resource-agents 1:4.0.0~rc1-4 / usr / lib / ocf / resource.d / heartbeat / redis

#!/bin/bash

. ${OCF_ROOT}/resource.d/heartbeat/.ocf-shellfuncs

: ${OCF_RESKEY_bin:=/usr/bin/redis-server}
: ${OCF_RESKEY_client_bin:=/usr/bin/redis-cli}
: ${OCF_RESKEY_user:=redis}
: ${OCF_RESKEY_rundir:=/var/run/redis}
: ${OCF_RESKEY_pidfile_name:=redis-server.pid}
: ${OCF_RESKEY_socket_name:=redis.sock}
: ${OCF_RESKEY_port:=6379}

if [ -z "$OCF_RESKEY_config" ]; then
	if [ -f "/etc/redis.conf" ]; then
		OCF_RESKEY_config="/etc/redis.conf"
	else
		OCF_RESKEY_config="/etc/redis/redis.conf"
	fi
fi

CHECK_SLAVE_STATE=0

REDIS_CHECK_DUMP="/usr/bin/redis-check-dump"
REDIS_SERVER="$OCF_RESKEY_bin"
REDIS_CLIENT="$OCF_RESKEY_client_bin"
REDIS_CONFIG="$OCF_RESKEY_config"
REDIS_USER="$OCF_RESKEY_user"
REDIS_RUNDIR="$OCF_RESKEY_rundir"
REDIS_PIDFILE="$OCF_RESKEY_rundir/$OCF_RESKEY_pidfile_name"
REDIS_SOCKET="$OCF_RESKEY_rundir/$OCF_RESKEY_socket_name"
REDIS_REPLICATION_PORT="$OCF_RESKEY_port"

if ! [ -f $REDIS_CHECK_DUMP ]; then
	REDIS_CHECK_DUMP="$(which redis-check-dump 2>/dev/null)"
fi

if [ -f "$REDIS_CONFIG" ]; then
	REDIS_DUMP_DIR="$(cat $REDIS_CONFIG | grep "^\s*dir\s" | awk '{ print $2 }' 2>/dev/null)"
	REDIS_DUMP_FILE="$(cat $REDIS_CONFIG | grep "^\s*dbfilename\s" | awk '{ print $2 }' 2>/dev/null)"
fi
: ${REDIS_DUMP_DIR:=/var/lib/redis/}
: ${REDIS_DUMP_FILE:=dump.rdb}

function meta_data() {
	cat <<EOI
<?xml version="1.0"?>
<!DOCTYPE resource-agent SYSTEM "ra-api-1.dtd">
<resource-agent name="redis">
<version>1.0</version>

<longdesc lang="en">
Resource agent script for redis server. 

This resource fully supports master/slave replication. The master preference of a node is determined by the 'slave_priority' parameter of the redis config.
When taking the resource from 'unmanaged' to 'managed', the currently active master will be given a priority of 1000 (plus 1 for each active connection). The default 'slave_priority' is 100, so the master will stay master. For a slave to become master after converting the resource to managed, set a slave_priority greater than 1000.
</longdesc>

<shortdesc lang="en">Redis server</shortdesc>

<parameters>
<parameter name="bin" unique="0" required="0">
<longdesc lang="en">
Path to \`redis-server\`
</longdesc>
<shortdesc lang="en">Path to \`redis-server\`</shortdesc>
<content type="string" default="${OCF_RESKEY_bin}" />
</parameter>

<parameter name="client_bin" unique="0" required="0">
<longdesc lang="en">
Path to \`redis-cli\`
</longdesc>
<shortdesc lang="en">Path to \`redis-cli\`</shortdesc>
<content type="string" default="${OCF_RESKEY_client_bin}" />
</parameter>

<parameter name="config" unique="1" required="0">
<longdesc lang="en">
Path to 'redis.conf'
</longdesc>
<shortdesc lang="en">Path to 'redis.conf'</shortdesc>
<content type="string" default="${OCF_RESKEY_config}" />
</parameter>

<parameter name="user" unique="0" required="0">
<longdesc lang="en">
User to run redis as
</longdesc>
<shortdesc lang="en">Redis user</shortdesc>
<content type="string" default="${OCF_RESKEY_user}" />
</parameter>

