test(conf): fix flaky config sync testcases

Before the recent changes these testcases relied on the fact that
peer nodes that are part of the test cluster are started one after
the other. This is now not the case in general with `emqx_cth_cluster`,
so now we need to additionally find the "longest running node" in the
cluster before running test logic.

apps/emqx_conf/test/emqx_conf_app_SUITE.erl

@@ -64,7 +64,8 @@ t_copy_new_data_dir(Config) ->
     ),
 
     %% 1. Start all nodes
-    [First | Rest] = Nodes = start_cluster(Cluster),
+    Nodes = start_cluster(Cluster),
+    [First | Rest] = sort_highest_uptime(Nodes),
     try
         NodeDataDir = erpc:call(First, emqx, data_dir, []),
         File = NodeDataDir ++ "/configs/cluster.hocon",
@@ -88,7 +89,8 @@ t_copy_deprecated_data_dir(Config) ->
     ),
 
     %% 1. Start all nodes
-    [First | Rest] = Nodes = start_cluster(Cluster),
+    Nodes = start_cluster(Cluster),
+    [First | Rest] = sort_highest_uptime(Nodes),
     try
         NodeDataDir = erpc:call(First, emqx, data_dir, []),
         File = NodeDataDir ++ "/configs/cluster-override.conf",
@@ -246,3 +248,11 @@ cluster(TC, Specs, Config) ->
 
 cluster_spec({Type, Num}) ->
     {Type, list_to_atom(atom_to_list(?MODULE) ++ integer_to_list(Num))}.
+
+sort_highest_uptime(Nodes) ->
+    Ranking = lists:sort([{-get_node_uptime(N), N} || N <- Nodes]),
+    element(2, lists:unzip(Ranking)).
+
+get_node_uptime(Node) ->
+    {Milliseconds, _} = erpc:call(Node, erlang, statistics, [wall_clock]),
+    Milliseconds.