diff --git a/basedpyright-code-budget.json b/basedpyright-code-budget.json index db49ae8f48b..3357212a6c8 100644 --- a/basedpyright-code-budget.json +++ b/basedpyright-code-budget.json @@ -99,7 +99,7 @@ "limit": 0 }, "reportUnknownArgumentType": { - "limit": 44530 + "limit": 44528 }, "reportUnknownLambdaType": { "limit": 109 @@ -138,9 +138,9 @@ "limit": 139 }, "reportUnusedImport": { - "limit": 545 + "limit": 544 }, "reportUnusedVariable": { - "limit": 146 + "limit": 145 } } diff --git a/litellm/cost_calculator.py b/litellm/cost_calculator.py index 4cd292b2416..572d76ad555 100644 --- a/litellm/cost_calculator.py +++ b/litellm/cost_calculator.py @@ -591,6 +591,7 @@ def cost_per_token( prompt_characters=prompt_characters, completion_characters=completion_characters, usage=usage_block, + service_tier=service_tier, vertex_location=vertex_location, ) elif cost_router == "cost_per_token": @@ -845,9 +846,11 @@ def _get_response_model(completion_response: object) -> str | None: _GEMINI_TRAFFIC_TYPE_TO_SERVICE_TIER: Final[dict] = { # ON_DEMAND_PRIORITY maps to "priority" — selects input_cost_per_token_priority, etc. "ON_DEMAND_PRIORITY": "priority", - # FLEX / BATCH maps to "flex" — selects input_cost_per_token_flex, etc. + # FLEX / BATCH / ON_DEMAND_FLEX maps to "flex" — selects input_cost_per_token_flex, etc. + # Vertex AI reports flex/shared-capacity traffic as ON_DEMAND_FLEX, not FLEX. "FLEX": "flex", "BATCH": "flex", + "ON_DEMAND_FLEX": "flex", # ON_DEMAND is standard pricing — no service_tier suffix applied "ON_DEMAND": None, } @@ -862,9 +865,9 @@ def _map_traffic_type_to_service_tier(traffic_type: str | None) -> str | None: trafficType values seen in practice ------------------------------------ - ON_DEMAND -> standard pricing (service_tier = None) - ON_DEMAND_PRIORITY -> priority pricing (service_tier = "priority") - FLEX / BATCH -> batch/flex pricing (service_tier = "flex") + ON_DEMAND -> standard pricing (service_tier = None) + ON_DEMAND_PRIORITY -> priority pricing (service_tier = "priority") + FLEX / BATCH / ON_DEMAND_FLEX -> batch/flex pricing (service_tier = "flex") """ if traffic_type is None: return None diff --git a/litellm/endpoints/speech/speech_to_completion_bridge/transformation.py b/litellm/endpoints/speech/speech_to_completion_bridge/transformation.py index a9429b673e4..fb66edbf272 100644 --- a/litellm/endpoints/speech/speech_to_completion_bridge/transformation.py +++ b/litellm/endpoints/speech/speech_to_completion_bridge/transformation.py @@ -8,6 +8,14 @@ if TYPE_CHECKING: from litellm.types.utils import ModelResponse +def _completion_response_cost(model_response: "ModelResponse") -> float | None: + hidden_params: Final = getattr(model_response, "_hidden_params", None) + if not isinstance(hidden_params, dict): + return None + response_cost: Final = hidden_params.get("response_cost") + return response_cost if isinstance(response_cost, float) else None + + class SpeechToCompletionBridgeTransformationHandler: def transform_request( self, @@ -123,4 +131,6 @@ class SpeechToCompletionBridgeTransformationHandler: # Create an httpx.Response object response: Final = httpx.Response(status_code=200, content=binary_data, headers=headers) - return HttpxBinaryResponseContent(response) + binary_response: Final = HttpxBinaryResponseContent(response) + binary_response.set_response_cost(_completion_response_cost(model_response)) + return binary_response diff --git a/litellm/integrations/callback_configs.json b/litellm/integrations/callback_configs.json index 6d2bcea8bae..7a2295a35ae 100644 --- a/litellm/integrations/callback_configs.json +++ b/litellm/integrations/callback_configs.json @@ -220,6 +220,12 @@ "ui_name": "Host URL", "description": "Langfuse host URL (default: https://cloud.langfuse.com)", "required": false + }, + "langfuse_environment": { + "type": "text", + "ui_name": "Tracing Environment", + "description": "Langfuse tracing environment (lowercase; falls back to LANGFUSE_TRACING_ENVIRONMENT)", + "required": false } }, "description": "Langfuse v2 Logging Integration" @@ -247,6 +253,12 @@ "ui_name": "Host URL", "description": "Langfuse host URL (default: https://cloud.langfuse.com)", "required": false + }, + "langfuse_environment": { + "type": "text", + "ui_name": "Tracing Environment", + "description": "Langfuse tracing environment (lowercase; falls back to LANGFUSE_TRACING_ENVIRONMENT)", + "required": false } }, "description": "Langfuse v3 OTEL Logging Integration" diff --git a/litellm/integrations/langfuse/langfuse.py b/litellm/integrations/langfuse/langfuse.py index da924a81e0c..d1a9125ac71 100644 --- a/litellm/integrations/langfuse/langfuse.py +++ b/litellm/integrations/langfuse/langfuse.py @@ -1,5 +1,6 @@ #### What this does #### # On success, logs events to Langfuse +import inspect import os import traceback from collections.abc import Callable, Iterable, Mapping @@ -21,6 +22,9 @@ from litellm.litellm_core_utils.core_helpers import ( reconstruct_model_name, safe_deep_copy, ) +from litellm.litellm_core_utils.initialize_dynamic_callback_params import ( + validate_langfuse_environment_value, +) from