Merge with default

7 files changed, 94 insertions, 23 deletions

diff --git a/packaging/RPMS/Fedora/rabbitmq-server.spec b/packaging/RPMS/Fedora/rabbitmq-server.spec
index 11f5f01c..bdd6c4a1 100644
--- a/packaging/RPMS/Fedora/rabbitmq-server.spec
+++ b/packaging/RPMS/Fedora/rabbitmq-server.spec
@@ -50,6 +50,7 @@ make install TARGET_DIR=%{_maindir} \
 
 mkdir -p %{buildroot}%{_localstatedir}/lib/rabbitmq/mnesia
 mkdir -p %{buildroot}%{_localstatedir}/log/rabbitmq
+mkdir -p %{buildroot}%{_localstatedir}/run/rabbitmq
 
 #Copy all necessary lib files etc.
 install -p -D -m 0755 %{S:1} %{buildroot}%{_initrddir}/rabbitmq-server
@@ -111,6 +112,7 @@ done
 %defattr(-,root,root,-)
 %attr(0750, rabbitmq, rabbitmq) %dir %{_localstatedir}/lib/rabbitmq
 %attr(0750, rabbitmq, rabbitmq) %dir %{_localstatedir}/log/rabbitmq
+%attr(0750, rabbitmq, rabbitmq) %dir %{_localstatedir}/run/rabbitmq
 %dir %{_sysconfdir}/rabbitmq
 %{_initrddir}/rabbitmq-server
 %config(noreplace) %{_sysconfdir}/logrotate.d/rabbitmq-server
diff --git a/packaging/common/rabbitmq-server.init b/packaging/common/rabbitmq-server.init
index d8a7a94d..e2815f04 100644
--- a/packaging/common/rabbitmq-server.init
+++ b/packaging/common/rabbitmq-server.init
@@ -24,6 +24,7 @@ DESC=rabbitmq-server
 USER=rabbitmq
 ROTATE_SUFFIX=
 INIT_LOG_DIR=/var/log/rabbitmq
+PID_FILE=/var/run/rabbitmq/pid
 
 LOCK_FILE= # This is filled in when building packages
 
@@ -40,9 +41,9 @@ start_rabbitmq () {
     else
         RETVAL=0
         set +e
-        setsid sh -c "$DAEMON > ${INIT_LOG_DIR}/startup_log \
-            2> ${INIT_LOG_DIR}/startup_err" &
-        $CONTROL wait >/dev/null 2>&1
+        setsid sh -c "RABBITMQ_PID_FILE=$PID_FILE $DAEMON > \
+            ${INIT_LOG_DIR}/startup_log 2> ${INIT_LOG_DIR}/startup_err" &
+        $CONTROL wait $PID_FILE >/dev/null 2>&1
         RETVAL=$?
         set -e
         case "$RETVAL" in
@@ -53,6 +54,7 @@ start_rabbitmq () {
                 fi
                 ;;
             *)
+                rm -f $PID_FILE
                 echo FAILED - check ${INIT_LOG_DIR}/startup_\{log, _err\}
                 RETVAL=1
                 ;;
@@ -68,6 +70,7 @@ stop_rabbitmq () {
         RETVAL=$?
         set -e
         if [ $RETVAL = 0 ] ; then
+            rm -f $PID_FILE
             if [ -n "$LOCK_FILE" ] ; then
                 rm -f $LOCK_FILE
             fi
diff --git a/packaging/common/rabbitmq-server.ocf b/packaging/common/rabbitmq-server.ocf
index d58c48ed..51e16517 100755
--- a/packaging/common/rabbitmq-server.ocf
+++ b/packaging/common/rabbitmq-server.ocf
@@ -29,6 +29,7 @@
 ##   OCF_RESKEY_log_base
 ##   OCF_RESKEY_mnesia_base
 ##   OCF_RESKEY_server_start_args
+##   OCF_RESKEY_pid_file
 
