Harden connection flapping aftercare

Keep sidebar retries boot-scoped, preserve immediate source-change cache rebuilds, and tighten session stream ownership before closing duplicate SSE sockets.

Co-Authored-By: Paperclip <noreply@paperclip.ing>

api/route_session_list_cache.py

@@ -175,6 +175,25 @@ def _session_list_cache_get(
         return None, False
 
 
+def _session_list_cache_stale_reason(key: tuple) -> str | None:
+    """Return why an existing cache entry is stale, if it is stale."""
+    now = time.monotonic()
+    current_stamp = _session_list_cache_resolved_source_stamp(key)
+    with _SESSIONS_CACHE_LOCK:
+        entry = _SESSIONS_CACHE.get(key)
+        if not entry:
+            return None
+        ts, stamp, _payload = entry
+        if stamp != current_stamp:
+            return "source"
+        ttl = _SESSIONS_CACHE_TTL_SECONDS
+        if _session_list_cache_streaming_freeze_marker() is not None:
+            ttl = _SESSIONS_CACHE_STREAMING_TTL_SECONDS
+        if (now - ts) >= ttl:
+            return "age"
+        return None
+
+
 def _session_list_cache_set(key: tuple, payload: dict) -> None:
     if not isinstance(payload, dict):
         return

api/routes.py

@@ -1668,6 +1668,7 @@ _session_list_runtime_sort_key = _route_session_list_cache._session_list_runtime
 _session_list_cache_set = _route_session_list_cache._session_list_cache_set
 _session_list_cache_source_stamp = _route_session_list_cache._session_list_cache_source_stamp
 _session_list_cache_state_db_fingerprint = _route_session_list_cache._session_list_cache_state_db_fingerprint
+_session_list_cache_stale_reason = _route_session_list_cache._session_list_cache_stale_reason
 _session_list_cache_streaming_freeze_marker = _route_session_list_cache._session_list_cache_streaming_freeze_marker
 
 _ROUTE_SESSION_LIST_CACHE_DYNAMIC_EXPORTS = {
@@ -2068,7 +2069,8 @@ def _get_cached_session_list_payload(
         return cached
 
     stale = cached  # now actually a stale payload when one exists, else None
-    if stale is not None:
+    stale_reason = _session_list_cache_stale_reason(key) if stale is not None else None
+    if stale is not None and stale_reason != "source":
         event, is_owner = _session_list_cache_claim_rebuild(key)
         if is_owner:
             if diag is not None:

static/messages.js

@@ -6456,7 +6456,7 @@ function _chatStreamActiveForSession(sid) {
 
 function _suspendSessionStreamForLiveChat(sid) {
   if (!sid) return;
-  if (_sessionStreamSessionId !== sid && !_sessionEventSource) return;
+  if (_sessionStreamSessionId !== sid) return;
   _sessionStreamHiddenSid = sid;
   stopSessionStream();
 }

static/sessions.js

@@ -4298,13 +4298,13 @@ async function _runRenderSessionListRefresh(opts, _gen){
   try{
     if(!($('sessionSearch').value||'').trim()) _contentSearchResults = [];
     const sessionListQS = _sessionListQueryString();
-    const sessionRequestOpts={
+    const sessionRequestOpts={timeoutToast:false};
-      timeoutToast:false,
+    if(!_sessionListHasLoadedOnce){
-      timeoutMs:_sessionListHasLoadedOnce?30000:_SESSION_LIST_BOOT_TIMEOUT_MS,
+      sessionRequestOpts.timeoutMs=_SESSION_LIST_BOOT_TIMEOUT_MS;
-      retries:1,
+      sessionRequestOpts.retries=1;
-      retryTimeouts:true,
+      sessionRequestOpts.retryTimeouts=true;
-      retryStatuses:[502,503,504],
+      sessionRequestOpts.retryStatuses=[502,503,504];
-    };
+    }
     const {sessData, projData}=await _loadSidebarSessionListPayload(sessionListQS, sessionRequestOpts);
     // Discard stale response — a newer renderSessionList() call superseded us.
     if (_gen !== _renderSessionListGen) return;

tests/test_issue4759_parallel_sidebar_boot_fetch.py

@@ -290,10 +290,6 @@ global.api = (url, opts) => {{
 (async () => {{
   const run = _loadSidebarSessionListPayload('?v=1', {{
     timeoutToast:false,
-    timeoutMs:90000,
-    retries:1,
-    retryTimeouts:true,
-    retryStatuses:[502,503,504],
   }});
 
   await Promise.resolve();

tests/test_performance_polling_hardening.py

@@ -24,7 +24,8 @@ def test_session_list_refreshes_are_coalesced_while_in_flight():
     assert "_renderSessionListQueuedRequest={" in src
     assert "opts:_mergeRenderSessionListOptions" in src
     assert "if (_gen !== _renderSessionListGen) return" in src
-    assert "api('/api/sessions' + sessionListQS,{timeoutToast:false})" in src
+    assert "const sessionRequestOpts={timeoutToast:false};" in src
+    assert "api('/api/sessions' + sessionListQS,sessionRequestOpts)" in src
     assert "api('/api/projects' + projectQS,{timeoutToast:false})" in src
 
 

tests/test_session_channel_option_x.py

@@ -452,6 +452,10 @@ def test_session_stream_pauses_while_chat_stream_is_active():
     assert "_sessionStreamHiddenSid = null;" in resume_src
     assert "startSessionStream(sid);" in resume_src
 
