*Do* start epmd ourselves if we want to impersonate a node. And just allow net_kernel:start() to fail and catch the error later.

3 files changed, 41 insertions, 16 deletions

diff --git a/src/rabbit_cli.erl b/src/rabbit_cli.erl
index a946c12a16..58f40a3ce9 100644
--- a/src/rabbit_cli.erl
+++ b/src/rabbit_cli.erl
@@ -17,7 +17,8 @@
 -module(rabbit_cli).
 -include("rabbit_cli.hrl").
 
--export([main/3, name_type/0, parse_arguments/4]).
+-export([main/3, start_distribution/0, start_distribution/1,
+         parse_arguments/4]).
 
 %%----------------------------------------------------------------------------
 
@@ -31,7 +32,8 @@
 -spec(main/3 :: (fun (([string()], string()) -> parse_result()),
                      fun ((atom(), atom(), [any()], [any()]) -> any()),
                          atom()) -> no_return()).
--spec(name_type/0 :: () -> 'shortnames' | 'longnames').
+-spec(start_distribution/0 :: () -> {'ok', pid()} | {'error', any()}).
+-spec(start_distribution/1 :: (string()) -> {'ok', pid()} | {'error', any()}).
 -spec(usage/1 :: (atom()) -> no_return()).
 -spec(parse_arguments/4 ::
         ([{atom(), [{string(), optdef()}]} | atom()],
@@ -42,6 +44,7 @@
 %%----------------------------------------------------------------------------
 
 main(ParseFun, DoFun, UsageMod) ->
+    error_logger:tty(false),
     {ok, [[NodeStr|_]|_]} = init:get_argument(nodename),
     {Command, Opts, Args} =
         case ParseFun(init:get_plain_arguments(), NodeStr) of
@@ -77,12 +80,6 @@ main(ParseFun, DoFun, UsageMod) ->
         {'EXIT', {badarg, _}} ->
             print_error("invalid parameter: ~p", [Args]),
             usage(UsageMod);
-        {error, {distribution_failed, _}} ->
-            print_error(
-              "Could not start distribution.~n"
-              "The most likely reason is that epmd has not been started by~n"
-              "the server. Make sure the server is running.~n", []),
-            rabbit_misc:quit(2);
         {error, {Problem, Reason}} when is_atom(Problem), is_binary(Reason) ->
             %% We handle this common case specially to avoid ~p since
             %% that has i18n issues
@@ -111,11 +108,11 @@ main(ParseFun, DoFun, UsageMod) ->
     end.
 
 start_distribution() ->
-    CtlNodeName = rabbit_misc:format("rabbitmq-cli-~s", [os:getpid()]),
-    case net_kernel:start([list_to_atom(CtlNodeName), name_type()]) of
-        {ok, _}         -> ok;
-        {error, Reason} -> throw({error, {distribution_failed, Reason}})
-    end.
+    start_distribution(list_to_atom(
+                         rabbit_misc:format("rabbitmq-cli-~s", [os:getpid()]))).
+
+start_distribution(Name) ->
+    net_kernel:start([Name, name_type()]).
 
 name_type() ->
     case os:getenv("RABBITMQ_USE_LONGNAME") of
diff --git a/src/rabbit_control_main.erl b/src/rabbit_control_main.erl
index e5a28b559d..5410185608 100644
--- a/src/rabbit_control_main.erl
+++ b/src/rabbit_control_main.erl
@@ -584,12 +584,18 @@ exit_loop(Port) ->
     end.
 
 become(BecomeNode) ->
+    %% We might not have started distribution if epmd was not running,
+    %% but we need it now even if we have to start epmd ourselves.
+    case erlang:is_alive() of
+        true  -> ok;
+        false -> rabbit_nodes:ensure_epmd(),
+                 rabbit_cli:start_distribution()
+    end,
     case net_adm:ping(BecomeNode) of
         pong -> exit({node_running, BecomeNode});
         pang -> io:format("  * Impersonating node: ~s...", [BecomeNode]),
-                error_logger:tty(false),
                 ok = net_kernel:stop(),
-                {ok, _} = net_kernel:start([BecomeNode,rabbit_cli:name_type()]),
+                {ok, _} = rabbit_cli:start_distribution(BecomeNode),
                 io:format(" done~n", []),
                 Dir = mnesia:system_info(directory),
                 io:format("  * Mnesia directory  : ~s~n", [Dir])
diff --git a/src/rabbit_nodes.erl b/src/rabbit_nodes.erl
index 7f7fcc3126..66d05428c6 100644
--- a/src/rabbit_nodes.erl
+++ b/src/rabbit_nodes.erl
@@ -18,7 +18,7 @@
 
 -export([names/1, diagnostics/1, make/1, parts/1, cookie_hash/0,
          is_running/2, is_process_running/2,
-         cluster_name/0, set_cluster_name/1]).
+         cluster_name/0, set_cluster_name/1, ensure_epmd/0]).
 
 -include_lib("kernel/include/inet.hrl").
 
@@ -41,6 +41,7 @@
 -spec(is_process_running/2 :: (node(), atom()) -> boolean()).
 -spec(cluster_name/0 :: () -> binary()).
 -spec(set_cluster_name/1 :: (binary()) -> 'ok').
+-spec(ensure_epmd/0 :: () -> 'ok').
 
 -endif.
 
@@ -77,6 +78,10 @@ diagnostics_node(Node) ->
     {Name, Host} = parts(Node),
     [{"~s:", [Node]} |
      case names(Host) of
+         {error, address} ->
+             [{"  * connection refused from epmd (port ~s) on ~s~n"
+               "    - has the server started it?~n",
+               [epmd_port(), Host]}];
          {error, Reason} ->
              [{"  * unable to connect to epmd (port ~s) on ~s: ~s~n",
                [epmd_port(), Host, rabbit_misc:format_inet_error(Reason)]}];
@@ -197,3 +202,20 @@ cluster_name_default() ->
 
 set_cluster_name(Name) ->
     rabbit_runtime_parameters:set_global(cluster_name, Name).
+
+ensure_epmd() ->
+    {ok, Root} = init:get_argument(root),
+    {ok, Prog} = init:get_argument(progname),
+    ID = random:uniform(1000000000),
+    Port = open_port(
+             {spawn_executable, filename:join([Root, "bin", Prog])},
+             [{args, ["-sname", rabbit_misc:format("epmd-starter-~b", [ID]),
+                      "-noshell", "-eval", "halt()."]},
+              exit_status, stderr_to_stdout, use_stdio]),
+    port_shutdown_loop(Port).
+
+port_shutdown_loop(Port) ->
+    receive
+        {Port, {exit_status, _Rc}} -> ok;
+        {Port, _}                  -> port_shutdown_loop(Port)
+    end.