 #######################################################################
 # Initialization:
@@ -42,10 +43,12 @@ OCF_RESKEY_server_default="/usr/sbin/rabbitmq-server"
 OCF_RESKEY_ctl_default="/usr/sbin/rabbitmqctl"
 OCF_RESKEY_nodename_default="rabbit@localhost"
 OCF_RESKEY_log_base_default="/var/log/rabbitmq"
+OCF_RESKEY_pid_file_default="/var/lib/rabbitmq/pid"
 : ${OCF_RESKEY_server=${OCF_RESKEY_server_default}}
 : ${OCF_RESKEY_ctl=${OCF_RESKEY_ctl_default}}
 : ${OCF_RESKEY_nodename=${OCF_RESKEY_nodename_default}}
 : ${OCF_RESKEY_log_base=${OCF_RESKEY_log_base_default}}
+: ${OCF_RESKEY_pid_file=${OCF_RESKEY_pid_file_default}}
 
 meta_data() {
     cat <<END
@@ -133,6 +136,14 @@ Additional arguments provided to the server on startup
 <content type="string" default="" />
 </parameter>
 
+<parameter name="pid_file" unique="0" required="0">
+<longdesc lang="en">
+Location of the file in which the pid will be stored
+</longdesc>
+<shortdesc lang="en">Pid file path</shortdesc>
+<content type="string" default="${OCF_RESKEY_pid_file_default}" />
+</parameter>
+
 </parameters>
 
 <actions>
@@ -164,6 +175,7 @@ RABBITMQ_CONFIG_FILE=$OCF_RESKEY_config_file
 RABBITMQ_LOG_BASE=$OCF_RESKEY_log_base
 RABBITMQ_MNESIA_BASE=$OCF_RESKEY_mnesia_base
 RABBITMQ_SERVER_START_ARGS=$OCF_RESKEY_server_start_args
+RABBITMQ_PID_FILE=$OCF_RESKEY_pid_file
 [ ! -z $RABBITMQ_NODENAME ] && NODENAME_ARG="-n $RABBITMQ_NODENAME"
 [ ! -z $RABBITMQ_NODENAME ]            && export RABBITMQ_NODENAME
 
@@ -174,6 +186,7 @@ export_vars() {
     [ ! -z $RABBITMQ_LOG_BASE ]            && export RABBITMQ_LOG_BASE
     [ ! -z $RABBITMQ_MNESIA_BASE ]         && export RABBITMQ_MNESIA_BASE
     [ ! -z $RABBITMQ_SERVER_START_ARGS ]   && export RABBITMQ_SERVER_START_ARGS
+    [ ! -z $RABBITMQ_PID_FILE ]            && export RABBITMQ_PID_FILE
 }
 
 rabbit_validate_partial() {
@@ -214,13 +227,13 @@ rabbit_status() {
 }
 
 rabbit_wait() {
-    rabbitmqctl_action "wait"
+    rabbitmqctl_action "wait" $1
 }
 
 rabbitmqctl_action() {
     local rc
     local action
-    action=$1
+    action=$@
     $RABBITMQ_CTL $NODENAME_ARG $action > /dev/null 2> /dev/null
     rc=$?
     case "$rc" in
@@ -252,9 +265,10 @@ rabbit_start() {
 
