Fix cluster status command and join_cluster error formatting

Author: dcorbacho

deps/rabbit/src/rabbit_core_ff.erl

@@ -205,8 +205,12 @@ mds_phase1_migration_post_enable(#{feature_name := FeatureName}) ->
     mds_migration_post_enable(FeatureName, Tables).
 
 mds_migration_enable(FeatureName, TablesAndOwners) ->
-    ok = ensure_khepri_cluster_matches_mnesia(FeatureName),
-    mds_migrate_tables_to_khepri(FeatureName, TablesAndOwners).
+    case ensure_khepri_cluster_matches_mnesia(FeatureName) of
+        ok ->
+            mds_migrate_tables_to_khepri(FeatureName, TablesAndOwners);
+        Error ->
+            Error
+    end.
 
 mds_migration_post_enable(FeatureName, TablesAndOwners) ->
     ?assert(rabbit_khepri:is_enabled(non_blocking)),
@@ -221,7 +225,12 @@ ensure_khepri_cluster_matches_mnesia(FeatureName) ->
        "Feature flag `~s`:   updating the Khepri cluster to match "
        "the Mnesia cluster",
        [FeatureName]),
-    rabbit_khepri:init_cluster().
+    try
+        rabbit_khepri:init_cluster()
+    catch
+        error:{khepri_mnesia_migration_ex, _, _} = Reason ->
+            {error, Reason}
+    end.
 
 mds_plugin_migration_enable(FeatureName, TablesAndOwners) ->
     global:set_lock({FeatureName, self()}),

deps/rabbit/src/rabbit_khepri.erl

@@ -472,9 +472,9 @@ get_sys_status(Proc) ->
 cli_cluster_status() ->
     case rabbit:is_running() of
         true ->
-            Nodes = nodes(),
-            [{nodes, [{disc, [N || N <- Nodes, rabbit_nodes:is_running(N)]}]},
-             {running_nodes, Nodes},
+            Nodes = locally_known_nodes(),
+            [{nodes, [{disc, Nodes}]},
+             {running_nodes, [N || N <- Nodes, rabbit_nodes:is_running(N)]},
              {cluster_name, rabbit_nodes:cluster_name()}];
         false ->
             []
@@ -498,6 +498,9 @@ init_cluster() ->
         _ = application:ensure_all_started(khepri_mnesia_migration),
         rabbit_log:debug("Khepri clustering: syncing cluster membership"),
         mnesia_to_khepri:sync_cluster_membership(?STORE_ID)
+    catch
+        error:{khepri_mnesia_migration_ex, _, _} = Reason ->
+            {error, Reason}
     after
         case IsRunning of
             true -> ok;

deps/rabbitmq_cli/lib/rabbitmq/cli/ctl/commands/join_cluster_command.ex

@@ -78,6 +78,16 @@ defmodule RabbitMQ.CLI.Ctl.Commands.JoinClusterCommand do
      "Error: cannot cluster node with itself: #{node_name}"}
   end
 
+  def output(
+        {:error,
+         {:khepri_mnesia_migration_ex, :all_mnesia_nodes_must_run,
+          %{all_nodes: nodes, running_nodes: running}}},
+        _opts
+      ) do
+    {:error, RabbitMQ.CLI.Core.ExitCodes.exit_software(),
+     "Error: all mnesia nodes must run to join the cluster, mnesia nodes: #{inspect(nodes)}, running nodes: #{inspect(running)}"}
+  end
+
   use RabbitMQ.CLI.DefaultOutput
 
   def banner([target_node], %{node: node_name}) do