fix(decisions): bill a deployment's query price and accept Parallel bases that end in a route

A proxy request logs through an object built before the router picks the deployment, so a query price on litellm_params never reached Decisions billing. Parallel Search accepts a base ending in /v1/search and shares PARALLEL_AI_API_BASE with Decisions, so Decisions now accepts that base and one ending in /v1/decisions.
This commit is contained in:
mateo-berri 2026-10-31 14:49:51 -07:00
parent 56039b9207
commit 6f332b8c80
3 changed files with 60 additions and 3 deletions

View file

@ -557,6 +557,7 @@ def prepare_call(
litellm_params={
"litellm_call_id": kwargs.get("litellm_call_id"),
"api_base": provider_config.get_complete_url(resolved_api_base, canonical_model),
"input_cost_per_query": kwargs.get("input_cost_per_query"),
},
custom_llm_provider=provider,
)

View file

@ -41,6 +41,9 @@ class ParallelAIDecisionsConfig(BaseDecisionsConfig):
)
return resolved_base, self.resolve_api_key(None)
def get_complete_url(self, api_base: str, model: str) -> str:
return super().get_complete_url(api_base.rstrip("/").removesuffix("/v1/search").removesuffix(self.path), model)
def _decisions_url(self, api_base: str | None) -> httpx.URL:
return httpx.URL(self.get_complete_url(api_base or self.get_default_api_base(), "decision"))

View file

@ -1,6 +1,7 @@
import json
from collections.abc import Mapping
from dataclasses import replace
from datetime import datetime
from typing import Final
import httpx
@ -23,7 +24,7 @@ from litellm.types.decisions import (
UnsupportedDecisionsRequest,
)
from litellm.types.utils import LlmProviders
from litellm.utils import ProviderConfigManager
from litellm.utils import ProviderConfigManager, Rules, function_setup
QUESTIONS: Final = {
"available": {"type": "noul", "instructions": "Is this product available?"},
@ -98,6 +99,12 @@ def test_provider_routes_systemone_questions_to_parallel_without_a_model_field()
assert config.get_complete_url("https://gateway.example/v1/", "decision") == "https://gateway.example/v1/decisions"
@pytest.mark.parametrize("suffix", ["", "/", "/v1", "/v1/", "/v1/decisions", "/v1/decisions/", "/v1/search", "/v1/search/"])
def test_base_carrying_a_parallel_route_reaches_the_decisions_endpoint(suffix: str) -> None:
complete_url: Final = ParallelAIDecisionsConfig().get_complete_url(f"https://api.parallel.ai{suffix}", "decision")
assert complete_url == "https://api.parallel.ai/v1/decisions"
def test_auth_replaces_conflicting_headers_without_forwarding_bearer_credentials() -> None:
headers: Final = ParallelAIDecisionsConfig().validate_environment(
{"Authorization": "Bearer old", "X-API-Key": "old", "CONTENT-TYPE": "text/plain", "X-Trace": "test"},
@ -226,6 +233,38 @@ def test_request_pricing_is_independent_of_token_count(tokens: int, call_type: s
assert cost == (rate, 0.0)
@pytest.mark.asyncio
async def test_deployment_query_price_replaces_the_catalog_price(
respx_mock: respx.MockRouter, monkeypatch: pytest.MonkeyPatch
) -> None:
monkeypatch.setattr(litellm, "disable_aiohttp_transport", True)
litellm.in_memory_llm_clients_cache.flush_cache()
respx_mock.post("https://api.parallel.ai/v1/decisions").respond(json=PAYLOAD)
price: Final = 25 * litellm.model_cost["parallel_ai/decision"]["input_cost_per_query"]
router: Final = litellm.Router(
model_list=[
{
"model_name": "webjev",
"litellm_params": {
"model": "parallel_ai/decision",
"api_key": "test-key",
"input_cost_per_query": price,
},
}
]
)
fields: Final = {
"model": "webjev",
"input": "Example",
"questions": [{"type": "predicate", "name": "available", "instructions": "Available?"}],
"litellm_call_id": "query-price-call",
}
logging_obj, data = function_setup("adecisions", Rules(), datetime.now(), **fields)
response: Final = await router.adecisions(**data, litellm_logging_obj=logging_obj)
assert response._hidden_params["response_cost"] == pytest.approx(price)
await GLOBAL_LOGGING_WORKER.clear_queue()
@pytest.mark.parametrize(
"payload", [{"answers": {}}, {"mode": "fast", "answers": {"available": {"type": "noul", "noul": "invalid"}}}]
)
@ -279,6 +318,9 @@ async def test_untrusted_base_cannot_receive_environment_key(
"https://api.parallel.ai:8443",
"https://attacker.example/api.parallel.ai",
"https://api.parallel.ai/other-tenant",
"https://api.parallel.ai/other-tenant/v1/search",
"https://attacker.example/v1/search",
"https://attacker.example/v1/decisions",
],
)
def test_environment_key_rejects_untrusted_destination_variants(api_base: str, monkeypatch: pytest.MonkeyPatch) -> None:
@ -288,7 +330,16 @@ def test_environment_key_rejects_untrusted_destination_variants(api_base: str, m
ParallelAIDecisionsConfig().resolve_connection(api_base, None)
@pytest.mark.parametrize("api_base", [None, "https://api.parallel.ai", "https://API.PARALLEL.AI:443/v1/"])
@pytest.mark.parametrize(
"api_base",
[
None,
"https://api.parallel.ai",
"https://API.PARALLEL.AI:443/v1/",
"https://api.parallel.ai/v1/decisions",
"https://api.parallel.ai/v1/search",
],
)
def test_default_destination_retains_environment_key(api_base: str | None, monkeypatch: pytest.MonkeyPatch) -> None:
monkeypatch.delenv("PARALLEL_AI_API_BASE", raising=False)
monkeypatch.setenv("PARALLEL_AI_API_KEY", "server-secret")
@ -297,7 +348,9 @@ def test_default_destination_retains_environment_key(api_base: str | None, monke
assert key == "server-secret"
@pytest.mark.parametrize("api_base", [None, "https://gateway.example/parallel/v1/"])
@pytest.mark.parametrize(
"api_base", [None, "https://gateway.example/parallel/v1/", "https://gateway.example/parallel/v1/search"]
)
def test_operator_base_retains_environment_key(api_base: str | None, monkeypatch: pytest.MonkeyPatch) -> None:
monkeypatch.setenv("PARALLEL_AI_API_BASE", "https://gateway.example/parallel")
monkeypatch.setenv("PARALLEL_AI_API_KEY", "server-secret")