diff --git a/heartbeat/galera b/heartbeat/galera index 4a313e24b..ba3de4b81 100755 --- a/heartbeat/galera +++ b/heartbeat/galera @@ -1,1023 +1,1030 @@ #!/bin/sh # # Copyright (c) 2014 David Vossel # All Rights Reserved. # # This program is free software; you can redistribute it and/or modify # it under the terms of version 2 of the GNU General Public License as # published by the Free Software Foundation. # # This program is distributed in the hope that it would be useful, but # WITHOUT ANY WARRANTY; without even the implied warranty of # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. # # Further, this software is distributed without any warranty that it is # free of the rightful claim of any third person regarding infringement # or the like. Any license provided herein, whether implied or # otherwise, applies only to this software file. Patent licenses, if # any, provided herein do not apply to combinations of this program with # other software, or any other product whatsoever. # # You should have received a copy of the GNU General Public License # along with this program; if not, write the Free Software Foundation, # Inc., 59 Temple Place - Suite 330, Boston MA 02111-1307, USA. # ## # README. # # This agent only supports being configured as a multistate Master # resource. # # Slave vs Master role: # # During the 'Slave' role, galera instances are in read-only mode and # will not attempt to connect to the cluster. This role exists only as # a means to determine which galera instance is the most up-to-date. The # most up-to-date node will be used to bootstrap a galera cluster that # has no current members. # # The galera instances will only begin to be promoted to the Master role # once all the nodes in the 'wsrep_cluster_address' connection address # have entered read-only mode. At that point the node containing the # database that is most current will be promoted to Master. Once the first # Master instance bootstraps the galera cluster, the other nodes will be # promoted to Master as well. # # Example: Create a galera cluster using nodes rhel7-node1 rhel7-node2 rhel7-node3 # # pcs resource create db galera enable_creation=true \ # wsrep_cluster_address="gcomm://rhel7-auto1,rhel7-auto2,rhel7-auto3" meta master-max=3 --master # # By setting the 'enable_creation' option, the database will be automatically # generated at startup. The meta attribute 'master-max=3' means that all 3 # nodes listed in the wsrep_cluster_address list will be allowed to connect # to the galera cluster and perform replication. # # NOTE: If you have more nodes in the pacemaker cluster then you wish # to have in the galera cluster, make sure to use location contraints to prevent # pacemaker from attempting to place a galera instance on a node that is # not in the 'wsrep_cluster_address" list. # ## ####################################################################### # Initialization: : ${OCF_FUNCTIONS_DIR=${OCF_ROOT}/lib/heartbeat} . ${OCF_FUNCTIONS_DIR}/ocf-shellfuncs if [ "$__OCF_ACTION" != "meta-data" ]; then . ${OCF_FUNCTIONS_DIR}/mysql-common.sh NODENAME=$(ocf_attribute_target) fi # It is common for some galera instances to store # check user that can be used to query status # in this file if [ -f "/etc/sysconfig/clustercheck" ]; then . /etc/sysconfig/clustercheck elif [ -f "/etc/default/clustercheck" ]; then . /etc/default/clustercheck fi # Parameter defaults OCF_RESKEY_wsrep_cluster_address_default="" OCF_RESKEY_cluster_host_map_default="" OCF_RESKEY_check_user_default="root" OCF_RESKEY_check_passwd_default="" : ${OCF_RESKEY_wsrep_cluster_address=${OCF_RESKEY_wsrep_cluster_address_default}} : ${OCF_RESKEY_cluster_host_map=${OCF_RESKEY_cluster_host_map_default}} : ${OCF_RESKEY_check_user=${OCF_RESKEY_check_user_default}} : ${OCF_RESKEY_check_passwd=${OCF_RESKEY_check_passwd_default}} ####################################################################### # Defaults: OCF_RESKEY_check_passwd_use_empty_default=0 : ${OCF_RESKEY_check_passwd_use_empty=${OCF_RESKEY_check_passwd_use_empty_default}} ####################################################################### usage() { cat < 1.0 Resource script for managing galera database. Manages a galera instance Location of the MySQL server binary MySQL server binary Location of the MySQL client binary MySQL client binary Configuration file MySQL config Directory containing databases MySQL datadir User running MySQL daemon MySQL user Group running MySQL daemon (for logfile and directory permissions) MySQL group The logfile to be used for mysqld. MySQL log file The pidfile to be used for mysqld. MySQL pid file The socket to be used for mysqld. MySQL socket If the MySQL database does not exist, it will be created Create the database if it does not exist Additional parameters which