scriptoriumadmin/litellm
1
0
Fork 0
mirror of https://github.com/BerriAI/litellm.git synced 2026-09-06 08:16:43 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 78

fix(prometheus): keep team alias and team wildcard names out of the other bucket

Browse source
This commit is contained in:
mateo-berri 2026-09-01 11:04:25 -07:00
parent 3b3099d78d
commit fc091c1248
2 changed files with 50 additions and 4 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

16
litellm/integrations/prometheus.py
View file

@ -169,10 +169,11 @@ def _get_proxy_llm_router() -> Router | None:
def _bounded_requested_model_label(requested_model: str | None) -> str | None:
"""
Bound ``requested_model`` label cardinality: names the router recognizes
(model names, deployment ids, aliases, routing groups) or matches via a
wildcard/pattern route keep their own label value; any other
client-supplied string collapses into the single ``other`` bucket. With no
router to vouch for the string, it also collapses to ``other``.
(model names, deployment ids, aliases, routing groups, team public model
names) or matches via a global or team wildcard/pattern route keep their
own label value; any other client-supplied string collapses into the
single ``other`` bucket. With no router to vouch for the string, it also
collapses to ``other``.
"""
if not requested_model:
return requested_model
@ -181,8 +182,15 @@ def _bounded_requested_model_label(requested_model: str | None) -> str | None:
return UNRECOGNIZED_REQUESTED_MODEL_LABEL
if llm_router.is_recognized_model(requested_model):
return requested_model
if requested_model in llm_router.team_public_model_names:
return requested_model
if llm_router.pattern_router.route(requested_model) is not None:
return requested_model
if any(
team_pattern_router.route(requested_model) is not None
for team_pattern_router in llm_router.team_pattern_routers.values()
):
return requested_model
return UNRECOGNIZED_REQUESTED_MODEL_LABEL

38
tests/test_litellm/integrations/test_prometheus_requested_model_cardinality.py
View file

@ -58,6 +58,24 @@ def router():
)
@pytest.fixture
def team_router():
return litellm.Router(
model_list=[
{
"model_name": "team-internal-gpt",
"litellm_params": {"model": "openai/gpt-4o-mini", "api_key": "fake-key"},
"model_info": {"team_id": "team-1", "team_public_model_name": "team-alias-gpt"},
},
{
"model_name": "team-internal-bedrock",
"litellm_params": {"model": "openai/*", "api_key": "fake-key"},
"model_info": {"team_id": "team-1", "team_public_model_name": "team-models/*"},
},
]
)
def _requested_model_values(metric) -> set[str]:
index = metric._labelnames.index("requested_model")
return {sample_key[index] for sample_key in metric._metrics}
@ -118,6 +136,26 @@ async def test_known_alias_and_wildcard_models_keep_their_own_labels(router):
}
@pytest.mark.asyncio
async def test_team_alias_and_team_wildcard_models_keep_their_own_labels(team_router):
logger = PrometheusLogger()
with patch("litellm.proxy.proxy_server.llm_router", team_router, create=True): # test-quality-ok: production reads proxy_server.llm_router lazily, no injection seam
await _fire_proxy_failure(logger, "team-alias-gpt")
await _fire_proxy_failure(logger, "team-models/gpt-4o-audio-preview")
await _fire_proxy_failure(logger, "agent-typo-hallucinated")
for metric in (
logger.litellm_proxy_failed_requests_metric,
logger.litellm_proxy_total_requests_metric,
):
assert _requested_model_values(metric) == {
"team-alias-gpt",
"team-models/gpt-4o-audio-preview",
UNRECOGNIZED_REQUESTED_MODEL_LABEL,
}
@pytest.mark.asyncio
async def test_unknown_models_collapse_to_other_when_router_is_unavailable():
logger = PrometheusLogger()