From fb6dd28ac3e1031cb052cf498ef2e483fd34e492 Mon Sep 17 00:00:00 2001
From: AJV20 <24819659+AJV20@users.noreply.github.com>
Date: Wed, 27 May 2026 12:19:42 -0400
Subject: [PATCH] fix: cap custom models endpoint probes

---
 CHANGELOG.md                                  |  4 ++++
 api/config.py                                 |  8 +++++++-
 tests/test_issue2540_models_endpoint_error.py | 19 +++++++++++++++++++
 3 files changed, 30 insertions(+), 1 deletion(-)

diff --git a/CHANGELOG.md b/CHANGELOG.md
index c365129f..c4a56117 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -3,6 +3,10 @@
 
 ## [Unreleased]
 
+### Fixed
+
+- Custom provider `/v1/models` discovery now uses a short per-endpoint timeout and gracefully skips slow or unreachable providers, reducing cold `/api/models` cache rebuild latency. (#3024)
+
 ## [v0.51.145] — 2026-05-26 — Release DQ (stage-batch27 — sidebar running-state preservation)
 
 ### Fixed
diff --git a/api/config.py b/api/config.py
index 103c9f21..34dd5e90 100644
--- a/api/config.py
+++ b/api/config.py
@@ -70,6 +70,12 @@ PROJECTS_FILE = STATE_DIR / "projects.json"
 
 logger = logging.getLogger(__name__)
 
+# Keep custom provider /v1/models probes below the frontend's generic request
+# timeout even when one upstream is slow or unreachable. The models cache rebuild
+# path probes configured custom endpoints serially, so each provider needs a
+# short hard cap and graceful degradation.
+CUSTOM_MODELS_ENDPOINT_TIMEOUT_SECONDS = 5.0
+
 
 def _env_mb_bytes(name: str, default_mb: int) -> int:
     """Parse an optional megabyte environment variable into bytes.
@@ -3653,7 +3659,7 @@ def get_available_models() -> dict:
                 req.add_header("User-Agent", "OpenAI/Python 1.0")
                 for k, v in headers.items():
                     req.add_header(k, v)
-                with urllib.request.urlopen(req, timeout=10) as response:  # nosec B310
+                with urllib.request.urlopen(req, timeout=CUSTOM_MODELS_ENDPOINT_TIMEOUT_SECONDS) as response:  # nosec B310
                     data = json.loads(response.read().decode("utf-8"))
                 return _extract_model_entries_from_payload(data, provider), None
             except urllib.error.HTTPError as exc:
diff --git a/tests/test_issue2540_models_endpoint_error.py b/tests/test_issue2540_models_endpoint_error.py
index 89dce0a0..b83cd9bc 100644
--- a/tests/test_issue2540_models_endpoint_error.py
+++ b/tests/test_issue2540_models_endpoint_error.py
@@ -112,6 +112,25 @@ def test_named_custom_provider_models_endpoint_5xx_preserves_status(monkeypatch,
     assert error["code"] == 502
     assert "returned 502" in error["message"]
 
+def test_named_custom_provider_models_endpoint_network_error_uses_short_timeout(monkeypatch, tmp_path):
+    observed_timeouts = []
+
+    def fake_urlopen(req, timeout=10):
+        observed_timeouts.append(timeout)
+        raise urllib.error.URLError("timed out")
+
+    monkeypatch.setattr(urllib.request, "urlopen", fake_urlopen)
+
+    with _ConfigState():
+        _configure_named_custom_provider(tmp_path, monkeypatch)
+        data = config.get_available_models()
+
+    group = _groups_by_provider(data)["custom:broken-proxy"]
+    assert group["models"] == []
+    assert group["models_endpoint_error"]["kind"] == "network"
+    assert observed_timeouts == [config.CUSTOM_MODELS_ENDPOINT_TIMEOUT_SECONDS]
+    assert max(observed_timeouts) <= 5.0
+
 
 def test_frontend_model_picker_renders_provider_endpoint_hint():
     ui = open("static/ui.js", encoding="utf-8").read()