[openstack agents] Volume re-attaching, v3 API, get hostname from corosync

heartbeat/openstack-cinder-volume

@@ -27,7 +27,6 @@
 # Inc., 59 Temple Place - Suite 330, Boston MA 02111-1307, USA.
 #
 
-
 #######################################################################
 # Initialization:
 
@@ -36,8 +35,14 @@
 
 # Defaults
 OCF_RESKEY_openstackcli_default="/usr/bin/openstack"
+OCF_RESKEY_node_id_cache_file_default="/tmp/node_id"
+OCF_RESKEY_volume_local_check_default="true"
+
+export attached_server_id=""
 
 : ${OCF_RESKEY_openstackcli=${OCF_RESKEY_openstackcli_default}}
+: ${OCF_RESKEY_node_id_cache_file=${OCF_RESKEY_node_id_cache_file_default}}
+: ${OCF_RESKEY_volume_local_check=${OCF_RESKEY_volume_local_check_default}}
 
 #######################################################################
 
@@ -74,6 +79,26 @@ Path to command line tools for openstack.
 <content type="string" default="${OCF_RESKEY_openstackcli_default}" />
 </parameter>
 
+<parameter name="node_id_cache_file">
+<longdesc lang="en">
+Path to Node ID cache file, used to avoid Openstack API calls:
+1. Is the local file written?
+2. Is openstack_id available as a node attribute?
+3. Can we get it from the API?
+</longdesc>
+<shortdesc lang="en">Path to Node ID cache file</shortdesc>
+<content type="string" default="${OCF_RESKEY_node_id_cache_file_default}" />
+</parameter>
+
+<parameter name="volume_local_check">
+<longdesc lang="en">
+This option allows the cluster to monitor the cinder volume presence without 
+calling the API.
+</longdesc>
+<shortdesc lang="en">Monitor cinder volume locally</shortdesc>
+<content type="boolean" default="${OCF_RESKEY_volume_local_check_default}" />
+</parameter>
+
 <parameter name="openrc" required="1">
 <longdesc lang="en">
 Valid Openstack credentials as openrc file from api_access/openrc.
@@ -103,8 +128,82 @@ Cinder volume identifier to use to attach the bloc storage.
 END
 }
 
+#
+# This is used to get the node ID from different sources:
+# 1. Is the local file written?
+# 2. Is openstack_id available as a node attribute?
+# 3. Can we get it from the API?
+#
+# When the ID is retrieved, the local cache file is written.
+# This prevents the agent to call the API each time the agent is used.
+#
+_get_node_id() {
+	local crm_node
+	local node
+	local node_id
+	local result
+
+	crm_node=$(crm_node -n)
+
+	#
+	# Use local cache
+	#
+	if [ -f $OCF_RESKEY_node_id_cache_file ] ; then
+		node_id=$(cat $OCF_RESKEY_node_id_cache_file)
+
+		if [ ! -z "$node_id" ] ; then
+			echo $node_id
+			return
+		fi
+	fi
+
+	#
+	# Query the attributes database
+	#
+	node_id=$(${HA_SBIN_DIR}/attrd_updater --query -n openstack_id -N $crm_node \
+		| tr ' ' '\n' \
+		| awk -F= '/value=/ {gsub("\"","");print $NF}')
+
+	if [ ! -z "$node_id" ] ; then
+		echo $node_id | awk '{print $1}'
+		echo $node_id | awk '{print $1}' > $OCF_RESKEY_node_id_cache_file
+		return
+	fi
+
+	#
+	# Use the API
+	#
+	node=$(crm_node -n | awk -F. '{print $1}')
+
+	result=$($OCF_RESKEY_openstackcli server list \
+		--format value --column ID --column Name \
+		| grep $node)
+
+	if [ $? -eq 0 ] ; then
+		echo $result | awk '{print $1}'
+		echo $result | awk '{print $1}' > $OCF_RESKEY_node_id_cache_file
+		return
+	fi
+
+	ocf_exit_reason "openstack_id attribute must be set for node $crm_node"
+	return $OCF_ERR_CONFIGURED
+}
+
 osvol_validate() {
+	local node_id
+
 	check_binary "$OCF_RESKEY_openstackcli"
+	check_binary "awk"
+	check_binary "tr"
+
+	. $OCF_RESKEY_openrc
+
+	node_id=$(_get_node_id)
+
+	if [ -z "$node_id" ] ; then
+		ocf_exit_reason "openstack_id attribute must be set for node $crm_node"
+		return $OCF_ERR_CONFIGURED
+	fi
 
 	if [ -z "$OCF_RESKEY_openrc" ]; then
 		ocf_exit_reason "openrc parameter not set"
@@ -116,39 +215,58 @@ osvol_validate() {
 		return $OCF_ERR_CONFIGURED
 	fi
 