are passed to the mysqld on startup. (e.g. --skip-external-locking or --skip-grant-tables) Additional parameters to pass to mysqld The galera cluster address. This takes the form of: gcomm://node,node,node Only nodes present in this node list will be allowed to start a galera instance. The galera node names listed in this address are expected to match valid pacemaker node names. If both names need to differ, you must provide a mapping in option cluster_host_map. Galera cluster address A mapping of pacemaker node names to galera node names. To be used when both pacemaker and galera names need to differ, (e.g. when galera names map to IP from a specific network interface) This takes the form of: pcmk1:node.1.galera;pcmk2:node.2.galera;pcmk3:node.3.galera where the galera resource started on node pcmk1 would be named node.1.galera in the wsrep_cluster_address Pacemaker to Galera name mapping Cluster check user. MySQL test user Cluster check user password. Empty passwords are ignored unless the parameter "check_passwd_use_empty" is set to 1. check password Use an empty "check_passwd" password. If this parameter is set to 1, "check_passwd" will be ignored and an empty password is used when calling the "mysql" client binary. check password use empty END } get_option_variable() { local key=$1 $MYSQL $MYSQL_OPTIONS_CHECK -e "SHOW VARIABLES like '$key';" | tail -1 } get_status_variable() { local key=$1 $MYSQL $MYSQL_OPTIONS_CHECK -e "show status like '$key';" | tail -1 } set_bootstrap_node() { local node=$(ocf_attribute_target $1) ${HA_SBIN_DIR}/crm_attribute -N $node -l reboot --name "${INSTANCE_ATTR_NAME}-bootstrap" -v "true" } clear_bootstrap_node() { ${HA_SBIN_DIR}/crm_attribute -N $NODENAME -l reboot --name "${INSTANCE_ATTR_NAME}-bootstrap" -D } is_bootstrap() { ${HA_SBIN_DIR}/crm_attribute -N $NODENAME -l reboot --name "${INSTANCE_ATTR_NAME}-bootstrap" --quiet 2>/dev/null } set_no_grastate() { ${HA_SBIN_DIR}/crm_attribute -N $NODENAME -l reboot --name "${INSTANCE_ATTR_NAME}-no-grastate" -v "true" } clear_no_grastate() { ${HA_SBIN_DIR}/crm_attribute -N $NODENAME -l reboot --name "${INSTANCE_ATTR_NAME}-no-grastate" -D } is_no_grastate() { local node=$(ocf_attribute_target $1) ${HA_SBIN_DIR}/crm_attribute -N $node -l reboot --name "${INSTANCE_ATTR_NAME}-no-grastate" --quiet 2>/dev/null } clear_last_commit() { ${HA_SBIN_DIR}/crm_attribute -N $NODENAME -l reboot --name "${INSTANCE_ATTR_NAME}-last-committed" -D } set_last_commit() { ${HA_SBIN_DIR}/crm_attribute -N $NODENAME -l reboot --name "${INSTANCE_ATTR_NAME}-last-committed" -v $1 } get_last_commit() { local node=$(ocf_attribute_target $1) if [ -z "$node" ]; then ${HA_SBIN_DIR}/crm_attribute -N $NODENAME -l reboot --name "${INSTANCE_ATTR_NAME}-last-committed" --quiet 2>/dev/null else ${HA_SBIN_DIR}/crm_attribute -N $node -l reboot --name "${INSTANCE_ATTR_NAME}-last-committed" --quiet 2>/dev/null fi } clear_safe_to_bootstrap() { ${HA_SBIN_DIR}/crm_attribute -N $NODENAME -l reboot --name "${INSTANCE_ATTR_NAME}-safe-to-bootstrap" -D } set_safe_to_bootstrap() { ${HA_SBIN_DIR}/crm_attribute -N $NODENAME -l reboot --name "${INSTANCE_ATTR_NAME}-safe-to-bootstrap" -v $1 } get_safe_to_bootstrap() { local node=$(ocf_attribute_target $1) if [ -z "$node" ]; then ${HA_SBIN_DIR}/crm_attribute -N $NODENAME -l reboot --name "${INSTANCE_ATTR_NAME}-safe-to-bootstrap" --quiet 2>/dev/null else ${HA_SBIN_DIR}/crm_attribute -N $node -l reboot --name "${INSTANCE_ATTR_NAME}-safe-to-bootstrap" --quiet 2>/dev/null fi } wait_for_sync() { local state=$(get_status_variable "wsrep_local_state") ocf_log info "Waiting for database to sync with the cluster. " while [ "$state" != "4" ]; do sleep 1 state=$(get_status_variable "wsrep_local_state") done ocf_log info "Database synced." } is_primary() { cluster_status=$(get_status_variable "wsrep_cluster_status") if [ "$cluster_status" = "Primary" ]; then return 0 fi if [ -z "$cluster_status" ]; then ocf_exit_reason "Unable to retrieve wsrep_cluster_status, verify check_user '$OCF_RESKEY_check_user' has permissions to view status" else ocf_log info "Galera instance wsrep_cluster_status=${cluster_status}" fi return 1 } is_readonly() { local res=$(get_option_variable "read_only") if ! ocf_is_true "$res"; then return 1 fi cluster_status=$(get_status_variable "wsrep_cluster_status") if ! [ "$cluster_status" = "Disconnected" ]; then return 1 fi return 0 } master_exists() { if [ "$__OCF_ACTION" = "demote" ]; then # We don't want to detect master instances during demote. # 1. we could be detecting ourselves as being master, which is no longer the case. # 2. we could be detecting other master instances that are in the process of shutting down. # by not detecting other master instances in "demote" we are deferring this check # to the next recurring monitor operation which will be much more accurate return 1 fi # determine if a master instance is already up and is healthy - ${HA_SBIN_DIR}/crm_mon --as-xml | grep "resource.