mirror of
https://github.com/BerriAI/litellm.git
synced 2026-09-05 08:07:05 +00:00
fix(spend_tracking): add missing_session_id: omit to leave SpendLogs.session_id null without a client session (#39458)
* fix(spend_tracking): leave SpendLogs.session_id null when no client session id was established Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> * chore(lint): ratchet basedpyright budget after session_id fix Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> * fix(spend_tracking): ignore trace ids as session ids Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> * fix(spend_tracking): gate null SpendLogs.session_id behind missing_session_id: omit Unset, generate and reject keep the legacy trace id fallback. omit records only metadata.session_id, the key Langfuse reads, so a trace id copied into litellm_session_id by get_litellm_params never becomes a session. Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> * fix(spend_tracking): stamp the omit decision on the request so a config reload cannot fabricate a session Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> * fix(spend_tracking): keep omit covering requests the pre-call stamp never reaches Router-model provider pass-through calls allm_passthrough_route directly and skips add_litellm_data_to_request, so those requests never run the pre-call helper and carry no omit stamp. Reading only the stamp made POST /anthropic/v1/messages write a fabricated uuid into SpendLogs.session_id under missing_session_id: omit while its Langfuse trace had no session, the exact divergence the policy exists to remove. The stamp now only pins omit on, and an unstamped request falls back to the configured policy, so a config reload still cannot fabricate a session for a request that was decided pre-call. * fix(spend_tracking): make the session-omission marker proxy-owned so clients cannot forge it * fix(spend_tracking): strip the client-sent omission marker from both metadata buckets before they merge Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> * fix(spend_tracking): strip the session-omission marker from both metadata buckets The pre-call policy ran before litellm_metadata is merged into metadata, so a client that planted the marker in litellm_metadata had it copied back into the route's own bucket after the strip and still got a null SpendLogs.session_id. --------- Co-authored-by: yassin <yassin@berri.ai> Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> Co-authored-by: mateo-berri <277851410+mateo-berri@users.noreply.github.com>
This commit is contained in:
devin-ai-integration[bot]
GitHub
parent
c2265b0ef3
commit
e046aee3d5
11 changed files with 519 additions and 720 deletions
|
|
@ -57,7 +57,7 @@
|
|||
"limit": 5601
|
||||
},
|
||||
"reportMissingTypeArgument": {
|
||||
"limit": 15288
|
||||
"limit": 15287
|
||||
},
|
||||
"reportMissingTypeStubs": {
|
||||
"limit": 40
|
||||
|
|
@ -105,10 +105,10 @@
|
|||
"limit": 109
|
||||
},
|
||||
"reportUnknownMemberType": {
|
||||
"limit": 38324
|
||||
"limit": 38323
|
||||
},
|
||||
"reportUnknownParameterType": {
|
||||
"limit": 19625
|
||||
"limit": 19624
|
||||
},
|
||||
"reportUnknownVariableType": {
|
||||
"limit": 29861
|
||||
|
|
|
|||
|
|
@ -1450,6 +1450,7 @@ SESSION_DEPLOYMENT_AFFINITY_TTL_METADATA_KEY: Final = "_session_deployment_affin
|
|||
CONSUMED_REQUEST_TAGS_METADATA_KEY: Final = "_consumed_request_tags"
|
||||
INTERNAL_CALL_ORIGIN_METADATA_KEY: Final = "internal_call_origin"
|
||||
SESSION_ID_GENERATED_METADATA_KEY: Final = "litellm_session_id_generated"
|
||||
SESSION_ID_OMITTED_METADATA_KEY: Final = "litellm_session_id_omitted"
|
||||
LITELLM_TRUNCATED_PAYLOAD_FIELD: Final = "litellm_truncated"
|
||||
LITELLM_TRUNCATION_DB_SAFEGUARD_NOTE: Final = (
|
||||
"Truncation is a DB storage safeguard. "
|
||||
|
|
|
|||
|
|
@ -2608,9 +2608,9 @@ class ConfigGeneralSettings(LiteLLMPydanticObjectBase):
|
|||
None,
|
||||
description="When set to True, rejects requests that contain client-side 'metadata.tags' to prevent users from influencing budgets by sending different tags. Tags can only be inherited from the API key metadata.",
|
||||
)
|
||||
missing_session_id: Literal["generate", "reject"] | None = Field(
|
||||
missing_session_id: Literal["generate", "reject", "omit"] | None = Field(
|
||||
None,
|
||||
description="What to do with LLM API requests that carry no session id (x-litellm-session-id header, metadata.session_id, etc.). 'generate' stamps one id into litellm_session_id, litellm_trace_id and metadata.session_id so SpendLogs and logging callbacks agree; 'reject' returns 400. Unset keeps the legacy behavior where SpendLogs falls back to the trace id while callbacks get no session id.",
|
||||
description="What to do with LLM API requests that carry no session id (x-litellm-session-id header, metadata.session_id, etc.). 'generate' stamps one id into litellm_session_id, litellm_trace_id and metadata.session_id so SpendLogs and logging callbacks agree; 'reject' returns 400; 'omit' leaves SpendLogs.session_id null, matching callbacks such as Langfuse that only record a client-established metadata.session_id. Unset keeps the legacy behavior where SpendLogs falls back to the trace id while callbacks get no session id.",
|
||||
)
|
||||
enable_public_model_hub: bool = Field(
|
||||
default=False,
|
||||
|
|
|
|||
|
|
@ -168,11 +168,8 @@ class _ProxyDBLogger(CustomLogger):
|
|||
"custom_llm_provider"
|
||||
) or request_data.get("custom_llm_provider", "")
|
||||
|
||||
# Propagate standard_logging_object and litellm_trace_id from the
|
||||
# Logging instance so that _get_session_id_for_spend_log uses the same
|
||||
# trace_id that Langfuse received (via async_failure_handler).
|
||||
# Without this, the DB session_id would be a random UUID that doesn't
|
||||
# match the Langfuse trace_id, making failed requests unsearchable.
|
||||
# Propagate standard_logging_object and litellm_trace_id from the Logging
|
||||
# instance so the failure row carries the same trace_id Langfuse received.
|
||||
_litellm_logging_obj: Final = request_data.get("litellm_logging_obj")
|
||||
if _litellm_logging_obj is not None:
|
||||
if not request_data.get("standard_logging_object"):
|
||||
|
|
|
|||
|
|
@ -25,6 +25,7 @@ from litellm.constants import (
|
|||
PRE_CALL_EXECUTED_GUARDRAILS_KEY,
|
||||
SESSION_DEPLOYMENT_AFFINITY_TTL_METADATA_KEY,
|
||||
SESSION_ID_GENERATED_METADATA_KEY,
|
||||
SESSION_ID_OMITTED_METADATA_KEY,
|
||||
)
|
||||
from litellm.litellm_core_utils.credential_accessor import CredentialAccessor
|
||||
from litellm.litellm_core_utils.initialize_dynamic_callback_params import (
|
||||
|
|
@ -733,12 +734,18 @@ def apply_missing_session_id_policy(
|
|||
general_settings: Mapping[str, object] | None,
|
||||
request: Request,
|
||||
) -> None:
|
||||
for metadata_key in ("metadata", "litellm_metadata"):
|
||||
if isinstance(client_metadata := data.get(metadata_key), dict):
|
||||
client_metadata.pop(SESSION_ID_OMITTED_METADATA_KEY, None)
|
||||
metadata: Final = data.get(_metadata_variable_name)
|
||||
policy: Final = general_settings.get("missing_session_id") if general_settings else None
|
||||
if policy is None or not _is_llm_inference_route(request):
|
||||
return
|
||||
metadata: Final = data.get(_metadata_variable_name)
|
||||
if not isinstance(metadata, dict):
|
||||
return
|
||||
if policy == "omit":
|
||||
metadata[SESSION_ID_OMITTED_METADATA_KEY] = True
|
||||
return
|
||||
if data.get("litellm_session_id") or metadata.get("session_id"):
|
||||
return
|
||||
match policy:
|
||||
|
|
@ -760,7 +767,8 @@ def apply_missing_session_id_policy(
|
|||
)
|
||||
case _:
|
||||
verbose_proxy_logger.warning(
|
||||
"Ignoring unknown general_settings.missing_session_id=%r; expected 'generate' or 'reject'", policy
|
||||
"Ignoring unknown general_settings.missing_session_id=%r; expected 'generate', 'reject' or 'omit'",
|
||||
policy,
|
||||
)
|
||||
|
||||
|
||||
|
|
|
|||
|
|
@ -40,6 +40,7 @@ from litellm._logging import verbose_proxy_logger
|
|||
from litellm._uuid import uuid
|
||||
from litellm.constants import (
|
||||
MAXIMUM_TRACEBACK_LINES_TO_LOG,
|
||||
SESSION_ID_OMITTED_METADATA_KEY,
|
||||
WEBSOCKET_CLOSE_REASON_MAX_BYTES,
|
||||
)
|
||||
from litellm.integrations.custom_guardrail import CustomGuardrail
|
||||
|
|
@ -581,8 +582,9 @@ class HttpPassThroughEndpointHelpers(BasePassthroughUtils):
|
|||
)
