fix: resolve provider icon sync failures (v1.6.1)

Three bugs prevented provider icons from reliably appearing in the
OpenWebUI model selector:

1. Wrong lookup order in _get_provider_icon: the hardcoded
   _PROVIDER_ICONS dict was always consulted before the dynamic
   OpenRouter registry. If OpenRouter changed any CDN path, the
   hardcoded (stale) URL was returned and served 404s while the
   registry — which always has current paths — was ignored for
   those providers. Fix: registry-first, hardcoded as fallback.

2. _provider_registry cached forever: a transient fetch failure
   at startup (network not ready, rate-limited, API hiccup) would
   leave the registry permanently empty until the pipe restarted.
   Non-200 responses were also silently ignored. Fix: added a 1h
   TTL (_PROVIDER_REGISTRY_TTL) and a log message for non-200s.

3. _icons_synced never reset: the set tracking "already synced"
   model IDs was never cleared between 5-minute model-cache
   refreshes. OWUI upserts models with the default data: icon
   after every pipes() call; the permanent _icons_synced state
   meant that any OWUI-overwritten icon was never restored. Fix:
   clear _icons_synced whenever the model cache is refreshed.

Tests: 555 passed (6 new assertions covering the three fixes).

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Author: sena-labs

CHANGELOG.md

@@ -7,6 +7,15 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [1.6.1] — 2026-05-08
+
+### Fixed
+
+- **Provider icon lookup order** — `_get_provider_icon` now consults the dynamic OpenRouter registry first and falls back to the hardcoded `_PROVIDER_ICONS` dict only when the registry is unavailable. Previously the hardcoded dict always took priority, meaning any CDN path change on OpenRouter's side would silently serve 404 URLs for the 13 built-in providers while the registry (which always has correct current URLs) was ignored for them.
+- **Provider registry now refreshes hourly** — `_load_provider_registry` previously cached the result for the entire lifetime of the `Pipe` instance. A transient network failure at startup (API down, rate-limited, not yet reachable) would permanently leave the registry empty until the pipe was restarted. Now the cache expires after `_PROVIDER_REGISTRY_TTL` (1 hour) and a fresh fetch is attempted automatically.
+- **Non-200 registry responses now logged** — HTTP 4xx/5xx responses from `GET /api/frontend/all-providers` were previously swallowed silently, making it impossible to diagnose why icons were missing. A `[OpenRouter Pipe] Provider registry returned HTTP {status}` message is now printed.
+- **`_icons_synced` cleared on model cache refresh** — the set of "already synced" model IDs was never reset between 5-minute model-cache cycles. OWUI upserts models (resetting their `profile_image_url` to the default `data:` icon) after every `pipes()` call; the permanent `_icons_synced` state meant the corrective re-sync was never retried. The set is now cleared whenever the model cache is refreshed, so any OWUI-overwritten icon is restored on the next sync pass.
+
 ## [1.6.0] — 2026-05-08
 
 ### Added

openrouter_pipe.py

@@ -3,7 +3,7 @@
 author: Sena Labs
 author_url: https://github.com/sena-labs
 funding_url: https://github.com/sponsors/sena-labs
-version: 1.6.0
+version: 1.6.1
 license: MIT
 icon_url: data:image/svg+xml;base64,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
 required_open_webui_version: 0.4.0
@@ -50,6 +50,11 @@
 # Cache TTL for model list (seconds)
 _MODELS_CACHE_TTL = 300.0  # 5 minutes
 
+# Cache TTL for the OpenRouter provider registry (seconds).
+# Refreshed periodically so transient fetch failures and CDN path changes
+# are recovered automatically without restarting the pipe.
+_PROVIDER_REGISTRY_TTL = 3600.0  # 1 hour
+
 # OpenRouter's frontend provider registry — gives us icon URLs for ~70 providers
 # (hosted SVG/PNG when available, gstatic favicons otherwise). Used as a
 # dynamic fallback when a model's author isn't in _PROVIDER_ICONS.
@@ -652,8 +657,9 @@ def __init__(self) -> None:
         # Track which model IDs already have icons synced (avoids repeated DB writes)
         self._icons_synced: set = set()
         # Lazy-loaded mirror of OpenRouter's provider registry (slug → icon URL).
-        # None = not attempted; {} = attempted but failed/empty (do not retry).
+        # Refreshed every _PROVIDER_REGISTRY_TTL seconds; None = not yet fetched.
         self._provider_registry: Optional[dict] = None
+        self._provider_registry_ts: float = 0.0
         # Lazy-loaded set of model IDs that have at least one ZDR endpoint.
         # None = not attempted; frozenset() = attempted but failed/empty.
         self._zdr_model_ids: Optional[frozenset] = None
@@ -874,6 +880,11 @@ def pipes(self) -> List[dict]:
         self._models_cache = models
         self._models_cache_ts = time.monotonic()
         self._models_cache_key = self._build_cache_key()
+        # Reset synced-set on every model cache refresh so _sync_model_icons
+        # re-checks all models. OWUI upserts models with the default data: icon
+        # after every pipes() call; clearing here ensures any overwritten icon
+        # is restored on the next sync pass.
+        self._icons_synced.clear()
 
         # Sync provider icons into Open WebUI's Models database
         if self.valves.SYNC_PROVIDER_ICONS:
@@ -1119,14 +1130,21 @@ def get_provider_icon(provider: str) -> Optional[str]:
         return _PROVIDER_ICONS.get(provider.lower())
 
     def _load_provider_registry(self) -> dict:
-        """Lazy-load OpenRouter's provider registry, cache for the pipe lifetime.
+        """Load OpenRouter's provider registry, refreshing every hour.
 