*id=\"${INSTANCE_ATTR_NAME}\".*role=\"Master\".*active=\"true\".*orphaned=\"false\".*failed=\"false\"" > /dev/null 2>&1 + ocf_version_cmp "$OCF_RESKEY_crm_feature_set" "3.1.0" + res=$? + if [ -z "$OCF_RESKEY_crm_feature_set" ] || [ $res -eq 2 ]; then + XMLOPT="--output-as=xml" + else + XMLOPT="--as-xml" + fi + ${HA_SBIN_DIR}/crm_mon -1 $XMLOPT | grep -q -i -E "resource.*id=\"${INSTANCE_ATTR_NAME}\".*role=\"(Promoted|Master)\".*active=\"true\".*orphaned=\"false\".*failed=\"false\"" return $? } clear_master_score() { local node=$(ocf_attribute_target $1) if [ -z "$node" ]; then $CRM_MASTER -D else $CRM_MASTER -D -N $node fi } set_master_score() { local node=$(ocf_attribute_target $1) if [ -z "$node" ]; then $CRM_MASTER -v 100 else $CRM_MASTER -N $node -v 100 fi } promote_everyone() { for node in $(echo "$OCF_RESKEY_wsrep_cluster_address" | sed 's/gcomm:\/\///g' | tr -d ' ' | tr -s ',' ' '); do local pcmk_node=$(galera_to_pcmk_name $node) if [ -z "$pcmk_node" ]; then ocf_log err "Could not determine pacemaker node from galera name <${node}>." return else node=$pcmk_node fi set_master_score $node done } greater_than_equal_long() { # there are values we need to compare in this script # that are too large for shell -gt to process echo | awk -v n1="$1" -v n2="$2" '{if (n1>=n2) printf ("true"); else printf ("false");}' | grep -q "true" } galera_to_pcmk_name() { local galera=$1 if [ -z "$OCF_RESKEY_cluster_host_map" ]; then echo $galera else echo "$OCF_RESKEY_cluster_host_map" | tr ';' '\n' | tr -d ' ' | sed 's/:/ /' | awk -F' ' '$2=="'"$galera"'" {print $1;exit}' fi } pcmk_to_galera_name() { local pcmk=$1 if [ -z "$OCF_RESKEY_cluster_host_map" ]; then echo $pcmk else echo "$OCF_RESKEY_cluster_host_map" | tr ';' '\n' | tr -d ' ' | sed 's/:/ /' | awk -F' ' '$1=="'"$pcmk"'" {print $2;exit}' fi } detect_first_master() { local best_commit=0 local last_commit=0 local missing_nodes=0 local nodes="" local nodes_recovered="" local all_nodes local best_node_gcomm local best_node local safe_to_bootstrap all_nodes=$(echo "$OCF_RESKEY_wsrep_cluster_address" | sed 's/gcomm:\/\///g' | tr -d ' ' | tr -s ',' ' ') best_node_gcomm=$(echo "$all_nodes" | sed 's/^.* \(.*\)$/\1/') best_node=$(galera_to_pcmk_name $best_node_gcomm) if [ -z "$best_node" ]; then ocf_log err "Could not determine initial best node from galera name <${best_node_gcomm}>." return fi # avoid selecting a recovered node as bootstrap if possible for node in $all_nodes; do local pcmk_node=$(galera_to_pcmk_name $node) if [ -z "$pcmk_node" ]; then ocf_log err "Could not determine pacemaker node from galera name <${node}>." return else node=$pcmk_node fi if is_no_grastate $node; then nodes_recovered="$nodes_recovered $node" else nodes="$nodes $node" fi done for node in $nodes_recovered $nodes; do safe_to_bootstrap=$(get_safe_to_bootstrap $node) if [ "$safe_to_bootstrap" = "1" ]; then # Galera marked the node as safe to boostrap during shutdown. Let's just # pick it as our bootstrap node. ocf_log info "Node <${node}> is marked as safe to bootstrap." best_node=$node # We don't need to wait for the other nodes to report state in this case missing_nodes=0 break fi last_commit=$(get_last_commit $node) if [ -z "$last_commit" ]; then ocf_log info "Waiting on node <${node}> to report database status before Master instances can start." missing_nodes=1 continue fi # this means -1, or that no commit has occured yet. if [ "$last_commit" = "18446744073709551615" ]; then last_commit="0" fi greater_than_equal_long "$last_commit" "$best_commit" if [ $? -eq 0 ]; then best_node=$(ocf_attribute_target $node) best_commit=$last_commit fi done if [ $missing_nodes -eq 1 ]; then return fi ocf_log info "Promoting $best_node to be our bootstrap node" set_master_score $best_node set_bootstrap_node $best_node } detect_safe_to_bootstrap() { local safe_to_bootstrap="" local uuid="" local seqno="" if [ -f ${OCF_RESKEY_datadir}/grastate.dat ]; then ocf_log info "attempting to read safe_to_bootstrap flag from ${OCF_RESKEY_datadir}/grastate.dat" safe_to_bootstrap=$(sed -n 's/^safe_to_bootstrap:\s*\(.*\)$/\1/p' < ${OCF_RESKEY_datadir}/grastate.dat) uuid=$(sed -n 's/^uuid:\s*\(.*\)$/\1/p' < ${OCF_RESKEY_datadir}/grastate.dat) seqno=$(sed -n 's/^seqno:\s*\(.*\)$/\1/p' < ${OCF_RESKEY_datadir}/grastate.dat) fi if [ -z "$uuid" ] || [ -z "$seqno" ] || \ [ "$uuid" = "00000000-0000-0000-0000-000000000000" ] || \ [ "$seqno" = "-1" ]; then clear_safe_to_bootstrap return fi if [ "$safe_to_bootstrap" = "1" ] || [ "$safe_to_bootstrap" = "0" ]; then set_safe_to_bootstrap $safe_to_bootstrap else clear_safe_to_bootstrap fi } detect_last_commit() { local last_commit local recover_args="--defaults-file=$OCF_RESKEY_config \ --pid-file=$OCF_RESKEY_pid \ --socket=$OCF_RESKEY_socket \ --datadir=$OCF_RESKEY_datadir" local recovery_file_regex='s/.