fix(#4756): refresh stale model catalogs on session visit

api/config.py

@@ -3732,6 +3732,7 @@ _available_models_cache: dict | None = None
 _available_models_cache_ts: float = 0.0
 _available_models_cache_source_fingerprint: dict | None = None
 _AVAILABLE_MODELS_CACHE_TTL: float = 86400.0  # 24 hours
+_SESSION_VISIT_MODELS_FRESHNESS_SECONDS: float = 300.0
 _available_models_cache_lock = threading.RLock()  # must be RLock: cold path refactoring moved slow work inside this lock, requiring re-entry
 _cache_build_cv = threading.Condition(_available_models_cache_lock)  # shares underlying RLock so notify_all() is safe inside with _available_models_cache_lock
 _cache_build_in_progress = False  # True while a cold path is actively building
@@ -5228,7 +5229,7 @@ def _read_visible_codex_cache_model_ids() -> list[str]:
     return ordered
 
 
-def get_available_models(*, prefer_cache: bool = False) -> dict:
+def get_available_models(*, prefer_cache: bool = False, force_refresh: bool = False) -> dict:
     """
     Return available models grouped by provider.
 
@@ -5252,6 +5253,10 @@ def get_available_models(*, prefer_cache: bool = False) -> dict:
     server-initiated wakeup turn (Option Z) takes so a cold catalog can never
     block the wakeup chat/start on a flaky network. A normal human request
     leaves this False and keeps the full live-discovery behaviour.
+
+    ``force_refresh=True`` is an internal escape hatch for bounded freshness
+    checks that need a real live rebuild while preserving the default cache
+    contract for every existing caller.
     """
     global _cache_build_in_progress, _available_models_cache, _available_models_cache_ts, _available_models_cache_source_fingerprint, _cache_build_cv
     # Config mtime check — must come before any config reads.
@@ -6678,10 +6683,12 @@ def get_available_models(*, prefer_cache: bool = False) -> dict:
     # so only one thread rebuilds while others wait.
     disk_groups = None
     stale_disk_groups = None
-    if _available_models_cache is None:
+    if _available_models_cache is None and not force_refresh:
         disk_groups = _load_models_cache_from_disk()
         if disk_groups is None:
             stale_disk_groups = _load_stale_models_cache_from_disk()
+    elif force_refresh:
+        stale_disk_groups = _load_stale_models_cache_from_disk()
 
     with _available_models_cache_lock:
         # If another thread is already building, wait for its result instead
@@ -6706,16 +6713,16 @@ def get_available_models(*, prefer_cache: bool = False) -> dict:
 
         # Serve from memory cache if fresh
         now = time.monotonic()
-        cached = _get_fresh_memory_models_cache(now)
-        if cached is not None:
-            return cached
+        if not force_refresh:
+            cached = _get_fresh_memory_models_cache(now)
+            if cached is not None:
+                return cached
 
         # Cold path: disk cache hit — use it (fast, no lock contention)
-        if disk_groups is not None:
+        if disk_groups is not None and not force_refresh:
             _available_models_cache = disk_groups
             _available_models_cache_ts = now
             _available_models_cache_source_fingerprint = _models_cache_source_fingerprint()
-            _save_models_cache_to_disk(disk_groups)
             return copy.deepcopy(disk_groups)
 
         # ── prefer_cache: NEVER run the live provider rebuild ────────────────
@@ -6933,6 +6940,43 @@ def get_available_models(*, prefer_cache: bool = False) -> dict:
         return copy.deepcopy(_static_models_catalog_without_live_probes())
 
