RAG-POC/tests/unit/test_llm.py

67 lines
1.8 KiB
Python

from __future__ import annotations
import httpx
from graphrag_platform.config import Settings
from graphrag_platform.llm import OpenAICompatibleClient
def test_llm_retries_rate_limit_and_honors_retry_after() -> None:
requests = 0
delays: list[float] = []
def handler(request: httpx.Request) -> httpx.Response:
nonlocal requests
requests += 1
if requests == 1:
return httpx.Response(
429,
headers={"Retry-After": "0.25"},
json={"error": "rate limited"},
request=request,
)
return httpx.Response(
200,
json={"choices": [{"message": {"content": "answer"}}]},
request=request,
)
settings = Settings(
_env_file=None,
llm_api_key="test-key",
llm_max_retries=2,
)
client = OpenAICompatibleClient(
settings,
transport=httpx.MockTransport(handler),
sleeper=delays.append,
jitter=lambda: 0,
)
assert client.complete_text("system", "user") == "answer"
assert requests == 2
assert delays == [0.25]
def test_llm_does_not_retry_authentication_failure() -> None:
requests = 0
def handler(request: httpx.Request) -> httpx.Response:
nonlocal requests
requests += 1
return httpx.Response(401, json={"error": "unauthorized"}, request=request)
client = OpenAICompatibleClient(
Settings(_env_file=None, llm_api_key="bad-key", llm_max_retries=3),
transport=httpx.MockTransport(handler),
sleeper=lambda _: None,
)
try:
client.complete_text("system", "user")
except httpx.HTTPStatusError:
pass
else:
raise AssertionError("Expected an HTTPStatusError")
assert requests == 1