     # Wait for the server to come up.
     # Let the CRM/LRM time us out if required
-    rabbit_wait
+    rabbit_wait $RABBITMQ_PID_FILE
     rc=$?
     if [ "$rc" != $OCF_SUCCESS ]; then
+        rm -f $RABBITMQ_PID_FILE
         ocf_log info "rabbitmq-server start failed: $rc"
         exit $OCF_ERR_GENERIC
     fi
@@ -285,6 +299,7 @@ rabbit_stop() {
         rabbit_status
         rc=$?
         if [ "$rc" = $OCF_NOT_RUNNING ]; then
+            rm -f $RABBITMQ_PID_FILE
             stop_wait=0
             break
         elif [ "$rc" != $OCF_SUCCESS ]; then
diff --git a/packaging/debs/Debian/debian/dirs b/packaging/debs/Debian/debian/dirs
index 625b7d41..5cf167d5 100644
--- a/packaging/debs/Debian/debian/dirs
+++ b/packaging/debs/Debian/debian/dirs
@@ -4,6 +4,7 @@ usr/sbin
 usr/share/man
 var/lib/rabbitmq/mnesia
 var/log/rabbitmq
+var/run/rabbitmq
 etc/logrotate.d
 etc/rabbitmq
 
diff --git a/packaging/debs/Debian/debian/postinst b/packaging/debs/Debian/debian/postinst
index b11340ef..ca531f14 100644
--- a/packaging/debs/Debian/debian/postinst
+++ b/packaging/debs/Debian/debian/postinst
@@ -32,6 +32,7 @@ fi
 
 chown -R rabbitmq:rabbitmq /var/lib/rabbitmq
 chown -R rabbitmq:rabbitmq /var/log/rabbitmq
+chown -R rabbitmq:rabbitmq /var/run/rabbitmq
 
 case "$1" in
     configure)
diff --git a/scripts/rabbitmq-server b/scripts/rabbitmq-server
index 2f80eb96..35b41d59 100755
--- a/scripts/rabbitmq-server
+++ b/scripts/rabbitmq-server
@@ -47,6 +47,7 @@ fi
 
 [ "x" = "x$RABBITMQ_MNESIA_DIR" ] && RABBITMQ_MNESIA_DIR=${MNESIA_DIR}
 [ "x" = "x$RABBITMQ_MNESIA_DIR" ] && RABBITMQ_MNESIA_DIR=${RABBITMQ_MNESIA_BASE}/${RABBITMQ_NODENAME}
+[ "x" = "x$RABBITMQ_PID_FILE" ] && RABBITMQ_PID_FILE=${RABBITMQ_MNESIA_DIR}/pid
 
 [ "x" = "x$RABBITMQ_PLUGINS_EXPAND_DIR" ] && RABBITMQ_PLUGINS_EXPAND_DIR=${PLUGINS_EXPAND_DIR}
 [ "x" = "x$RABBITMQ_PLUGINS_EXPAND_DIR" ] && RABBITMQ_PLUGINS_EXPAND_DIR=${RABBITMQ_MNESIA_BASE}/${RABBITMQ_NODENAME}-plugins-expand
@@ -67,6 +68,9 @@ fi
 RABBITMQ_START_RABBIT=
 [ "x" = "x$RABBITMQ_ALLOW_INPUT" ] && RABBITMQ_START_RABBIT='-noinput'
 
+mkdir -p $(dirname ${RABBITMQ_PID_FILE})
+echo $$ > ${RABBITMQ_PID_FILE}
+
 RABBITMQ_EBIN_ROOT="${RABBITMQ_HOME}/ebin"
 if [ "x" = "x$RABBITMQ_NODE_ONLY" ]; then
     if erl \
diff --git a/src/rabbit_control.erl b/src/rabbit_control.erl
index e8afed0c..b9e550c9 100644
--- a/src/rabbit_control.erl
+++ b/src/rabbit_control.erl
@@ -20,7 +20,6 @@
 -export([start/0, stop/0, action/5, diagnostics/1, log_action/3]).
 
 -define(RPC_TIMEOUT, infinity).
--define(WAIT_FOR_VM_ATTEMPTS, 5).
 
