Merge branch 'master' into rabbitmq-server-479

7 files changed, 89 insertions, 13 deletions

diff --git a/Makefile b/Makefile
index 1afbfeec9f..aca411be26 100644
--- a/Makefile
+++ b/Makefile
@@ -262,6 +262,9 @@ $(SOURCE_DIST): $(ERLANG_MK_RECURSIVE_DEPS_LIST)
 		   grep -qs -E "^[[:blank:]]*include[[:blank:]]+(erlang\.mk|.*/erlang\.mk)$$" $@/deps/$$(basename $$dep)/erlang.mk; then \
 			echo "include ../../erlang.mk" > $@/deps/$$(basename $$dep)/erlang.mk; \
 		fi; \
+		sed -E -i.bak "s|^[[:blank:]]*include[[:blank:]]+\.\./.*erlang.mk$$|include ../../erlang.mk|" \
+		 $@/deps/$$(basename $$dep)/Makefile && \
+		rm $@/deps/$$(basename $$dep)/Makefile.bak; \
 		if test -f "$$dep/license_info"; then \
 			cp "$$dep/license_info" "$@/deps/licensing/license_info_$$(basename "$$dep")"; \
 			cat "$$dep/license_info" >> $@/LICENSE; \
diff --git a/docs/rabbitmqctl.1.xml b/docs/rabbitmqctl.1.xml
index b9e478cb53..fb4959b276 100644
--- a/docs/rabbitmqctl.1.xml
+++ b/docs/rabbitmqctl.1.xml
@@ -1966,8 +1966,8 @@
                 <listitem><para>
                     The new memory limit at which flow control is
                     triggered, expressed in bytes as an integer number
-                    greater than or equal to 0 or as string with memory unit
-                    (e.g. 1024MB). Available units are:
+                    greater than or equal to 0 or as a string with memory units
+                    (e.g. 512M or 1G). Available units are:
                     k, kiB: kibibytes (2^10 bytes)
                     M, MiB: mebibytes (2^20)
                     G, GiB: gibibytes (2^30)
@@ -1979,6 +1979,36 @@
             </variablelist>
           </listitem>
         </varlistentry>
+        <varlistentry>
+          <term><cmdsynopsis><command>set_disk_free_limit</command> <arg choice="req"><replaceable>disk_limit</replaceable></arg></cmdsynopsis></term>
+          <listitem>
+            <variablelist>
+              <varlistentry>
+                <term>disk_limit</term>
+                <listitem><para>
+                    Lower bound limit as an integer in bytes or a string with memory units (see vm_memory_high_watermark),
+                    e.g. 512M or 1G. Once free disk space reaches the limit, a disk alarm will be set.
+                </para></listitem>
+              </varlistentry>
+            </variablelist>
+          </listitem>
+        </varlistentry>
+        <varlistentry>
+          <term><cmdsynopsis><command>set_disk_free_limit mem_relative</command> <arg choice="req"><replaceable>fraction</replaceable></arg></cmdsynopsis></term>
+          <listitem>
+            <variablelist>
+              <varlistentry>
+                <term>fraction</term>
+                <listitem><para>
+                    Limit relative to the total amount available RAM 
+                    as a non-negative floating point number. 
+                    Values lower than 1.0 can be dangerous and 
+                    should be used carefully.
+                </para></listitem>
+              </varlistentry>
+            </variablelist>
+          </listitem>
+        </varlistentry>
       </variablelist>
     </refsect2>
   </refsect1>
diff --git a/scripts/rabbitmq-server-ha.ocf b/scripts/rabbitmq-server-ha.ocf
index 1e4ebd0f1e..376c661a45 100755
--- a/scripts/rabbitmq-server-ha.ocf
+++ b/scripts/rabbitmq-server-ha.ocf
@@ -13,6 +13,9 @@
 #
 # See usage() function below for more details ...
 #
+# Note that the script uses set_rabbitmq_policy.sh script located in the
+# same directory to setup RabbitMQ policies.
+#
 #######################################################################
 # Initialization:
 