<parameter name="rundir" unique="1" required="0">
<longdesc lang="en">
Directory to store socket and pid file in
</longdesc>
<shortdesc lang="en">Redis var/run dir</shortdesc>
<content type="string" default="${OCF_RESKEY_rundir}"/>
</parameter>

<parameter name="pidfile_name" unique="0" required="0">
<longdesc lang="en">
The filename to use for the pidfile. Will be created in the rundir. 
Should only be a basename, not a full path.
</longdesc>
<shortdesc lang="en">Redis pidfile name</shortdesc>
<content type="string" default="${OCF_RESKEY_pidfile_name}"/>
</parameter>

<parameter name="socket_name" unique="0" required="0">
<longdesc lang="en">
The filename to use for the socket. Will be crated in the rundir. 
Should only be a basename, not a full path.
</longdesc>
<shortdesc lang="en">Redis socket name</shortdesc>
<content type="string" default="${OCF_RESKEY_socket_name}"/>
</parameter>

<parameter name="port" unique="0" required="0">
<longdesc lang="en">
Port for replication client to connect to on remote server
</longdesc>
<shortdesc lang="en">Replication port</shortdesc>
<content type="string" default="${OCF_RESKEY_port}"/>
</parameter>

<parameter name="wait_last_known_master" unique="0" required="0">
<longdesc lang="en">
During redis cluster bootstrap, wait for the last known master to be
promoted before allowing any other instances in the cluster to be
promoted. This lessens the risk of data loss when persistent data
is in use.
</longdesc>
<shortdesc lang="en">Wait for last known master</shortdesc>
<content type="boolean" default="false"/>
</parameter>
</parameters>

<actions>
<action name="start" timeout="120" />
<action name="stop" timeout="120" />
<action name="status" timeout="60" />
<action name="monitor" depth="0" timeout="60" interval="45" />
<action name="monitor" role="Master" depth="0" timeout="60" interval="20" />
<action name="monitor" role="Slave" depth="0" timeout="60" interval="60" />
<action name="promote" timeout="120" />
<action name="demote" timeout="120" />
<action name="notify" timeout="90" />
<action name="validate-all" timeout="5" />
<action name="meta-data" timeout="5" />
</actions>
</resource-agent>
EOI
}

INSTANCE_ATTR_NAME=`echo ${OCF_RESOURCE_INSTANCE}| awk -F : '{print $1}'`
CRM_ATTR_REPL_INFO="${HA_SBIN_DIR}/crm_attribute --type crm_config --name ${INSTANCE_ATTR_NAME}_REPL_INFO -s redis_replication"
MASTER_HOST=""
MASTER_ACTIVE_CACHED=""
MASTER_ACTIVE=""

master_is_active()
{
	if [ -z "$MASTER_ACTIVE_CACHED" ]; then
		# determine if a master instance is already up and is healthy
		crm_mon --as-xml | grep "resource.*id=\"${OCF_RESOURCE_INSTANCE}\".*role=\"Master\".*active=\"true\".*orphaned=\"false\".*failed=\"false\"" > /dev/null 2>&1
		MASTER_ACTIVE=$?
		MASTER_ACTIVE_CACHED="true"
	fi
	return $MASTER_ACTIVE
}

function set_master()
{
	MASTER_HOST="$1"
	${CRM_ATTR_REPL_INFO} -v "$1" -q
}

function last_known_master()
{
	if [ -z "$MASTER_HOST" ]; then
		MASTER_HOST="$(${CRM_ATTR_REPL_INFO} --query  -q  2>/dev/null)"
	fi
	echo "$MASTER_HOST"
}

function crm_master_reboot() {
	"${HA_SBIN_DIR}/crm_master" -l reboot "$@"
}

function calculate_score()
{
	perf_score="$1"
	connected_clients="$2"

    	if ocf_is_true "$OCF_RESKEY_wait_last_known_master"; then
		# only set perferred score by slave_priority if
		# we are not waiting for the last known master. Otherwise
		# we want the agent to have complete control over the scoring.
		perf_score=""
		connected_clients="0"
	fi

	if [[ -z "$perf_score" ]]; then
		if [[ "$(last_known_master)" == "$NODENAME" ]]; then
			perf_score=1000
		else
			perf_score=1
		fi
	fi
	perf_score=$(( perf_score + connected_clients ))
	echo "$perf_score"
}