 
+def _models_cache_file_age_seconds(cache_path: Path, now: float) -> float | None:
+    try:
+        return max(0.0, now - cache_path.stat().st_mtime)
+    except OSError:
+        return None
+
+
+def get_available_models_for_session_visit() -> dict:
+    """Return /api/models with a short session-visit freshness horizon."""
+    global _available_models_cache, _available_models_cache_ts, _available_models_cache_source_fingerprint
+    cache_path = _get_models_cache_path()
+    cache_age = _models_cache_file_age_seconds(cache_path, time.time())
+    disk_cached = None
+    if cache_age is not None and cache_age < _SESSION_VISIT_MODELS_FRESHNESS_SECONDS:
+        now_mono = time.monotonic()
+        with _available_models_cache_lock:
+            cached = _get_fresh_memory_models_cache(now_mono)
+            if cached is not None:
+                return cached
+        disk_cached = _load_models_cache_from_disk()
+        if disk_cached is not None:
+            with _available_models_cache_lock:
+                _available_models_cache = copy.deepcopy(disk_cached)
+                _available_models_cache_ts = time.monotonic()
+                _available_models_cache_source_fingerprint = _models_cache_source_fingerprint()
+            return copy.deepcopy(disk_cached)
+
+    stale_cached = disk_cached or _load_stale_models_cache_from_disk()
+    try:
+        return get_available_models(force_refresh=True)
+    except Exception:
+        logger.debug("session-visit models refresh failed", exc_info=True)
+        if stale_cached is not None:
+            return copy.deepcopy(stale_cached)
+        return get_available_models(prefer_cache=True)
+
+
 # ── Static file path ─────────────────────────────────────────────────────────
 _INDEX_HTML_PATH = REPO_ROOT / "static" / "index.html"
 

api/routes.py

@@ -2278,6 +2278,7 @@ from api.config import (
     _resolve_cli_toolsets,
     _INDEX_HTML_PATH,
     get_available_models,
+    get_available_models_for_session_visit,
     _provider_is_known_or_configured,
     IMAGE_EXTS,
     MD_EXTS,
@@ -9158,6 +9159,11 @@ def handle_get(handler, parsed) -> bool:
         # the detached rebuild worker (and the legacy synchronous rebuild),
         # which the request-thread wrapper could not reach. See
         # api.config.get_available_models cold path + profile_scope_for_detached_worker.
+        freshness = parse_qs(parsed.query or "").get("freshness", [""])[0].strip().lower()
+        if freshness == "session_visit":
+            return j(handler, get_available_models_for_session_visit())
+        if freshness:
+            return bad(handler, f"unknown models freshness: {freshness}", status=400)
         return j(handler, get_available_models())
 
     if parsed.path == "/api/models/live":

static/sessions.js

@@ -1192,6 +1192,14 @@ async function loadSession(sid){
     return loadSession(continuationSid,{...opts,skipLineageResolve:true,skipContinuationResolve:true,force:true});
   }
   S.session=data.session;
+  if(typeof populateModelDropdown==='function'){
+    const modelRefreshSid=sid;
+    const modelRefreshPromise=Promise.resolve().then(()=>{
+      if(_loadingSessionId!==modelRefreshSid) return;
+      return populateModelDropdown({freshness:'session_visit'});
+    }).catch(()=>{});
+    if(typeof window!=='undefined') window._modelDropdownReady=modelRefreshPromise;
+  }
   // Loading a real existing session abandons any pre-session toolset override
   // staged on the empty composer — clear it so it can't leak into a later New
   // Chat started from this session (#4490 follow-up).

