Merge branch 'rabbitmq-server-398' into stable

5 files changed, 35 insertions, 13 deletions

diff --git a/docs/rabbitmqctl.1.xml b/docs/rabbitmqctl.1.xml
index fe24a47376..1ecf8d4d3a 100644
--- a/docs/rabbitmqctl.1.xml
+++ b/docs/rabbitmqctl.1.xml
@@ -1829,6 +1829,21 @@
         </varlistentry>
 
         <varlistentry>
+          <term><cmdsynopsis><command>node_health_check</command></cmdsynopsis></term>
+          <listitem>
+            <para>
+              Health check of the RabbitMQ node. Verifies the rabbit application is
+              running, list_queues and list_channels return, and alarms are not set.
+            </para>
+            <para role="example-prefix">For example:</para>
+            <screen role="example">rabbitmqctl node_health_check -n rabbit@stringer</screen>
+            <para role="example">
+              This command performs a health check on the RabbitMQ node.
+            </para>
+          </listitem>
+        </varlistentry>
+
+        <varlistentry>
           <term><cmdsynopsis><command>environment</command></cmdsynopsis></term>
           <listitem>
             <para>
diff --git a/include/rabbit_cli.hrl b/include/rabbit_cli.hrl
index 2e687e2eb8..7f5db6053b 100644
--- a/include/rabbit_cli.hrl
+++ b/include/rabbit_cli.hrl
@@ -47,8 +47,6 @@
 -define(OFFLINE_DEF, {?OFFLINE_OPT, flag}).
 -define(ONLINE_DEF, {?ONLINE_OPT, flag}).
 
--define(RPC_TIMEOUT, infinity).
-
 %% Subset of standartized exit codes from sysexits.h, see
 %% https://github.com/rabbitmq/rabbitmq-server/issues/396 for discussion.
 -define(EX_OK         ,  0).
diff --git a/src/rabbit_cli.erl b/src/rabbit_cli.erl
index 4aad3c0916..6679d9329e 100644
--- a/src/rabbit_cli.erl
+++ b/src/rabbit_cli.erl
@@ -255,14 +255,10 @@ print_badrpc_diagnostics(Nodes) ->
 %% a timeout unless we set our ticktime to be the same. So let's do
 %% that.
 rpc_call(Node, Mod, Fun, Args) ->
-    rpc_call(Node, Mod, Fun, Args, ?RPC_TIMEOUT).
+    rabbit_misc:rpc_call(Node, Mod, Fun, Args).
 
 rpc_call(Node, Mod, Fun, Args, Timeout) ->
-    case rpc:call(Node, net_kernel, get_net_ticktime, [], Timeout) of
-        {badrpc, _} = E -> E;
-        Time            -> net_kernel:set_net_ticktime(Time, 0),
-                           rpc:call(Node, Mod, Fun, Args, Timeout)
-    end.
+    rabbit_misc:rpc_call(Node, Mod, Fun, Args, Timeout).
 
 rpc_call(Node, Mod, Fun, Args, Ref, Pid, Timeout) ->
-    rpc_call(Node, Mod, Fun, Args++[Ref, Pid], Timeout).
+    rabbit_misc:rpc_call(Node, Mod, Fun, Args, Ref, Pid, Timeout).
diff --git a/src/rabbit_control_main.erl b/src/rabbit_control_main.erl
index 4cb7eb094b..b805d21e48 100644
--- a/src/rabbit_control_main.erl
+++ b/src/rabbit_control_main.erl
@@ -17,12 +17,13 @@
 -module(rabbit_control_main).
 -include("rabbit.hrl").
 -include("rabbit_cli.hrl").
+-include("rabbit_misc.hrl").
 
 -export([start/0, stop/0, parse_arguments/2, action/5, action/6,
          sync_queue/1, cancel_sync_queue/1, become/1,
          purge_queue/1]).
 
--import(rabbit_cli, [rpc_call/4, rpc_call/5, rpc_call/7]).
+-import(rabbit_misc, [rpc_call/4, rpc_call/5, rpc_call/7]).
 
 -define(EXTERNAL_CHECK_INTERVAL, 1000).
 
@@ -83,6 +84,7 @@
          report,
          set_cluster_name,
          eval,
+         node_health_check,
 
          close_connection,
          {trace_on, [?VHOST_DEF]},
@@ -111,7 +113,7 @@
         [stop, stop_app, start_app, wait, reset, force_reset, rotate_logs,
          join_cluster, change_cluster_node_type, update_cluster_nodes,
          forget_cluster_node, rename_cluster_node, cluster_status, status,
-         environment, eval, force_boot, help]).
+         environment, eval, force_boot, help, node_health_check]).
 
 -define(COMMANDS_WITH_TIMEOUT,
         [list_user_permissions, list_policies, list_queues, list_exchanges,
@@ -547,6 +549,17 @@ action(eval, Node, [Expr], _Opts, _Inform) ->
 action(help, _Node, _Args, _Opts, _Inform) ->
     io:format("~s", [rabbit_ctl_usage:usage()]);
 
+action(node_health_check, Node, _Args, _Opts, Inform) ->
+    Inform("Checking health of node ~p", [Node]),
+    try
+        rabbit_health_check:node(Node),
+        io:format("Health check passed~n")
+    catch
+        {node_is_ko, ErrorMsg, ErrorCode} ->
+            io:format("Heath check failed:~n~s~n", [ErrorMsg]),
+            halt(ErrorCode)
+    end;
+
 action(Command, Node, Args, Opts, Inform) ->
     %% For backward compatibility, run commands accepting a timeout with
     %% the default timeout.
diff --git a/src/rabbit_plugins_main.erl b/src/rabbit_plugins_main.erl
index 4aeed4826c..e248989a7a 100644
--- a/src/rabbit_plugins_main.erl
+++ b/src/rabbit_plugins_main.erl
@@ -173,7 +173,7 @@ format_plugins(Node, Pattern, Opts, #cli{all      = All,
 
     EnabledImplicitly = Implicit -- Enabled,
     {StatusMsg, Running} =
-        case rabbit_cli:rpc_call(Node, rabbit_plugins, active, []) of
+        case rabbit_misc:rpc_call(Node, rabbit_plugins, active, []) of
             {badrpc, _} -> {"[failed to contact ~s - status not shown]", []};
             Active      -> {"* = running on ~s", Active}
         end,
@@ -279,7 +279,7 @@ sync(Node, ForceOnline, #cli{file = File}) ->
 
 rpc_call(Node, Online, Mod, Fun, Args) ->
     io:format("~nApplying plugin configuration to ~s...", [Node]),
-    case rabbit_cli:rpc_call(Node, Mod, Fun, Args) of
+    case rabbit_misc:rpc_call(Node, Mod, Fun, Args) of
         {ok, [], []} ->
             io:format(" nothing to do.~n", []);
         {ok, Start, []} ->