|
||||
|
||||
# Set internal keys after merging client-supplied metadata so a request
|
||||
# body that mirrors them cannot clobber the authenticated key or the
|
||||
# real parent span.
|
||||
# body that mirrors them cannot clobber the authenticated key, the real
|
||||
# parent span, or the proxy's own session-id decision.
|
||||
_metadata.pop(SESSION_ID_OMITTED_METADATA_KEY, None)
|
||||
_metadata["user_api_key"] = user_api_key_dict.api_key
|
||||
_metadata["litellm_parent_otel_span"] = user_api_key_dict.parent_otel_span
|
||||
_metadata["user_api_key_budget_reservation"] = user_api_key_dict.budget_reservation
|
||||
|
|
|
|||
|
|
@ -15,6 +15,7 @@ from litellm.constants import (
|
|||
LITELLM_TRUNCATED_PAYLOAD_FIELD,
|
||||
LITELLM_TRUNCATION_DB_SAFEGUARD_NOTE,
|
||||
REDACTED_BY_LITELM_STRING,
|
||||
SESSION_ID_OMITTED_METADATA_KEY,
|
||||
)
|
||||
from litellm.constants import (
|
||||
MAX_STRING_LENGTH_PROMPT_IN_DB as DEFAULT_MAX_STRING_LENGTH_PROMPT_IN_DB,
|
||||
|
|
@ -578,7 +579,9 @@ def get_logging_payload(kwargs, response_obj, start_time, end_time) -> SpendLogs
|
|||
),
|
||||
session_id=_get_session_id_for_spend_log(
|
||||
kwargs=kwargs,
|
||||
metadata=metadata,
|
||||
standard_logging_payload=standard_logging_payload,
|
||||
omit_when_missing=_omits_session_id_when_missing(metadata),
|
||||
),
|
||||
request_duration_ms=_get_request_duration_ms(start_time, end_time),
|
||||
status=_get_status_for_spend_log(
|
||||
|
|
@ -602,26 +605,39 @@ def get_logging_payload(kwargs, response_obj, start_time, end_time) -> SpendLogs
|
|||
raise e
|
||||
|
||||
|
||||
def _omits_session_id_when_missing(metadata: Mapping[str, object] | None) -> bool:
|
||||
"""The pre-call stamp pins `omit` on for the requests that carry it, so a config reload between pre-call and spend
|
||||
logging cannot fabricate a session. `apply_missing_session_id_policy` drops any client-supplied copy of the key
|
||||
from both metadata buckets before stamping, which the merge of `litellm_metadata` into `metadata` makes
|
||||
necessary, so a caller cannot forge it. Requests that never reach the pre-call helper, router-model
|
||||
passthrough among them, carry no stamp, so they fall back to the configured policy and `omit` still covers their
|
||||
spend logs."""
|
||||
if metadata is not None and metadata.get(SESSION_ID_OMITTED_METADATA_KEY):
|
||||
return True
|
||||
|
||||
from litellm.proxy.proxy_server import general_settings
|
||||
|
||||
return general_settings.get("missing_session_id") == "omit"
|
||||
|
||||
|
||||
def _get_session_id_for_spend_log(
|
||||
kwargs: dict,
|
||||
kwargs: Mapping[str, object],
|
||||
metadata: Mapping[str, object] | None,
|
||||
standard_logging_payload: StandardLoggingPayload | None,
|
||||
) -> str:
|
||||
"""
|
||||
Get the session id for the spend log.
|
||||
omit_when_missing: bool,
|
||||
) -> str | None:
|
||||
"""Under `omit` only `metadata.session_id`, the key Langfuse reads, counts as a session; `litellm_session_id` may
|
||||
be a copied trace id."""
|
||||
if omit_when_missing:
|
||||
session_id: Final = metadata.get("session_id") if metadata else None
|
||||
return str(session_id) if session_id else None
|
||||
|
||||
This ensures each spend log is associated with a unique session id.
|
||||
|
||||
"""
|
||||
from litellm._uuid import uuid
|
||||
|
||||
if standard_logging_payload is not None and standard_logging_payload.get("trace_id") is not None:
|
||||
return str(standard_logging_payload.get("trace_id"))
|
||||
|
||||
# Users can dynamically set the trace_id for each request by passing `litellm_trace_id` in kwargs
|
||||
if kwargs.get("litellm_trace_id") is not None:
|
||||
return str(kwargs.get("litellm_trace_id"))
|
||||
|
||||
# Ensure we always have a session id, if none is provided
|
||||
return str(uuid.uuid4())
|
||||
|
||||
|
||||
|
|
|
|||
|
|
@ -5730,3 +5730,38 @@ async def test_pass_through_request_leaves_cost_router_logger_working():
|
|||
verbose_logger.removeHandler(recorder)
|
||||
|
||||
assert not raised, f"cost router logger raised on the passthrough logging params: {raised[0].exc_info}"
|
||||
|
||||
|
||||
@pytest.mark.parametrize("client_metadata_key", ["litellm_metadata", "metadata"])
|
||||
def test_passthrough_client_cannot_forge_session_id_omission(client_metadata_key: str):
|
||||
"""The omit marker is proxy-owned: only the pre-call policy may set it. A pass-through body that carries
|
||||
it in its own metadata must not null out SpendLogs.session_id on a request the proxy never omitted."""
|
||||
from litellm.constants import SESSION_ID_OMITTED_METADATA_KEY
|
||||
from litellm.proxy.spend_tracking.spend_tracking_utils import _get_session_id_for_spend_log
|
||||
|
||||
mock_request = MagicMock(spec=Request)
|
||||
mock_request.method = "POST"
|
||||
mock_request.url = "http://0.0.0.0:4000/gemini/v1beta/models/gemini-2.5-flash:generateContent"
|
||||
mock_request.headers = Headers({})
|
||||
mock_request.scope = {}
|
||||
|
||||
kwargs = HttpPassThroughEndpointHelpers._init_kwargs_for_pass_through_endpoint(
|
||||
request=mock_request,
|
||||
user_api_key_dict=UserAPIKeyAuth(api_key="hashed-key"),
|
||||
passthrough_logging_payload=MagicMock(),
|
||||
logging_obj=MagicMock(),
|
||||
_parsed_body={client_metadata_key: {SESSION_ID_OMITTED_METADATA_KEY: True}},
|
||||
litellm_call_id="lit-6694-call-id",
|
||||
)
|
||||
|
||||
metadata = kwargs["litellm_params"]["metadata"]
|
||||
assert SESSION_ID_OMITTED_METADATA_KEY not in metadata
|
||||
assert (
|
||||
_get_session_id_for_spend_log(
|
||||
kwargs={},
|
||||
metadata=metadata,
|
||||
standard_logging_payload={"trace_id": "per-call-random-trace-id"},
|
||||
omit_when_missing=bool(metadata.get(SESSION_ID_OMITTED_METADATA_KEY)),
|
||||
)
|
||||
== "per-call-random-trace-id"
|
||||
)
|
||||
|
|
|
|||
|
|
@ -14,6 +14,7 @@ from litellm.constants import (
|
|||
LITELLM_TRUNCATED_PAYLOAD_FIELD,
|
||||
LITELLM_TRUNCATION_DB_SAFEGUARD_NOTE,
|
||||
REDACTED_BY_LITELM_STRING,
|
||||
SESSION_ID_OMITTED_METADATA_KEY,
|
||||
)
|
||||
from litellm.litellm_core_utils.safe_json_dumps import safe_dumps
|
||||
from litellm.proxy.spend_tracking.spend_tracking_utils import (
|
||||
|
|
@ -21,6 +22,7 @@ from litellm.proxy.spend_tracking.spend_tracking_utils import (
|
|||
_get_proxy_server_request_for_spend_logs_payload,
|
||||
_get_request_duration_ms,
|
||||
_get_response_for_spend_logs_payload,
|
||||
_get_session_id_for_spend_log,
|
||||
_get_spend_logs_metadata,
|
||||
_get_vector_store_request_for_spend_logs_payload,
|
||||
_is_master_key,
|
||||
|
|
@ -33,6 +35,7 @@ from litellm.proxy.spend_tracking.spend_tracking_utils import (
|
|||
get_logging_payload,
|
||||
get_spend_logs_id,
|
||||
)
|
||||
from litellm.proxy._types import SpendLogsPayload
|
||||
from litellm.proxy.utils import hash_token
|
||||
from litellm.types.utils import (
|
||||
StandardLoggingHiddenParams,
|
||||
|
|
@ -74,6 +77,110 @@ def test_get_logging_payload_maps_openai_cached_tokens_to_cache_read_input_token
|
|||
assert additional_usage_values["prompt_tokens_details"]["cached_tokens"] == 123
|
||||
|
||||
|
||||
_TRACE_ONLY_STANDARD_LOGGING: Final = cast(
|
||||
StandardLoggingPayload,
|
||||
{
|
||||
"trace_id": "trace-abc",
|
||||
"session_id": "trace-abc",
|
||||
"metadata": {},
|
||||
"model_map_information": None,
|
||||
"request_tags": [],
|
||||
},
|
||||
)
|
||||
|
||||
|
||||
def _trace_only_session_id(omit_when_missing: bool) -> str | None:
|
||||
"""get_litellm_params copies metadata.trace_id into litellm_session_id, so every field echoes the trace id."""