static/ui.js

@@ -1992,6 +1992,8 @@ function _persistSessionModelCorrection(model, provider, opts){
   });
   return opts&&opts.propagateErrors ? request : request.catch(()=>{});
 }
+let _modelDropdownRequestSeq=0;
+
 function _applySessionModelFallback(sel){
   if(!sel) return null;
   const configuredDefault=String(window._defaultModel||'').trim();
@@ -2014,10 +2016,15 @@ function _applySessionModelFallback(sel){
 async function populateModelDropdown(opts={}){
   const sel=$('modelSelect');
   if(!sel) return;
+  const requestSeq=++_modelDropdownRequestSeq;
   try{
-    const _modelsRes=await fetch(new URL('api/models',document.baseURI||location.href).href,{credentials:'include'});
+    const modelsUrl=new URL('api/models',document.baseURI||location.href);
+    if(opts&&opts.freshness) modelsUrl.searchParams.set('freshness',opts.freshness);
+    const _modelsRes=await fetch(modelsUrl.href,{credentials:'include'});
+    if(requestSeq!==_modelDropdownRequestSeq) return;
     if(_redirectIfUnauth(_modelsRes)) return;
     const data=await _modelsRes.json();
+    if(requestSeq!==_modelDropdownRequestSeq) return;
     // Store active provider globally so the send path can warn on mismatch
     window._activeProvider=data.active_provider||null;
     // Store default model so newSession() can apply it (#872).
@@ -2112,8 +2119,9 @@ async function populateModelDropdown(opts={}){
     }
     // Kick off a background live-model fetch for the active provider.
     // This runs after the static list is already shown (no blocking flicker).
-    if(data.active_provider) _fetchLiveModels(data.active_provider, sel);
+    if(data.active_provider) _fetchLiveModels(data.active_provider, sel, requestSeq);
   }catch(e){
+    if(requestSeq!==_modelDropdownRequestSeq) return;
     // API unavailable -- keep the hardcoded HTML options as fallback
     console.warn('Failed to load models from server:',e.message);
     if(typeof syncModelChip==='function') syncModelChip();
@@ -2203,10 +2211,12 @@ function _addLiveModelsToSelect(provider, models, sel){
   return added;
 }
 
-async function _fetchLiveModels(provider, sel){
+async function _fetchLiveModels(provider, sel, requestSeq=null){
   if(!provider||!sel) return;
+  if(requestSeq!==null&&requestSeq!==_modelDropdownRequestSeq) return;
   // Already fetched — apply cached models to this select element (#872)
   if(_liveModelCache[provider]){
+    if(requestSeq!==null&&requestSeq!==_modelDropdownRequestSeq) return;
     const added=_addLiveModelsToSelect(provider,_liveModelCache[provider],sel);
     if(added>0 && typeof syncModelChip==='function') syncModelChip();
     return;
@@ -2216,10 +2226,13 @@ async function _fetchLiveModels(provider, sel){
     const url=new URL('api/models/live',document.baseURI||location.href);
     url.searchParams.set('provider',provider);
     const _liveRes=await fetch(url.href,{credentials:'include'});
+    if(requestSeq!==null&&requestSeq!==_modelDropdownRequestSeq) return;
     if(_redirectIfUnauth(_liveRes)) return;
     const data=await _liveRes.json();
+    if(requestSeq!==null&&requestSeq!==_modelDropdownRequestSeq) return;
     if(!data.models||!data.models.length) return;
     _liveModelCache[provider]=data.models;
+    if(requestSeq!==null&&requestSeq!==_modelDropdownRequestSeq) return;
     const added=_addLiveModelsToSelect(provider,data.models,sel);
     if(added>0){
       if(typeof syncModelChip==='function') syncModelChip();