litellm.litellm_core_utils.redact_messages import redact_user_api_key_info from litellm.llms.custom_httpx.http_handler import _get_httpx_client from litellm.secret_managers.main import str_to_bool @@ -140,6 +144,7 @@ class LangFuseLogger: langfuse_public_key=None, langfuse_secret=None, langfuse_host=None, + langfuse_environment: str | None = None, flush_interval=1, allow_env_credentials: bool = True, ): @@ -159,6 +164,10 @@ class LangFuseLogger: if not (self.langfuse_host.startswith("http://") or self.langfuse_host.startswith("https://")): # add http:// if unset, assume communicating over private network - e.g. render self.langfuse_host = "http://" + self.langfuse_host + _env_override: Final = str(langfuse_environment).strip() if langfuse_environment is not None else None + self.langfuse_environment = _env_override or os.getenv("LANGFUSE_TRACING_ENVIRONMENT") + if self.langfuse_environment: + validate_langfuse_environment_value(self.langfuse_environment) self.langfuse_release = os.getenv("LANGFUSE_RELEASE") self.langfuse_debug = os.getenv("LANGFUSE_DEBUG") self.langfuse_flush_interval = LangFuseLogger._get_langfuse_flush_interval(flush_interval) @@ -182,6 +191,8 @@ class LangFuseLogger: } self.langfuse_sdk_version: str = langfuse.version.__version__ + if "environment" in inspect.signature(Langfuse.__init__).parameters: + parameters["environment"] = self.langfuse_environment if Version(self.langfuse_sdk_version) >= Version("2.6.0"): parameters["sdk_integration"] = "litellm" self.Langfuse: Langfuse = self.safe_init_langfuse_client(parameters) diff --git a/litellm/integrations/langfuse/langfuse_handler.py b/litellm/integrations/langfuse/langfuse_handler.py index f4dd80f91f5..8a407f71b3b 100644 --- a/litellm/integrations/langfuse/langfuse_handler.py +++ b/litellm/integrations/langfuse/langfuse_handler.py @@ -1,3 +1,5 @@ +import os + """ This file contains the LangFuseHandler class @@ -108,6 +110,7 @@ class LangFuseHandler: langfuse_public_key=credentials.get("langfuse_public_key"), langfuse_secret=credentials.get("langfuse_secret") or credentials.get("langfuse_secret_key"), langfuse_host=credentials.get("langfuse_host"), + langfuse_environment=credentials.get("langfuse_environment"), allow_env_credentials=credentials.get("langfuse_host") is None, ) in_memory_dynamic_logger_cache.set_cache( @@ -135,8 +138,29 @@ class LangFuseHandler: or standard_callback_dynamic_params.get("langfuse_secret_key"), langfuse_public_key=standard_callback_dynamic_params.get("langfuse_public_key"), langfuse_host=standard_callback_dynamic_params.get("langfuse_host"), + langfuse_environment=LangFuseHandler._meaningful_dynamic_environment(standard_callback_dynamic_params), ) + @staticmethod + def _meaningful_dynamic_environment( + standard_callback_dynamic_params: StandardCallbackDynamicParams, + ) -> str | None: + """Return the per-request environment only when it changes behavior. + + Empty/whitespace values and values equal to the deployment-wide + LANGFUSE_TRACING_ENVIRONMENT fallback are treated as absent so an + environment-only override that matches the default does not mint a + duplicate SDK client (each client costs threads and counts against + MAX_LANGFUSE_INITIALIZED_CLIENTS). + """ + raw = standard_callback_dynamic_params.get("langfuse_environment") + if raw is None: + return None + value = str(raw).strip() + if not value or value == os.getenv("LANGFUSE_TRACING_ENVIRONMENT"): + return None + return value + @staticmethod def _dynamic_langfuse_credentials_are_passed( standard_callback_dynamic_params: StandardCallbackDynamicParams, @@ -153,6 +177,7 @@ class LangFuseHandler: or standard_callback_dynamic_params.get("langfuse_public_key") is not None or standard_callback_dynamic_params.get("langfuse_secret") is not None or standard_callback_dynamic_params.get("langfuse_secret_key") is not None + or LangFuseHandler._meaningful_dynamic_environment(standard_callback_dynamic_params) is not None ): return True return False diff --git a/litellm/integrations/langfuse/langfuse_otel.py b/litellm/integrations/langfuse/langfuse_otel.py index 3d044c3ea15..a96fac32c2a 100644 --- a/litellm/integrations/langfuse/langfuse_otel.py +++ b/litellm/integrations/langfuse/langfuse_otel.py @@ -231,7 +231,10 @@ class LangfuseOtelLogger(OpenTelemetry): from litellm.integrations.arize._utils import safe_set_attribute from litellm.litellm_core_utils.safe_json_dumps import safe_dumps - langfuse_environment: Final = os.environ.get("LANGFUSE_TRACING_ENVIRONMENT") + dynamic_params: Final = kwargs.get("standard_callback_dynamic_params") + langfuse_environment: Final = ( + dynamic_params.get("langfuse_environment") if dynamic_params else None + ) or os.environ.get("LANGFUSE_TRACING_ENVIRONMENT") if langfuse_environment: safe_set_attribute( span, diff --git a/litellm/litellm_core_utils/initialize_dynamic_callback_params.py b/litellm/litellm_core_utils/initialize_dynamic_callback_params.py index 3b42ca4eaaf..65c5b0d9799 100644 --- a/litellm/litellm_core_utils/initialize_dynamic_callback_params.py +++ b/litellm/litellm_core_utils/initialize_dynamic_callback_params.py @@ -1,3 +1,4 @@ +import re from collections.abc import Iterator, Mapping from typing import Any, Final @@ -45,12 +46,29 @@ def validate_no_callback_env_reference(param: str, value: object, *, source: str _raise_env_reference_error(param, source=source) +# Langfuse rejects events whose environment does not match this pattern +# (lowercase alphanumerics, hyphens, underscores; no "langfuse" prefix). +# Validating here fails fast at config/init time instead of silently +# dropping every trace server-side. +LANGFUSE_ENVIRONMENT_PATTERN: Final = r"^(?!langfuse)[a-z0-9-_]+$" + + +def validate_langfuse_environment_value(value: str) -> None: + if not re.match(LANGFUSE_ENVIRONMENT_PATTERN, value): + raise ValueError( + f"Invalid langfuse_environment {value!r}: must be lowercase " + "alphanumerics/hyphens/underscores and must not start with " + f"'langfuse' (pattern {LANGFUSE_ENVIRONMENT_PATTERN})" + ) + + # Hardcoded list of supported callback params to avoid runtime inspection issues with TypedDict _supported_callback_params: Final[tuple[str, ...]] = ( "langfuse_public_key", "langfuse_secret", "langfuse_secret_key", "langfuse_host", + "langfuse_environment", "langfuse_prompt_version", "langsmith_api_key", "langsmith_project", diff --git a/litellm/litellm_core_utils/litellm_logging.py b/litellm/litellm_core_utils/litellm_logging.py index 626af7530a4..fd2200c59cc 100644 --- a/litellm/litellm_core_utils/litellm_logging.py +++ b/litellm/litellm_core_utils/litellm_logging.py @@ -1590,7 +1590,7 @@ class Logging(LiteLLMLoggingBaseClass): if transformed_result is not None: result = transformed_result - if isinstance(result, BaseModel) and hasattr(result, "_hidden_params"): + if isinstance(result, (BaseModel, HttpxBinaryResponseContent)) and hasattr(result, "_hidden_params"): hidden_params: Final = getattr(result, "_hidden_params", {}) if ( "response_cost" in hidden_params and hidden_params["response_cost"] is not None diff --git a/litellm/llms/vertex_ai/cost_calculator.py b/litellm/llms/vertex_ai/cost_calculator.py index 23cb1e5b580..8b00fc2e925 100644 --- a/litellm/llms/vertex_ai/cost_calculator.py +++ b/litellm/llms/vertex_ai/cost_calculator.py @@ -64,6 +64,7 @@ def cost_per_character( usage: Usage, prompt_characters: float | None = None, completion_characters: float | None = None, + service_tier: str | None = None, vertex_location: str | None = None, ) -> tuple[float, float]: """ @@ -74,6 +75,8 @@ def cost_per_character( - custom_llm_provider: str, "vertex_ai-*" - prompt_characters: float, the number of input characters - completion_characters: float, the number of output characters + - service_tier: optional tier derived from Gemini trafficType + ("priority" for ON_DEMAND_PRIORITY, "flex" for FLEX/batch). - vertex_location: the Vertex AI location serving the request; non-global locations apply the model's regional-endpoint uplift multiplier @@ -92,6 +95,7 @@ def cost_per_character( model=model, custom_llm_provider=custom_llm_provider, usage=usage, + service_tier=service_tier, ) else: try: @@ -123,6 +127,7 @@ def cost_per_character( model=model, custom_llm_provider=custom_llm_provider, usage=usage, + service_tier=service_tier, ) ## CALCULATE OUTPUT COST @@ -131,6 +136,7 @@ def cost_per_character( model=model, custom_llm_provider=custom_llm_provider, usage=usage, + service_tier=service_tier, ) else: completion_tokens: Final = usage.completion_tokens @@ -162,6 +168,7 @@ def cost_per_character( model=model, custom_llm_provider=custom_llm_provider, usage=usage, + service_tier=service_tier, ) vertex_uplift: Final = get_vertex_regional_endpoint_uplift(model_info, vertex_location) diff --git a/litellm/main.py b/litellm/main.py index 98f92e50599..8ee102f5d07 100644 --- a/litellm/main.py +++ b/litellm/main.py @@ -8013,7 +8013,7 @@ def speech( if max_retries is None: max_retries = litellm.num_retries or openai.DEFAULT_MAX_RETRIES - litellm_params_dict: Final = get_litellm_params(**kwargs) + litellm_params_dict: Final = get_litellm_params(metadata=metadata, api_key=api_key or dynamic_api_key, **kwargs) # Get provider-specific text-to-speech config and map parameters text_to_speech_provider_config = ProviderConfigManager.get_provider_text_to_speech_config( diff --git a/litellm/proxy/_types.py b/litellm/proxy/_types.py index 61acac3ff74..727aea6f539 100644 --- a/litellm/proxy/_types.py +++ b/litellm/proxy/_types.py @@ -20,6 +20,7 @@ from typing_extensions import NotRequired, ReadOnly, Required, TypedDict from litellm._uuid import uuid from litellm.constants import DEFAULT_STAGGER_WINDOW_SECONDS, MCP_STDIO_ALLOWED_COMMANDS from litellm.litellm_core_utils.initialize_dynamic_callback_params import ( + validate_langfuse_environment_value, validate_no_callback_env_reference, ) from litellm.types.integrations.compression_interception import ( @@ -2027,6 +2028,8 @@ class AddTeamCallback(LiteLLMPydanticObjectBase): raise ValueError(f"Invalid callback variable: {key}. Must be one of {valid_keys}") callback_vars[key] = str(value) validate_no_callback_env_reference(key, callback_vars[key], source="key/team