function set_score()
{
	local score="$1"

    	if ocf_is_true "$OCF_RESKEY_wait_last_known_master" && ! master_is_active; then
		local last_master="$(last_known_master)"
		if [ -n "$last_master" ] && [[ "$last_master" != "$NODENAME" ]]; then
			ocf_log info "Postponing setting master score for ${NODENAME} until last known master instance [${last_master}] is promoted"
			return
		fi
	fi

	ocf_log debug "monitor: Setting master score to '$score'"
	crm_master_reboot -v "$score"
}

function redis_client() {
	ocf_log debug "redis_client: '$REDIS_CLIENT' -s '$REDIS_SOCKET' $*"
	if [ -n "$clientpasswd" ]; then
		"$REDIS_CLIENT" -s "$REDIS_SOCKET" -a "$clientpasswd" "$@" | sed 's/\r//'
	else 
		"$REDIS_CLIENT" -s "$REDIS_SOCKET" "$@" | sed 's/\r//'
	fi
}

function simple_status() {
	local pid

	if ! [ -f "$REDIS_PIDFILE" ]; then
		return $OCF_NOT_RUNNING
	fi

	pid="$(<"$REDIS_PIDFILE")"
	pidof "$REDIS_SERVER" | grep -q "\<$pid\>" || return $OCF_NOT_RUNNING

	ocf_log debug "monitor: redis-server running under pid $pid"

	return $OCF_SUCCESS
}

function monitor() {
	local res

	simple_status
	res=$?
	if (( res != OCF_SUCCESS )); then
		return $res
	fi

	typeset -A info
	while read line; do
		[[ "$line" == "#"* ]] && continue
		[[ "$line" != *":"* ]] && continue
		IFS=':' read -r key value <<< "$line"
		info[$key]="$value"
	done < <(redis_client info)
	if [[ -z "${info[role]}" ]]; then
		ocf_log err "monitor: Could not get role from \`$REDIS_CLIENT -s $REDIS_SOCKET info\`"
		return $OCF_ERR_GENERIC
	fi

	if ocf_is_ms; then
		# Here we see if a score has already been set.
		# If score isn't set we the redis setting 'slave_priority'.
		# If that isn't set, we default to 1000 for a master, and 1 for slave.
		# We then add 1 for each connected client
		score="$(crm_master_reboot --get-value --quiet 2>/dev/null)"
		if [[ -z "$score" ]]; then
			score=$(calculate_score "${info[slave_priority]}" "${info[connected_clients]}")
			set_score "$score"
		fi

		if [[ "${info[role]}" == "master" ]]; then
			if ocf_is_probe; then
				set_master "$NODENAME"
			fi
			return $OCF_RUNNING_MASTER
		fi

		if [ "$CHECK_SLAVE_STATE" -eq 1 ]; then
			if [[ "${info[master_link_status]}" != "up" ]]; then
				ocf_log info "monitor: Slave mode link has not yet been established (link=${info[master_link_status]})"
				return $OCF_ERR_GENERIC
			fi
			if [[ "${info[master_host]}" != "$(last_known_master)" ]]; then
				ocf_log err "monitor: Slave mode current master does not match running master. current=${info[master_host]}, running=$(last_known_master)"
				return $OCF_ERR_GENERIC
			fi
		fi
	fi
	return $OCF_SUCCESS
}

function check_dump_file()
{
	if ! have_binary "$REDIS_CHECK_DUMP"; then
		return 0
	fi
	$REDIS_CHECK_DUMP ${REDIS_DUMP_DIR}/${REDIS_DUMP_FILE} 2>&1
}

function start() {
	monitor
	status=$?

	if (( status == OCF_SUCCESS )) || (( status == OCF_RUNNING_MASTER )); then
		ocf_log info "start: redis is already running"
		return $OCF_SUCCESS
	fi

	[[ ! -d "$REDIS_RUNDIR" ]] && mkdir -p "$REDIS_RUNDIR"
	chown -R "$REDIS_USER" "$REDIS_RUNDIR"
	if have_binary "restorecon"; then
		restorecon -Rv "$REDIS_RUNDIR"
	fi