*WSREP\:.*position\s*recovery.*--log_error='\''\([^'\'']*\)'\''.*/\1/p' local recovered_position_regex='s/.*WSREP\:\s*[R|r]ecovered\s*position.*\:\(.*\)\s*$/\1/p' # codership/galera#354 # Some ungraceful shutdowns can leave an empty gvwstate.dat on # disk. This will prevent galera to join the cluster if it is # configured to attempt PC recovery. Removing that file makes the # node fall back to the normal, unoptimized joining process. if [ -f ${OCF_RESKEY_datadir}/gvwstate.dat ] && \ [ ! -s ${OCF_RESKEY_datadir}/gvwstate.dat ]; then ocf_log warn "empty ${OCF_RESKEY_datadir}/gvwstate.dat detected, removing it to prevent PC recovery failure at next restart" rm -f ${OCF_RESKEY_datadir}/gvwstate.dat fi ocf_log info "attempting to detect last commit version by reading ${OCF_RESKEY_datadir}/grastate.dat" last_commit="$(cat ${OCF_RESKEY_datadir}/grastate.dat | sed -n 's/^seqno.\s*\(.*\)\s*$/\1/p')" if [ -z "$last_commit" ] || [ "$last_commit" = "-1" ]; then local tmp=$(mktemp) chown $OCF_RESKEY_user:$OCF_RESKEY_group $tmp # if we pass here because grastate.dat doesn't exist, # try not to bootstrap from this node if possible if [ ! -f ${OCF_RESKEY_datadir}/grastate.dat ]; then set_no_grastate fi ocf_log info "now attempting to detect last commit version using 'mysqld_safe --wsrep-recover'" $SU - $OCF_RESKEY_user -s /bin/sh -c \ "${OCF_RESKEY_binary} $recover_args --wsrep-recover --log-error=$tmp 2>/dev/null" last_commit="$(cat $tmp | sed -n $recovered_position_regex | tail -1)" if [ -z "$last_commit" ]; then # Galera uses InnoDB's 2pc transactions internally. If # server was stopped in the middle of a replication, the # recovery may find a "prepared" XA transaction in the # redo log, and mysql won't recover automatically local recovery_file="$(cat $tmp | sed -n $recovery_file_regex)" if [ -e $recovery_file ]; then cat $recovery_file | grep -q -E '\[ERROR\]\s+Found\s+[0-9]+\s+prepared\s+transactions!' 2>/dev/null if [ $? -eq 0 ]; then # we can only rollback the transaction, but that's OK # since the DB will get resynchronized anyway ocf_log warn "local node <${NODENAME}> was not shutdown properly. Rollback stuck transaction with --tc-heuristic-recover" $SU - $OCF_RESKEY_user -s /bin/sh -c \ "${OCF_RESKEY_binary} $recover_args --wsrep-recover \ --tc-heuristic-recover=rollback --log-error=$tmp 2>/dev/null" last_commit="$(cat $tmp | sed -n $recovered_position_regex | tail -1)" if [ ! -z "$last_commit" ]; then ocf_log warn "State recovered. force SST at next restart for full resynchronization" rm -f ${OCF_RESKEY_datadir}/grastate.dat # try not to bootstrap from this node if possible set_no_grastate fi fi fi fi rm -f $tmp fi if [ ! -z "$last_commit" ]; then ocf_log info "Last commit version found: $last_commit" set_last_commit $last_commit return $OCF_SUCCESS else ocf_exit_reason "Unable to detect last known write sequence number" clear_last_commit return $OCF_ERR_GENERIC fi } # For galera, promote is really start galera_promote() { local rc local extra_opts local bootstrap local safe_to_bootstrap master_exists if [ $? -eq 0 ]; then # join without bootstrapping extra_opts="--wsrep-cluster-address=${OCF_RESKEY_wsrep_cluster_address}" else bootstrap=$(is_bootstrap) if ocf_is_true $bootstrap; then # The best node for bootstrapping wasn't cleanly shutdown. Allow # bootstrapping anyways if [ "$(get_safe_to_bootstrap)" = "0" ]; then sed -ie 's/^\(safe_to_bootstrap:\) 0/\1 1/' ${OCF_RESKEY_datadir}/grastate.dat ocf_log info "safe_to_bootstrap in ${OCF_RESKEY_datadir}/grastate.dat set to 1 on node ${NODENAME}" fi ocf_log info "Node <${NODENAME}> is bootstrapping the cluster" extra_opts="--wsrep-cluster-address=gcomm://" else ocf_exit_reason "Failure, Attempted to promote Master instance of $OCF_RESOURCE_INSTANCE before bootstrap node has been detected." clear_last_commit return $OCF_ERR_GENERIC fi fi galera_monitor if [ $? -eq $OCF_RUNNING_MASTER ]; then if ocf_is_true $bootstrap; then promote_everyone clear_bootstrap_node ocf_log info "boostrap node already up, promoting the rest of the galera instances." fi clear_safe_to_bootstrap clear_last_commit return $OCF_SUCCESS fi # last commit/safe_to_bootstrap flag are no longer relevant once promoted clear_last_commit clear_safe_to_bootstrap mysql_common_prepare_dirs mysql_common_start "$extra_opts" rc=$? if [ $rc != $OCF_SUCCESS ]; then return $rc fi galera_monitor rc=$? if [ $rc != $OCF_SUCCESS -a $rc != $OCF_RUNNING_MASTER ]; then ocf_exit_reason "Failed initial monitor action" return $rc fi is_readonly if [ $? -eq 0 ]; then ocf_exit_reason "Failure. Master instance started in read-only