-	. $OCF_RESKEY_openrc
-
-	if ! $OCF_RESKEY_openstackcli volume list|grep -q $OCF_RESKEY_volume_id ; then
-		ocf_exit_reason "Volume $OCF_RESKEY_volume_id not found"
-		return $OCF_ERR_CONFIGURED
-	fi
-
 	return $OCF_SUCCESS
 }
 
 osvol_monitor() {
 	local result
-	local attached_server_id
 	local node_id
+	local short_volume_id
+	local fdisk_command
+
+	if ocf_is_true $OCF_RESKEY_volume_local_check ; then
+		#
+		# Is the volue attached?
+		# We check the local devices
+		#
+		short_volume_id=$(echo $OCF_RESKEY_volume_id | awk '{print substr($0, 0, 20)}')
+		if uname | grep -q Linux ; then
+			fdisk_command="fdisk -l"
+		else
+			fdisk_command="fdisk"
+		fi
 
-	# Is the volue attached?
-	result=$($OCF_RESKEY_openstackcli volume show \
-		--column status \
-		--column attachments \
-		--format value \
-		$OCF_RESKEY_volume_id)
-
-	if [[ "$result" =~ "available" ]] ; then
-		ocf_log warn "$OCF_RESKEY_volume_id is not attached to any instance"
-		return $OCF_NOT_RUNNING
+		$fdisk_command /dev/disk/by-id/virtio-$short_volume_id 1>/dev/null 2>&1
+		if [ $? -eq 0 ] ; then
+			return $OCF_SUCCESS
+		else
+			ocf_log warn "$OCF_RESKEY_volume_id is not attached to instance $(_get_node_id)"
+			return $OCF_NOT_RUNNING
+		fi
 	else
-		attached_server_id=$(echo $result|head -n1|awk -F "'" '{print $4}')
-		ocf_log info "$OCF_RESKEY_volume_id is attached to instance $attached_server_id"
-		node_id=$(${HA_SBIN_DIR}/attrd_updater --query -n openstack_id -N $HOSTNAME \
-			| awk '{gsub("value=","") ; gsub("\"","") ; print $NF}')
-
-		if [ "$node_id" != "$attached_server_id" ] ; then
+		#
+		# Is the volue attached?
+		# We use the API
+		#
+		result=$($OCF_RESKEY_openstackcli volume show \
+			--column status \
+			--column attachments \
+			--format value \
+			$OCF_RESKEY_volume_id)
+
+		if echo "$result" | grep -q available ; then
+			ocf_log warn "$OCF_RESKEY_volume_id is not attached to any instance"
 			return $OCF_NOT_RUNNING
+		else
+			export attached_server_id=$(echo $result|head -n1|awk -F "'" '{print $4}')
+			ocf_log info "$OCF_RESKEY_volume_id is attached to instance $attached_server_id"
+
+			# Compare node_id and the id of the node the volume is attached to
+			node_id=$(_get_node_id)
+
+			if [ "$node_id" != "$attached_server_id" ] ; then
+				return $OCF_NOT_RUNNING
+			fi
 		fi
 	fi
 
@@ -158,14 +276,20 @@ osvol_monitor() {
 osvol_stop() {
 	local node_id
 
+	#
+	# Is the volume already attached?
+	#
 	osvol_monitor
 	if [ $? = $OCF_NOT_RUNNING ]; then
 		ocf_log info "Volume $OCF_RESKEY_volume_id already available"
 		return $OCF_SUCCESS
 	fi
 