|
||||
return _get_session_id_for_spend_log(
|
||||
kwargs={"litellm_trace_id": "trace-abc", "litellm_session_id": "trace-abc"},
|
||||
metadata={"trace_id": "trace-abc"},
|
||||
standard_logging_payload=_TRACE_ONLY_STANDARD_LOGGING,
|
||||
omit_when_missing=omit_when_missing,
|
||||
)
|
||||
|
||||
|
||||
def test_omit_leaves_session_id_none_when_only_a_trace_id_exists():
|
||||
assert _trace_only_session_id(omit_when_missing=True) is None
|
||||
|
||||
|
||||
def test_omit_leaves_session_id_none_without_any_ids():
|
||||
assert (
|
||||
_get_session_id_for_spend_log(kwargs={}, metadata=None, standard_logging_payload=None, omit_when_missing=True)
|
||||
is None
|
||||
)
|
||||
|
||||
|
||||
def test_omit_records_metadata_session_id():
|
||||
session_id: Final = _get_session_id_for_spend_log(
|
||||
kwargs={"litellm_session_id": "chain-1"},
|
||||
metadata={"trace_id": "chain-1", "session_id": "chain-1"},
|
||||
standard_logging_payload=_TRACE_ONLY_STANDARD_LOGGING,
|
||||
omit_when_missing=True,
|
||||
)
|
||||
assert session_id == "chain-1"
|
||||
|
||||
|
||||
def test_legacy_policy_keeps_trace_id_fallback():
|
||||
assert _trace_only_session_id(omit_when_missing=False) == "trace-abc"
|
||||
generated: Final = _get_session_id_for_spend_log(
|
||||
kwargs={}, metadata=None, standard_logging_payload=None, omit_when_missing=False
|
||||
)
|
||||
assert len(str(generated)) == 36
|
||||
|
||||
|
||||
@pytest.mark.parametrize(
|
||||
("request_metadata", "expected"),
|
||||
[
|
||||
({"trace_id": "trace-abc"}, "trace-abc"),
|
||||
({"trace_id": "trace-abc", SESSION_ID_OMITTED_METADATA_KEY: True}, None),
|
||||
({"trace_id": "trace-abc", "session_id": "chain-1", SESSION_ID_OMITTED_METADATA_KEY: True}, "chain-1"),
|
||||
],
|
||||
)
|
||||
def test_get_logging_payload_reads_omit_decision_stamped_on_request(
|
||||
request_metadata: dict[str, object], expected: str | None
|
||||
):
|
||||
"""The pre-call stamp, not the live general_settings, decides the policy, so a config reload between
|
||||
pre-call and spend logging cannot fabricate a session for a request accepted under `omit`."""
|
||||
with patch( # test-quality-ok: proves log time ignores proxy config; general_settings is yaml, not an HTTP boundary
|
||||
"litellm.proxy.proxy_server.general_settings", {"missing_session_id": "generate"}
|
||||
):
|
||||
payload: SpendLogsPayload = get_logging_payload(
|
||||
kwargs={
|
||||
"model": "gpt-4o-mini",
|
||||
"litellm_trace_id": "trace-abc",
|
||||
"litellm_params": {"litellm_session_id": "trace-abc", "metadata": request_metadata},
|
||||
"standard_logging_object": _TRACE_ONLY_STANDARD_LOGGING,
|
||||
},
|
||||
response_obj=litellm.ModelResponse(id="chatcmpl-test", choices=[]),
|
||||
start_time=datetime.datetime.now(timezone.utc),
|
||||
end_time=datetime.datetime.now(timezone.utc),
|
||||
)
|
||||
assert payload["session_id"] == expected
|
||||
|
||||
|
||||
@pytest.mark.parametrize("policy", ["omit", "generate", None])
|
||||
def test_get_logging_payload_applies_omit_to_requests_that_carry_no_stamp(policy: str | None):
|
||||
"""Router-model passthrough calls `allm_passthrough_route` directly and never reaches the pre-call helper that
|
||||
stamps the omit decision, so an unstamped request falls back to the configured policy. Without that fallback
|
||||
`missing_session_id: omit` would fabricate a uuid session id on every passthrough spend log while its Langfuse
|
||||
trace has none, which is the divergence the policy exists to remove."""
|
||||
with patch( # test-quality-ok: general_settings is proxy config, loaded from yaml, not an HTTP boundary
|
||||
"litellm.proxy.proxy_server.general_settings", {} if policy is None else {"missing_session_id": policy}
|
||||
):
|
||||
payload: SpendLogsPayload = get_logging_payload(
|
||||
kwargs={
|
||||
"model": "claude-opus-4",
|
||||
"litellm_trace_id": "trace-abc",
|
||||
"litellm_params": {"litellm_session_id": "trace-abc", "metadata": {"trace_id": "trace-abc"}},
|
||||
"standard_logging_object": _TRACE_ONLY_STANDARD_LOGGING,
|
||||
},
|
||||
response_obj=litellm.ModelResponse(id="chatcmpl-test", choices=[]),
|
||||
start_time=datetime.datetime.now(timezone.utc),
|
||||
end_time=datetime.datetime.now(timezone.utc),
|
||||
)
|
||||
assert payload["session_id"] == (None if policy == "omit" else "trace-abc")
|
||||
|
||||
|
||||
def test_get_logging_payload_preserves_anthropic_cache_read_input_tokens():
|
||||
additional_usage_values = _get_additional_usage_values_for_usage(
|
||||
litellm.Usage(
|
||||
|
|
@ -277,9 +384,7 @@ def test_sanitize_request_body_for_spend_logs_payload_long_string():
|
|||
from litellm.constants import MAX_STRING_LENGTH_PROMPT_IN_DB
|
||||
|
||||
# Create a string longer than MAX_STRING_LENGTH_PROMPT_IN_DB (2048)
|
||||
long_string = (
|
||||
"a" * 3000
|
||||
) # Create a string longer than MAX_STRING_LENGTH_PROMPT_IN_DB
|
||||
long_string = "a" * 3000 # Create a string longer than MAX_STRING_LENGTH_PROMPT_IN_DB
|
||||
request_body = {"text": long_string, "normal_text": "short text"}
|
||||
sanitized = _sanitize_request_body_for_spend_logs_payload(request_body)
|
||||
|
||||
|
|
@ -329,9 +434,7 @@ def test_sanitize_request_body_for_spend_logs_payload_nested_list():
|
|||
|
||||
# Create a string longer than MAX_STRING_LENGTH_PROMPT_IN_DB
|
||||
long_string = "a" * (MAX_STRING_LENGTH_PROMPT_IN_DB + 500)
|
||||
request_body = {
|
||||
"items": [{"text": long_string}, {"text": "short"}, [{"text": long_string}]]
|
||||
}
|
||||
request_body = {"items": [{"text": long_string}, {"text": "short"}, [{"text": long_string}]]}
|
||||
sanitized = _sanitize_request_body_for_spend_logs_payload(request_body)
|
||||
|
||||
# Calculate expected lengths based on actual MAX_STRING_LENGTH_PROMPT_IN_DB
|
||||
|
|
@ -415,14 +518,10 @@ def test_sanitize_request_body_for_spend_logs_payload_circular_reference():
|
|||
|
||||
# Test that it handles circular reference without infinite recursion
|
||||
sanitized = _sanitize_request_body_for_spend_logs_payload(a)
|
||||
assert sanitized == {
|
||||
"b": {"a": {}}
|
||||
} # Should return empty dict for circular reference
|
||||
assert sanitized == {"b": {"a": {}}} # Should return empty dict for circular reference
|
||||
|
||||
|
||||
@patch(
|
||||
"litellm.proxy.spend_tracking.spend_tracking_utils._should_store_prompts_and_responses_in_spend_logs"
|
||||
)
|
||||
@patch("litellm.proxy.spend_tracking.spend_tracking_utils._should_store_prompts_and_responses_in_spend_logs")
|
||||
def test_get_vector_store_request_for_spend_logs_payload_store_prompts_true(
|
||||
mock_should_store,
|
||||
):
|
||||
|
|
@ -431,27 +530,16 @@ def test_get_vector_store_request_for_spend_logs_payload_store_prompts_true(
|
|||
|
||||
# Sample vector store request metadata
|
||||
vector_store_request = [
|
||||
{
|
||||
"vector_store_search_response": {
|
||||
"data": [
|
||||
{"content": [{"text": "sensitive information", "type": "text"}]}
|
||||
]
|
||||
}
|
||||
}
|
||||
{"vector_store_search_response": {"data": [{"content": [{"text": "sensitive information", "type": "text"}]}]}}
|
||||
]
|
||||
|
||||
# When store_prompts is True, the original data should be returned unchanged
|
||||
result = _get_vector_store_request_for_spend_logs_payload(vector_store_request)
|
||||
assert result == vector_store_request
|
||||
assert (
|
||||
result[0]["vector_store_search_response"]["data"][0]["content"][0]["text"]
|
||||
== "sensitive information"
|
||||
)
|
||||
assert result[0]["vector_store_search_response"]["data"][0]["content"][0]["text"] == "sensitive information"
|
||||
|
||||
|
||||
@patch(
|
||||
"litellm.proxy.spend_tracking.spend_tracking_utils._should_store_prompts_and_responses_in_spend_logs"
|
||||
)
|
||||
@patch("litellm.proxy.spend_tracking.spend_tracking_utils._should_store_prompts_and_responses_in_spend_logs")
|
||||
def test_get_vector_store_request_for_spend_logs_payload_store_prompts_false(
|
||||
mock_should_store,
|
||||
):
|
||||
|
|
@ -460,32 +548,18 @@ def test_get_vector_store_request_for_spend_logs_payload_store_prompts_false(
|
|||
|
||||
# Sample vector store request metadata
|
||||
vector_store_request = [
|
||||
{
|
||||
"vector_store_search_response": {
|
||||
"data": [
|
||||
{"content": [{"text": "sensitive information", "type": "text"}]}
|
||||
]
|
||||
}
|
||||
}
|
||||
{"vector_store_search_response": {"data": [{"content": [{"text": "sensitive information", "type": "text"}]}]}}
|
||||
]
|
||||
|
||||
# When store_prompts is False, text should be redacted
|
||||
result = _get_vector_store_request_for_spend_logs_payload(vector_store_request)
|
||||
assert result is not None
|
||||
assert (
|
||||
result[0]["vector_store_search_response"]["data"][0]["content"][0]["text"]
|
||||
== REDACTED_BY_LITELM_STRING
|
||||
)
|
||||
assert result[0]["vector_store_search_response"]["data"][0]["content"][0]["text"] == REDACTED_BY_LITELM_STRING
|
||||
# Ensure other fields are unchanged
|
||||
assert (
|
||||
result[0]["vector_store_search_response"]["data"][0]["content"][0]["type"]
|
||||
== "text"
|
||||
)
|
||||
assert result[0]["vector_store_search_response"]["data"][0]["content"][0]["type"] == "text"
|
||||
|
||||
|
||||
@patch(
|
||||
"litellm.proxy.spend_tracking.spend_tracking_utils._should_store_prompts_and_responses_in_spend_logs"
|
||||
)
|
||||
@patch("litellm.proxy.spend_tracking.spend_tracking_utils._should_store_prompts_and_responses_in_spend_logs")
|
||||
def test_get_vector_store_request_for_spend_logs_payload_null_input(mock_should_store):
|
||||
# When input is None
|
||||
mock_should_store.return_value = False
|
||||
|
|
@ -493,9 +567,7 @@ def test_get_vector_store_request_for_spend_logs_payload_null_input(mock_should_
|
|||
assert result is None
|
||||
|
||||
|
||||
@patch(
|
||||
"litellm.proxy.spend_tracking.spend_tracking_utils._should_store_prompts_and_responses_in_spend_logs"
|
||||
)
|
||||
@patch("litellm.proxy.spend_tracking.spend_tracking_utils._should_store_prompts_and_responses_in_spend_logs")
|
||||
def test_get_messages_for_spend_logs_realtime_returns_messages(mock_should_store):
|
||||
"""
|
||||
Test that _get_messages_for_spend_logs_payload returns messages
|
||||
|
|
@ -522,9 +594,7 @@ def test_get_messages_for_spend_logs_realtime_returns_messages(mock_should_store
|
|||
assert parsed[1]["content"] == "What is the weather today?"
