From 4e4935a64a3f395a30d5f19be31700dbdd829835 Mon Sep 17 00:00:00 2001
From: DhruvTilva <tilvadhruv8@gmail.com>
Date: Mon, 29 Jun 2026 09:45:03 +0100
Subject: [PATCH] fix(llm): enforce API timeout in the secondary LLM dispatch
 path (#1442)
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

_call_llm (used by the dedup LLM tiebreaker) built its Anthropic and
OpenAI-compatible clients with max_retries but no timeout, so requests
on this path silently ignored GRAPHIFY_API_TIMEOUT — unlike the primary
extraction paths (_call_openai_compat / _call_claude) which already pass
both. Add timeout=_resolve_api_timeout() to both constructors.

The PR branch self-neutralized: a v8 merge resolved the conflict in
favor of the max_retries-bearing line and dropped the original one-line
fix, so it is re-applied here on top of current v8 with max_retries
preserved. Adds regression coverage for both _call_llm branches, which
were previously untested.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---
 graphify/llm.py            |  4 +--
 tests/test_llm_backends.py | 59 ++++++++++++++++++++++++++++++++++++++
 2 files changed, 61 insertions(+), 2 deletions(-)

diff --git a/graphify/llm.py b/graphify/llm.py
index 396f9dc1..c0d2efa2 100644
--- a/graphify/llm.py
+++ b/graphify/llm.py
@@ -1905,7 +1905,7 @@ def _call_llm(
             import anthropic
         except ImportError as exc:
             raise ImportError(_backend_pkg_hint("anthropic", "anthropic")) from exc
-        client = anthropic.Anthropic(api_key=key, base_url=cfg["base_url"], max_retries=_resolve_max_retries())
+        client = anthropic.Anthropic(api_key=key, base_url=cfg["base_url"], timeout=_resolve_api_timeout(), max_retries=_resolve_max_retries())
         resp = client.messages.create(
             model=mdl,
             max_tokens=max_tokens,
@@ -1988,7 +1988,7 @@ def _call_llm(
         from openai import OpenAI
     except ImportError as exc:
         raise ImportError(_backend_pkg_hint("openai", "openai")) from exc
-    client = OpenAI(api_key=key, base_url=cfg["base_url"], max_retries=_resolve_max_retries())
+    client = OpenAI(api_key=key, base_url=cfg["base_url"], timeout=_resolve_api_timeout(), max_retries=_resolve_max_retries())
     kwargs: dict = {
         "model": mdl,
         "messages": [{"role": "user", "content": prompt}],
diff --git a/tests/test_llm_backends.py b/tests/test_llm_backends.py
index acb04e95..79f64027 100644
--- a/tests/test_llm_backends.py
+++ b/tests/test_llm_backends.py
@@ -1028,3 +1028,62 @@ def test_openai_compat_client_built_with_retries(monkeypatch):
         "user msg", temperature=0, max_completion_tokens=4096, backend="kimi",
     )
     assert ctor_kwargs.get("max_retries", 0) >= 5, ctor_kwargs
+
+
+def test_call_llm_claude_client_built_with_timeout_and_retries(monkeypatch):
+    """The secondary dispatch path (_call_llm, used by the dedup tiebreaker)
+    must build its Anthropic client with both timeout and max_retries, matching
+    the primary extraction path — #1442. Previously _call_llm passed neither
+    (then only max_retries), so GRAPHIFY_API_TIMEOUT was silently ignored here."""
+    import sys
+    import types
+
+    ctor_kwargs = {}
+
+    class _FakeMessages:
+        def create(self, **_):
+            return types.SimpleNamespace(content=[types.SimpleNamespace(text="ok")])
+
+    class _FakeAnthropic:
+        def __init__(self, *_, **kwargs):
+            ctor_kwargs.update(kwargs)
+            self.messages = _FakeMessages()
+
+    fake_module = types.ModuleType("anthropic")
+    fake_module.Anthropic = _FakeAnthropic
+    monkeypatch.setitem(sys.modules, "anthropic", fake_module)
+    monkeypatch.setattr(llm, "_get_backend_api_key", lambda _b: "fake-key")
+    monkeypatch.setenv("GRAPHIFY_API_TIMEOUT", "1")
+    monkeypatch.delenv("GRAPHIFY_MAX_RETRIES", raising=False)
+
+    assert llm._call_llm("hi", backend="claude") == "ok"
+    assert ctor_kwargs.get("timeout") == 1.0, ctor_kwargs
+    assert ctor_kwargs.get("max_retries", 0) >= 5, ctor_kwargs
+
+
+def test_call_llm_openai_compat_client_built_with_timeout_and_retries(monkeypatch):
+    """Same #1442 fix for the OpenAI-compatible branch of _call_llm."""
+    import sys
+    import types
+
+    ctor_kwargs = {}
+
+    class _FakeOpenAI:
+        def __init__(self, *_, **kwargs):
+            ctor_kwargs.update(kwargs)
+            self.chat = self
+            self.completions = self
+
+        def create(self, **_):
+            return _fake_openai_response("ok", finish_reason="stop", completion_tokens=1)
+
+    fake_module = types.ModuleType("openai")
+    fake_module.OpenAI = _FakeOpenAI
+    monkeypatch.setitem(sys.modules, "openai", fake_module)
+    monkeypatch.setattr(llm, "_get_backend_api_key", lambda _b: "fake-key")
+    monkeypatch.setenv("GRAPHIFY_API_TIMEOUT", "1")
+    monkeypatch.delenv("GRAPHIFY_MAX_RETRIES", raising=False)
+
+    llm._call_llm("hi", backend="kimi")
+    assert ctor_kwargs.get("timeout") == 1.0, ctor_kwargs
+    assert ctor_kwargs.get("max_retries", 0) >= 5, ctor_kwargs