Fix await_startup/1. Change its design from "figure out what apps we should be running and wait for them" to "wait for the boot marker to go away".bug26295

2 files changed, 21 insertions, 27 deletions

diff --git a/src/app_utils.erl b/src/app_utils.erl
index e321888d..87e6fa0b 100644
--- a/src/app_utils.erl
+++ b/src/app_utils.erl
@@ -17,7 +17,7 @@
 
 -export([load_applications/1, start_applications/1, start_applications/2,
          stop_applications/1, stop_applications/2, app_dependency_order/2,
-         wait_for_applications/1, app_dependencies/1]).
+         app_dependencies/1]).
 
 -ifdef(use_specs).
 
@@ -28,7 +28,6 @@
 -spec stop_applications([atom()])                   -> 'ok'.
 -spec start_applications([atom()], error_handler()) -> 'ok'.
 -spec stop_applications([atom()], error_handler())  -> 'ok'.
--spec wait_for_applications([atom()])               -> 'ok'.
 -spec app_dependency_order([atom()], boolean())     -> [digraph:vertex()].
 -spec app_dependencies(atom())                      -> [atom()].
 
@@ -69,10 +68,6 @@ stop_applications(Apps, ErrorHandler) ->
                         ErrorHandler,
                         Apps).
 
-
-wait_for_applications(Apps) ->
-    [wait_for_application(App) || App <- Apps], ok.
-
 app_dependency_order(RootApps, StripUnreachable) ->
     {ok, G} = rabbit_misc:build_acyclic_graph(
                 fun ({App, _Deps}) -> [{App, App}] end,
@@ -93,13 +88,6 @@ app_dependency_order(RootApps, StripUnreachable) ->
 %%---------------------------------------------------------------------------
 %% Private API
 
-wait_for_application(Application) ->
-    case lists:keymember(Application, 1, rabbit_misc:which_applications()) of
-         true  -> ok;
-         false -> timer:sleep(1000),
-                  wait_for_application(Application)
-    end.
-
 load_applications(Worklist, Loaded) ->
     case queue:out(Worklist) of
         {empty, _WorkList} ->
diff --git a/src/rabbit.erl b/src/rabbit.erl
index a4c460ae..191f04a4 100644
--- a/src/rabbit.erl
+++ b/src/rabbit.erl
@@ -365,10 +365,11 @@ start_it(StartFun) ->
 stop() ->
     case whereis(rabbit_boot) of
         undefined -> ok;
-        _         -> await_startup()
+        _         -> await_startup(true)
     end,
     rabbit_log:info("Stopping RabbitMQ~n"),
-    stop_apps(app_shutdown_order()).
+    Apps = ?APPS ++ rabbit_plugins:active(),
+    stop_apps(app_utils:app_dependency_order(Apps, true)).
 
 stop_and_halt() ->
     try
@@ -410,7 +411,20 @@ run_cleanup_steps(Apps) ->
     ok.
 
 await_startup() ->
-    app_utils:wait_for_applications(app_startup_order()).
+    await_startup(false).
+
+await_startup(HaveSeenRabbitBoot) ->
+    %% We don't take absence of rabbit_boot as evidence we've started,
+    %% since there's a small window before it is registered.
+    case whereis(rabbit_boot) of
+        undefined -> case HaveSeenRabbitBoot orelse is_running() of
+                         true  -> ok;
+                         false -> timer:sleep(100),
+                                  await_startup(false)
+                     end;
+        _         -> timer:sleep(100),
+                     await_startup(true)
+    end.
 
 status() ->
     S1 = [{pid,                  list_to_integer(os:getpid())},
@@ -461,6 +475,9 @@ listeners() ->
                   ip_address = IP,
                   port       = Port} <- Listeners, Node =:= node()].
 
+%% TODO this only determines if the rabbit application has started,
+%% not if it is running, never mind plugins. It would be nice to have
+%% more nuance here.
 is_running() -> is_running(node()).
 
 is_running(Node) -> rabbit_nodes:is_process_running(Node, rabbit).
@@ -508,17 +525,6 @@ stop(_State) ->
     ok.
 
 %%---------------------------------------------------------------------------
-%% application life cycle
-
-app_startup_order() ->
-    ok = app_utils:load_applications(?APPS),
-    app_utils:app_dependency_order(?APPS, false).
-
-app_shutdown_order() ->
-    Apps = ?APPS ++ rabbit_plugins:active(),
-    app_utils:app_dependency_order(Apps, true).
-
-%%---------------------------------------------------------------------------
 %% boot step logic
 
 run_boot_steps() ->