|
||||
|
||||
|
||||
@patch(
|
||||
"litellm.proxy.spend_tracking.spend_tracking_utils._should_store_prompts_and_responses_in_spend_logs"
|
||||
)
|
||||
@patch("litellm.proxy.spend_tracking.spend_tracking_utils._should_store_prompts_and_responses_in_spend_logs")
|
||||
def test_get_messages_for_spend_logs_strips_null_bytes(mock_should_store):
|
||||
"""Regression for PostgreSQL 22P05: NUL bytes must be stripped from messages."""
|
||||
mock_should_store.return_value = True
|
||||
|
|
@ -541,9 +611,7 @@ def test_get_messages_for_spend_logs_strips_null_bytes(mock_should_store):
|
|||
assert parsed[0]["content"] == "helloworld"
|
||||
|
||||
|
||||
@patch(
|
||||
"litellm.proxy.spend_tracking.spend_tracking_utils._should_store_prompts_and_responses_in_spend_logs"
|
||||
)
|
||||
@patch("litellm.proxy.spend_tracking.spend_tracking_utils._should_store_prompts_and_responses_in_spend_logs")
|
||||
def test_get_messages_for_spend_logs_realtime_empty_when_disabled(mock_should_store):
|
||||
"""
|
||||
Test that _get_messages_for_spend_logs_payload returns '{}' for realtime calls
|
||||
|
|
@ -561,9 +629,7 @@ def test_get_messages_for_spend_logs_realtime_empty_when_disabled(mock_should_st
|
|||
assert result == "{}"
|
||||
|
||||
|
||||
@patch(
|
||||
"litellm.proxy.spend_tracking.spend_tracking_utils._should_store_prompts_and_responses_in_spend_logs"
|
||||
)
|
||||
@patch("litellm.proxy.spend_tracking.spend_tracking_utils._should_store_prompts_and_responses_in_spend_logs")
|
||||
def test_get_messages_for_spend_logs_non_realtime_returns_empty(mock_should_store):
|
||||
"""
|
||||
Test that _get_messages_for_spend_logs_payload returns '{}' for non-realtime
|
||||
|
|
@ -581,9 +647,7 @@ def test_get_messages_for_spend_logs_non_realtime_returns_empty(mock_should_stor
|
|||
assert result == "{}"
|
||||
|
||||
|
||||
@patch(
|
||||
"litellm.proxy.spend_tracking.spend_tracking_utils._should_store_prompts_and_responses_in_spend_logs"
|
||||
)
|
||||
@patch("litellm.proxy.spend_tracking.spend_tracking_utils._should_store_prompts_and_responses_in_spend_logs")
|
||||
def test_get_response_for_spend_logs_payload_truncates_large_base64(mock_should_store):
|
||||
from litellm.constants import MAX_STRING_LENGTH_PROMPT_IN_DB
|
||||
|
||||
|
|
@ -611,9 +675,7 @@ def test_get_response_for_spend_logs_payload_truncates_large_base64(mock_should_
|
|||
assert parsed["data"][0]["other_field"] == "value"
|
||||
|
||||
|
||||
@patch(
|
||||
"litellm.proxy.spend_tracking.spend_tracking_utils._should_store_prompts_and_responses_in_spend_logs"
|
||||
)
|
||||
@patch("litellm.proxy.spend_tracking.spend_tracking_utils._should_store_prompts_and_responses_in_spend_logs")
|
||||
def test_get_response_for_spend_logs_payload_strips_null_bytes(mock_should_store):
|
||||
"""Regression for PostgreSQL 22P05: NUL bytes must be stripped from response."""
|
||||
mock_should_store.return_value = True
|
||||
|
|
@ -626,18 +688,14 @@ def test_get_response_for_spend_logs_payload_strips_null_bytes(mock_should_store
|
|||
assert json.loads(response_json)["content"] == "answerhere"
|
||||
|
||||
|
||||
@patch(
|
||||
"litellm.proxy.spend_tracking.spend_tracking_utils._should_store_prompts_and_responses_in_spend_logs"
|
||||
)
|
||||
@patch("litellm.proxy.spend_tracking.spend_tracking_utils._should_store_prompts_and_responses_in_spend_logs")
|
||||
def test_get_response_for_spend_logs_payload_truncates_large_embedding(
|
||||
mock_should_store,
|
||||
):
|
||||
from litellm.constants import MAX_STRING_LENGTH_PROMPT_IN_DB
|
||||
|
||||
mock_should_store.return_value = True
|
||||
embedding_values = [
|
||||
round(i * 0.0001, 6) for i in range(MAX_STRING_LENGTH_PROMPT_IN_DB + 500)
|
||||
]
|
||||
embedding_values = [round(i * 0.0001, 6) for i in range(MAX_STRING_LENGTH_PROMPT_IN_DB + 500)]
|
||||
large_embedding = json.dumps(embedding_values)
|
||||
payload = cast(
|
||||
StandardLoggingPayload,
|
||||
|
|
@ -685,9 +743,7 @@ def test_truncation_includes_db_safeguard_note():
|
|||
)
|
||||
|
||||
|
||||
@patch(
|
||||
"litellm.proxy.spend_tracking.spend_tracking_utils._should_store_prompts_and_responses_in_spend_logs"
|
||||
)
|
||||
@patch("litellm.proxy.spend_tracking.spend_tracking_utils._should_store_prompts_and_responses_in_spend_logs")
|
||||
def test_response_truncation_logs_info_message(mock_should_store):
|
||||
"""
|
||||
Test that when response is truncated before DB storage, an info log is emitted
|
||||
|
|
@ -702,18 +758,14 @@ def test_response_truncation_logs_info_message(mock_should_store):
|
|||
{"response": {"data": [{"content": large_text}]}},
|
||||
)
|
||||
|
||||
with patch(
|
||||
"litellm.proxy.spend_tracking.spend_tracking_utils.verbose_proxy_logger"
|
||||
) as mock_logger:
|
||||
with patch("litellm.proxy.spend_tracking.spend_tracking_utils.verbose_proxy_logger") as mock_logger:
|
||||
_get_response_for_spend_logs_payload(payload)
|
||||
mock_logger.info.assert_called_once()
|
||||
log_msg = mock_logger.info.call_args[0][0]
|
||||
assert "response was truncated" in log_msg
|
||||
|
||||
|
||||
@patch(
|
||||
"litellm.proxy.spend_tracking.spend_tracking_utils._should_store_prompts_and_responses_in_spend_logs"
|
||||
)
|
||||
@patch("litellm.proxy.spend_tracking.spend_tracking_utils._should_store_prompts_and_responses_in_spend_logs")
|
||||
def test_request_body_truncation_logs_info_message(mock_should_store):
|
||||
"""
|
||||
Test that when request body is truncated before DB storage, an info log is emitted.
|
||||
|
|
@ -722,18 +774,10 @@ def test_request_body_truncation_logs_info_message(mock_should_store):
|
|||
|
||||
mock_should_store.return_value = True
|
||||
large_prompt = "C" * (MAX_STRING_LENGTH_PROMPT_IN_DB + 500)
|
||||
litellm_params = {
|
||||
"proxy_server_request": {
|
||||
"body": {"messages": [{"role": "user", "content": large_prompt}]}
|
||||
}
|
||||
}
|
||||
litellm_params = {"proxy_server_request": {"body": {"messages": [{"role": "user", "content": large_prompt}]}}}
|
||||
|
||||
with patch(
|
||||
"litellm.proxy.spend_tracking.spend_tracking_utils.verbose_proxy_logger"
|
||||
) as mock_logger:
|
||||
_get_proxy_server_request_for_spend_logs_payload(
|
||||
metadata={}, litellm_params=litellm_params, kwargs={}
|
||||
)
|
||||
with patch("litellm.proxy.spend_tracking.spend_tracking_utils.verbose_proxy_logger") as mock_logger:
|
||||
_get_proxy_server_request_for_spend_logs_payload(metadata={}, litellm_params=litellm_params, kwargs={})
|
||||
mock_logger.info.assert_called_once()
|
||||
log_msg = mock_logger.info.call_args[0][0]
|
||||
assert "request body was truncated" in log_msg
|
||||
|
|
@ -870,14 +914,10 @@ def test_get_logging_payload_api_key_preserved_when_standard_logging_payload_is_
|
|||
)
|
||||
|
||||
# The api_key should be hashed (not the raw key)
|
||||
assert (
|
||||
payload["api_key"] != test_api_key
|
||||
), "api_key should be hashed, not the raw key"
|
||||
assert payload["api_key"] != test_api_key, "api_key should be hashed, not the raw key"
|
||||
|
||||
# The api_key should be a valid hash (64 character hex string for SHA256)
|
||||
assert (
|
||||
len(payload["api_key"]) == 64
|
||||
), f"Expected 64 character hash, got {len(payload['api_key'])} characters"
|
||||
assert len(payload["api_key"]) == 64, f"Expected 64 character hash, got {len(payload['api_key'])} characters"
|
||||
|
||||
# Verify other fields are set correctly
|
||||
assert payload["model"] == "openai/gpt-4.1"
|
||||
|
|
@ -1019,9 +1059,7 @@ async def test_api_key_preserved_through_failure_hook_to_database():
|
|||
|
||||
assert payload_api_key is not None, "🚨 CRITICAL: payload['api_key'] is None!"