-	node_id=$(${HA_SBIN_DIR}/attrd_updater --query -n openstack_id -N $HOSTNAME \
-		| awk '{gsub("value=","") ; gsub("\"","") ; print $NF}')
+	node_id=$(_get_node_id)
+
+	#
+	# Unmout the volume
+	#
 	if ! $OCF_RESKEY_openstackcli server remove volume $node_id $OCF_RESKEY_volume_id ; then
 		ocf_log error "Couldn't remove volume $OCF_RESKEY_volume_id from instance $node_id"
 		return $OCF_ERR_GENERIC
@@ -178,14 +302,33 @@ osvol_stop() {
 osvol_start() {
 	local node_id
 
+	#
+	# Is the volume already attached?
+	#
 	osvol_monitor
 	if [ $? = $OCF_SUCCESS ]; then
 		ocf_log info "$OCF_RESKEY_volume_id already attached"
 		return $OCF_SUCCESS
 	fi
 
-	node_id=$(${HA_SBIN_DIR}/attrd_updater --query -n openstack_id -N $HOSTNAME \
-		| awk '{gsub("value=","") ; gsub("\"","") ; print $NF}')
+	#
+	# Unmout it from another node
+	# TODO: make it optional in case multi-attachment is allowed by Cinder
+	#
+	if [ ! -z $attached_server_id ] ; then
+		if ! $OCF_RESKEY_openstackcli server remove volume $attached_server_id $OCF_RESKEY_volume_id ; then
+			ocf_log error "Couldn't remove volume $OCF_RESKEY_volume_id from instance $attached_server_id"
+			return $OCF_ERR_GENERIC
+		fi
+	fi
+
+	export attached_server_id=""
+
+	node_id=$(_get_node_id)
+
+	#
+	# Attach the volume
+	#
 	$OCF_RESKEY_openstackcli server add volume $node_id $OCF_RESKEY_volume_id
 	if [ $? != $OCF_SUCCESS ]; then
 		ocf_log error "Couldn't add volume $OCF_RESKEY_volume_id to instance $node_id"

heartbeat/openstack-floating-ip

@@ -132,7 +132,7 @@ osflip_validate() {
 		return $OCF_ERR_CONFIGURED
 	fi
 
-	${HA_SBIN_DIR}/attrd_updater --query -n openstack_ports -N $HOSTNAME > /dev/null 2>&1
+	${HA_SBIN_DIR}/attrd_updater --query -n openstack_ports -N $(crm_node -n) > /dev/null 2>&1
 	if [ $? -ne 0 ] ; then
 		ocf_log warn "attr_updater failed to get openstack_ports attribute of node $OCF_RESOURCE_INSTANCE"
 		return $OCF_ERR_GENERIC
@@ -147,8 +147,10 @@ osflip_monitor() {
 	local node_port_ids
 	local port
 	local buffer
+	local crm_node
 
-	node_port_ids=$(${HA_SBIN_DIR}/attrd_updater --query -n openstack_ports -N $HOSTNAME \
+	crm_node=$(crm_node -n)
+	node_port_ids=$(${HA_SBIN_DIR}/attrd_updater --query -n openstack_ports -N $crm_node \
 		| awk -F= '{gsub("\"","");print $NF}' \
 		| tr ',' ' ' \
 		| awk -F: '{print $NF}')
@@ -162,13 +164,13 @@ osflip_monitor() {
 	for port in $node_port_ids ; do
 		if echo $result | grep -q $port ; then
 			floating_ip=$(echo $result | awk '/floating_ip_address/ {print $2}')
-			${HA_SBIN_DIR}/attrd_updater ${OCF_RESKEY_delay} -n openstack_floating_ip -v $floating_ip
+			${HA_SBIN_DIR}/attrd_updater ${OCF_RESKEY_delay} -S status -n openstack_floating_ip -v $floating_ip
 
 			return $OCF_SUCCESS
 		fi
 	done
 
-	${HA_SBIN_DIR}/attrd_updater ${OCF_RESKEY_delay} -D -n openstack_floating_ip
+	${HA_SBIN_DIR}/attrd_updater ${OCF_RESKEY_delay} -D -S state -n openstack_floating_ip
 	ocf_log warn "$OCF_RESKEY_ip_id is not attached to any fixed address"
 	return $OCF_NOT_RUNNING
 }
@@ -207,9 +209,13 @@ osflip_start() {
 	fi
 