mode, check configuration." return $OCF_ERR_GENERIC fi is_primary if [ $? -ne 0 ]; then ocf_exit_reason "Failure. Master instance started, but is not in Primary mode." return $OCF_ERR_GENERIC fi if ocf_is_true $bootstrap; then promote_everyone clear_bootstrap_node # clear attribute no-grastate. if last shutdown was # not clean, we cannot be extra-cautious by requesting a SST # since this is the bootstrap node clear_no_grastate ocf_log info "Bootstrap complete, promoting the rest of the galera instances." else # if this is not the bootstrap node, make sure this instance # syncs with the rest of the cluster before promotion returns. wait_for_sync # sync is done, clear info about last startup clear_no_grastate fi ocf_log info "Galera started" return $OCF_SUCCESS } galera_demote() { mysql_common_stop rc=$? if [ $rc -ne $OCF_SUCCESS ] && [ $rc -ne $OCF_NOT_RUNNING ]; then ocf_exit_reason "Failed to stop Master galera instance during demotion to Master" return $rc fi # if this node was previously a bootstrap node, that is no longer the case. clear_bootstrap_node clear_last_commit clear_no_grastate clear_safe_to_bootstrap # Clear master score here rather than letting pacemaker do so once # demote finishes. This way a promote cannot take place right # after this demote even if pacemaker is requested to do so. It # will first have to run a start/monitor op, to reprobe the state # of the other galera nodes and act accordingly. clear_master_score # record last commit for next promotion detect_safe_to_bootstrap detect_last_commit rc=$? return $rc } galera_start() { local rc local galera_node galera_node=$(pcmk_to_galera_name $NODENAME) if [ -z "$galera_node" ]; then ocf_exit_reason "Could not determine galera name from pacemaker node <${NODENAME}>." return $OCF_ERR_CONFIGURED fi echo $OCF_RESKEY_wsrep_cluster_address | grep -q -F $galera_node if [ $? -ne 0 ]; then ocf_exit_reason "local node <${NODENAME}> (galera node <${galera_node}>) must be a member of the wsrep_cluster_address <${OCF_RESKEY_wsrep_cluster_address}> to start this galera instance" return $OCF_ERR_CONFIGURED fi galera_monitor if [ $? -eq $OCF_RUNNING_MASTER ]; then ocf_exit_reason "master galera instance started outside of the cluster's control" return $OCF_ERR_GENERIC fi mysql_common_prepare_dirs detect_safe_to_bootstrap detect_last_commit rc=$? if [ $rc -ne $OCF_SUCCESS ]; then return $rc fi master_exists if [ $? -eq 0 ]; then ocf_log info "Master instances are already up, setting master score so this instance will join galera cluster." set_master_score $NODENAME else clear_master_score detect_first_master fi return $OCF_SUCCESS } galera_monitor() { local rc local galera_node local status_loglevel="err" # Set loglevel to info during probe if ocf_is_probe; then status_loglevel="info" fi mysql_common_status $status_loglevel rc=$? if [ $rc -eq $OCF_NOT_RUNNING ]; then last_commit=$(get_last_commit $node) if [ -n "$last_commit" ]; then # if last commit is set, this instance is considered started in slave mode rc=$OCF_SUCCESS master_exists if [ $? -ne 0 ]; then detect_first_master else # a master instance exists and is healthy, promote this # local read only instance # so it can join the master galera cluster. set_master_score fi fi return $rc elif [ $rc -ne $OCF_SUCCESS ]; then return $rc fi # if we make it here, mysql is running. Check cluster status now. galera_node=$(pcmk_to_galera_name $NODENAME) if [ -z "$galera_node" ]; then ocf_exit_reason "Could not determine galera name from pacemaker node <${NODENAME}>." return $OCF_ERR_CONFIGURED fi echo $OCF_RESKEY_wsrep_cluster_address | grep -q -F $galera_node if [ $? -ne 0 ]; then ocf_exit_reason "local node <${NODENAME}> (galera node <${galera_node}>) is started, but is not a member of the wsrep_cluster_address <${OCF_RESKEY_wsrep_cluster_address}>" return $OCF_ERR_GENERIC fi is_primary if [ $? -eq 0 ]; then if ocf_is_probe; then # restore master score during probe # if we detect this is a master instance set_master_score fi rc=$OCF_RUNNING_MASTER else ocf_exit_reason "local node <${NODENAME}> is started, but not in primary mode. Unknown state." rc=$OCF_ERR_GENERIC fi return $rc } galera_stop() { local rc # make sure the process is stopped mysql_common_stop rc=$1 clear_safe_to_bootstrap clear_last_commit clear_master_score clear_bootstrap_node clear_no_grastate return $rc } galera_validate() { if ! ocf_is_ms; then ocf_exit_reason "Galera must be configured as a multistate Master/Slave resource." return $OCF_ERR_CONFIGURED fi if [ -z "$OCF_RESKEY_wsrep_cluster_address" ]; then ocf_exit_reason "Galera must be configured with a wsrep_cluster_address value." return $OCF_ERR_CONFIGURED fi mysql_common_validate } case "$1" in meta-data) meta_data exit $OCF_SUCCESS;; usage|help) usage exit $OCF_SUCCESS;; esac