|
||||
|
||||
assert (
|
||||
payload_api_key == hashed_key
|
||||
), f"🚨 CRITICAL: Expected api_key={hashed_key}, got {payload_api_key}"
|
||||
assert payload_api_key == hashed_key, f"🚨 CRITICAL: Expected api_key={hashed_key}, got {payload_api_key}"
|
||||
|
||||
# Verify token parameter matches
|
||||
assert data["token"] == hashed_key, f"Token parameter should be {hashed_key}"
|
||||
|
|
@ -1066,9 +1104,7 @@ def test_get_logging_payload_includes_agent_id_from_kwargs():
|
|||
end_time=end_time,
|
||||
)
|
||||
|
||||
assert (
|
||||
payload["agent_id"] == test_agent_id
|
||||
), f"Expected agent_id '{test_agent_id}', got '{payload.get('agent_id')}'"
|
||||
assert payload["agent_id"] == test_agent_id, f"Expected agent_id '{test_agent_id}', got '{payload.get('agent_id')}'"
|
||||
|
||||
|
||||
@patch("litellm.proxy.proxy_server.master_key", None)
|
||||
|
|
@ -1093,9 +1129,7 @@ def test_get_logging_payload_includes_overhead_in_spend_logs_metadata():
|
|||
startTime=1234567890.0,
|
||||
endTime=1234567891.0,
|
||||
completionStartTime=None,
|
||||
model_map_information=StandardLoggingModelInformation(
|
||||
model_map_key="gpt-3.5-turbo", model_map_value=None
|
||||
),
|
||||
model_map_information=StandardLoggingModelInformation(model_map_key="gpt-3.5-turbo", model_map_value=None),
|
||||
model="gpt-3.5-turbo",
|
||||
model_id="model-123",
|
||||
model_group="openai",
|
||||
|
|
@ -1173,9 +1207,9 @@ def test_get_logging_payload_includes_overhead_in_spend_logs_metadata():
|
|||
metadata = json.loads(metadata_json)
|
||||
|
||||
# Verify overhead is stored directly in metadata
|
||||
assert (
|
||||
metadata.get("litellm_overhead_time_ms") == test_overhead_ms
|
||||
), f"Expected overhead '{test_overhead_ms}', got '{metadata.get('litellm_overhead_time_ms')}'"
|
||||
assert metadata.get("litellm_overhead_time_ms") == test_overhead_ms, (
|
||||
f"Expected overhead '{test_overhead_ms}', got '{metadata.get('litellm_overhead_time_ms')}'"
|
||||
)
|
||||
|
||||
|
||||
@patch("litellm.proxy.proxy_server.master_key", None)
|
||||
|
|
@ -1228,9 +1262,7 @@ def test_get_logging_payload_handles_missing_overhead_gracefully():
|
|||
startTime=1234567890.0,
|
||||
endTime=1234567891.0,
|
||||
completionStartTime=None,
|
||||
model_map_information=StandardLoggingModelInformation(
|
||||
model_map_key="gpt-3.5-turbo", model_map_value=None
|
||||
),
|
||||
model_map_information=StandardLoggingModelInformation(model_map_key="gpt-3.5-turbo", model_map_value=None),
|
||||
model="gpt-3.5-turbo",
|
||||
model_id="model-123",
|
||||
model_group="openai",
|
||||
|
|
@ -1309,14 +1341,12 @@ def test_get_logging_payload_handles_missing_overhead_gracefully():
|
|||
metadata = json.loads(metadata_json)
|
||||
|
||||
# When overhead is None, litellm_overhead_time_ms should be None or not present
|
||||
assert (
|
||||
metadata.get("litellm_overhead_time_ms") is None
|
||||
), "litellm_overhead_time_ms should be None when overhead is not provided"
|
||||
assert metadata.get("litellm_overhead_time_ms") is None, (
|
||||
"litellm_overhead_time_ms should be None when overhead is not provided"
|
||||
)
|
||||
|
||||
|
||||
@patch(
|
||||
"litellm.proxy.spend_tracking.spend_tracking_utils._should_store_prompts_and_responses_in_spend_logs"
|
||||
)
|
||||
@patch("litellm.proxy.spend_tracking.spend_tracking_utils._should_store_prompts_and_responses_in_spend_logs")
|
||||
def test_spend_logs_redacts_request_and_response_when_turn_off_message_logging_enabled(
|
||||
mock_should_store,
|
||||
):
|
||||
|
|
@ -1347,9 +1377,7 @@ def test_spend_logs_redacts_request_and_response_when_turn_off_message_logging_e
|
|||
)
|
||||
|
||||
parsed_request = json.loads(request_result)
|
||||
assert parsed_request["messages"] == [
|
||||
{"role": "user", "content": "redacted-by-litellm"}
|
||||
]
|
||||
assert parsed_request["messages"] == [{"role": "user", "content": "redacted-by-litellm"}]
|
||||
assert parsed_request["model"] == "gpt-4"
|
||||
|
||||
# Test response redaction - use dict response to verify redaction
|
||||
|
|
@ -1368,9 +1396,7 @@ def test_spend_logs_redacts_request_and_response_when_turn_off_message_logging_e
|
|||
{"response": response_dict},
|
||||
)
|
||||
|
||||
response_result = _get_response_for_spend_logs_payload(
|
||||
payload=payload, kwargs=kwargs
|
||||
)
|
||||
response_result = _get_response_for_spend_logs_payload(payload=payload, kwargs=kwargs)
|
||||
|
||||
# When redaction is enabled and response is a dict (not ModelResponse),
|
||||
# perform_redaction redacts content in-place within the choices structure
|
||||
|
|
@ -1415,30 +1441,22 @@ def test_should_store_prompts_and_responses_in_spend_logs_case_insensitive_strin
|
|||
# When env var is True, should return True
|
||||
mock_get_secret_bool.return_value = True
|
||||
result = _should_store_prompts_and_responses_in_spend_logs()
|
||||
assert (
|
||||
result is True
|
||||
), f"Expected True (from env var) for '{false_value}', got {result}"
|
||||
assert result is True, f"Expected True (from env var) for '{false_value}', got {result}"
|
||||
|
||||
# When env var is False, should return False
|
||||
mock_get_secret_bool.return_value = False
|
||||
result = _should_store_prompts_and_responses_in_spend_logs()
|
||||
assert (
|
||||
result is False
|
||||
), f"Expected False (from env var) for '{false_value}', got {result}"
|
||||
assert result is False, f"Expected False (from env var) for '{false_value}', got {result}"
|
||||
|
||||
# Test when general_settings doesn't have the key at all
|
||||
with patch("litellm.proxy.proxy_server.general_settings", {}):
|
||||
mock_get_secret_bool.return_value = True
|
||||
result = _should_store_prompts_and_responses_in_spend_logs()
|
||||
assert (
|
||||
result is True
|
||||
), "Expected True (from env var) when key missing, got False"
|
||||
assert result is True, "Expected True (from env var) when key missing, got False"
|
||||
|
||||
mock_get_secret_bool.return_value = False
|
||||
result = _should_store_prompts_and_responses_in_spend_logs()
|
||||
assert (
|
||||
result is False
|
||||
), "Expected False (from env var) when key missing, got True"
|
||||
assert result is False, "Expected False (from env var) when key missing, got True"
|
||||
|
||||
|
||||
def test_get_spend_logs_metadata_guardrail_info_fallback_from_metadata():
|
||||
|
|
@ -1831,9 +1849,7 @@ def test_get_logging_payload_includes_retry_info_in_spend_logs_metadata():
|
|||
startTime=1234567890.0,
|
||||
endTime=1234567891.0,
|
||||
completionStartTime=None,
|
||||
model_map_information=StandardLoggingModelInformation(
|
||||
model_map_key="gpt-3.5-turbo", model_map_value=None
|
||||
),
|
||||
model_map_information=StandardLoggingModelInformation(model_map_key="gpt-3.5-turbo", model_map_value=None),
|
||||
model="gpt-3.5-turbo",
|
||||
model_id="model-123",
|
||||
model_group="openai",
|
||||
|
|
@ -1897,12 +1913,10 @@ def test_get_logging_payload_includes_retry_info_in_spend_logs_metadata():
|
|||
|
||||
metadata = json.loads(payload["metadata"])
|
||||
|
||||
assert (
|
||||
metadata.get("attempted_retries") == 2
|
||||
), f"Expected attempted_retries=2, got {metadata.get('attempted_retries')}"
|
||||
assert (
|
||||
metadata.get("max_retries") == 3
|
||||
), f"Expected max_retries=3, got {metadata.get('max_retries')}"
|
||||
assert metadata.get("attempted_retries") == 2, (
|
||||
f"Expected attempted_retries=2, got {metadata.get('attempted_retries')}"
|
||||
)
|
||||
assert metadata.get("max_retries") == 3, f"Expected max_retries=3, got {metadata.get('max_retries')}"
|
||||
|
||||
|
||||
@patch("litellm.proxy.proxy_server.master_key", None)
|
||||
|
|
@ -1930,9 +1944,7 @@ def test_get_logging_payload_handles_missing_retry_info_gracefully():
|
|||
startTime=1234567890.0,
|
||||
endTime=1234567891.0,
|
||||
completionStartTime=None,
|
||||
model_map_information=StandardLoggingModelInformation(
|
||||
model_map_key="gpt-3.5-turbo", model_map_value=None
|
||||
),
|
||||
model_map_information=StandardLoggingModelInformation(model_map_key="gpt-3.5-turbo", model_map_value=None),
|
||||
model="gpt-3.5-turbo",
|
||||
model_id="model-123",
|
||||
model_group="openai",
|
||||
|
|
@ -1996,20 +2008,14 @@ def test_get_logging_payload_handles_missing_retry_info_gracefully():
|
|||
|
||||
metadata = json.loads(payload["metadata"])
|
||||
|
||||
assert (
|
||||
metadata.get("attempted_retries") is None
|
||||
), "attempted_retries should be None when not provided"
|
||||
assert (
|
||||
metadata.get("max_retries") is None
|
||||
), "max_retries should be None when not provided"
|
||||
assert metadata.get("attempted_retries") is None, "attempted_retries should be None when not provided"
|
||||
assert metadata.get("max_retries") is None, "max_retries should be None when not provided"
|
||||
|
||||
|
||||
def test_get_request_duration_ms_normal():
|
||||
"""Test that request duration is correctly computed in milliseconds."""