callback metadata") + if key == "langfuse_environment": + validate_langfuse_environment_value(callback_vars[key]) return values diff --git a/litellm/proxy/common_utils/callback_config_validation.py b/litellm/proxy/common_utils/callback_config_validation.py index 680cc226d18..7ee3bd8d829 100644 --- a/litellm/proxy/common_utils/callback_config_validation.py +++ b/litellm/proxy/common_utils/callback_config_validation.py @@ -14,11 +14,36 @@ _NEWRELIC_VAR_PREFIX: Final = "newrelic_" def callback_config_error(callback_name: str | None, callback_vars: Mapping[str, str] | None) -> str | None: - if callback_name != _NEWRELIC_CALLBACK or not callback_vars: + if not callback_vars: + return None + env_error: Final = _langfuse_environment_error(callback_vars) + if env_error is not None: + return env_error + if callback_name != _NEWRELIC_CALLBACK: return None return _newrelic_config_error(callback_vars) +def _langfuse_environment_error(callback_vars: Mapping[str, str]) -> str | None: + """Reject langfuse_environment values Langfuse ingestion would drop. + + Accepting an invalid value here would 200 the config write and then + silently lose every trace for that key/team at request time. + """ + value: Final = callback_vars.get("langfuse_environment") + if value is None: + return None + from litellm.litellm_core_utils.initialize_dynamic_callback_params import ( + validate_langfuse_environment_value, + ) + + try: + validate_langfuse_environment_value(value) + except ValueError as e: + return str(e) + return None + + def logging_metadata_config_error(metadata: Mapping[str, object] | None) -> str | None: """Validate every ``logging`` entry of a team/key metadata payload.""" if not metadata: diff --git a/litellm/proxy/management_endpoints/team_callback_endpoints.py b/litellm/proxy/management_endpoints/team_callback_endpoints.py index 9a2ec38d627..08346983f32 100644 --- a/litellm/proxy/management_endpoints/team_callback_endpoints.py +++ b/litellm/proxy/management_endpoints/team_callback_endpoints.py @@ -262,6 +262,7 @@ async def add_team_callbacks( - langfuse_secret_key: The secret key for the Langfuse callback - langfuse_secret: The secret for the Langfuse callback - langfuse_host: The host for the Langfuse callback + - langfuse_environment: The tracing environment for the Langfuse callback (lowercase; falls back to LANGFUSE_TRACING_ENVIRONMENT) - gcs_bucket_name: The name of the GCS bucket - gcs_path_service_account: The path to the GCS service account - langsmith_api_key: The API key for the Langsmith callback diff --git a/litellm/types/integrations/langfuse.py b/litellm/types/integrations/langfuse.py index 066cd760d74..6742aefea39 100644 --- a/litellm/types/integrations/langfuse.py +++ b/litellm/types/integrations/langfuse.py @@ -1,10 +1,11 @@ -from typing_extensions import TypedDict +from typing_extensions import ReadOnly, TypedDict class LangfuseLoggingConfig(TypedDict): langfuse_secret: str | None langfuse_public_key: str | None langfuse_host: str | None + langfuse_environment: ReadOnly[str | None] class LangfuseUsageDetails(TypedDict): diff --git a/litellm/types/llms/openai.py b/litellm/types/llms/openai.py index 4a6c4a5bbb5..45f6b5c55a9 100644 --- a/litellm/types/llms/openai.py +++ b/litellm/types/llms/openai.py @@ -107,7 +107,17 @@ EmbeddingInput = str | list[str] class HttpxBinaryResponseContent(_HttpxBinaryResponseContent): - _hidden_params: dict = {} + _hidden_params: dict + + def __init__(self, response: httpx.Response) -> None: + super().__init__(response) + self._hidden_params = {} # mutable-ok: mutable-dict contract shared with ModelResponse logging consumers + + def set_response_cost(self, response_cost: float | None) -> None: + if response_cost is None: + self._hidden_params.pop("response_cost", None) + return + self._hidden_params["response_cost"] = response_cost class NotGiven: diff --git a/litellm/types/utils.py b/litellm/types/utils.py index b5caea0e778..ef3586f2559 100644 --- a/litellm/types/utils.py +++ b/litellm/types/utils.py @@ -3278,6 +3278,7 @@ class StandardCallbackDynamicParams(TypedDict, total=False): langfuse_secret: str | None langfuse_secret_key: str | None langfuse_host: str | None + langfuse_environment: ReadOnly[str | None] # Langfuse prompt version langfuse_prompt_version: int | None diff --git a/ruff-strict-budget.json b/ruff-strict-budget.json index b24335ad112..149c44ed083 100644 --- a/ruff-strict-budget.json +++ b/ruff-strict-budget.json @@ -108,7 +108,7 @@ "limit": 3 }, "F401": { - "limit": 14 + "limit": 13 }, "LOG015": { "limit": 5 @@ -171,7 +171,7 @@ "limit": 175 }, "RUF012": { - "limit": 240 + "limit": 239 }, "RUF015": { "limit": 8 @@ -183,7 +183,7 @@ "limit": 4 }, "RUF059": { - "limit": 67 + "limit": 66 }, "RUF100": { "limit": 0 diff --git a/tests/test_litellm/integrations/test_langfuse.py b/tests/test_litellm/integrations/test_langfuse.py index 747f733a46d..f153ec1193c 100644 --- a/tests/test_litellm/integrations/test_langfuse.py +++ b/tests/test_litellm/integrations/test_langfuse.py @@ -1179,6 +1179,14 @@ def test_max_langfuse_clients_limit(): class _RecordingLangfuse: last_parameters: Optional[dict] = None + def __init__(self, environment=None, **parameters): + type(self).last_parameters = {"environment": environment, **parameters} + self.client = MagicMock() + + +class _RecordingLangfuseWithoutEnvironment: + last_parameters: Optional[dict] = None + def __init__(self, **parameters): type(self).last_parameters = parameters self.client = MagicMock() @@ -1195,6 +1203,62 @@ def _build_langfuse_logger(monkeypatch) -> LangFuseLogger: ) +def test_langfuse_environment_is_passed_to_sdk_client(monkeypatch): + monkeypatch.setenv("LANGFUSE_MOCK", "false") + monkeypatch.delenv("LANGFUSE_TRACING_ENVIRONMENT", raising=False) + monkeypatch.setattr(litellm, "initialized_langfuse_clients", 0) + with patch("langfuse.Langfuse", _RecordingLangfuse): + logger = LangFuseLogger( + langfuse_public_key="pk-env", + langfuse_secret="sk-env", + langfuse_host="https://test.langfuse.com", + langfuse_environment="staging", + ) + assert logger.langfuse_environment == "staging" + assert _RecordingLangfuse.last_parameters["environment"] == "staging" + + +def test_langfuse_environment_falls_back_to_deployment_env_var(monkeypatch): + monkeypatch.setenv("LANGFUSE_MOCK", "false") + monkeypatch.setenv("LANGFUSE_TRACING_ENVIRONMENT", "deployment-wide") + monkeypatch.setattr(litellm, "initialized_langfuse_clients", 0) + with patch("langfuse.Langfuse", _RecordingLangfuse): + logger = LangFuseLogger( + langfuse_public_key="pk-env", + langfuse_secret="sk-env", + langfuse_host="https://test.langfuse.com", + ) + assert logger.langfuse_environment == "deployment-wide" + assert _RecordingLangfuse.last_parameters["environment"] == "deployment-wide" + + +def test_langfuse_environment_omitted_for_old_sdk_versions(monkeypatch): + monkeypatch.setenv("LANGFUSE_MOCK", "false") + monkeypatch.setattr(litellm, "initialized_langfuse_clients", 0) + with patch("langfuse.Langfuse", _RecordingLangfuseWithoutEnvironment): + LangFuseLogger( + langfuse_public_key="pk-env", + langfuse_secret="sk-env", + langfuse_host="https://test.langfuse.com", + langfuse_environment="staging", + ) + assert "environment" not in _RecordingLangfuseWithoutEnvironment.last_parameters + + +def test_dynamic_langfuse_environment_triggers_dynamic_logger(): + from litellm.integrations.langfuse.langfuse_handler import LangFuseHandler + from litellm.types.utils import StandardCallbackDynamicParams + + params = StandardCallbackDynamicParams(langfuse_environment="team-a-env") + + assert LangFuseHandler._dynamic_langfuse_credentials_are_passed(params) is True + + config = LangFuseHandler.get_dynamic_langfuse_logging_config( + standard_callback_dynamic_params=params + ) + assert config["langfuse_environment"] == "team-a-env" + + def test_langfuse_sdk_client_survives_httpx_cache_eviction(monkeypatch): import gc import weakref @@ -1408,3 +1472,52 @@ def test_update_trace_keys_matches_whole_keys_not_substrings(): ) assert "input" not in trace_params + + +def test_langfuse_environment_is_coerced_and_validated(monkeypatch): + monkeypatch.setenv("LANGFUSE_MOCK", "false") + monkeypatch.delenv("LANGFUSE_TRACING_ENVIRONMENT", raising=False) + monkeypatch.setattr(litellm, "initialized_langfuse_clients", 0) + with patch("langfuse.Langfuse", _RecordingLangfuse): + logger = LangFuseLogger( + langfuse_public_key="pk-env", + langfuse_secret="sk-env", + langfuse_host="https://test.langfuse.com", + langfuse_environment=123, # non-string: must coerce, not crash + ) + assert logger.langfuse_environment == "123" + + with pytest.raises(ValueError, match="langfuse_environment"): + LangFuseLogger( + langfuse_public_key="pk-env", + langfuse_secret="sk-env", + langfuse_host="https://test.langfuse.com", + langfuse_environment="Production", + ) + + +def test_langfuse_empty_environment_falls_back_and_is_not_dynamic(monkeypatch): + from litellm.integrations.langfuse.langfuse_handler import LangFuseHandler + from litellm.types.utils import StandardCallbackDynamicParams + + monkeypatch.setenv("LANGFUSE_TRACING_ENVIRONMENT", "production") + + # '' falls back to the deployment env var at init + monkeypatch.setenv("LANGFUSE_MOCK", "false") + monkeypatch.setattr(litellm, "initialized_langfuse_clients", 0) + with patch("langfuse.Langfuse", _RecordingLangfuse): + logger = LangFuseLogger( + langfuse_public_key="pk-env", + langfuse_secret="sk-env", + langfuse_host="https://test.langfuse.com", + langfuse_environment="", + ) + assert logger.langfuse_environment == "production" + + # env-only params that add nothing do not select a dynamic logger + for redundant in ["", " ", "production"]: + params = StandardCallbackDynamicParams(langfuse_environment=redundant) + assert LangFuseHandler._dynamic_langfuse_credentials_are_passed(params) is False + + params = StandardCallbackDynamicParams(langfuse_environment="team-a-prod") + assert LangFuseHandler._dynamic_langfuse_credentials_are_passed(params) is True diff --git a/tests/test_litellm/integrations/test_langfuse_otel.py b/tests/test_litellm/integrations/test_langfuse_otel.py index 89607494920..0a9ce55fe16 100644 --- a/tests/test_litellm/integrations/test_langfuse_otel.py +++ b/tests/test_litellm/integrations/test_langfuse_otel.py @@ -137,6 +137,32 @@ class TestLangfuseOtelIntegration: mock_span, "langfuse.environment", test_env ) + def test_set_langfuse_environment_attribute_prefers_dynamic_param(self): + """Per-key/team langfuse_environment beats the deployment env var.""" + + class _RecordingSpan: + def __init__(self): + self.attributes = {} + + def set_attribute(self, key, value): + self.attributes[key] = value + + span = _RecordingSpan() + mock_kwargs = { + "standard_callback_dynamic_params": { + "langfuse_environment": "team-a-env" + } + } + + with patch.dict( + os.environ, {"LANGFUSE_TRACING_ENVIRONMENT": "deployment-wide"} + ): + LangfuseOtelLogger._set_langfuse_specific_attributes( + span, mock_kwargs, {} + ) + + assert span.attributes["langfuse.environment"] == "team-a-env" + def test_extract_langfuse_metadata_basic(self): """Ensure metadata is correctly pulled from litellm_params.""" metadata_in = {"generation_name": "my-gen", "custom": "data"} diff --git a/tests/test_litellm/litellm_core_utils/test_initialize_dynamic_callback_params.py b/tests/test_litellm/litellm_core_utils/test_initialize_dynamic_callback_params.py index 9b2bd5e2585..fe965f75f8f 100644 --- a/tests/test_litellm/litellm_core_utils/test_initialize_dynamic_callback_params.py +++ b/tests/test_litellm/litellm_core_utils/test_initialize_dynamic_callback_params.py @@ -233,3 +233,18 @@ def test_trusted_vars_overlay_uses_shared_parser_semantics(): ) assert params.get("newrelic_api_key") == "12345" + + +def test_validate_langfuse_environment_value(): + import pytest + + from litellm.litellm_core_utils.initialize_dynamic_callback_params import ( + validate_langfuse_environment_value, + ) + + validate_langfuse_environment_value("team-a-prod") + validate_langfuse_environment_value("staging_2") + + for bad in ["Production", "langfuse-eu", "", "team a"]: + with pytest.raises(ValueError, match="langfuse_environment"): + validate_langfuse_environment_value(bad) diff --git a/tests/test_litellm/llms/gemini/test_cost_calculator.py b/tests/test_litellm/llms/gemini/test_cost_calculator.py index 2f6f9a556c0..f1d62800337 100644 --- a/tests/test_litellm/llms/gemini/test_cost_calculator.py +++ b/tests/test_litellm/llms/gemini/test_cost_calculator.py @@ -363,6 +363,36 @@ def test_gemini_image_generation_cost_no_web_search_when_absent(monkeypatch): assert cost_zero == cost_none +@pytest.mark.parametrize( + "traffic_type, expected_service_tier", + [ + ("ON_DEMAND", None), + ("ON_DEMAND_PRIORITY", "priority"), + ("FLEX", "flex"), + ("BATCH", "flex"), + # Vertex AI reports flex/shared-capacity traffic as ON_DEMAND_FLEX. + ("ON_DEMAND_FLEX", "flex"), + # trafficType is matched case-insensitively. + ("on_demand_flex", "flex"), + (None, None), + ("SOMETHING_UNKNOWN", None), + ], +) +def test_map_traffic_type_to_service_tier( + traffic_type: str | None, expected_service_tier: str | None +): + """ + Gemini/Vertex usageMetadata.trafficType maps to the LiteLLM service_tier + that selects flex/priority cost keys. ON_DEMAND_FLEX (Vertex's flex opt-in + value) must map to "flex" so flex-tier requests are not billed as standard. + """ + from litellm.cost_calculator import _map_traffic_type_to_service_tier + + assert ( + _map_traffic_type_to_service_tier(traffic_type) == expected_service_tier + ) + + @pytest.mark.parametrize( "model,custom_llm_provider,expected_cache_read_cost", [ @@ -378,7 +408,7 @@ def test_flash_alias_cache_read_is_ten_percent_of_input( monkeypatch, model, custom_llm_provider, expected_cache_read_cost ): monkeypatch.setenv("LITELLM_LOCAL_MODEL_COST_MAP", "True") - litellm.model_cost = litellm.get_model_cost_map(url="") + monkeypatch.setattr(litellm, "model_cost", litellm.get_model_cost_map(url="")) model_info = litellm.get_model_info( model=model, custom_llm_provider=custom_llm_provider @@ -399,7 +429,7 @@ def test_flash_alias_cache_read_is_ten_percent_of_input( ) def test_flash_latest_alias_spellings_price_identically(monkeypatch, prefixed, bare): monkeypatch.setenv("LITELLM_LOCAL_MODEL_COST_MAP", "True") - litellm.model_cost = litellm.get_model_cost_map(url="") + monkeypatch.setattr(litellm, "model_cost", litellm.get_model_cost_map(url="")) prefixed_entry = litellm.model_cost[prefixed] bare_entry = litellm.model_cost[bare] diff --git a/tests/test_litellm/proxy/common_utils/test_callback_config_validation.py b/tests/test_litellm/proxy/common_utils/test_callback_config_validation.py new file mode 100644 index 00000000000..5a06bb92059 --- /dev/null +++ b/tests/test_litellm/proxy/common_utils/test_callback_config_validation.py @@ -0,0 +1,12 @@ +from litellm.proxy.common_utils.callback_config_validation import ( + callback_config_error, +) + + +def test_callback_config_error_rejects_invalid_langfuse_environment(): + for callback in ["langfuse", "langfuse_otel"]: + error = callback_config_error(callback, {"langfuse_environment": "Production"}) + assert error is not None and "langfuse_environment" in error + + assert callback_config_error("langfuse", {"langfuse_environment": "team-a-prod"}) is