+    suspend_src = _js_function_decl(js, "_suspendSessionStreamForLiveChat")
+    assert "if (_sessionStreamSessionId !== sid) return;" in suspend_src
+    assert "stopSessionStream();" in suspend_src
+
 
 def test_frontend_busy_race_gate_obsoleted_by_option_z_pivot():
     """Per the Option Z PIVOT note baked into the handler body, the browser

tests/test_session_sidebar_cache.py

@@ -198,7 +198,7 @@ def test_session_list_cache_follower_wait_stage_when_rebuild_inflight(monkeypatc
     assert "session_list_cache_hit" in owner_diag.stages or "session_list_cache_stored" in owner_diag.stages
 
 
-def test_session_list_cache_follower_reuses_stale_payload_during_slow_rebuild(monkeypatch):
+def test_session_list_cache_source_changed_owner_rebuilds_while_follower_reuses_stale(monkeypatch):
     routes._session_list_cache_clear()
 
     key = routes._session_list_cache_key(
@@ -216,10 +216,10 @@ def test_session_list_cache_follower_reuses_stale_payload_during_slow_rebuild(mo
             stamp,
             payload,
         )
-    # Simulate the #4834 path: state.db/WAL/fingerprint changes after the
+    # Simulate state.db/WAL/fingerprint changing after the stale payload was
-    # stale payload was cached. Followers must still be able to use that stale
+    # cached. The owner must rebuild synchronously so committed external state is
-    # payload while the owner rebuild is blocked; otherwise sidebar polling can
+    # visible immediately, but followers can still use stale while that rebuild
-    # pile up behind a slow rebuild.
+    # is blocked; otherwise sidebar polling can pile up behind a slow rebuild.
     monkeypatch.setattr(
         routes,
         "_session_list_cache_source_stamp",
@@ -265,17 +265,15 @@ def test_session_list_cache_follower_reuses_stale_payload_during_slow_rebuild(mo
         owner_thread.join(2.0)
         follower_thread.join(2.0)
 
-    assert owner_result["payload"] == _session_cache_payload("stale")
+    assert owner_result["payload"] == _session_cache_payload("fresh")
     assert follower_result["payload"] == _session_cache_payload("stale")
-    assert "session_list_cache_stale_background_rebuild" in owner_diag.stages
+    assert "session_list_cache_rebuild_owner" in owner_diag.stages
-    assert (
+    assert "session_list_cache_wait_stale_fallback" in follower_diag.stages
-        "session_list_cache_stale_return" in follower_diag.stages
-        or "session_list_cache_stale_background_rebuild" in follower_diag.stages
-    )
 
 
 def test_session_list_cache_owner_returns_stale_and_rebuilds_in_background(monkeypatch):
     routes._session_list_cache_clear()
+    monkeypatch.setattr(routes, "_session_list_cache_source_stamp", lambda _key: ("stable",))
 
     key = routes._session_list_cache_key(
         active_profile="default",
@@ -292,8 +290,6 @@ def test_session_list_cache_owner_returns_stale_and_rebuilds_in_background(monke
             stamp,
             payload,
         )
-    monkeypatch.setattr(routes, "_session_list_cache_source_stamp", lambda _key: ("changed",))
-
     started = threading.Event()
     release = threading.Event()
     diag = _StageRecorder()
@@ -327,6 +323,7 @@ def test_session_list_cache_owner_returns_stale_and_rebuilds_in_background(monke
 
 def test_session_list_cache_stale_background_rebuild_failure_releases_owner(monkeypatch):
     routes._session_list_cache_clear()
+    monkeypatch.setattr(routes, "_session_list_cache_source_stamp", lambda _key: ("stable",))
 
     key = routes._session_list_cache_key(
         active_profile="default",
@@ -343,8 +340,6 @@ def test_session_list_cache_stale_background_rebuild_failure_releases_owner(monk
             stamp,
             payload,
         )
-    monkeypatch.setattr(routes, "_session_list_cache_source_stamp", lambda _key: ("changed",))
-
     started = threading.Event()
     logged = []
 

tests/test_session_sidebar_resilience.py

@@ -56,7 +56,7 @@ def test_sessions_and_projects_load_independently_so_projects_failure_cannot_bla
     assert "_applySessionListPayload(sessData,projData)" in block
 
 
-def test_sessions_api_uses_longer_timeout_and_timeout_retry_for_boot_refresh():
+def test_sessions_api_uses_longer_timeout_and_timeout_retry_for_boot_refresh_only():
     sessions_src = _sessions_js()
     workspace_src = _workspace_js()
     helper_start = sessions_src.find("async function _loadSidebarSessionListPayload")
@@ -65,11 +65,13 @@ def test_sessions_api_uses_longer_timeout_and_timeout_retry_for_boot_refresh():
     assert helper_end > helper_start
     helper = sessions_src[helper_start:helper_end]
 
-    assert "timeoutMs:_sessionListHasLoadedOnce?30000:_SESSION_LIST_BOOT_TIMEOUT_MS" in sessions_src
+    assert "const sessionRequestOpts={timeoutToast:false};" in sessions_src
+    assert "if(!_sessionListHasLoadedOnce){" in sessions_src
+    assert "sessionRequestOpts.timeoutMs=_SESSION_LIST_BOOT_TIMEOUT_MS;" in sessions_src
     assert "const sessData = await api('/api/sessions' + sessionListQS,sessionRequestOpts);" in helper
     assert "api('/api/sessions' + sessionListQS,{timeoutToast:false})" not in helper
-    assert "retryTimeouts:true" in sessions_src
+    assert "sessionRequestOpts.retryTimeouts=true" in sessions_src
-    assert "retryStatuses:[502,503,504]" in sessions_src
+    assert "sessionRequestOpts.retryStatuses=[502,503,504]" in sessions_src
     assert "retryTimeouts" in workspace_src
     assert "retryStatuses" in workspace_src