|
||||
start = datetime.datetime(2025, 1, 1, 0, 0, 0, tzinfo=timezone.utc)
|
||||
end = datetime.datetime(
|
||||
2025, 1, 1, 0, 0, 2, 500000, tzinfo=timezone.utc
|
||||
) # 2.5s later
|
||||
end = datetime.datetime(2025, 1, 1, 0, 0, 2, 500000, tzinfo=timezone.utc) # 2.5s later
|
||||
result = _get_request_duration_ms(start, end)
|
||||
assert result == 2500
|
||||
|
||||
|
|
@ -2039,9 +2045,7 @@ def test_get_logging_payload_includes_request_duration_ms():
|
|||
"litellm_params": {"api_base": "https://api.openai.com"},
|
||||
"standard_logging_object": None,
|
||||
}
|
||||
response_obj = {
|
||||
"usage": {"prompt_tokens": 10, "completion_tokens": 5, "total_tokens": 15}
|
||||
}
|
||||
response_obj = {"usage": {"prompt_tokens": 10, "completion_tokens": 5, "total_tokens": 15}}
|
||||
|
||||
with (
|
||||
patch("litellm.proxy.proxy_server.master_key", None),
|
||||
|
|
@ -2107,16 +2111,12 @@ def test_sanitize_request_body_strips_secret_fields():
|
|||
}
|
||||
sanitized = _sanitize_request_body_for_spend_logs_payload(request_body)
|
||||
|
||||
assert (
|
||||
"secret_fields" not in sanitized
|
||||
), "secret_fields must be stripped from the sanitized request body"
|
||||
assert "secret_fields" not in sanitized, "secret_fields must be stripped from the sanitized request body"
|
||||
assert sanitized["model"] == "gpt-4"
|
||||
assert sanitized["messages"] == [{"role": "user", "content": "hi"}]
|
||||
|
||||
|
||||
@patch(
|
||||
"litellm.proxy.spend_tracking.spend_tracking_utils._should_store_prompts_and_responses_in_spend_logs"
|
||||
)
|
||||
@patch("litellm.proxy.spend_tracking.spend_tracking_utils._should_store_prompts_and_responses_in_spend_logs")
|
||||
def test_proxy_server_request_payload_excludes_secret_fields(mock_should_store):
|
||||
"""
|
||||
End-to-end test: when the proxy_server_request body contains
|
||||
|
|
@ -2140,14 +2140,10 @@ def test_proxy_server_request_payload_excludes_secret_fields(mock_should_store):
|
|||
}
|
||||
}
|
||||
|
||||
result = _get_proxy_server_request_for_spend_logs_payload(
|
||||
metadata={}, litellm_params=litellm_params, kwargs={}
|
||||
)
|
||||
result = _get_proxy_server_request_for_spend_logs_payload(metadata={}, litellm_params=litellm_params, kwargs={})
|
||||
parsed = json.loads(result)
|
||||
|
||||
assert (
|
||||
"secret_fields" not in parsed
|
||||
), "secret_fields must never appear in the spend-log proxy_server_request column"
|
||||
assert "secret_fields" not in parsed, "secret_fields must never appear in the spend-log proxy_server_request column"
|
||||
assert parsed["model"] == "gpt-4"
|
||||
assert parsed["messages"] == [{"role": "user", "content": "hello"}]
|
||||
|
||||
|
|
@ -2176,10 +2172,7 @@ def test_redact_prompt_leaks_strips_input_value_python_repr():
|
|||
|
||||
|
||||
def test_redact_prompt_leaks_strips_input_value_json():
|
||||
error_text = (
|
||||
'{"error":{"message":"validation failed",'
|
||||
'"input":[{"role":"user","content":"top-secret-content"}]}}'
|
||||
)
|
||||
error_text = '{"error":{"message":"validation failed","input":[{"role":"user","content":"top-secret-content"}]}}'
|
||||
redacted = _redact_prompt_leaks_in_error_string(error_text)
|
||||
assert "top-secret-content" not in redacted
|
||||
assert REDACTED_BY_LITELM_STRING in redacted
|
||||
|
|
@ -2203,9 +2196,7 @@ def test_redact_prompt_leaks_empty_string():
|
|||
assert _redact_prompt_leaks_in_error_string("") == ""
|
||||
|
||||
|
||||
@patch(
|
||||
"litellm.proxy.spend_tracking.spend_tracking_utils._should_store_prompts_and_responses_in_spend_logs"
|
||||
)
|
||||
@patch("litellm.proxy.spend_tracking.spend_tracking_utils._should_store_prompts_and_responses_in_spend_logs")
|
||||
def test_sanitize_error_information_redacts_when_not_storing_prompts(
|
||||
mock_should_store,
|
||||
):
|
||||
|
|
@ -2233,9 +2224,7 @@ def test_sanitize_error_information_redacts_when_not_storing_prompts(
|
|||
assert sanitized["llm_provider"] == "openai"
|
||||
|
||||
|
||||
@patch(
|
||||
"litellm.proxy.spend_tracking.spend_tracking_utils._should_store_prompts_and_responses_in_spend_logs"
|
||||
)
|
||||
@patch("litellm.proxy.spend_tracking.spend_tracking_utils._should_store_prompts_and_responses_in_spend_logs")
|
||||
def test_sanitize_error_information_skips_redaction_when_storing_prompts(
|
||||
mock_should_store,
|
||||
):
|
||||
|
|
@ -2246,9 +2235,7 @@ def test_sanitize_error_information_skips_redaction_when_storing_prompts(
|
|||
"error_class": "RateLimitError",
|
||||
"llm_provider": "openai",
|
||||
"traceback": "",
|
||||
"error_message": (
|
||||
'OpenAIException - {"error":{"input":[{"role":"user","content":"kept"}]}}'
|
||||
),
|
||||
"error_message": ('OpenAIException - {"error":{"input":[{"role":"user","content":"kept"}]}}'),
|
||||
}
|
||||
|
||||
sanitized = _sanitize_error_information_for_spend_logs(error_info)
|
||||
|
|
@ -2259,9 +2246,7 @@ def test_sanitize_error_information_skips_redaction_when_storing_prompts(
|
|||
assert REDACTED_BY_LITELM_STRING not in sanitized["error_message"]
|
||||
|
||||
|
||||
@patch(
|
||||
"litellm.proxy.spend_tracking.spend_tracking_utils._should_store_prompts_and_responses_in_spend_logs"
|
||||
)
|
||||
@patch("litellm.proxy.spend_tracking.spend_tracking_utils._should_store_prompts_and_responses_in_spend_logs")
|
||||
def test_sanitize_error_information_caps_size_regardless_of_prompt_flag(
|
||||
mock_should_store,
|
||||
):
|
||||
|
|
@ -2292,9 +2277,7 @@ def test_sanitize_error_information_none_passthrough():
|
|||
assert _sanitize_error_information_for_spend_logs(None) is None
|
||||
|
||||
|
||||
@patch(
|
||||
"litellm.proxy.spend_tracking.spend_tracking_utils._should_store_prompts_and_responses_in_spend_logs"
|
||||
)
|
||||
@patch("litellm.proxy.spend_tracking.spend_tracking_utils._should_store_prompts_and_responses_in_spend_logs")
|
||||
def test_sanitize_error_information_reproduces_lit_2992(mock_should_store):
|
||||
# Mirrors the reproduced row body from LIT-2992 — a RateLimitError whose
|
||||
# message embeds 178 pydantic validation errors, each carrying a full
|
||||
|
|
@ -2335,10 +2318,7 @@ def test_redact_prompt_leaks_handles_nested_multimodal_content():