galera_validate rc=$? LSB_STATUS_STOPPED=3 if [ $rc -ne 0 ]; then case "$1" in stop) exit $OCF_SUCCESS;; monitor) exit $OCF_NOT_RUNNING;; status) exit $LSB_STATUS_STOPPED;; *) exit $rc;; esac fi if [ -z "${OCF_RESKEY_check_passwd}" ]; then # This value is automatically sourced from /etc/sysconfig/checkcluster if available OCF_RESKEY_check_passwd=${MYSQL_PASSWORD} fi if [ -z "${OCF_RESKEY_check_user}" ]; then # This value is automatically sourced from /etc/sysconfig/checkcluster if available OCF_RESKEY_check_user=${MYSQL_USERNAME} fi : ${OCF_RESKEY_check_user="root"} MYSQL_OPTIONS_CHECK="-nNE --user=${OCF_RESKEY_check_user}" if ocf_is_true "${OCF_RESKEY_check_passwd_use_empty}"; then MYSQL_OPTIONS_CHECK="$MYSQL_OPTIONS_CHECK --password=" elif [ -n "${OCF_RESKEY_check_passwd}" ]; then MYSQL_OPTIONS_CHECK="$MYSQL_OPTIONS_CHECK --password=${OCF_RESKEY_check_passwd}" fi # This value is automatically sourced from /etc/sysconfig/checkcluster if available if [ -n "${MYSQL_HOST}" ]; then MYSQL_OPTIONS_CHECK="$MYSQL_OPTIONS_CHECK -h ${MYSQL_HOST}" fi # This value is automatically sourced from /etc/sysconfig/checkcluster if available if [ -n "${MYSQL_PORT}" ]; then MYSQL_OPTIONS_CHECK="$MYSQL_OPTIONS_CHECK -P ${MYSQL_PORT}" fi # What kind of method was invoked? case "$1" in start) galera_start;; stop) galera_stop;; status) mysql_common_status err;; monitor) galera_monitor;; promote) galera_promote;; demote) galera_demote;; validate-all) exit $OCF_SUCCESS;; *) usage exit $OCF_ERR_UNIMPLEMENTED;; esac # vi:sw=4:ts=4:et: diff --git a/heartbeat/redis.in b/heartbeat/redis.in index da7230a49..7e534db4a 100755 --- a/heartbeat/redis.in +++ b/heartbeat/redis.in @@ -1,769 +1,776 @@ #!@BASH_SHELL@ # # Resource agent script for redis server. # # Copyright (c) 2013 Patrick Hemmer # All Rights Reserved. # # This program is free software; you can redistribute it and/or modify # it under the terms of version 2 of the GNU General Public License as # published by the Free Software Foundation. # # This program is distributed in the hope that it would be useful, but # WITHOUT ANY WARRANTY; without even the implied warranty of # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. # # Further, this software is distributed without any warranty that it is # free of the rightful claim of any third person regarding infringement # or the like. Any license provided herein, whether implied or # otherwise, applies only to this software file. Patent licenses, if # any, provided herein do not apply to combinations of this program with # other software, or any other product whatsoever. # # You should have received a copy of the GNU General Public License # along with this program; if not, write the Free Software Foundation, # Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA # ####################################################################### # Initialization: : ${OCF_FUNCTIONS_DIR=${OCF_ROOT}/lib/heartbeat} . ${OCF_FUNCTIONS_DIR}/ocf-shellfuncs # Parameter defaults OCF_RESKEY_bin_default="/usr/bin/redis-server" OCF_RESKEY_client_bin_default="/usr/bin/redis-cli" if [ -f "/etc/redis.conf" ]; then OCF_RESKEY_config_default="/etc/redis.conf" else OCF_RESKEY_config_default="/etc/redis/redis.conf" fi OCF_RESKEY_user_default="redis" OCF_RESKEY_rundir_default="/var/run/redis" OCF_RESKEY_pidfile_name_default="redis-server.pid" OCF_RESKEY_socket_name_default="redis.sock" OCF_RESKEY_port_default="6379" OCF_RESKEY_tunnel_host_default="127.0.0.1" OCF_RESKEY_tunnel_port_map_default="" OCF_RESKEY_wait_last_known_master_default="false" : ${OCF_RESKEY_bin=${OCF_RESKEY_bin_default}} : ${OCF_RESKEY_client_bin=${OCF_RESKEY_client_bin_default}} : ${OCF_RESKEY_config=${OCF_RESKEY_config_default}} : ${OCF_RESKEY_user=${OCF_RESKEY_user_default}} : ${OCF_RESKEY_rundir=${OCF_RESKEY_rundir_default}} : ${OCF_RESKEY_pidfile_name=${OCF_RESKEY_pidfile_name_default}} : ${OCF_RESKEY_socket_name=${OCF_RESKEY_socket_name_default}} : ${OCF_RESKEY_port=${OCF_RESKEY_port_default}} : ${OCF_RESKEY_tunnel_host=${OCF_RESKEY_tunnel_host_default}} : ${OCF_RESKEY_tunnel_port_map=${OCF_RESKEY_tunnel_port_map_default}} : ${OCF_RESKEY_wait_last_known_master=${OCF_RESKEY_wait_last_known_master_default}} 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 [ -z "$REDIS_CHECK_DUMP" ]; then REDIS_CHECK_DUMP="$(which redis-check-rdb 2>/dev/null)" fi if [ -r "$REDIS_CONFIG" ]; then REDIS_DUMP_DIR="$(grep "^\s*dir\s" < "$REDIS_CONFIG" | awk '{ print $2 }' 2>/dev/null)" REDIS_DUMP_FILE="$(grep "^\s*dbfilename\s" < "$REDIS_CONFIG" | awk '{ print $2 }' 2>/dev/null)" fi : ${REDIS_DUMP_DIR:=/var/lib/redis/} : ${REDIS_DUMP_FILE:=dump.rdb} redis_meta_data() { cat < 1.0 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. Redis server Path to \`redis-server\` Path to \`redis-server\` Path to \`redis-cli\` Path to \`redis-cli\` Path to 'redis.conf' Path to 'redis.conf' User to run redis as Redis user Directory to store socket and pid file in Redis var/run dir The filename to use for the pidfile. Will be created in the rundir. Should only be a basename, not a full path. Redis pidfile name The filename to use for the socket. Will be crated in the rundir. Should only be a basename, not a full path. Redis socket name Port for replication client to connect to on remote server Replication port When replication traffic is tunnelled, this is the host to target to forward outgoing traffic to the redis master. The resource agent configures the redis slave to target the master via tunnel_host:tunnel_port. Note that in order to enable replication traffic tunneling, parameter {tunnel_port_map} must be populated. Tunnel host for replication traffic A mapping of pacemaker node names to redis port number. To be used when redis servers need to tunnel replication traffic. On every node where the redis resource is running, the redis server listens to a different port. Each redis server can access its peers for replication traffic via a tunnel accessible at {tunnel_host}:port. The mapping the form of: pcmk1-name:port-for-redis1;pcmk2-name:port-for-redis2;pcmk3-name:port-for-redis3 where the redis resource started on node pcmk1-name would listen on port port-for-redis1 Mapping of Redis server name to redis port 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. Wait for last known master 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 ${INSTANCE_ATTR_NAME}_REPL_INFO -s redis_replication" MASTER_HOST="" MASTER_ACTIVE_CACHED="" MASTER_ACTIVE="" CLI_HAVE_AUTH_WARNING=0 CLI_HAVE_ARG_NO_AUTH_WARNING=0 CLI_HAVE_ENV_AUTH=0 redis_cli_features() { CLI_VER=$("$REDIS_CLIENT" -v | awk '{print $NF}') # Starting with 4.0.10 there is a warning on stderr when using a pass # Starting with 5.0.0 there is an argument to silence the warning: --no-auth-warning # Starting with 5.0.3 there is an option to use REDISCLI_AUTH evironment variable for password, no warning in this case ocf_version_cmp $CLI_VER 5.0.3 res=$? if [[ res -ge 1 ]]; then CLI_HAVE_ENV_AUTH=1 fi ocf_version_cmp $CLI_VER 5.0.0 res=$? if [[ res -ge 1 ]]; then CLI_HAVE_ARG_NO_AUTH_WARNING=1 fi ocf_version_cmp $CLI_VER 4.0.10 res=$? if [[ res -ge 1 ]]; then CLI_HAVE_AUTH_WARNING=1 fi } 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 + ocf_version_cmp "$OCF_RESKEY_crm_feature_set" "3.1.0" + res=$? + if [ -z "$OCF_RESKEY_crm_feature_set" ] || [ $res -eq 2 ]; then + XMLOPT="--output-as=xml" + else + XMLOPT="--as-xml" + fi + crm_mon -1 $XMLOPT | grep -q -i -E "resource.*id=\"${OCF_RESOURCE_INSTANCE}\".* role=\"(Promoted|Master)\".* active=\"true\".* orphaned=\"false\".* failed=\"false\"" MASTER_ACTIVE=$? MASTER_ACTIVE_CACHED="true" fi return $MASTER_ACTIVE } set_master() { MASTER_HOST="$1" ${CRM_ATTR_REPL_INFO} -v "$1" -q } last_known_master() { if [ -z "$MASTER_HOST" ]; then MASTER_HOST="$(${CRM_ATTR_REPL_INFO} --query -q 2>/dev/null)" fi echo "$MASTER_HOST" } crm_master_reboot() { local node node=$(ocf_attribute_target) "${HA_SBIN_DIR}/crm_master" -N "$node" -l reboot "$@" } 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" } set_score() { local score local last_master score="$1" if ocf_is_true "$OCF_RESKEY_wait_last_known_master" && ! master_is_active; then 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" } redis_client() { ocf_log debug "redis_client: '$REDIS_CLIENT' -s '$REDIS_SOCKET' $*" if [ -n "$clientpasswd" ]; then # Consider redis-cli features to choose optimal password passing method and warning filtering workaround if [[ CLI_HAVE_ENV_AUTH -eq 1 ]]; then REDISCLI_AUTH=$clientpasswd "$REDIS_CLIENT" -s "$REDIS_SOCKET" "$@" | sed 's/\r//' elif [[ CLI_HAVE_ARG_NO_AUTH_WARNING -eq 1 ]]; then "$REDIS_CLIENT" -s "$REDIS_SOCKET" --no-auth-warning -a "$clientpasswd" "$@" | sed 's/\r//' elif [[ CLI_HAVE_AUTH_WARNING -eq 1 ]]; then ("$REDIS_CLIENT" -s "$REDIS_SOCKET" -a "$clientpasswd" "$@" 2>&1 >&3 3>&- | grep -v "Using a password" >&2 3>&-) 3>&1 | sed 's/\r//' else "$REDIS_CLIENT" -s "$REDIS_SOCKET" -a "$clientpasswd" "$@" | sed 's/\r//' fi else "$REDIS_CLIENT" -s "$REDIS_SOCKET" "$@" | sed 's/\r//' fi } simple_status() { local pid if ! [ -f "$REDIS_PIDFILE" ]; then return $OCF_NOT_RUNNING fi pid="$(<"$REDIS_PIDFILE")" pidof $(basename "$REDIS_SERVER") | grep -q "\<$pid\>" || return $OCF_NOT_RUNNING ocf_log debug "monitor: redis-server running under pid $pid" return $OCF_SUCCESS } redis_monitor() { local res local master_name local last_known_master_port simple_status res=$? 