biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Metrics] Hide deprecated metrics with gpu_ prefix (#24245)

Browse Source 
Signed-off-by: Mark McLoughlin <markmc@redhat.com>
This commit is contained in:
Mark McLoughlin
2025-09-16 03:15:57 +01:00
committed by GitHub
parent 3c96e7b8a1
commit 2942970d44
2 changed files with 61 additions and 44 deletions
Download Patch File Download Diff File Expand all files Collapse all files

21
tests/entrypoints/openai/test_metrics.py
View File

@@ -232,6 +232,9 @@ EXPECTED_METRICS_V1 = [
"vllm:gpu_cache_usage_perc",
"vllm:gpu_prefix_cache_queries",
"vllm:gpu_prefix_cache_hits",
"vllm:kv_cache_usage_perc",
"vllm:prefix_cache_queries",
"vllm:prefix_cache_hits",
"vllm:num_preemptions_total",
"vllm:prompt_tokens_total",
"vllm:generation_tokens_total",
@@ -277,6 +280,9 @@ EXPECTED_METRICS_V1 = [
]
HIDDEN_DEPRECATED_METRICS: list[str] = [
"vllm:gpu_cache_usage_perc",
"vllm:gpu_prefix_cache_queries",
"vllm:gpu_prefix_cache_hits",
"vllm:time_per_output_token_seconds_sum",
"vllm:time_per_output_token_seconds_bucket",
"vllm:time_per_output_token_seconds_count",
@@ -307,7 +313,7 @@ async def test_abort_metrics_reset(server: RemoteOpenAIServer,
client: openai.AsyncClient, use_v1: bool):
running_requests, waiting_requests, kv_cache_usage = (
_get_running_metrics_from_api(server))
_get_running_metrics_from_api(server, use_v1))
# Expect no running requests or kvcache usage
assert running_requests == 0
@@ -330,7 +336,7 @@ async def test_abort_metrics_reset(server: RemoteOpenAIServer,
# Check that we have running requests
running_requests, waiting_requests, kv_cache_usage = (
_get_running_metrics_from_api(server))
_get_running_metrics_from_api(server, use_v1))
# Expect running requests and kvcache usage
assert running_requests > 0
@@ -349,7 +355,7 @@ async def test_abort_metrics_reset(server: RemoteOpenAIServer,
# 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))
_get_running_metrics_from_api(server, use_v1))
assert running_requests_after == 0,\
(f"Expected 0 running requests after abort, got "
@@ -362,7 +368,7 @@ async def test_abort_metrics_reset(server: RemoteOpenAIServer,
f"{kv_cache_usage_after}")
def _get_running_metrics_from_api(server: RemoteOpenAIServer):
def _get_running_metrics_from_api(server: RemoteOpenAIServer, use_v1: bool):
"""Return (running_count, waiting_count, kv_cache_usage)"""
response = requests.get(server.url_for("metrics"))
@@ -371,6 +377,9 @@ def _get_running_metrics_from_api(server: RemoteOpenAIServer):
# 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")
for family in text_string_to_metric_families(response.text):
if family.name == "vllm:num_requests_running":
for sample in family.samples:
@@ -382,9 +391,9 @@ def _get_running_metrics_from_api(server: RemoteOpenAIServer):
if sample.name == "vllm:num_requests_waiting":
waiting_requests = sample.value
break
elif family.name == "vllm:gpu_cache_usage_perc":
elif family.name == kv_cache_usage_metric:
for sample in family.samples:
if sample.name == "vllm:gpu_cache_usage_perc":
if sample.name == kv_cache_usage_metric:
kv_cache_usage = sample.value
break