None + assert callback_config_error("langfuse", {"langfuse_public_key": "pk"}) is None diff --git a/tests/test_litellm/test_cost_calculator.py b/tests/test_litellm/test_cost_calculator.py index dc2fbe3ed73..4117b2fce6d 100644 --- a/tests/test_litellm/test_cost_calculator.py +++ b/tests/test_litellm/test_cost_calculator.py @@ -2619,6 +2619,49 @@ def test_completion_cost_anthropic_auto_tier_uses_served_priority_rate(_local_mo assert cost == pytest.approx(expected_priority) +def test_completion_cost_vertex_ai_gemini_flex_traffic_type(_local_model_cost_map): + """ + Vertex AI flex-tier billing regression for issue #37647. + + Vertex Gemini 3.x models route through ``cost_per_character`` (the + ``cost_router`` token-path gate only matches "gemini-2"), and its token + fallbacks dropped ``service_tier``. A response served with + ``trafficType=ON_DEMAND_FLEX`` must be billed at the flex rate, not the + standard rate. + """ + from litellm import completion_cost + + model = "gemini-3-test-flex-tier-cost-model" + litellm.register_model( + model_cost={ + model: { + "input_cost_per_token": 1.5e-6, + "output_cost_per_token": 9e-6, + "input_cost_per_token_flex": 7.5e-7, + "output_cost_per_token_flex": 4.5e-6, + "litellm_provider": "vertex_ai", + "max_tokens": 8192, + } + } + ) + + def _cost_for_traffic_type(traffic_type): + usage = Usage(prompt_tokens=1000, completion_tokens=500, total_tokens=1500) + response = ModelResponse(usage=usage, model=model) + response._hidden_params["provider_specific_fields"] = {"traffic_type": traffic_type} + return completion_cost( + completion_response=response, + model=model, + custom_llm_provider="vertex_ai", + ) + + standard_cost = _cost_for_traffic_type("ON_DEMAND") + flex_cost = _cost_for_traffic_type("ON_DEMAND_FLEX") + + assert standard_cost == pytest.approx(1000 * 1.5e-6 + 500 * 9e-6) + assert flex_cost == pytest.approx(1000 * 7.5e-7 + 500 * 4.5e-6) + + def test_completion_cost_non_string_service_tier_defers_to_served_tier(_local_model_cost_map): """ Regression: a non-string request-level ``service_tier`` (reachable via diff --git a/tests/test_litellm/test_main.py b/tests/test_litellm/test_main.py index 8f2b06be4b3..3eea47bcd5a 100644 --- a/tests/test_litellm/test_main.py +++ b/tests/test_litellm/test_main.py @@ -1,7 +1,12 @@ +import asyncio +import base64 import contextlib import copy import json import os +from collections.abc import Mapping +from dataclasses import dataclass +from typing import Final import httpx import pytest @@ -14,6 +19,9 @@ from unittest.mock import MagicMock, patch import litellm from litellm import main as litellm_main +from litellm.integrations.custom_logger import CustomLogger +from litellm.litellm_core_utils.core_helpers import get_litellm_metadata_from_kwargs +from litellm.types.utils import Usage async def _async_fake_bedrock_image_details(image_url): @@ -2957,3 +2965,109 @@ async def test_acompletion_resolves_provider_from_api_base(): ) assert response.choices[0].message.content == "resolved" + + +@dataclass(frozen=True, slots=True) +class _RecordedSpeechSuccess: + call_type: str | None + spend_metadata: Mapping[str, object] + response_cost: float | None + logged_response_cost: float | None + + +def _record_speech_success(payload: dict[str, object]) -> _RecordedSpeechSuccess: + call_type: Final = payload.get("call_type") + response_cost: Final = payload.get("response_cost") + logging_payload: Final = payload.get("standard_logging_object") + logged_cost: Final = logging_payload.get("response_cost") if isinstance(logging_payload, dict) else None + return _RecordedSpeechSuccess( + call_type=call_type if isinstance(call_type, str) else None, + spend_metadata=get_litellm_metadata_from_kwargs(payload), + response_cost=response_cost if isinstance(response_cost, float) else None, + logged_response_cost=logged_cost if isinstance(logged_cost, float) else None, + ) + + +class _SuccessEventRecorder(CustomLogger): + def __init__(self) -> None: + super().__init__() + self.events: list[_RecordedSpeechSuccess] = [] # mutable-ok: test recorder of success-callback events + + async def async_log_success_event( + self, kwargs: dict[str, object], response_obj: object, start_time: object, end_time: object + ) -> None: + self.events.append(_record_speech_success(kwargs)) + + +async def _wait_for_success_event(recorder: _SuccessEventRecorder, call_type: str) -> _RecordedSpeechSuccess: + for _ in range(100): + if (event := next((e for e in recorder.events if e.call_type == call_type), None)) is not None: + return event + await asyncio.sleep(0.05) + pytest.fail(f"no {call_type} success event; got {[e.call_type for e in recorder.events]}") + + +def _gemini_tts_generate_content_response() -> dict[str, object]: + return { + "candidates": [ + { + "content": { + "parts": [ + { + "inlineData": { + "mimeType": "audio/L16;codec=pcm;rate=24000", + "data": base64.b64encode(b"pcm-audio-bytes").decode(), + } + } + ], + "role": "model", + }, + "finishReason": "STOP", + "index": 0, + } + ], + "usageMetadata": { + "promptTokenCount": 5, + "candidatesTokenCount": 