Fix rotation providers never degrading: don't record success on lazy streams

For streaming requests providers returned a lazy generator and immediately
called record_success(), resetting the failure counter to 0 BEFORE the
rotation handler primed the stream and the upstream returned 400/429. Every
attempt did success(reset->0) then failure(->1), so the disable threshold (3)
was never reached and a consistently-failing provider was never disabled.

Guard the premature record_success() so it does not fire for streaming
(codex, openai, runpod, coderai). The caller already records success after
the stream is primed/consumed, so the failure counter now accumulates and a
failing provider is correctly disabled/cooled-down. Bump to 0.99.74.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

aisbf/__init__.py

@@ -55,7 +55,7 @@ from .auth.qwen import QwenOAuth2
 from .handlers import RequestHandler, RotationHandler, AutoselectHandler
 from .utils import count_messages_tokens, split_messages_into_chunks, get_max_request_tokens_for_model
 
-__version__ = "0.99.73"
+__version__ = "0.99.74"
 __all__ = [
     # Config
     "config",

aisbf/providers/coderai.py

@@ -496,7 +496,11 @@ class CoderAIProviderHandler(BaseProviderHandler):
                 raise RuntimeError(f"[{self.provider_id}] No active CoderAI broker session; direct fallback not allowed with broker_preferred=True")
 
             response = self.client.chat.completions.create(**payload)
-            self.record_success()
+            # Streaming returns a lazy iterator; recording success here would
+            # prematurely reset the failure counter before the stream is consumed.
+            # The caller records success after priming/consuming the stream.
+            if not stream:
+                self.record_success()
             return response
         except Exception:
             self.record_failure()

aisbf/providers/codex.py

@@ -858,7 +858,14 @@ class CodexProviderHandler(BaseProviderHandler):
                 )
             
             logger.info(f"CodexProviderHandler: Response received")
-            self.record_success()
+            # For streaming requests the response is a lazy async generator: the
+            # upstream HTTP request has not run yet, so we cannot know if it
+            # succeeded. Recording success here would prematurely reset the
+            # failure counter before priming surfaces an immediate provider error
+            # (e.g. 400/429), preventing the provider from ever being disabled.
+            # The caller records success after the stream is primed/consumed.
+            if not stream:
+                self.record_success()
             
             if AISBF_DEBUG:
                 logger.info(f"=== RAW CODEX RESPONSE ===")

aisbf/providers/openai.py

@@ -169,7 +169,11 @@ class OpenAIProviderHandler(BaseProviderHandler):
 
             response = self.client.chat.completions.create(**request_params)
             logging.info(f"OpenAIProviderHandler: Response received: {response}")
-            self.record_success()
+            # Streaming returns a lazy iterator; the upstream call has not been
+            # consumed yet, so recording success here would prematurely reset the
+            # failure counter. The caller records success after priming/consuming.
+            if not stream:
+                self.record_success()
             
             # Dump raw response if AISBF_DEBUG is enabled
             if AISBF_DEBUG:

aisbf/providers/runpod.py

@@ -86,7 +86,11 @@ class RunpodProviderHandler(BaseProviderHandler):
         await self._ensure_non_public_resource_ready()
         delegate = self._build_delegate_handler(self._wrapper_mode)
         response = await delegate.handle_request(model, messages, max_tokens=max_tokens, temperature=temperature, stream=stream, tools=tools, tool_choice=tool_choice)
-        self.record_success()
+        # For streaming the delegate returns a lazy iterator; recording success
+        # here would prematurely reset the failure counter before the stream is
+        # consumed. The caller records success after priming/consuming the stream.
+        if not stream:
+            self.record_success()
         return response
 
     async def get_models(self) -> List[Model]:

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "aisbf"
-version = "0.99.73"
+version = "0.99.74"
 description = "AISBF - AI Service Broker Framework || AI Should Be Free - A modular proxy server for managing multiple AI provider integrations"
 readme = "README.md"
 license = "GPL-3.0-or-later"

setup.py

@@ -106,7 +106,7 @@ class InstallCommand(_install):
 
 setup(
     name="aisbf",
-    version="0.99.73",
+    version="0.99.74",
     author="AISBF Contributors",
     author_email="stefy@nexlab.net",
     description="AISBF - AI Service Broker Framework || AI Should Be Free - A modular proxy server for managing multiple AI provider integrations",