- _retry.py: replace `assert last_error is not None` with explicit `if last_error is None: raise RuntimeError(...)` so the original rate-limit error is preserved under `python -O`. - openai.py: drop the `_retry_factory` alias chain; rename the bound decorator to `_retry_429` to match the pattern in mistral.py. - mistral.py: comment the imports so future reviewers understand why `from mistralai.client import …` is the canonical path on 2.x (no top-level `__init__.py`; no `mistralai.models` subpackage either). - docs/configuration.md: add `OPENAI_GENERATION_MODEL` and `OLLAMA_GENERATION_MODEL` rows to the env-var reference table. - test_mistral.py: add direct unit test for the `_is_rate_limit` predicate (429 → True, 500 → False, missing-attr → False). - test_registry.py: stub `mistralai.client.Mistral` in the registry picker test, mirroring the Ollama sibling, so the test doesn't depend on the SDK accepting arbitrary keys. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
79 lines
2.8 KiB
Python
79 lines
2.8 KiB
Python
"""Shared rate-limit retry helper for provider modules.
|
|
|
|
OpenAI and Mistral both retry on 429 with the same exponential-backoff curve;
|
|
extracting the loop here keeps the two provider modules thin and lets future
|
|
providers (Bedrock throttling, etc.) reuse the same primitive.
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
import logging
|
|
from collections.abc import Awaitable, Callable
|
|
from functools import wraps
|
|
from typing import Any, TypeVar
|
|
|
|
import anyio
|
|
|
|
logger = logging.getLogger(__name__)
|
|
|
|
MAX_RETRIES = 5
|
|
INITIAL_RETRY_DELAY = 2.0
|
|
MAX_RETRY_DELAY = 60.0
|
|
|
|
T = TypeVar("T")
|
|
|
|
|
|
def retry_on_rate_limit(
|
|
exception_type: type[BaseException],
|
|
is_rate_limit: Callable[[BaseException], bool] = lambda _exc: True,
|
|
*,
|
|
provider_name: str = "provider",
|
|
) -> Callable[[Callable[..., Awaitable[T]]], Callable[..., Awaitable[T]]]:
|
|
"""Build a decorator that retries on rate-limit exceptions.
|
|
|
|
Args:
|
|
exception_type: Catch this exception class (e.g. ``openai.RateLimitError``,
|
|
``mistralai.client.errors.SDKError``).
|
|
is_rate_limit: Predicate that decides whether a caught exception is
|
|
actually a rate-limit (vs. some other error of the same class).
|
|
Defaults to "always True" — appropriate when ``exception_type`` is
|
|
already a rate-limit-specific class.
|
|
provider_name: Used in log messages so operators can tell which
|
|
provider exhausted retries.
|
|
"""
|
|
|
|
def decorator(func: Callable[..., Awaitable[T]]) -> Callable[..., Awaitable[T]]:
|
|
@wraps(func)
|
|
async def wrapper(*args: Any, **kwargs: Any) -> T:
|
|
retry_delay = INITIAL_RETRY_DELAY
|
|
last_error: BaseException | None = None
|
|
|
|
for attempt in range(1, MAX_RETRIES + 1):
|
|
try:
|
|
return await func(*args, **kwargs)
|
|
except exception_type as e:
|
|
if not is_rate_limit(e):
|
|
raise
|
|
last_error = e
|
|
if attempt < MAX_RETRIES:
|
|
logger.warning(
|
|
"%s rate limit hit (attempt %d/%d), retrying in %.1fs...",
|
|
provider_name,
|
|
attempt,
|
|
MAX_RETRIES,
|
|
retry_delay,
|
|
)
|
|
await anyio.sleep(retry_delay)
|
|
retry_delay = min(retry_delay * 2, MAX_RETRY_DELAY)
|
|
|
|
logger.error(
|
|
"%s rate limit exceeded after %d attempts", provider_name, MAX_RETRIES
|
|
)
|
|
if last_error is None: # pragma: no cover — loop above always sets this
|
|
raise RuntimeError("retry loop exited without capturing an error")
|
|
raise last_error
|
|
|
|
return wrapper
|
|
|
|
return decorator
|