	# check for 0 byte database dump file. This is an unrecoverable start
	# condition that we can avoid by deleting the 0 byte database file.
	if [ -f "${REDIS_DUMP_DIR}/${REDIS_DUMP_FILE}" ]; then
		local size="$(stat --format "%s" ${REDIS_DUMP_DIR}/${REDIS_DUMP_FILE})"
		if [ "$?" -eq "0" ] && [ "$size" -eq "0" ]; then
			ocf_log notice "Detected 0 byte ${REDIS_DUMP_FILE}, deleting zero length file to avoid start failure."
			rm -f ${REDIS_DUMP_DIR}/${REDIS_DUMP_FILE}
		fi
	fi

	ocf_log info "start: $REDIS_SERVER --daemonize yes --unixsocket '$REDIS_SOCKET' --pidfile '$REDIS_PIDFILE'"
	output="$(su "$REDIS_USER" -s /bin/sh -c "cd '$REDIS_RUNDIR'; exec '$REDIS_SERVER' '$REDIS_CONFIG' --daemonize yes --unixsocket '$REDIS_SOCKET' --pidfile '$REDIS_PIDFILE'" 2>&1)"

	while true; do
		# wait for redis to start
		typeset -A info
		while read line; do
			[[ "$line" == "#"* ]] && continue
			[[ "$line" != *":"* ]] && continue
			IFS=':' read -r key value <<< "$line"
			info[$key]="$value"
		done < <(redis_client info)

		if (( info[loading] == 0 )); then
			break
		elif (( info[loading] == 1 )); then
			sleep "${info[loading_eta_seconds]}"
		elif pidof "$REDIS_SERVER" >/dev/null; then
			# unknown error, but the process still exists.
			# This check is mainly because redis daemonizes before it starts listening, causing `redis-cli` to fail
			#   See https://github.com/antirez/redis/issues/2368
			# It's possible that the `pidof` will pick up a different redis, but in that case, the start operation will just time out
			sleep 1
		else
			check_output="$(check_dump_file)"
			ocf_log err "start: Unknown error waiting for redis to start. redis-check-dump output=${check_output//$'\n'/; }"
			return $OCF_ERR_GENERIC
		fi
	done

	ocf_is_ms && demote # pacemaker expects resources to start in slave mode

	monitor
	status=$?
	if (( status == OCF_SUCCESS )) || (( status == OCF_RUNNING_MASTER )); then
		return $OCF_SUCCESS
	fi

	check_output="$(check_dump_file)"
	ocf_log err "start: Unknown error starting redis. redis-server output=${output//$'\n'/; } redis-check-dump output=${check_output//$'\n'/; }"
	return $status
}

function stop() {
	monitor
	status=$?

	if (( status == OCF_NOT_RUNNING )); then
		ocf_log info "stop: redis is already stopped"
		crm_master_reboot -D
		return $OCF_SUCCESS
	fi

	pid="$(<"$REDIS_PIDFILE")"
	kill -TERM "$pid"

	while true; do
		simple_status
		status=$?
		if (( status == OCF_NOT_RUNNING )); then
			crm_master_reboot -D
			return $OCF_SUCCESS
		fi
		sleep 1
	done
}

function promote() {
	monitor
	status=$?

	if (( status == OCF_RUNNING_MASTER )); then
		ocf_log info "promote: Already running as master"
		set_master "$NODENAME"
		return $OCF_SUCCESS
	elif (( status != OCF_SUCCESS )); then
		ocf_log err "promote: Node is not running as a slave"
		return $OCF_ERR_GENERIC
	fi

	redis_client slaveof no one

	monitor
	status=$?
	if (( status == OCF_RUNNING_MASTER )); then
		set_master "$NODENAME"
		return $OCF_SUCCESS
	fi

	ocf_log err "promote: Unknown error while promoting to master (status=$status)"
	return $OCF_ERR_GENERIC
}

function demote() {
	local master_host
	local master_port

	CHECK_SLAVE_STATE=1
	monitor
	status=$?

	if (( status == OCF_SUCCESS )); then
		ocf_log info "demote: Already running as slave"
		return $OCF_SUCCESS
	elif (( status == OCF_NOT_RUNNING )); then
		ocf_log err "demote: Failed to demote, redis not running."
		return $OCF_NOT_RUNNING
	fi

	master_host="$(last_known_master)"
	master_port="${REDIS_REPLICATION_PORT}"

