fix(openai): Add integ tests for error handling

Author: Murat Kaan Meral

tests_integ/models/test_model_openai.py

@@ -1,11 +1,13 @@
 import os
+import unittest.mock
 
 import pydantic
 import pytest
 
 import strands
 from strands import Agent, tool
 from strands.models.openai import OpenAIModel
+from strands.types.exceptions import ContextWindowOverflowException, ModelThrottledException
 from tests_integ.models import providers
 
 # these tests only run if we have the openai api key
@@ -167,3 +169,55 @@ def tool_with_image_return():
     # 'user', but this message with role 'tool' contains an image URL."
     # See https://github.com/strands-agents/sdk-python/issues/320 for additional details
     agent("Run the the tool and analyze the image")
+
+
+def test_context_window_overflow_integration():
+    """Integration test for context window overflow with OpenAI.
+
+    This test verifies that when a request exceeds the model's context window,
+    the OpenAI model properly raises a ContextWindowOverflowException.
+    """
+    # Use gpt-4o-mini which has a smaller context window to make this test more reliable
+    mini_model = OpenAIModel(
+        model_id="gpt-4o-mini-2024-07-18",
+        client_args={
+            "api_key": os.getenv("OPENAI_API_KEY"),
+        },
+    )
+
+    agent = Agent(model=mini_model)
+
+    # Create a very long text that should exceed context window
+    # This text is designed to be long enough to exceed context but not hit token rate limits
+    long_text = (
+        "This text is longer than context window, but short enough to not get caught in token rate limit. " * 6800
+    )
+
+    # This should raise ContextWindowOverflowException which gets handled by conversation manager
+    # The agent should attempt to reduce context and retry
+    with pytest.raises(ContextWindowOverflowException):
+        agent(long_text)
+
+
+def test_rate_limit_throttling_integration_no_retries(model):
+    """Integration test for rate limit handling with retries disabled.
+
+    This test verifies that when a request exceeds OpenAI's rate limits,
+    the model properly raises a ModelThrottledException. We disable retries
+    to avoid waiting for the exponential backoff during testing.
+    """
+    # Patch the event loop constants to disable retries for this test
+    with unittest.mock.patch("strands.event_loop.event_loop.MAX_ATTEMPTS", 1):
+        agent = Agent(model=model)
+
+        # Create a message that's very long to trigger token-per-minute rate limits
+        # This should be large enough to exceed TPM limits immediately
+        very_long_text = "Really long text " * 20000
+
+        # This should raise ModelThrottledException without retries
+        with pytest.raises(ModelThrottledException) as exc_info:
+            agent(very_long_text)
+
+        # Verify it's a rate limit error
+        error_message = str(exc_info.value).lower()
+        assert "rate limit" in error_message or "tokens per min" in error_message