rpc: remove connections from a container during server shutdown

Current logic that prevent it is not needed since parallel_for_each runs
on an entire container before relinquish control during server shutdown,
so there is no problem invalidating iterators.

rpc/rpc.hh

@@ -303,7 +303,6 @@ public:
         resource_limits _limits;
         semaphore _resources_available;
         std::unordered_set<lw_shared_ptr<connection>> _conns;
-        bool _stopping = false;
         promise<> _ss_stopped;
         seastar::gate _reply_gate;
         server_options _options;
@@ -314,7 +313,6 @@ public:
         server(protocol& proto, server_options opts, server_socket, resource_limits memory_limit = resource_limits());
         void accept();
         future<> stop() {
-            _stopping = true; // prevents closed connections to be deleted from _conns
             _ss.abort_accept();
             _ss = server_socket();
             _resources_available.broken();

rpc/rpc_impl.hh

@@ -900,11 +900,7 @@ future<> protocol<Serializer, MsgType>::server::connection::process() {
         this->_error = true;
         return this->stop_send_loop().then_wrapped([this] (future<> f) {
             f.ignore_ready_future();
-            if (!this->_server._stopping) {
-                // if server is stopping do not remove connection
-                // since it may invalidate _conns iterators
-                this->_server._conns.erase(this->shared_from_this());
-            }
+            this->_server._conns.erase(this->shared_from_this());
             this->_stopped.set_value();
         });
     }).finally([conn_ptr = this->shared_from_this()] {