rabbitmqctl status is not adequate to wait for the server as it can return

successfully when the vm has started but not the app. The app can then fail.
Therefore introduce a new command to wait for the app to start.

Note that this subcommand contains a timeout to wait for the VM to
start, but will wait indefinitely for the app to start once the VM
has.

3 files changed, 48 insertions, 15 deletions

diff --git a/docs/rabbitmqctl.1.xml b/docs/rabbitmqctl.1.xml
index bd9fee7d..5c090e5a 100644
--- a/docs/rabbitmqctl.1.xml
+++ b/docs/rabbitmqctl.1.xml
@@ -158,6 +158,28 @@
         </varlistentry>
 
         <varlistentry>
+          <term><cmdsynopsis><command>wait</command></cmdsynopsis></term>
+          <listitem>
+            <para>
+              Wait for the RabbitMQ application to start.
+            </para>
+            <para>
+              This command will wait for the RabbitMQ application to
+              start at the node. As long as the Erlang node is up but
+              the RabbitMQ application is down it will wait
+              indefinitely. If the node itself goes down, or takes too
+              long to come up, it will fail.
+            </para>
+            <para role="example-prefix">For example:</para>
+            <screen role="example">rabbitmqctl wait</screen>
+            <para role="example">
+              This command will return when the RabbitMQ node has
+              started up.
+            </para>
+          </listitem>
+        </varlistentry>
+
+        <varlistentry>
           <term><cmdsynopsis><command>status</command></cmdsynopsis></term>
           <listitem>
             <para>
diff --git a/packaging/common/rabbitmq-server.init b/packaging/common/rabbitmq-server.init
index 54fd39b7..8ef1000b 100644
--- a/packaging/common/rabbitmq-server.init
+++ b/packaging/common/rabbitmq-server.init
@@ -22,7 +22,6 @@ DAEMON=/usr/sbin/${NAME}
 CONTROL=/usr/sbin/rabbitmqctl
 DESC=rabbitmq-server
 USER=rabbitmq
-TIMEOUT=10
 ROTATE_SUFFIX=
 INIT_LOG_DIR=/var/log/rabbitmq
 
@@ -40,7 +39,7 @@ start_rabbitmq () {
         set +e
         nohup $DAEMON > ${INIT_LOG_DIR}/startup_log \
             2> ${INIT_LOG_DIR}/startup_err &
-        wait_for_rabbitmq
+        $CONTROL wait >/dev/null 2>&1
         case "$?" in
             0)
                 echo SUCCESS
@@ -59,18 +58,6 @@ start_rabbitmq () {
     fi
 }
 
-wait_for_rabbitmq() {
-    WAITED=0
-    while [ $WAITED != $TIMEOUT ]; do
-        if status_rabbitmq quiet ; then
-            return 0
-        fi
-        sleep 1
-        WAITED=`expr $WAITED + 1`
-    done
-    return 1
-}
-
 stop_rabbitmq () {
     set +e
     status_rabbitmq quiet
diff --git a/src/rabbit_control.erl b/src/rabbit_control.erl
index 80483097..a7d07b0f 100644
--- a/src/rabbit_control.erl
+++ b/src/rabbit_control.erl
@@ -20,6 +20,7 @@
 -export([start/0, stop/0, action/5, diagnostics/1]).
 
 -define(RPC_TIMEOUT, infinity).
+-define(WAIT_FOR_VM_TIMEOUT, 5000).
 
 -define(QUIET_OPT, "-q").
 -define(NODE_OPT, "-n").
@@ -297,7 +298,30 @@ action(list_permissions, Node, [], Opts, Inform) ->
     VHost = proplists:get_value(?VHOST_OPT, Opts),
     Inform("Listing permissions in vhost ~p", [VHost]),
     display_list(call(Node, {rabbit_auth_backend_internal,
-                             list_vhost_permissions, [VHost]})).
+                             list_vhost_permissions, [VHost]}));
+
+action(wait, Node, [], _Opts, Inform) ->
+    Inform("Waiting for ~p", [Node]),
+    wait_for_application(Node, ?WAIT_FOR_VM_TIMEOUT).
+
+wait_for_application(_Node, NodeTimeout) when NodeTimeout =< 0 ->
+    {badrpc, nodedown};
+
+wait_for_application(Node, NodeTimeout) ->
+    case call(Node, {application, which_applications, []}) of
+        {badrpc, nodedown} -> wait_for_application0(Node, NodeTimeout - 1000);
+        {badrpc, _} = E    -> E;
+        Apps               -> case proplists:is_defined(rabbit, Apps) of
+                                  %% We've seen the node up; if it goes down
+                                  %% die immediately.
+                                  false -> wait_for_application0(Node, 0);
+                                  true  -> ok
+                              end
+    end.
+
+wait_for_application0(Node, NodeTimeout) ->
+    timer:sleep(1000),
+    wait_for_application(Node, NodeTimeout).
 
 default_if_empty(List, Default) when is_list(List) ->
     if List == [] ->