mirror of
https://github.com/bytedance/deer-flow.git
synced 2026-10-26 21:26:39 +00:00
1278c5192b
* fix(models): honor HTTP-date Retry-After hints in Claude retries * refactor(models): share bounded HTTP-date retry parsing * test(agents): move retry date regression to middleware suite
69 lines
3.0 KiB
Python
69 lines
3.0 KiB
Python
"""HTTP-date retry hints through the public Claude model and real SDK."""
|
|
|
|
import asyncio
|
|
from datetime import UTC, datetime, timedelta
|
|
from email.utils import format_datetime
|
|
from types import SimpleNamespace
|
|
|
|
import anthropic
|
|
import httpx
|
|
import pytest
|
|
|
|
from deerflow.models import claude_provider as provider
|
|
|
|
|
|
@pytest.fixture
|
|
def fixed_clock(monkeypatch):
|
|
now = datetime(2026, 1, 1, tzinfo=UTC)
|
|
monkeypatch.setattr(provider, "time", SimpleNamespace(time=lambda: now.timestamp()))
|
|
return now
|
|
|
|
|
|
@pytest.mark.parametrize("seconds, expected", [(-60, 0), (0, 0), (60, 60000), (86400, 86400000), (86401, 2400)])
|
|
def test_date_delay_bounds(fixed_clock, seconds, expected):
|
|
raw = format_datetime(fixed_clock + timedelta(seconds=seconds), usegmt=True)
|
|
error = SimpleNamespace(response=httpx.Response(429, headers={"Retry-After": raw}))
|
|
assert provider.ClaudeChatModel._calc_backoff_ms(1, error) == expected
|
|
|
|
|
|
@pytest.mark.parametrize("raw", ["Thu, 99 Jan 2026 00:10:00 GMT", "Fri, 31 Dec 9999 23:59:59 GMT", "not a date"])
|
|
def test_unusable_dates_keep_local_backoff(fixed_clock, raw):
|
|
error = SimpleNamespace(response=httpx.Response(503, headers={"Retry-After": raw}))
|
|
assert provider.ClaudeChatModel._calc_backoff_ms(2, error) == 4800
|
|
|
|
|
|
@pytest.mark.asyncio
|
|
@pytest.mark.parametrize("mode", ["sync", "async"])
|
|
@pytest.mark.parametrize("status", [429, 503])
|
|
@pytest.mark.parametrize("raw", ["Thu, 01 Jan 2026 00:01:00 GMT", "Thursday, 01-Jan-26 00:01:00 GMT", "Thu Jan 1 00:01:00 2026"])
|
|
async def test_public_call_waits_for_http_date(monkeypatch, fixed_clock, mode, status, raw):
|
|
requests = []
|
|
waits = []
|
|
|
|
def respond(request):
|
|
requests.append(request)
|
|
if len(requests) == 1:
|
|
return httpx.Response(status, headers={"Retry-After": raw}, json={"type": "error", "error": {"type": "rate_limit_error" if status == 429 else "api_error", "message": "Try later"}})
|
|
return httpx.Response(
|
|
200, json={"id": "msg-test", "type": "message", "role": "assistant", "model": "claude-sonnet-4-6", "content": [{"type": "text", "text": "recovered"}], "stop_reason": "end_turn", "usage": {"input_tokens": 1, "output_tokens": 1}}
|
|
)
|
|
|
|
async def sleep(delay):
|
|
waits.append(delay)
|
|
|
|
provider.time.sleep = waits.append
|
|
monkeypatch.setattr(asyncio, "sleep", sleep)
|
|
model = provider.ClaudeChatModel(model="claude-sonnet-4-6", anthropic_api_key="sk-ant-fake", max_tokens=1024, max_retries=0, retry_max_attempts=2)
|
|
transport = httpx.MockTransport(respond)
|
|
if mode == "sync":
|
|
with anthropic.Anthropic(api_key="sk-ant-fake", max_retries=0, http_client=httpx.Client(transport=transport)) as client:
|
|
model._client = client
|
|
result = model.invoke("hi")
|
|
else:
|
|
async with anthropic.AsyncAnthropic(api_key="sk-ant-fake", max_retries=0, http_client=httpx.AsyncClient(transport=transport)) as client:
|
|
model._async_client = client
|
|
result = await model.ainvoke("hi")
|
|
assert result.content == "recovered"
|
|
assert len(requests) == 2
|
|
assert waits == [60.0]
|