         Returns ``{slug: icon_url}`` (with each slug also indexed under its
         hyphen-stripped variant so e.g. ``x-ai`` resolves to the registry's
-        ``xai`` entry). Network failures are silent — a single empty dict is
-        cached and the pipe falls back to the hardcoded ``_PROVIDER_ICONS``.
+        ``xai`` entry). Network failures are logged and an empty dict is
+        returned; the pipe falls back to ``_PROVIDER_ICONS`` in that case.
+        The cache is automatically refreshed after ``_PROVIDER_REGISTRY_TTL``
+        seconds so transient failures and CDN path changes are recovered
+        without restarting the pipe.
         """
-        if self._provider_registry is not None:
+        now = time.monotonic()
+        if (
+            self._provider_registry is not None
+            and (now - self._provider_registry_ts) < _PROVIDER_REGISTRY_TTL
+        ):
             return self._provider_registry
 
         registry: dict = {}
@@ -1153,28 +1171,37 @@ def _load_provider_registry(self) -> dict:
                         compact = slug.replace("-", "")
                         if compact and compact != slug:
                             registry.setdefault(compact, icon)
+                else:
+                    print(
+                        f"[OpenRouter Pipe] Provider registry returned HTTP "
+                        f"{resp.status_code} — provider icons may be incomplete"
+                    )
             finally:
                 resp.close()
         except Exception as exc:  # pragma: no cover
             print(f"[OpenRouter Pipe] Provider registry fetch failed: {exc}")
 
         self._provider_registry = registry
+        self._provider_registry_ts = now
         return registry
 
     def _get_provider_icon(self, provider_key: str) -> Optional[str]:
         """Resolve a provider icon URL using the layered fallback chain.
 
-        Order: hardcoded ``_PROVIDER_ICONS`` → registry exact match →
-        registry hyphen-stripped match. Returns ``None`` if no source has it.
+        Order: registry exact match → registry hyphen-stripped →
+        hardcoded ``_PROVIDER_ICONS``. The registry is authoritative because
+        it always reflects OpenRouter's current CDN paths; the hardcoded dict
+        is a reliable offline fallback when the registry is unavailable.
+        Returns ``None`` if no source has it.
         """
         if not provider_key:
             return None
         key = provider_key.lower()
-        icon = _PROVIDER_ICONS.get(key)
+        registry = self._load_provider_registry()
+        icon = registry.get(key) or registry.get(key.replace("-", ""))
         if icon:
             return icon
-        registry = self._load_provider_registry()
-        return registry.get(key) or registry.get(key.replace("-", "")) or None
+        return _PROVIDER_ICONS.get(key)
 
     def _parse_provider_filter(self) -> Optional[set]:
         """Parse MODEL_PROVIDERS valve into a set of lowercase provider names."""

test_pipe.py

@@ -44,6 +44,7 @@
 _format_citation_list = mod._format_citation_list
 _OWUI_INTERNAL_KEYS = mod._OWUI_INTERNAL_KEYS
 _is_owui_managed_icon = mod._is_owui_managed_icon
+_PROVIDER_REGISTRY_TTL = mod._PROVIDER_REGISTRY_TTL
 
 # ── Helpers ───────────────────────────────────────────────────────────────────
 _PASS = 0
@@ -2108,11 +2109,12 @@ def _counting_reg_get(url, *args, **kwargs):
 _pipe_lookup = Pipe()
 _pipe_lookup.valves = Pipe.Valves(OPENROUTER_API_KEY="test-key")
 with patch.object(_pipe_lookup._session, "get", side_effect=_counting_reg_get):
-    # Hardcoded fast path — registry never consulted
+    # Registry is consulted first; openai is in both registry and hardcoded dict —
+    # the registry URL wins and matches the hardcoded one.
     _icon_openai = _pipe_lookup._get_provider_icon("openai")
     _assert(
         _icon_openai == "https://openrouter.ai/images/icons/OpenAI.svg",
-        "_get_provider_icon: hardcoded dict hit returns OpenAI icon",
+        "_get_provider_icon: registry-first hit returns OpenAI icon",
     )
 