@@ -92,6 +95,11 @@ UEND
 }
 
 meta_data() {
+    # The EXTENDED_OCF_PARAMS parameter below does not exist by default
+    # and hence converted to an empty string unless overridden. It
+    # could be used by an extention script to add new parameters. For
+    # example see https://review.openstack.org/#/c/249180/10
+
     cat <<END
 <?xml version="1.0"?>
 <!DOCTYPE resource-agent SYSTEM "ra-api-1.dtd">
@@ -268,6 +276,8 @@ If too many timeouts happen in a raw, the monitor call will return with error.
 <content type="string" default="${OCF_RESKEY_max_rabbitmqctl_timeouts_default}" />
 </parameter>
 
+$EXTENDED_OCF_PARAMS
+
 </parameters>
 
 <actions>
@@ -1734,11 +1744,10 @@ action_promote() {
                         ocf_log info "${LH} action end."
                         exit $OCF_FAILED_MASTER
                     fi
-                    ocf_log info "${LH} Setting HA policy for all queues"
-                    ${OCF_RESKEY_ctl} set_policy ha-all "." '{"ha-mode":"all", "ha-sync-mode":"automatic"}' --apply-to all --priority 0
-                    ${OCF_RESKEY_ctl} set_policy heat_rpc_expire "^heat-engine-listener\\." '{"expires":3600000,"ha-mode":"all","ha-sync-mode":"automatic"}' --apply-to all --priority 1
-                    ${OCF_RESKEY_ctl} set_policy results_expire "^results\\." '{"expires":3600000,"ha-mode":"all","ha-sync-mode":"automatic"}' --apply-to all --priority 1
-                    ${OCF_RESKEY_ctl} set_policy tasks_expire "^tasks\\." '{"expires":3600000,"ha-mode":"all","ha-sync-mode":"automatic"}' --apply-to all --priority 1
+
+                    local set_policy_path="$(dirname $0)/set_rabbitmq_policy.sh"
+                    . $set_policy_path
+
                     # create timestamp file
                     ocf_log info "${LH} Updating start timestamp"
                     ocf_run crm_attribute -N $THIS_PCMK_NODE -l reboot --name 'rabbit-start-time' --update $(now)
diff --git a/scripts/set_rabbitmq_policy.sh b/scripts/set_rabbitmq_policy.sh
new file mode 100755
index 0000000000..a88b0c417a
--- /dev/null
+++ b/scripts/set_rabbitmq_policy.sh
@@ -0,0 +1,5 @@
+# This script is called by rabbitmq-server-ha.ocf during RabbitMQ
+# cluster start up. It is a convenient place to set your cluster
+# policy here, for example:
+# ${OCF_RESKEY_ctl} set_policy ha-all "." '{"ha-mode":"all", "ha-sync-mode":"automatic"}' --apply-to all --priority 0
+
diff --git a/src/rabbit_control_main.erl b/src/rabbit_control_main.erl
index 91a079da93..c064e82ac6 100644
--- a/src/rabbit_control_main.erl
+++ b/src/rabbit_control_main.erl
@@ -88,6 +88,7 @@
          {trace_on, [?VHOST_DEF]},
          {trace_off, [?VHOST_DEF]},
          set_vm_memory_high_watermark,
+         set_disk_free_limit,
          help
         ]).
 
@@ -426,6 +427,27 @@ action(set_vm_memory_high_watermark, Node, ["absolute", Arg], _Opts, Inform) ->
             {error_string, "Unable to parse absolute memory limit value ~p", [Arg]}
     end;
 