tests/test_issue4756_session_visit_model_refresh.py

@@ -0,0 +1,281 @@
+"""Regression coverage for #4756 session-visit model catalog freshness."""
+
+from __future__ import annotations
+
+import io
+import json
+import os
+import time
+from pathlib import Path
+from urllib.parse import urlparse
+
+REPO = Path(__file__).resolve().parent.parent
+
+
+def _catalog(label: str) -> dict:
+    return {
+        "active_provider": "openai",
+        "default_model": label,
+        "configured_model_badges": {},
+        "groups": [
+            {
+                "provider": "OpenAI",
+                "provider_id": "openai",
+                "models": [{"id": label, "label": label}],
+            }
+        ],
+        "aliases": {},
+    }
+
+
+def _reset_models_memory_cache(monkeypatch):
+    import api.config as cfg
+
+    monkeypatch.setattr(cfg, "_available_models_cache", None, raising=False)
+    monkeypatch.setattr(cfg, "_available_models_cache_ts", 0.0, raising=False)
+    monkeypatch.setattr(cfg, "_available_models_cache_source_fingerprint", None, raising=False)
+    monkeypatch.setattr(cfg, "_cache_build_in_progress", False, raising=False)
+
+
+def test_session_visit_fresh_profile_cache_returns_without_live_rebuild(tmp_path, monkeypatch):
+    import api.config as cfg
+
+    _reset_models_memory_cache(monkeypatch)
+    disk_catalog = _catalog("cached-model")
+    cache_path = tmp_path / "models_cache.profile.json"
+    cache_path.write_text("{}", encoding="utf-8")
+    now = time.time()
+    os.utime(cache_path, (now, now))
+
+    monkeypatch.setattr(cfg, "_SESSION_VISIT_MODELS_FRESHNESS_SECONDS", 300.0, raising=False)
+    monkeypatch.setattr(cfg, "_get_models_cache_path", lambda: cache_path)
+    monkeypatch.setattr(cfg, "_load_models_cache_from_disk", lambda: disk_catalog)
+    monkeypatch.setattr(cfg, "_load_stale_models_cache_from_disk", lambda: None)
+    monkeypatch.setattr(cfg, "_models_cache_source_fingerprint", lambda: {"profile": "demo"})
+
+    def _unexpected_live_rebuild(**_kwargs):
+        raise AssertionError("fresh session-visit cache must not run a live rebuild")
+
+    monkeypatch.setattr(cfg, "get_available_models", _unexpected_live_rebuild)
+
+    assert cfg.get_available_models_for_session_visit() == disk_catalog
+
+
+def test_session_visit_ignores_recently_warmed_memory_when_disk_cache_is_stale(tmp_path, monkeypatch):
+    import api.config as cfg
+
+    _reset_models_memory_cache(monkeypatch)
+    stale_catalog = _catalog("stale-model")
+    rebuilt_catalog = _catalog("rebuilt-model")
+    cache_path = tmp_path / "models_cache.profile.json"
+    cache_path.write_text("{}", encoding="utf-8")
+    old = time.time() - 600.0
+    os.utime(cache_path, (old, old))
+    calls = []
+
+    monkeypatch.setattr(cfg, "_SESSION_VISIT_MODELS_FRESHNESS_SECONDS", 300.0, raising=False)
+    monkeypatch.setattr(cfg, "_get_models_cache_path", lambda: cache_path)
+    monkeypatch.setattr(cfg, "_load_models_cache_from_disk", lambda: stale_catalog)
+    monkeypatch.setattr(cfg, "_load_stale_models_cache_from_disk", lambda: stale_catalog)
+    monkeypatch.setattr(cfg, "_models_cache_source_fingerprint", lambda: {"profile": "demo"})
+    monkeypatch.setattr(cfg, "_available_models_cache", stale_catalog, raising=False)
+    monkeypatch.setattr(cfg, "_available_models_cache_ts", time.monotonic(), raising=False)
+    monkeypatch.setattr(cfg, "_available_models_cache_source_fingerprint", {"profile": "demo"}, raising=False)
+
+    def _live_rebuild(**kwargs):
+        calls.append(kwargs)
+        return rebuilt_catalog
+
+    monkeypatch.setattr(cfg, "get_available_models", _live_rebuild)
+
+    assert cfg.get_available_models_for_session_visit() == rebuilt_catalog
+    assert calls == [{"force_refresh": True}]
+
+