	# The elected master has to remain a slave during startup.
	# During this period a placeholder master host is assigned.
	if [ -z "$master_host" ] || [[ "$master_host" == "$NODENAME" ]]; then
		CHECK_SLAVE_STATE=0
		master_host="no-such-master"
	elif ! master_is_active; then
		# no master has been promoted yet. we'll be notified when the
		# master starts.
		CHECK_SLAVE_STATE=0
		master_host="no-such-master"
	fi

	ocf_log info "demote: Setting master to '$master_host'"

	redis_client slaveof "$master_host" "$master_port"

	# Wait forever for the slave to connect to the master and finish the 
	# sync. Timeout is controlled by Pacemaker "op start timeout=XX".
	#
	# hint: redis master_link_status will only come "up" when 
	#       the SYNC with the master has completed.
	#       This can take an arbitraty time (data) and should 
	#       only be parametrized by the start operation timeout
	#	by the administrator, not by this resource agent code
	while true; do 
		# Wait infinite if replication is syncing
		# Then start/demote operation timeout determines timeout
		monitor
		status=$?
		if (( status == OCF_SUCCESS )); then
			return $OCF_SUCCESS
		fi
		
		sleep 1
	done

	ocf_log err "demote: Unexpected error setting slave mode (status=$status)"
	return $OCF_ERR_GENERIC
}

function notify() {
	mode="${OCF_RESKEY_CRM_meta_notify_type}-${OCF_RESKEY_CRM_meta_notify_operation}"
	case "$mode" in
		post-demote|post-promote) # change the master
			monitor
			status=$?
			if (( status == OCF_SUCCESS )); then # were a slave
				# calling demote updates the slave's connection
				# to the newly appointed Master instance.
				demote
			fi
			;;
	esac
	return $OCF_SUCCESS
}

function validate() {
	if [[ -x "$REDIS_SERVER" ]]; then
		ocf_log err "validate: $REDIS_SERVER does not exist or is not executable"
		return $OCF_ERR_INSTALLED
	fi
	if [[ -x "$REDIS_CLIENT" ]]; then
		ocf_log err "validate: $REDIS_CLIENT does not exist or is not executable"
		return $OCF_ERR_INSTALLED
	fi
	if [[ -f "$REDIS_CONFIG" ]]; then
		ocf_log err "validate: $REDIS_CONFIG does not exist"
		return $OCF_ERR_CONFIGURED
	fi
	if !  getent passwd "$REDIS_USER" &>/dev/null; then
		ocf_log err "validate: $REDIS_USER is not a valid user"
		return $OCF_ERR_CONFIGURED
	fi
}

NODENAME=$(ocf_local_nodename)
if [ -f "$REDIS_CONFIG" ]; then
	clientpasswd="$(cat $REDIS_CONFIG | sed -n -e  's/^\s*requirepass\s*\(.*\)\s*$/\1/p' | tail -n 1)"
fi

ocf_log debug "action=${1:-$__OCF_ACTION} notify_type=${OCF_RESKEY_CRM_meta_notify_type} notify_operation=${OCF_RESKEY_CRM_meta_notify_operation} master_host=${OCF_RESKEY_CRM_meta_notify_master_uname} slave_host=${OCF_RESKEY_CRM_meta_notify_slave_uname} promote_host=${OCF_RESKEY_CRM_meta_notify_promote_uname} demote_host=${OCF_RESKEY_CRM_meta_notify_demote_uname}; params: bin=${OCF_RESKEY_bin} client_bin=${OCF_RESKEY_client_bin} config=${OCF_RESKEY_config} user=${OCF_RESKEY_user} rundir=${OCF_RESKEY_rundir} port=${OCF_RESKEY_port}"

case "${1:-$__OCF_ACTION}" in
	status|monitor)
		monitor
		;;
	start)
		start
		;;
	stop)
		stop
		;;
	restart)
		stop && start
		;;
	promote)
		promote
		;;
	demote)
		demote
		;;
	notify)
		notify
		;;
	meta-data)
		meta_data
		;;
	validate-all)
		validate
		;;
	*)
		echo "Usage: $0 {monitor|start|stop|restart|promote|demote|notify|validate-all|meta-data}"
		exit $OCF_ERR_UNIMPLEMENTED
		;;
esac
status=$?
ocf_log debug "exit_status=$status"
exit $status