|
|||
# Multi-modal payload: 'content' is itself a list. The depth-1 regex
|
||||
# would stop at the inner '['; the parser-based scanner must walk
|
||||
# through balanced nested brackets.
|
||||
error_text = (
|
||||
'{"error":{"messages":[{"role":"user",'
|
||||
'"content":[{"type":"text","text":"top-secret-multimodal"}]}]}}'
|
||||
)
|
||||
error_text = '{"error":{"messages":[{"role":"user","content":[{"type":"text","text":"top-secret-multimodal"}]}]}}'
|
||||
redacted = _redact_prompt_leaks_in_error_string(error_text)
|
||||
assert "top-secret-multimodal" not in redacted
|
||||
assert REDACTED_BY_LITELM_STRING in redacted
|
||||
|
|
@ -2347,9 +2327,7 @@ def test_redact_prompt_leaks_handles_nested_multimodal_content():
|
|||
def test_redact_prompt_leaks_handles_bracket_in_prompt_text():
|
||||
# Prompt text contains a literal '[' — the depth-1 regex would close
|
||||
# the outer ']' prematurely. The parser must respect string quoting.
|
||||
error_text = (
|
||||
'{"error":{"input":[{"role":"user","content":"secret[123 still secret"}]}}'
|
||||
)
|
||||
error_text = '{"error":{"input":[{"role":"user","content":"secret[123 still secret"}]}}'
|
||||
redacted = _redact_prompt_leaks_in_error_string(error_text)
|
||||
assert "secret[123" not in redacted
|
||||
assert "still secret" not in redacted
|
||||
|
|
@ -2368,8 +2346,7 @@ def test_redact_prompt_leaks_handles_escaped_quote_in_prompt_text():
|
|||
def test_redact_prompt_leaks_handles_nested_input_python_repr():
|
||||
# Python dict-repr with nested list inside 'input' — single quotes.
|
||||
error_text = (
|
||||
"validation error: {'input': [{'role': 'user', "
|
||||
"'content': [{'type': 'text', 'text': 'leaked-nested-text'}]}]}"
|
||||
"validation error: {'input': [{'role': 'user', 'content': [{'type': 'text', 'text': 'leaked-nested-text'}]}]}"
|
||||
)
|
||||
redacted = _redact_prompt_leaks_in_error_string(error_text)
|
||||
assert "leaked-nested-text" not in redacted
|
||||
|
|
@ -2385,9 +2362,7 @@ def test_redact_prompt_leaks_handles_unterminated_value():
|
|||
assert REDACTED_BY_LITELM_STRING in redacted
|
||||
|
||||
|
||||
@patch(
|
||||
"litellm.proxy.spend_tracking.spend_tracking_utils._should_store_prompts_and_responses_in_spend_logs"
|
||||
)
|
||||
@patch("litellm.proxy.spend_tracking.spend_tracking_utils._should_store_prompts_and_responses_in_spend_logs")
|
||||
def test_sanitize_error_information_redacts_traceback_when_not_storing_prompts(
|
||||
mock_should_store,
|
||||
):
|
||||
|
|
@ -2419,9 +2394,7 @@ def test_sanitize_error_information_redacts_traceback_when_not_storing_prompts(
|
|||
assert "ValueError: invalid request" in sanitized["traceback"]
|
||||
|
||||
|
||||
@patch(
|
||||
"litellm.proxy.spend_tracking.spend_tracking_utils._should_store_prompts_and_responses_in_spend_logs"
|
||||
)
|
||||
@patch("litellm.proxy.spend_tracking.spend_tracking_utils._should_store_prompts_and_responses_in_spend_logs")
|
||||
def test_sanitize_error_information_skips_traceback_redaction_when_storing_prompts(
|
||||
mock_should_store,
|
||||
):
|
||||
|
|
@ -2431,9 +2404,7 @@ def test_sanitize_error_information_skips_traceback_redaction_when_storing_promp
|
|||
"error_code": "500",
|
||||
"error_class": "ValueError",
|
||||
"llm_provider": "",
|
||||
"traceback": (
|
||||
'raise ValueError({"input":[{"role":"user","content":"tb-kept"}]})'
|
||||
),
|
||||
"traceback": ('raise ValueError({"input":[{"role":"user","content":"tb-kept"}]})'),
|
||||
"error_message": "invalid request",
|
||||
}
|
||||
|
||||
|
|
@ -2448,20 +2419,14 @@ def test_redact_prompt_leaks_strips_prompt_key_completions_payload():
|
|||
# /v1/completions echoes the user input under the top-level 'prompt' key
|
||||
# rather than 'messages'. Without 'prompt' coverage the body would survive
|
||||
# the redactor when store_prompts_in_spend_logs is False.
|
||||
error_text = (
|
||||
'{"error":{"message":"validation failed",'
|
||||
'"prompt":"super-secret-completion-text"}}'
|
||||
)
|
||||
error_text = '{"error":{"message":"validation failed","prompt":"super-secret-completion-text"}}'
|
||||
redacted = _redact_prompt_leaks_in_error_string(error_text)
|
||||
assert "super-secret-completion-text" not in redacted
|
||||
assert REDACTED_BY_LITELM_STRING in redacted
|
||||
|
||||
|
||||
def test_redact_prompt_leaks_strips_prompt_key_python_repr():
|
||||
error_text = (
|
||||
"{'model': 'gpt-3.5-turbo-instruct', "
|
||||
"'prompt': 'leaked-completion-prompt-body'}"
|
||||
)
|
||||
error_text = "{'model': 'gpt-3.5-turbo-instruct', 'prompt': 'leaked-completion-prompt-body'}"
|
||||
redacted = _redact_prompt_leaks_in_error_string(error_text)
|
||||
assert "leaked-completion-prompt-body" not in redacted
|
||||
assert REDACTED_BY_LITELM_STRING in redacted
|
||||
|
|
@ -2495,11 +2460,7 @@ def test_redact_prompt_leaks_strips_pydantic_input_value_list():
|
|||
|
||||
|
||||
def test_redact_prompt_leaks_strips_pydantic_input_value_dict():
|
||||
error_text = (
|
||||
"[type=dict_type, "
|
||||
"input_value={'role': 'user', 'content': 'leaked-dict-content'}, "
|
||||
"input_type=dict]"
|
||||
)
|
||||
error_text = "[type=dict_type, input_value={'role': 'user', 'content': 'leaked-dict-content'}, input_type=dict]"
|
||||
redacted = _redact_prompt_leaks_in_error_string(error_text)
|
||||
assert "leaked-dict-content" not in redacted
|
||||
assert REDACTED_BY_LITELM_STRING in redacted
|
||||
|
|
@ -2541,9 +2502,7 @@ def test_redact_prompt_leaks_combined_quoted_key_and_pydantic_assignment():
|
|||
assert redacted.count(REDACTED_BY_LITELM_STRING) >= 2
|
||||
|
||||
|
||||
@patch(
|
||||
"litellm.proxy.spend_tracking.spend_tracking_utils._should_store_prompts_and_responses_in_spend_logs"
|
||||
)
|
||||
@patch("litellm.proxy.spend_tracking.spend_tracking_utils._should_store_prompts_and_responses_in_spend_logs")
|
||||
def test_sanitize_error_information_redacts_pydantic_assignment_form(
|
||||
mock_should_store,
|
||||
):
|
||||
|
|
@ -2741,9 +2700,7 @@ def test_get_spend_logs_metadata_non_sk_raw_key_hashed():
|
|||
|
||||
def test_get_spend_logs_metadata_already_hashed_unchanged_with_provenance():
|
||||
already_hashed = hash_token("sk-some-key")
|
||||
meta = _get_spend_logs_metadata(
|
||||
{"user_api_key": already_hashed, "user_api_key_hash": already_hashed}
|
||||
)
|
||||
meta = _get_spend_logs_metadata({"user_api_key": already_hashed, "user_api_key_hash": already_hashed})
|
||||
assert meta["user_api_key"] == already_hashed
|
||||
assert hash_token(already_hashed) != meta["user_api_key"] # no double-hash
|
||||
|
||||
|
|
@ -2758,9 +2715,7 @@ def test_get_spend_logs_metadata_already_hashed_no_provenance_is_rehashed():
|
|||
def test_get_spend_logs_metadata_provenance_bypass_requires_hash_match():
|
||||
already_hashed = hash_token("sk-some-key")
|
||||
different_hash = hash_token("sk-other-key")
|
||||
meta = _get_spend_logs_metadata(
|
||||
{"user_api_key": already_hashed, "user_api_key_hash": different_hash}
|
||||
)
|
||||
meta = _get_spend_logs_metadata({"user_api_key": already_hashed, "user_api_key_hash": different_hash})
|
||||
assert meta["user_api_key"] == hash_token(already_hashed)
|
||||
|
||||
|
||||
|
|
@ -2797,16 +2752,12 @@ def test_get_logging_payload_uses_recovered_combined_usage_on_failure():
|
|||
"model": "anthropic/claude-haiku-4-5",
|
||||
"call_type": "acompletion",
|
||||
"litellm_params": {"metadata": {"user_api_key": "sk-test"}},
|
||||
"combined_usage_object": Usage(
|
||||
prompt_tokens=30, completion_tokens=1, total_tokens=31
|
||||
),
|
||||
"combined_usage_object": Usage(prompt_tokens=30, completion_tokens=1, total_tokens=31),
|
||||
}
|
||||
response_obj = Exception("MidStreamFallbackError: read timeout")
|
||||
now = datetime.datetime.now(timezone.utc)
|
||||
|
||||
payload = get_logging_payload(
|
||||
kwargs=kwargs, response_obj=response_obj, start_time=now, end_time=now
|
||||
)
|
||||
payload = get_logging_payload(kwargs=kwargs, response_obj=response_obj, start_time=now, end_time=now)
|
||||
|
||||
assert payload["prompt_tokens"] == 30
|
||||
assert payload["completion_tokens"] == 1
|
||||
|
|
@ -2825,9 +2776,7 @@ def test_get_logging_payload_failure_without_recovered_usage_is_zero():
|
|||
response_obj = Exception("BadRequestError")
|
||||
now = datetime.datetime.now(timezone.utc)
|
