Skip to content

Independent Auditing of AI Agents. Run by human or the agent itself, to answer the most crucial question in the AI Agent Economy. Is the agent doing what is supposed to do? With iFixAi you can have this answer in less than 120 seconds.

fix: preserve exhausted-quota errors and stop MiniMax retries (#269)

Rudy Mizrahi Celeklid18a3b9a395f
diff --git a/ifixai/providers/base.py b/ifixai/providers/base.py
index db63b93..0b95ff6 100644
--- a/ifixai/providers/base.py
+++ b/ifixai/providers/base.py
@@ -137,6 +137,7 @@ _FATAL_ERROR_MARKERS: tuple[str, ...] = (
_QUOTA_EXHAUSTED_MARKERS: tuple[str, ...] = (
"insufficient_quota",
"insufficient credits",
+ "insufficient balance",
"payment required",
)
diff --git a/ifixai/providers/minimax.py b/ifixai/providers/minimax.py
index 0c8163b..b967573 100644
--- a/ifixai/providers/minimax.py
+++ b/ifixai/providers/minimax.py
@@ -15,6 +15,7 @@ from ifixai.providers.base import (
ProviderRateLimitError,
ProviderResponseError,
ProviderTimeoutError,
+ is_fatal_provider_error,
raise_if_truncated,
)
from ifixai.providers.secrets import scrub_secrets
@@ -279,6 +280,8 @@ class MiniMaxProvider(ChatProvider):
ProviderRateLimitError,
ProviderTimeoutError,
) as exc:
+ if is_fatal_provider_error(exc):
+ raise
last_error = exc
if attempt == config.max_retries:
raise
@@ -317,7 +320,7 @@ class MiniMaxProvider(ChatProvider):
raise ProviderRateLimitError(
provider="minimax",
endpoint=endpoint,
- details="HTTP 429: rate limited",
+ details=f"HTTP 429: {scrub_secrets(body[:500])}",
)
if response.status in RETRYABLE_HTTP_STATUS_CODES:
raise ProviderOverloadedError(
diff --git a/ifixai/tests/test_minimax_exhausted_quota.py b/ifixai/tests/test_minimax_exhausted_quota.py
new file mode 100644
index 0000000..10efa5d
--- /dev/null
+++ b/ifixai/tests/test_minimax_exhausted_quota.py
@@ -0,0 +1,72 @@
+import pytest
+from aiohttp import web
+
+from ifixai.core.types import ChatMessage, ProviderConfig
+from ifixai.providers.base import ProviderRateLimitError, is_fatal_provider_error
+from ifixai.providers.minimax import MiniMaxProvider
+
+
+ {"code": "insufficient_quota", "message": "owned quota exhausted"},
+ {"code": 1008, "message": "insufficient balance"},
+])
+async def test_minimax_exhausted_quota_is_fatal_without_retrying(monkeypatch, error):
+ requests = []
+
+ async def no_sleep(*args):
+ return None
+
+ async def reply(request):
+ requests.append(await request.json())
+ return web.json_response({'error': error}, status=429)
+
+ monkeypatch.setattr('ifixai.providers.minimax.asyncio.sleep', no_sleep)
+ app = web.Application()
+ app.router.add_post('/chat/completions', reply)
+ runner = web.AppRunner(app)
+ await runner.setup()
+ site = web.TCPSite(runner, '127.0.0.1', 0)
+ await site.start()
+ endpoint = f'http://127.0.0.1:{site._server.sockets[0].getsockname()[1]}'
+ provider = MiniMaxProvider()
+ try:
+ with pytest.raises(ProviderRateLimitError) as caught:
+ await provider.send_message([ChatMessage(content='hello')], ProviderConfig(provider='minimax', endpoint=endpoint, max_retries=1))
+ assert is_fatal_provider_error(caught.value)
+ assert len(requests) == 1
+ finally:
+ await provider.aclose()
+ await runner.cleanup()
+
+
+async def test_minimax_temporary_throttle_still_recovers(monkeypatch):
+ requests = []
+ sleeps = []
+
+ async def no_sleep(delay):
+ sleeps.append(delay)
+
+ async def reply(request):
+ requests.append(await request.json())
+ if len(requests) == 1:
+ return web.json_response({'error': {'code': 1002, 'message': 'rate limit exceeded'}}, status=429)
+ return web.json_response({'choices': [{'message': {'content': 'owned recovered reply'}}]})
+
+ monkeypatch.setattr('ifixai.providers.minimax.asyncio.sleep', no_sleep)
+ app = web.Application()
+ app.router.add_post('/chat/completions', reply)
+ runner = web.AppRunner(app)
+ await runner.setup()
+ site = web.TCPSite(runner, '127.0.0.1', 0)
+ await site.start()
+ endpoint = f'http://127.0.0.1:{site._server.sockets[0].getsockname()[1]}'
+ provider = MiniMaxProvider()
+ try:
+ assert await provider.send_message([ChatMessage(content='hello')], ProviderConfig(provider='minimax', endpoint=endpoint, max_retries=1)) == 'owned recovered reply'
+ assert len(requests) == 2
+ assert sleeps == [1]
+ finally:
+ await provider.aclose()
+ await runner.cleanup()