60, + "totalTokenCount": 65, + "promptTokensDetails": [{"modality": "TEXT", "tokenCount": 5}], + "candidatesTokensDetails": [{"modality": "AUDIO", "tokenCount": 60}], + }, + "modelVersion": "gemini-2.5-flash-preview-tts", + } + + +@pytest.mark.asyncio +async def test_aspeech_gemini_bridge_keeps_proxy_metadata_for_spend_tracking( + respx_mock: respx.MockRouter, monkeypatch: pytest.MonkeyPatch +) -> None: + monkeypatch.setattr(litellm, "disable_aiohttp_transport", True) + monkeypatch.delenv("GEMINI_API_KEY", raising=False) + monkeypatch.delenv("GOOGLE_API_KEY", raising=False) + recorder: Final = _SuccessEventRecorder() + monkeypatch.setattr(litellm, "callbacks", [recorder]) + mock_route: Final = respx_mock.post( + url__regex=r"https://generativelanguage\.googleapis\.com/v1beta/models/gemini-2\.5-flash-preview-tts:generateContent.*" + ).mock(return_value=httpx.Response(200, json=_gemini_tts_generate_content_response())) + + await litellm.aspeech( + model="gemini/gemini-2.5-flash-preview-tts", + input="spend tracking check", + voice="Kore", + api_key="fake-gemini-key", + metadata={"user_api_key": "hashed-virtual-key", "user_api_key_user_id": "user-1"}, + ) + + assert mock_route.called + assert mock_route.calls.last.request.headers["x-goog-api-key"] == "fake-gemini-key" + speech_event: Final = await _wait_for_success_event(recorder, call_type="aspeech") + assert speech_event.spend_metadata["user_api_key"] == "hashed-virtual-key" + assert speech_event.spend_metadata["user_api_key_user_id"] == "user-1" + expected_prompt_cost, expected_completion_cost = litellm.cost_per_token( + model="gemini/gemini-2.5-flash-preview-tts", + usage_object=Usage(prompt_tokens=5, completion_tokens=60, total_tokens=65), + ) + expected_cost: Final = expected_prompt_cost + expected_completion_cost + assert expected_cost > 0 + assert speech_event.response_cost == pytest.approx(expected_cost) + assert speech_event.logged_response_cost == pytest.approx(expected_cost) diff --git a/tests/test_litellm/types/llms/test_types_llms_openai.py b/tests/test_litellm/types/llms/test_types_llms_openai.py index 3966677e928..42719ce838b 100644 --- a/tests/test_litellm/types/llms/test_types_llms_openai.py +++ b/tests/test_litellm/types/llms/test_types_llms_openai.py @@ -7,6 +7,7 @@ import pytest import json import litellm +from litellm.types.llms.openai import HttpxBinaryResponseContent def test_generic_event(): @@ -522,3 +523,34 @@ class TestOpenAIFileObjectBatchGuardrailSerialization: page = FileListPage(object="list", data=[self._file_object()], has_more=False) assert "litellm_batch_guardrail" not in page.model_dump(mode="json")["data"][0] + + +def _binary_content(payload: bytes) -> HttpxBinaryResponseContent: + import httpx + + return HttpxBinaryResponseContent(httpx.Response(200, content=payload)) + + +def test_httpx_binary_response_content_hidden_params_are_per_instance(): + first = _binary_content(b"first") + second = _binary_content(b"second") + + first._hidden_params["response_cost"] = 0.5 + + assert second._hidden_params == {} + + +def test_set_response_cost_none_leaves_hidden_params_empty(): + binary_response = _binary_content(b"audio") + + binary_response.set_response_cost(None) + + assert "response_cost" not in binary_response._hidden_params + + binary_response.set_response_cost(0.25) + + assert binary_response._hidden_params["response_cost"] == 0.25 + + binary_response.set_response_cost(None) + + assert "response_cost" not in binary_response._hidden_params diff --git a/ui/litellm-dashboard/src/components/callback_info_helpers.tsx b/ui/litellm-dashboard/src/components/callback_info_helpers.tsx index 3906aa744f7..4b6f6233fe8 100644 --- a/ui/litellm-dashboard/src/components/callback_info_helpers.tsx +++ b/ui/litellm-dashboard/src/components/callback_info_helpers.tsx @@ -109,6 +109,7 @@ export const CALLBACK_CONFIGS: CallbackConfig[] = [ langfuse_public_key: "text", langfuse_secret_key: "password", langfuse_host: "text", + langfuse_environment: "text", }, description: "Langfuse v2 Logging Integration", }, @@ -121,6 +122,7 @@ export const CALLBACK_CONFIGS: CallbackConfig[] = [ langfuse_public_key: "text", langfuse_secret_key: "password", langfuse_host: "text", + langfuse_environment: "text", }, description: "Langfuse v3 OTEL Logging Integration", }, diff --git a/ui/litellm-dashboard/src/lib/http/schema.d.ts b/ui/litellm-dashboard/src/lib/http/schema.d.ts index d660d6b5aca..4ed460da833 100644 --- a/ui/litellm-dashboard/src/lib/http/schema.d.ts +++ b/ui/litellm-dashboard/src/lib/http/schema.d.ts @@ -14836,6 +14836,7 @@ export interface paths { * - langfuse_secret_key: The secret key for the Langfuse callback * - langfuse_secret: The secret for the Langfuse callback * - langfuse_host: The host for the Langfuse callback + * - langfuse_environment: The tracing environment for the Langfuse callback (lowercase; falls back to LANGFUSE_TRACING_ENVIRONMENT) * - gcs_bucket_name: The name of the GCS bucket * - gcs_path_service_account: The path to the GCS service account * - langsmith_api_key: The API key for the Langsmith callback