 	# Get port_id from subnet_id
-	node_port_ids=$(${HA_SBIN_DIR}/attrd_updater --query -n openstack_ports -N $HOSTNAME \
+	node_port_ids=$(${HA_SBIN_DIR}/attrd_updater --query -n openstack_ports -N $(crm_node -n) \
 		| awk '{gsub("value=","") ; gsub("\"","") ; print $NF}')
-	node_port_id=$(echo $node_port_ids|tr ',' '\n'|awk -F: "/$OCF_RESKEY_subnet_id/ {print \$2}")
+
+	node_port_id=$(echo $node_port_ids \
+		| tr ',' '\n' \
+		| awk -F: "/$OCF_RESKEY_subnet_id/ {print \$2}")
+
 	ocf_log info "Moving IP address $OCF_RESKEY_ip_id to port ID $node_port_id"
 
 	$OCF_RESKEY_openstackcli floating ip set --port $node_port_id $OCF_RESKEY_ip_id

heartbeat/openstack-info

@@ -115,6 +115,10 @@ OSInfoStats() {
 		--format value --column ID --column Name \
 		| grep $node)
 
+	if [ $? -ne 0 ] ; then
+		ocf_exit_reason "cannot find $node in instances list"
+		exit $OCF_ERR_GENERIC
+	fi
 	node_id=$(echo $result|awk '{print $1}')
 
 	# Nova data: flavor, AZ…
@@ -148,14 +152,26 @@ OSInfoStats() {
 			value="$value,$p"
 		fi
 	done
-	${HA_SBIN_DIR}/attrd_updater ${OCF_RESKEY_delay} -n openstack_ports -v $value
+	${HA_SBIN_DIR}/attrd_updater ${OCF_RESKEY_delay} -n openstack_ports -v "$value"
 
 	if [ ! -z "$OS_REGION_NAME" ] ; then
-		${HA_SBIN_DIR}/attrd_updater ${OCF_RESKEY_delay} -n openstack_region -v $OS_REGION_NAME
+		${HA_SBIN_DIR}/attrd_updater ${OCF_RESKEY_delay} -n openstack_region -v "$OS_REGION_NAME"
+	fi
+
+	if [ ! -z "$OS_TENANT_ID" ] ; then
+		${HA_SBIN_DIR}/attrd_updater ${OCF_RESKEY_delay} -n openstack_tenant_id -v "$OS_TENANT_ID"
+
+		if [ ! -z "$OS_TENANT_NAME" ] ; then
+			${HA_SBIN_DIR}/attrd_updater ${OCF_RESKEY_delay} -n openstack_tenant_name -v "$OS_TENANT_NAME"
+		fi
+	else
+		${HA_SBIN_DIR}/attrd_updater ${OCF_RESKEY_delay} -n openstack_project_id -v "$OS_PROJECT_ID"
+
+		if [ ! -z "$OS_PROJECT_NAME" ] ; then
+			${HA_SBIN_DIR}/attrd_updater ${OCF_RESKEY_delay} -n openstack_project_name -v "$OS_PROJECT_NAME"
+		fi
 	fi
 
-	${HA_SBIN_DIR}/attrd_updater ${OCF_RESKEY_delay} -n openstack_tenant_name -v $OS_TENANT_NAME
-	${HA_SBIN_DIR}/attrd_updater ${OCF_RESKEY_delay} -n openstack_tenant_id -v $OS_TENANT_ID
 }
 
 OSInfo_usage() {
@@ -181,6 +197,8 @@ OSInfo_stop() {
 	${HA_SBIN_DIR}/attrd_updater ${OCF_RESKEY_delay} -D -n openstack_region
 	${HA_SBIN_DIR}/attrd_updater ${OCF_RESKEY_delay} -D -n openstack_tenant_id
 	${HA_SBIN_DIR}/attrd_updater ${OCF_RESKEY_delay} -D -n openstack_tenant_name
+	${HA_SBIN_DIR}/attrd_updater ${OCF_RESKEY_delay} -D -n openstack_project_id
+	${HA_SBIN_DIR}/attrd_updater ${OCF_RESKEY_delay} -D -n openstack_project_name
 	exit $OCF_SUCCESS
 }