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 -G --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]}" != "$(last_known_master)" ]]; then if [ -n "${OCF_RESKEY_tunnel_port_map}" ]; then master_name=$(port_to_redis_node ${info[master_port]}) last_known_master_port=$(redis_node_to_port $(last_known_master)) if [[ "${info[master_host]}" != "${OCF_RESKEY_tunnel_host}" ]] || [[ "${info[master_port]}" != "${last_known_master_port}" ]]; then ocf_log err "monitor: Slave mode current tunnelled connection to redis server does not match running master. tunnelled='${info[master_host]}:${info[master_port]} (${master_name})', running='$(last_known_master)'" return $OCF_ERR_GENERIC fi else ocf_log err "monitor: Slave mode current master does not match running master. current=${info[master_host]}, running=$(last_known_master)" return $OCF_ERR_GENERIC fi fi fi fi return $OCF_SUCCESS } redis_node_to_port() { local node=$1 echo "$OCF_RESKEY_tunnel_port_map" | tr ';' '\n' | tr -d ' ' | sed 's/:/ /' | awk -F' ' '$1=="'"$node"'" {print $2;exit}' } port_to_redis_node() { local port=$1 echo "$OCF_RESKEY_tunnel_port_map" | tr ';' '\n' | tr -d ' ' | sed 's/:/ /' | awk -F' ' '$2=="'"$port"'" {print $1;exit}' } get_tunnel_port_from_master() { local master_name=$1 crm_attribute --node "$master_name" -l forever --name ${INSTANCE_ATTR_NAME}-tunnel-port --query -q 2>/dev/null } get_master_from_tunnel_port() { local master_name=$1 crm_attribute --node "$master_name" -l forever --name ${INSTANCE_ATTR_NAME}-tunnel-port --query -q 2>/dev/null } check_dump_file() { if ! have_binary "$REDIS_CHECK_DUMP"; then return 0 fi $REDIS_CHECK_DUMP ${REDIS_DUMP_DIR}/${REDIS_DUMP_FILE} 2>&1 } redis_start() { local size redis_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 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 $(basename "$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 $OCF_ERR_GENERIC fi done while ! [ -s "$REDIS_PIDFILE" ]; do ocf_log debug "start: Waiting for pid file '$REDIS_PIDFILE' to appear" sleep 1 done ocf_is_ms && redis_demote # pacemaker expects resources to start in slave mode redis_monitor status=$? 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 $status } redis_stop() { redis_monitor status=$? 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 $OCF_SUCCESS fi sleep 1 done } redis_promote() { redis_monitor status=$? 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 $OCF_ERR_GENERIC fi redis_client slaveof no one redis_monitor status=$? 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 $OCF_ERR_GENERIC } redis_demote() { local master_host local master_port local tunnel_port # client kill is only supported in Redis 2.8.12 or greater version=$(redis_client -v | awk '{print $NF}') ocf_version_cmp "$version" "2.8.11" client_kill=$? CHECK_SLAVE_STATE=1 redis_monitor status=$? 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 if [ -n "${OCF_RESKEY_tunnel_port_map}" ]; then # master_host can be the special marker "no-such-master" # while a master is being selected. In this case, no # tunnel port is returned, but this is not fatal. tunnel_port=$(redis_node_to_port "$master_host") if [ -n "$tunnel_port" ]; then ocf_log info "demote: Setting master to '$master_host' via local tunnel '${OCF_RESKEY_tunnel_host}' on port '$tunnel_port'" master_host="${OCF_RESKEY_tunnel_host}" master_port="$tunnel_port" fi else ocf_log info "demote: Setting master to '$master_host'" fi 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 if [ "$client_kill" -eq 2 ]; then redis_client CLIENT PAUSE 2000 fi redis_monitor status=$? if (( status == OCF_SUCCESS )); then if [ "$client_kill" -eq 2 ]; then redis_client CLIENT KILL type normal fi return $OCF_SUCCESS fi sleep 1 done ocf_log err "demote: Unexpected error setting slave mode (status=$status)" return $OCF_ERR_GENERIC } redis_notify() { mode="${OCF_RESKEY_CRM_meta_notify_type}-${OCF_RESKEY_CRM_meta_notify_operation}" case "$mode" in post-demote|post-promote) # change the master redis_monitor status=$? if (( status == OCF_SUCCESS )); then # were a slave # calling demote updates the slave's connection # to the newly appointed Master instance. redis_demote fi ;; esac return $OCF_SUCCESS } redis_validate() { if [[ ! -x "$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 $OCF_ERR_CONFIGURED fi } if [ "$__OCF_ACTION" != "meta-data" ]; then NODENAME=$(ocf_attribute_target) fi if [ -r "$REDIS_CONFIG" ]; then clientpasswd="$(sed -n -e 's/^\s*requirepass\s*\(.*\)\s*$/\1/p' < $REDIS_CONFIG | tail -n 1)" fi if [ "$__OCF_ACTION" = "start" ]; then redis_validate || exit $? fi redis_cli_features ocf_log debug "action=${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) redis_monitor ;; start) redis_start ;; stop) redis_stop ;; restart) redis_stop && redis_start ;; promote) redis_promote ;; demote) redis_demote ;; notify) redis_notify ;; meta-data) redis_meta_data ;; validate-all) redis_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