+action(set_disk_free_limit, Node, [Arg], _Opts, Inform) ->
+    case rabbit_resource_monitor_misc:parse_information_unit(Arg) of
+        {ok, Limit} ->
+            Inform("Setting disk free limit on ~p to ~p bytes", [Node, Limit]),
+            rpc_call(Node, rabbit_disk_monitor, set_disk_free_limit, [Limit]);
+        {error, parse_error} ->
+            {error_string, "Unable to parse disk free limit value ~p", [Arg]}
+    end;
+
+action(set_disk_free_limit, Node, ["mem_relative", Arg], _Opts, Inform) ->
+    Frac = list_to_float(case string:chr(Arg, $.) of
+                             0 -> Arg ++ ".0";
+                             _ -> Arg
+                         end),
+    Inform("Setting disk free limit on ~p to ~p of total RAM", [Node, Frac]),
+    rpc_call(Node, 
+             rabbit_disk_monitor, 
+             set_disk_free_limit, 
+             [{mem_relative, Frac}]);
+
+
 action(set_permissions, Node, [Username, CPerm, WPerm, RPerm], Opts, Inform) ->
     VHost = proplists:get_value(?VHOST_OPT, Opts),
     Inform("Setting permissions for user \"~s\" in vhost \"~s\"",
@@ -667,10 +689,10 @@ read_pid_file(PidFile, Wait) ->
 
 become(BecomeNode) ->
     error_logger:tty(false),
-    ok = net_kernel:stop(),
     case net_adm:ping(BecomeNode) of
         pong -> exit({node_running, BecomeNode});
-        pang -> io:format("  * Impersonating node: ~s...", [BecomeNode]),
+        pang -> ok = net_kernel:stop(),
+                io:format("  * Impersonating node: ~s...", [BecomeNode]),
                 {ok, _} = rabbit_cli:start_distribution(BecomeNode),
                 io:format(" done~n", []),
                 Dir = mnesia:system_info(directory),
diff --git a/src/rabbit_resource_monitor_misc.erl b/src/rabbit_resource_monitor_misc.erl
index 88f3bd1121..f90b8ce310 100644
--- a/src/rabbit_resource_monitor_misc.erl
+++ b/src/rabbit_resource_monitor_misc.erl
@@ -30,10 +30,12 @@ parse_information_unit(Value) when is_integer(Value) -> {ok, Value};
 parse_information_unit(Value) when is_list(Value) ->
     case re:run(Value,
                 "^(?<VAL>[0-9]+)(?<UNIT>kB|KB|MB|GB|kb|mb|gb|Kb|Mb|Gb|kiB|KiB|MiB|GiB|kib|mib|gib|KIB|MIB|GIB|k|K|m|M|g|G)?$",
-                [{capture, all_names, list}]) of
+                [{capture, all_but_first, list}]) of
     	{match, [[], _]} ->
             {ok, list_to_integer(Value)};
-        {match, [Unit, Num]} ->
+        {match, [Num]} ->
+            {ok, list_to_integer(Num)};
+        {match, [Num, Unit]} ->
             Multiplier = case Unit of
                              KiB when KiB =:= "k";  KiB =:= "kiB"; KiB =:= "K"; KiB =:= "KIB"; KiB =:= "kib" -> 1024;
                              MiB when MiB =:= "m";  MiB =:= "MiB"; MiB =:= "M"; MiB =:= "MIB"; MiB =:= "mib" -> 1024*1024;
diff --git a/src/rabbit_upgrade.erl b/src/rabbit_upgrade.erl
index daf39b8acc..0a85ef3e7d 100644
--- a/src/rabbit_upgrade.erl
+++ b/src/rabbit_upgrade.erl
@@ -100,7 +100,12 @@ ensure_backup_taken() ->
                      false -> ok = take_backup();
                      _     -> ok
                  end;
-        true  -> throw({error, previous_upgrade_failed})
+        true  ->
+          error("Found lock file at ~s.
+            Either previous upgrade is in progress or has failed.
+            Database backup path: ~s",
+            [lock_filename(), backup_dir()]),
+          throw({error, previous_upgrade_failed})
     end.
 
 take_backup() ->
@@ -108,7 +113,7 @@ take_backup() ->
     case rabbit_mnesia:copy_db(BackupDir) of
         ok         -> info("upgrades: Mnesia dir backed up to ~p~n",
                            [BackupDir]);
-        {error, E} -> throw({could_not_back_up_mnesia_dir, E})
+        {error, E} -> throw({could_not_back_up_mnesia_dir, E, BackupDir})
     end.
 
 ensure_backup_removed() ->