+def test_session_visit_stale_profile_cache_revalidates_with_live_rebuild(tmp_path, monkeypatch):
+    import api.config as cfg
+
+    _reset_models_memory_cache(monkeypatch)
+    stale_catalog = _catalog("stale-model")
+    rebuilt_catalog = _catalog("rebuilt-model")
+    cache_path = tmp_path / "models_cache.profile.json"
+    cache_path.write_text("{}", encoding="utf-8")
+    old = time.time() - 600.0
+    os.utime(cache_path, (old, old))
+    calls = []
+
+    monkeypatch.setattr(cfg, "_SESSION_VISIT_MODELS_FRESHNESS_SECONDS", 300.0, raising=False)
+    monkeypatch.setattr(cfg, "_get_models_cache_path", lambda: cache_path)
+    monkeypatch.setattr(cfg, "_load_models_cache_from_disk", lambda: stale_catalog)
+    monkeypatch.setattr(cfg, "_load_stale_models_cache_from_disk", lambda: stale_catalog)
+
+    def _live_rebuild(**kwargs):
+        calls.append(kwargs)
+        return rebuilt_catalog
+
+    monkeypatch.setattr(cfg, "get_available_models", _live_rebuild)
+
+    assert cfg.get_available_models_for_session_visit() == rebuilt_catalog
+    assert calls == [{"force_refresh": True}]
+
+
+def test_default_disk_hit_does_not_restamp_stale_cache_for_session_visit(tmp_path, monkeypatch):
+    import api.config as cfg
+
+    _reset_models_memory_cache(monkeypatch)
+    stale_catalog = _catalog("stale-model")
+    rebuilt_catalog = _catalog("rebuilt-model")
+    cache_path = tmp_path / "models_cache.profile.json"
+    cache_path.write_text("{}", encoding="utf-8")
+    old = time.time() - 600.0
+    os.utime(cache_path, (old, old))
+    refresh_calls = []
+
+    monkeypatch.setattr(cfg, "_SESSION_VISIT_MODELS_FRESHNESS_SECONDS", 300.0, raising=False)
+    monkeypatch.setattr(cfg, "_get_models_cache_path", lambda: cache_path)
+    monkeypatch.setattr(cfg, "_load_models_cache_from_disk", lambda: stale_catalog)
+    monkeypatch.setattr(cfg, "_load_stale_models_cache_from_disk", lambda: stale_catalog)
+    monkeypatch.setattr(cfg, "_models_cache_source_fingerprint", lambda: {"profile": "demo"})
+    monkeypatch.setattr(cfg, "_cfg_mtime", 0.0, raising=False)
+    monkeypatch.setattr(cfg, "_save_models_cache_to_disk", lambda _cache: (_ for _ in ()).throw(
+        AssertionError("plain disk hits must not rewrite the models cache file")
+    ))
+
+    assert cfg.get_available_models() == stale_catalog
+
+    def _live_rebuild(**kwargs):
+        refresh_calls.append(kwargs)
+        return rebuilt_catalog
+
+    monkeypatch.setattr(cfg, "get_available_models", _live_rebuild)
+
+    assert cfg.get_available_models_for_session_visit() == rebuilt_catalog
+    assert refresh_calls == [{"force_refresh": True}]
+
+
+def test_session_visit_live_rebuild_failure_falls_back_to_cached_catalog(tmp_path, monkeypatch):
+    import api.config as cfg
+
+    _reset_models_memory_cache(monkeypatch)
+    stale_catalog = _catalog("fallback-model")
+    cache_path = tmp_path / "models_cache.profile.json"
+    cache_path.write_text("{}", encoding="utf-8")
+    old = time.time() - 600.0
+    os.utime(cache_path, (old, old))
+
+    monkeypatch.setattr(cfg, "_SESSION_VISIT_MODELS_FRESHNESS_SECONDS", 300.0, raising=False)
+    monkeypatch.setattr(cfg, "_get_models_cache_path", lambda: cache_path)
+    monkeypatch.setattr(cfg, "_load_models_cache_from_disk", lambda: None)
+    monkeypatch.setattr(cfg, "_load_stale_models_cache_from_disk", lambda: stale_catalog)
+
+    def _failing_live_rebuild(**kwargs):
+        assert kwargs == {"force_refresh": True}
+        raise RuntimeError("provider unavailable")
+
+    monkeypatch.setattr(cfg, "get_available_models", _failing_live_rebuild)
+
+    assert cfg.get_available_models_for_session_visit() == stale_catalog
+
+
+class _FakeHandler:
+    def __init__(self):
+        self.status = None
+        self.sent_headers = []
