From ddf13f70e9ba8b3536757c5af7ebda0b019c4351 Mon Sep 17 00:00:00 2001
From: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
Date: Fri, 18 Sep 2026 12:06:36 +0000
Subject: [PATCH] feature: add unit tests for generic_call_api_with_retries
 retry/backoff paths

Co-Authored-By: Gilad Samuels <gilad.samuels@cognition.ai>
---
 tests/llm/test_llm_utils.py | 126 ++++++++++++++++++++++++++++++++++++
 1 file changed, 126 insertions(+)

diff --git a/tests/llm/test_llm_utils.py b/tests/llm/test_llm_utils.py
index d7c42853d..ab3078a51 100644
--- a/tests/llm/test_llm_utils.py
+++ b/tests/llm/test_llm_utils.py
@@ -216,6 +216,132 @@ def test_retry_parse_raises():
         llm_utils.retry(mock_chat, llm_utils.Discussion(), 3, parser_raises)
 
 
+class FakeRateLimitError(Exception):
+    pass
+
+
+class FakeAPIError(Exception):
+    def __init__(self, message="api error", status_code=None):
+        super().__init__(message)
+        self.status_code = status_code
+
+
+def fake_get_status_code(exception):
+    return getattr(exception, "status_code", None)
+
+
+@pytest.fixture
+def recorded_sleeps(monkeypatch):
+    sleeps = []
+    monkeypatch.setattr(llm_utils.time, "sleep", lambda delay: sleeps.append(delay))
+    return sleeps
+
+
+def call_with_retries(client_function, **kwargs):
+    params = dict(
+        client_function=client_function,
+        api_params={},
+        is_response_valid_fn=lambda response: response == "valid",
+        rate_limit_exceptions=(FakeRateLimitError,),
+        api_error_exceptions=(FakeAPIError,),
+        get_status_code_fn=fake_get_status_code,
+        max_retries=4,
+        initial_retry_delay_seconds=2,
+        max_retry_delay_seconds=10,
+    )
+    params.update(kwargs)
+    return llm_utils.generic_call_api_with_retries(**params)
+
+
+def test_generic_call_api_succeeds_without_sleeping(recorded_sleeps):
+    client = Mock(return_value="valid")
+
+    assert call_with_retries(client) == "valid"
+    assert client.call_count == 1
+    assert recorded_sleeps == []
+
+
+def test_generic_call_api_retries_rate_limit_error(recorded_sleeps):
+    client = Mock(side_effect=[FakeRateLimitError("slow down"), "valid"])
+
+    assert call_with_retries(client) == "valid"
+    assert client.call_count == 2
+    assert recorded_sleeps == [4]  # initial_delay * 2 ** attempt, attempt == 1
+
+
+@pytest.mark.parametrize("status_code", [429, 500, 503])
+def test_generic_call_api_retries_retriable_status_codes(recorded_sleeps, status_code):
+    client = Mock(side_effect=[FakeAPIError(status_code=status_code), "valid"])
+
+    assert call_with_retries(client) == "valid"
+    assert client.call_count == 2
+    assert recorded_sleeps == [4]
+
+
+@pytest.mark.parametrize("status_code", [400, 401, 404, None])
+def test_generic_call_api_reraises_non_retriable_status_codes(recorded_sleeps, status_code):
+    client = Mock(side_effect=FakeAPIError(status_code=status_code))
+
+    with pytest.raises(FakeAPIError):
+        call_with_retries(client)
+
+    assert client.call_count == 1
+    assert recorded_sleeps == []
+
+
+def test_generic_call_api_reraises_unexpected_exception(recorded_sleeps):
+    client = Mock(side_effect=ValueError("boom"))
+
+    with pytest.raises(ValueError):
+        call_with_retries(client)
+
+    assert client.call_count == 1
+    assert recorded_sleeps == []
+
+
+def test_generic_call_api_retries_invalid_response(recorded_sleeps):
+    client = Mock(side_effect=["invalid", "invalid", "valid"])
+
+    assert call_with_retries(client) == "valid"
+    assert client.call_count == 3
+    # invalid content uses attempt - 1 as the exponent
+    assert recorded_sleeps == [2, 4]
+
+
+def test_generic_call_api_raises_runtime_error_when_retries_exhausted(recorded_sleeps):
+    client = Mock(return_value="invalid")
+
+    with pytest.raises(RuntimeError, match="failed after 4 retries"):
+        call_with_retries(client)
+
+    assert client.call_count == 4
+    # no sleep on the last attempt, which gives up instead of retrying
+    assert recorded_sleeps == [2, 4, 8]
+
+
+def test_generic_call_api_raises_runtime_error_after_repeated_rate_limits(recorded_sleeps):
+    client = Mock(side_effect=FakeRateLimitError("slow down"))
+
+    with pytest.raises(RuntimeError):
+        call_with_retries(client)
+
+    assert client.call_count == 4
+    assert recorded_sleeps == [4, 8, 10]  # capped by max_retry_delay_seconds
+
+
+def test_generic_call_api_reraises_when_status_code_fn_fails(recorded_sleeps):
+    client = Mock(side_effect=FakeAPIError(status_code=429))
+
+    def failing_status_fn(exception):
+        raise AttributeError("no status code")
+
+    with pytest.raises(FakeAPIError):
+        call_with_retries(client, get_status_code_fn=failing_status_fn)
+
+    assert client.call_count == 1
+    assert recorded_sleeps == []
+
+
 def test_extract_code_blocks():
     text = """\
 This is some text.