diff --git a/apps/emqx/src/emqx_router_helper.erl b/apps/emqx/src/emqx_router_helper.erl
index 78cf62d6c..8d96bf81d 100644
--- a/apps/emqx/src/emqx_router_helper.erl
+++ b/apps/emqx/src/emqx_router_helper.erl
@@ -146,13 +146,18 @@ handle_info({mnesia_table_event, Event}, State) ->
     ?SLOG(debug, #{msg => "unexpected_mnesia_table_event", event => Event}),
     {noreply, State};
 handle_info({nodedown, Node}, State = #{nodes := Nodes}) ->
-    global:trans(
-        {?LOCK, self()},
-        fun() ->
-            mria:transaction(?ROUTE_SHARD, fun ?MODULE:cleanup_routes/1, [Node])
-        end
-    ),
-    ok = mria:dirty_delete(?ROUTING_NODE, Node),
+    case mria_rlog:role() of
+        core ->
+            global:trans(
+                {?LOCK, self()},
+                fun() ->
+                    mria:transaction(?ROUTE_SHARD, fun ?MODULE:cleanup_routes/1, [Node])
+                end
+            ),
+            ok = mria:dirty_delete(?ROUTING_NODE, Node);
+        replicant ->
+            ok
+    end,
     ?tp(emqx_router_helper_cleanup_done, #{node => Node}),
     {noreply, State#{nodes := lists:delete(Node, Nodes)}, hibernate};
 handle_info({membership, {mnesia, down, Node}}, State) ->
diff --git a/apps/emqx/src/emqx_router_sup.erl b/apps/emqx/src/emqx_router_sup.erl
index d0e5ea05a..0fa48d9d2 100644
--- a/apps/emqx/src/emqx_router_sup.erl
+++ b/apps/emqx/src/emqx_router_sup.erl
@@ -41,4 +41,9 @@ init([]) ->
         hash,
         {emqx_router, start_link, []}
     ]),
-    {ok, {{one_for_all, 0, 1}, [Helper, RouterPool]}}.
+    SupFlags = #{
+        strategy => one_for_one,
+        intensity => 10,
+        period => 100
+    },
+    {ok, {SupFlags, [Helper, RouterPool]}}.
diff --git a/changes/ce/fix-11388.en.md b/changes/ce/fix-11388.en.md
new file mode 100644
index 000000000..835155585
--- /dev/null
+++ b/changes/ce/fix-11388.en.md
@@ -0,0 +1,6 @@
+Increase `emqx_router_sup` restart intensity.
+
+The goal is to tolerate occasional crashes that can happen under relatively normal conditions
+and don't seem critical to shutdown the whole app (emqx).
+For example, mria write/delete call delegated from a replicant to a core node by `emqx_router_helper` may fail,
+if the core node is being stopped / restarted / not ready.