+        self.body = bytearray()
+        self.wfile = self
+        self.rfile = io.BytesIO(b"")
+        self.headers = {"Content-Length": "0"}
+        self.request = None
+
+    def send_response(self, status):
+        self.status = status
+
+    def send_header(self, name, value):
+        self.sent_headers.append((name, value))
+
+    def end_headers(self):
+        pass
+
+    def write(self, data):
+        self.body.extend(data)
+
+    def json_body(self):
+        return json.loads(bytes(self.body).decode("utf-8"))
+
+
+def test_models_route_session_visit_freshness_uses_bounded_helper(monkeypatch):
+    import api.routes as routes
+
+    expected = _catalog("route-model")
+    calls = []
+
+    def _session_visit_catalog():
+        calls.append("session_visit")
+        return expected
+
+    monkeypatch.setattr(routes, "get_available_models_for_session_visit", _session_visit_catalog)
+
+    handler = _FakeHandler()
+    parsed = urlparse("http://example.com/api/models?freshness=session_visit")
+    routes.handle_get(handler, parsed)
+
+    assert handler.status == 200
+    assert handler.json_body()["default_model"] == "route-model"
+    assert calls == ["session_visit"]
+
+
+def _read_static(name: str) -> str:
+    return (REPO / "static" / name).read_text(encoding="utf-8")
+
+
+def _extract_function_body(src: str, signature: str) -> str:
+    idx = src.find(signature)
+    if idx == -1:
+        raise AssertionError(f"signature {signature!r} not found")
+    header_end = src.find("){", idx)
+    if header_end == -1:
+        raise AssertionError(f"function body start for {signature!r} not found")
+    open_idx = header_end + 1
+    depth = 0
+    for i in range(open_idx, len(src)):
+        if src[i] == "{":
+            depth += 1
+        elif src[i] == "}":
+            depth -= 1
+            if depth == 0:
+                return src[idx : i + 1]
+    raise AssertionError(f"unbalanced braces in {signature!r}")
+
+
+def test_populate_model_dropdown_accepts_session_visit_freshness_and_guards_stale_responses():
+    body = _extract_function_body(_read_static("ui.js"), "async function populateModelDropdown(")
+    live_tail = _extract_function_body(_read_static("ui.js"), "async function _fetchLiveModels(")
+
+    assert "modelsUrl.searchParams.set('freshness',opts.freshness)" in body
+    assert "const requestSeq=++_modelDropdownRequestSeq" in body
+    assert body.count("requestSeq!==_modelDropdownRequestSeq") >= 3
+    assert "_fetchLiveModels(data.active_provider, sel, requestSeq)" in body
+    assert live_tail.count("requestSeq!==null&&requestSeq!==_modelDropdownRequestSeq") >= 4
+
+
+def test_load_session_schedules_async_session_visit_model_refresh_after_metadata_load():
+    body = _extract_function_body(_read_static("sessions.js"), "async function loadSession(")
+
+    assign_idx = body.index("S.session=data.session")
+    promise_idx = body.index("const modelRefreshPromise=Promise.resolve().then(")
+    ready_idx = body.index("window._modelDropdownReady=modelRefreshPromise")
+    refresh_idx = body.index("populateModelDropdown({freshness:'session_visit'})")
+    stale_guard_idx = body.index("_loadingSessionId!==modelRefreshSid")
+
+    assert assign_idx < promise_idx < refresh_idx < ready_idx
+    assert promise_idx < stale_guard_idx < refresh_idx
+
+
+def test_boot_model_dropdown_reuses_inflight_session_visit_refresh():
+    body = _extract_function_body(_read_static("boot.js"), "const _startBootModelDropdown=()=>")
+
+    ready_idx = body.index("const ready=window._modelDropdownReady;")
+    reuse_idx = body.index("if(ready&&typeof ready.then==='function') return ready;")
+    hydrate_idx = body.index("const next=_hydrateBootModelDropdown();")
+
+    assert ready_idx < reuse_idx < hydrate_idx