     # Slug not in dict but in registry (exact)
@@ -2139,7 +2141,7 @@ def _counting_reg_get(url, *args, **kwargs):
     # Empty/None provider key
     _assert(_pipe_lookup._get_provider_icon("") is None, "_get_provider_icon: empty key → None")
 
-# 25l. Registry network failure → cached empty dict, no retry, dict still works
+# 25l. Registry network failure → cached empty dict, no retry within TTL window
 _pipe_fail = Pipe()
 _pipe_fail.valves = Pipe.Valves(OPENROUTER_API_KEY="test-key")
 
@@ -2151,9 +2153,9 @@ def _failing_reg_get(*args, **kwargs):
 
 with patch.object(_pipe_fail._session, "get", side_effect=_failing_reg_get):
     _r_fail = _pipe_fail._load_provider_registry()
-    _r_fail_2 = _pipe_fail._load_provider_registry()
+    _r_fail_2 = _pipe_fail._load_provider_registry()  # within TTL — no second fetch
 _assert(_r_fail == {}, "registry: network failure → empty dict")
-_assert(_fail_call_count == 1, "registry: failure does not retry (cached empty)")
+_assert(_fail_call_count == 1, "registry: no retry within TTL window after failure")
 
 # Hardcoded dict still works after registry failure
 _assert(
@@ -2165,16 +2167,65 @@ def _failing_reg_get(*args, **kwargs):
     "_get_provider_icon: x-ai falls back to None when registry failed",
 )
 
-# 25m. Registry HTTP non-200 → empty dict
+# 25m. Registry HTTP non-200 → empty dict + log message
 _mock_reg_403 = MagicMock()
 _mock_reg_403.status_code = 403
 _mock_reg_403.json.return_value = {"data": []}
 
 _pipe_403 = Pipe()
 _pipe_403.valves = Pipe.Valves(OPENROUTER_API_KEY="test-key")
-with patch.object(_pipe_403._session, "get", return_value=_mock_reg_403):
-    _r_403 = _pipe_403._load_provider_registry()
-_assert(_r_403 == {}, "registry: HTTP 403 → empty dict (no parse, no retry)")
+_log_403_msgs = []
+with patch("builtins.print", side_effect=lambda *a, **kw: _log_403_msgs.append(" ".join(str(x) for x in a))):
+    with patch.object(_pipe_403._session, "get", return_value=_mock_reg_403):
+        _r_403 = _pipe_403._load_provider_registry()
+_assert(_r_403 == {}, "registry: HTTP 403 → empty dict")
+_assert(
+    any("403" in m for m in _log_403_msgs),
+    "registry: HTTP 403 logs a warning message",
+)
+
+# 25n. Registry TTL expiry forces re-fetch
+_pipe_ttl = Pipe()
+_pipe_ttl.valves = Pipe.Valves(OPENROUTER_API_KEY="test-key")
+
+_ttl_call_count = 0
+def _ttl_reg_get(url, *args, **kwargs):
+    global _ttl_call_count
+    if "all-providers" in url:
+        _ttl_call_count += 1
+    return _mock_reg_resp  # reuse payload mock
+
+with patch.object(_pipe_ttl._session, "get", side_effect=_ttl_reg_get):
+    _pipe_ttl._load_provider_registry()               # first fetch
+    _pipe_ttl._provider_registry_ts -= _PROVIDER_REGISTRY_TTL + 1  # expire TTL
+    _pipe_ttl._load_provider_registry()               # should re-fetch
+_assert(_ttl_call_count == 2, "registry: re-fetches after TTL expiry")
+
+# 25o. _icons_synced cleared on model cache refresh
+_pipe_sync_clear = Pipe()
+_pipe_sync_clear.valves = Pipe.Valves(OPENROUTER_API_KEY="test-key", SYNC_PROVIDER_ICONS=False)
+_pipe_sync_clear._models_cache = None
+
+_mock_models_resp_sc = MagicMock()
+_mock_models_resp_sc.status_code = 200
+_mock_models_resp_sc.json.return_value = {"data": [{"id": "openai/gpt-4o", "name": "GPT-4o"}]}
+
+with patch.object(_pipe_sync_clear._session, "get", return_value=_mock_models_resp_sc):
+    _pipe_sync_clear.pipes()
+
+# Populate _icons_synced to simulate prior sync
+_pipe_sync_clear._icons_synced.add("openai/gpt-4o")
+_assert(len(_pipe_sync_clear._icons_synced) == 1, "_icons_synced: populated before cache expire")
+
+# Expire cache and call pipes() again — _icons_synced must be cleared
+_pipe_sync_clear._models_cache_ts = 0.0
+with patch.object(_pipe_sync_clear._session, "get", return_value=_mock_models_resp_sc):
+    _pipe_sync_clear.pipes()
+
+_assert(
+    len(_pipe_sync_clear._icons_synced) == 0,
+    "_icons_synced: cleared on model cache refresh (allows re-sync after OWUI upsert)",
+)
 
 # ── 26. _stream_response() edge cases ────────────────────────────────────────