||||
|
||||
payload = get_logging_payload(
|
||||
kwargs=kwargs, response_obj=response_obj, start_time=now, end_time=now
|
||||
)
|
||||
payload = get_logging_payload(kwargs=kwargs, response_obj=response_obj, start_time=now, end_time=now)
|
||||
|
||||
assert payload["total_tokens"] == 0
|
||||
|
||||
|
|
@ -2853,9 +2802,7 @@ def test_get_logging_payload_sets_litellm_call_id_for_correlation():
|
|||
}
|
||||
now = datetime.datetime.now(timezone.utc)
|
||||
|
||||
payload = get_logging_payload(
|
||||
kwargs=kwargs, response_obj=response_obj, start_time=now, end_time=now
|
||||
)
|
||||
payload = get_logging_payload(kwargs=kwargs, response_obj=response_obj, start_time=now, end_time=now)
|
||||
metadata = json.loads(payload["metadata"])
|
||||
|
||||
assert payload["request_id"] == provider_response_id
|
||||
|
|
@ -2882,9 +2829,7 @@ def test_get_logging_payload_litellm_call_id_falls_back_to_litellm_params():
|
|||
}
|
||||
now = datetime.datetime.now(timezone.utc)
|
||||
|
||||
payload = get_logging_payload(
|
||||
kwargs=kwargs, response_obj=response_obj, start_time=now, end_time=now
|
||||
)
|
||||
payload = get_logging_payload(kwargs=kwargs, response_obj=response_obj, start_time=now, end_time=now)
|
||||
|
||||
assert json.loads(payload["metadata"])["litellm_call_id"] == trace_call_id
|
||||
|
||||
|
|
@ -2901,14 +2846,10 @@ def test_get_logging_payload_litellm_call_id_when_response_has_no_id():
|
|||
"litellm_call_id": trace_call_id,
|
||||
"litellm_params": {"metadata": {"user_api_key": "sk-test"}},
|
||||
}
|
||||
response_obj = {
|
||||
"usage": {"prompt_tokens": 1, "completion_tokens": 1, "total_tokens": 2}
|
||||
}
|
||||
response_obj = {"usage": {"prompt_tokens": 1, "completion_tokens": 1, "total_tokens": 2}}
|
||||
now = datetime.datetime.now(timezone.utc)
|
||||
|
||||
payload = get_logging_payload(
|
||||
kwargs=kwargs, response_obj=response_obj, start_time=now, end_time=now
|
||||
)
|
||||
payload = get_logging_payload(kwargs=kwargs, response_obj=response_obj, start_time=now, end_time=now)
|
||||
|
||||
assert json.loads(payload["metadata"])["litellm_call_id"] == trace_call_id
|
||||
assert payload["request_id"] == trace_call_id
|
||||
|
|
@ -2932,9 +2873,7 @@ def test_get_logging_payload_cache_hit_keeps_raw_litellm_call_id():
|
|||
}
|
||||
now = datetime.datetime.now(timezone.utc)
|
||||
|
||||
payload = get_logging_payload(
|
||||
kwargs=kwargs, response_obj=response_obj, start_time=now, end_time=now
|
||||
)
|
||||
payload = get_logging_payload(kwargs=kwargs, response_obj=response_obj, start_time=now, end_time=now)
|
||||
|
||||
assert json.loads(payload["metadata"])["litellm_call_id"] == trace_call_id
|
||||
assert "_cache_hit" in payload["request_id"]
|
||||
|
|
@ -3074,9 +3013,7 @@ def test_get_logging_payload_hashes_bearer_prefixed_api_key():
|
|||
assert not payload["api_key"].startswith("Bearer"), (
|
||||
f"api_key column contains plaintext Bearer key: {payload['api_key']}"
|
||||
)
|
||||
assert not payload["api_key"].startswith("sk-"), (
|
||||
f"api_key column contains unhashed key: {payload['api_key']}"
|
||||
)
|
||||
assert not payload["api_key"].startswith("sk-"), f"api_key column contains unhashed key: {payload['api_key']}"
|
||||
|
||||
metadata_dict = json.loads(payload["metadata"])
|
||||
assert not metadata_dict["user_api_key"].startswith("Bearer"), (
|
||||
|
|
@ -3747,9 +3684,7 @@ def test_get_logging_payload_includes_fallback_info_in_spend_logs_metadata():
|
|||
startTime=1234567890.0,
|
||||
endTime=1234567891.0,
|
||||
completionStartTime=None,
|
||||
model_map_information=StandardLoggingModelInformation(
|
||||
model_map_key="gpt-3.5-turbo", model_map_value=None
|
||||
),
|
||||
model_map_information=StandardLoggingModelInformation(model_map_key="gpt-3.5-turbo", model_map_value=None),
|
||||
model="gpt-3.5-turbo",
|
||||
model_id="model-123",
|
||||
model_group="openai",
|
||||
|
|
@ -3813,12 +3748,12 @@ def test_get_logging_payload_includes_fallback_info_in_spend_logs_metadata():
|
|||
|
||||
metadata = json.loads(payload["metadata"])
|
||||
|
||||
assert (
|
||||
metadata.get("attempted_fallbacks") == 2
|
||||
), f"Expected attempted_fallbacks=2, got {metadata.get('attempted_fallbacks')}"
|
||||
assert (
|
||||
metadata.get("original_model_group") == "azure-gpt-fallback"
|
||||
), f"Expected original_model_group=azure-gpt-fallback, got {metadata.get('original_model_group')}"
|
||||
assert metadata.get("attempted_fallbacks") == 2, (
|
||||
f"Expected attempted_fallbacks=2, got {metadata.get('attempted_fallbacks')}"
|
||||
)
|
||||
assert metadata.get("original_model_group") == "azure-gpt-fallback", (
|
||||
f"Expected original_model_group=azure-gpt-fallback, got {metadata.get('original_model_group')}"
|
||||
)
|
||||
|
||||
|
||||
def test_get_logging_payload_handles_missing_fallback_info_gracefully():
|
||||
|
|
@ -3844,9 +3779,7 @@ def test_get_logging_payload_handles_missing_fallback_info_gracefully():
|
|||
startTime=1234567890.0,
|
||||
endTime=1234567891.0,
|
||||
completionStartTime=None,
|
||||
model_map_information=StandardLoggingModelInformation(
|
||||
model_map_key="gpt-3.5-turbo", model_map_value=None
|
||||
),
|
||||
model_map_information=StandardLoggingModelInformation(model_map_key="gpt-3.5-turbo", model_map_value=None),
|
||||
model="gpt-3.5-turbo",
|
||||
model_id="model-123",
|
||||
model_group="openai",
|
||||
|
|
@ -3910,12 +3843,10 @@ def test_get_logging_payload_handles_missing_fallback_info_gracefully():
|
|||
|
||||
metadata = json.loads(payload["metadata"])
|
||||
|
||||
assert (
|
||||
metadata.get("attempted_fallbacks") is None
|
||||
), "attempted_fallbacks should be None when not provided"
|
||||
assert (
|
||||
metadata.get("original_model_group") is None
|
||||
), "original_model_group should be None when not provided"
|
||||
assert metadata.get("attempted_fallbacks") is None, "attempted_fallbacks should be None when not provided"
|
||||
assert metadata.get("original_model_group") is None, "original_model_group should be None when not provided"
|
||||
|
||||
|
||||
@pytest.mark.parametrize("bucket", ["metadata", "litellm_metadata"])
|
||||
def test_injected_cache_breakpoints_survive_into_spend_log_metadata(bucket):
|
||||
"""The injection marker only gates savings if it reaches the spend-log row.
|
||||
|
|
|
|||
File diff suppressed because it is too large
Load diff
4
ui/litellm-dashboard/src/lib/http/schema.d.ts
generated
vendored
4
ui/litellm-dashboard/src/lib/http/schema.d.ts
generated
vendored
|
|
@ -25779,9 +25779,9 @@ export interface components {
|
|||
mcp_xff_num_trusted_hops?: number | null;
|
||||
/**
|
||||
* Missing Session Id
|
||||
* @description What to do with LLM API requests that carry no session id (x-litellm-session-id header, metadata.session_id, etc.). 'generate' stamps one id into litellm_session_id, litellm_trace_id and metadata.session_id so SpendLogs and logging callbacks agree; 'reject' returns 400. Unset keeps the legacy behavior where SpendLogs falls back to the trace id while callbacks get no session id.
|
||||
* @description What to do with LLM API requests that carry no session id (x-litellm-session-id header, metadata.session_id, etc.). 'generate' stamps one id into litellm_session_id, litellm_trace_id and metadata.session_id so SpendLogs and logging callbacks agree; 'reject' returns 400; 'omit' leaves SpendLogs.session_id null, matching callbacks such as Langfuse that only record a client-established metadata.session_id. Unset keeps the legacy behavior where SpendLogs falls back to the trace id while callbacks get no session id.
|
||||
*/
|
||||
missing_session_id?: ("generate" | "reject") | null;
|
||||
missing_session_id?: ("generate" | "reject" | "omit") | null;
|
||||
/**
|
||||
* Model List Healthy Only
|
||||
* @description When true, `/models`, `/v1/models/{id}` and `/model/info` hide models whose backing deployments are all unhealthy, for every caller, without needing `healthy_only=true` per request. Requires `background_health_checks: true`, and keeps deployment health state cached without turning on `enable_health_check_routing`, so routing is unaffected. With no health state nothing is hidden. Hiding is presentation-only, a hidden model can still be called.
|
||||
|
|
|
|||
Loading…
Add table
Reference in a new issue