vllm:request_prompt_tokens_count",
-    "vllm:request_generation_tokens_sum",
-    "vllm:request_generation_tokens_bucket",
-    "vllm:request_generation_tokens_count",
-    "vllm:request_params_n_sum",
-    "vllm:request_params_n_bucket",
-    "vllm:request_params_n_count",
-    "vllm:request_params_max_tokens_sum",
-    "vllm:request_params_max_tokens_bucket",
-    "vllm:request_params_max_tokens_count",
-    "vllm:iteration_tokens_total",
-    "vllm:num_preemptions_total",
-    "vllm:prompt_tokens_total",
-    "vllm:generation_tokens_total",
-    "vllm:request_success_total",
-    "vllm:cache_config_info",
-    # labels in cache_config_info
-    "block_size",
-    "cache_dtype",
-    "cpu_offload_gb",
-    "enable_prefix_caching",
-    "gpu_memory_utilization",
-    "num_cpu_blocks",
-    "num_gpu_blocks",
-    "num_gpu_blocks_override",
-    "sliding_window",
-    "swap_space_bytes",
-]
-
 EXPECTED_METRICS_V1 = [
     "vllm:num_requests_running",
     "vllm:num_requests_waiting",
@@ -304,17 +229,21 @@ HIDDEN_DEPRECATED_METRICS: list[str] = [
 
 @pytest.mark.asyncio
 async def test_metrics_exist(
-    server: RemoteOpenAIServer, client: openai.AsyncClient, use_v1: bool
+    server: RemoteOpenAIServer,
+    client: openai.AsyncClient,
 ):
     # sending a request triggers the metrics to be logged.
     await client.completions.create(
-        model=MODEL_NAME, prompt="Hello, my name is", max_tokens=5, temperature=0.0
+        model=MODEL_NAME,
+        prompt="Hello, my name is",
+        max_tokens=5,
+        temperature=0.0,
     )
 
     response = requests.get(server.url_for("metrics"))
     assert response.status_code == HTTPStatus.OK
 
-    for metric in EXPECTED_METRICS_V1 if use_v1 else EXPECTED_METRICS:
+    for metric in EXPECTED_METRICS_V1:
         if metric in HIDDEN_DEPRECATED_METRICS and not server.show_hidden_metrics:
             continue
         assert metric in response.text
@@ -322,10 +251,11 @@ async def test_metrics_exist(
 
 @pytest.mark.asyncio
 async def test_abort_metrics_reset(
-    server: RemoteOpenAIServer, client: openai.AsyncClient, use_v1: bool
+    server: RemoteOpenAIServer,
+    client: openai.AsyncClient,
 ):
     running_requests, waiting_requests, kv_cache_usage = _get_running_metrics_from_api(
-        server, use_v1
+        server
     )
 
     # Expect no running requests or kvcache usage
@@ -351,7 +281,7 @@ async def test_abort_metrics_reset(
 
     # Check that we have running requests
     running_requests, waiting_requests, kv_cache_usage = _get_running_metrics_from_api(
-        server, use_v1
+        server
     )
 
     # Expect running requests and kvcache usage
@@ -371,7 +301,7 @@ async def test_abort_metrics_reset(
 
     # Verify running and waiting requests counts and KV cache usage are zero
     running_requests_after, waiting_requests_after, kv_cache_usage_after = (
-        _get_running_metrics_from_api(server, use_v1)
+        _get_running_metrics_from_api(server)
     )
 
     assert running_requests_after == 0, (
@@ -385,7 +315,7 @@ async def test_abort_metrics_reset(
     )
 
 
-def _get_running_metrics_from_api(server: RemoteOpenAIServer, use_v1: bool):
+def _get_running_metrics_from_api(server: RemoteOpenAIServer):
     """Return (running_count, waiting_count, kv_cache_usage)"""
 
     response = requests.get(server.url_for("metrics"))
@@ -394,9 +324,7 @@ def _get_running_metrics_from_api(server: RemoteOpenAIServer, use_v1: bool):
     # Verify running and waiting requests counts and KV cache usage are zero
     running_requests, waiting_requests, kv_cache_usage = None, None, None
 
-    kv_cache_usage_metric = (
-        "vllm:kv_cache_usage_perc" if use_v1 else "vllm:gpu_cache_usage_perc"
-    )
+    kv_cache_usage_metric = "vllm:kv_cache_usage_perc"
 
     for family in text_string_to_metric_families(response.text):
         if family.name == "vllm:num_requests_running":
@@ -422,7 +350,7 @@ def _get_running_metrics_from_api(server: RemoteOpenAIServer, use_v1: bool):
     return running_requests, waiting_requests, kv_cache_usage
 
 
-def test_metrics_exist_run_batch(use_v1: bool):
+def test_metrics_exist_run_batch():
     input_batch = """{"custom_id": "request-0", "method": "POST", "url": "/v1/embeddings", "body": {"model": "intfloat/multilingual-e5-small", "input": "You are a helpful assistant."}}"""  # noqa: E501
 
     base_url = "0.0.0.0"
@@ -452,7 +380,6 @@ def test_metrics_exist_run_batch(use_v1: bool):
                 "--port",
                 port,
             ],
-            env={"VLLM_USE_V1": "1"},
         )
 
         def is_server_up(url):