 -define(QUIET_OPT, "-q").
 -define(NODE_OPT, "-n").
@@ -193,9 +192,9 @@ action(force_cluster, Node, ClusterNodeSs, _Opts, Inform) ->
            [Node, ClusterNodes]),
     rpc_call(Node, rabbit_mnesia, force_cluster, [ClusterNodes]);
 
-action(wait, Node, [], _Opts, Inform) ->
+action(wait, Node, [PidFile], _Opts, Inform) ->
     Inform("Waiting for ~p", [Node]),
-    wait_for_application(Node, ?WAIT_FOR_VM_ATTEMPTS);
+    wait_for_application(Node, PidFile, Inform);
 
 action(status, Node, [], _Opts, Inform) ->
     Inform("Status of node ~p", [Node]),
@@ -356,23 +355,69 @@ action(report, Node, _Args, _Opts, Inform) ->
 
 %%----------------------------------------------------------------------------
 
-wait_for_application(Node, Attempts) ->
+wait_for_application(Node, PidFile, Inform) ->
+    Pid = wait_and_read_pid_file(PidFile),
+    Inform("pid is ~s", [Pid]),
+    wait_for_application(Node, Pid).
+
+wait_for_application(Node, Pid) ->
+    case process_up(Pid) of
+        true  -> case node_up(Node) of
+                     true  -> ok;
+                     false -> timer:sleep(1000),
+                              wait_for_application(Node, Pid)
+                 end;
+        false -> {error, process_not_running}
+    end.
+
+wait_and_read_pid_file(PidFile) ->
+    case file:read_file(PidFile) of
+        {ok,    Bin}    -> string:strip(binary_to_list(Bin), right, $\n);
+        {error, enoent} -> timer:sleep(500),
+                           wait_and_read_pid_file(PidFile);
+        {error, _} = E  -> exit({error, {could_not_read_pid, E}})
+    end.
+
+node_up(Node) ->
     case rpc_call(Node, application, which_applications, [infinity]) of
-        {badrpc, _} = E -> case Attempts of
-                               0 -> E;
-                               _ -> wait_for_application0(Node, Attempts - 1)
-                           end;
-        Apps            -> case proplists:is_defined(rabbit, Apps) of
-                               %% We've seen the node up; if it goes down
-                               %% die immediately.
-                               true  -> ok;
-                               false -> wait_for_application0(Node, 0)
-                           end
+        {badrpc, _} -> false;
+        Apps        -> proplists:is_defined(rabbit, Apps)
     end.
 
-wait_for_application0(Node, Attempts) ->
-    timer:sleep(1000),
-    wait_for_application(Node, Attempts).
+% Test using some OS clunkiness since we shouldn't trust
+% rpc:call(os, getpid, []) at this point
+process_up(Pid) ->
+    with_os([{unix, fun () ->
+                            system("ps -p " ++ Pid
+                                   ++ " >/dev/null 2>&1") =:= 0
+                    end},
+             {win32, fun () ->
+                             Res = os:cmd("tasklist /nh /fi \"pid eq " ++
+                                          Pid ++ "\" 2>&1"),
+                             case re:run(Res, "erl\\.exe", [{capture, none}]) of
+                                 match -> true;
+                                 _     -> false
+                             end
+                     end}]).
+
+with_os(Handlers) ->
+    {OsFamily, _} = os:type(),
+    case proplists:get_value(OsFamily, Handlers) of
+        undefined -> throw({unsupported_os, OsFamily});
+        Handler   -> Handler()
+    end.
+
+% Like system(3)
+system(Cmd) ->
+    ShCmd = "sh -c '" ++ escape_quotes(Cmd) ++ "'",
+    Port = erlang:open_port({spawn, ShCmd}, [exit_status,nouse_stdio]),
+    receive {Port, {exit_status, Status}} -> Status end.
+
+% Escape the quotes in a shell command so that it can be used in "sh -c 'cmd'"
+escape_quotes(Cmd) ->
+    lists:flatten(lists:map(fun ($') -> "'\\''"; (Ch) -> Ch end, Cmd)).
+
+%%----------------------------------------------------------------------------
 
 default_if_empty(List, Default) when is_list(List) ->
     if List == [] -> Default;