#!@BASH_SHELL@
#
#
# MariaDB
#
# Description:  Manages a MariaDB Promotable database as Linux-HA resource
#
# Authors:  Alan Robertson:               DB2 Script
#           Jakub Janczak:                rewrite as MySQL
#           Andrew Beekhof:               cleanup and import
#           Sebastian Reitenbach:         add OpenBSD defaults, more cleanup
#           Narayan Newton:               add Gentoo/Debian defaults
#           Marian Marinov, Florian Haas: add replication capability
#           Yves Trudeau, Baron Schwartz: add VIP support and improve replication
#           Nils Carlson:                 add GTID support and semi-sync support
#
# Support:  users@clusterlabs.org
# License:  GNU General Public License (GPL)
#
# (c) 2002-2005 International Business Machines, Inc.
#     2005-2010 Linux-HA contributors
#
# See usage() function below for more details...
#
# OCF instance parameters:
#   OCF_RESKEY_binary
#   OCF_RESKEY_client_binary
#   OCF_RESKEY_config
#   OCF_RESKEY_datadir
#   OCF_RESKEY_user
#   OCF_RESKEY_group
#   OCF_RESKEY_node_list
#   OCF_RESKEY_test_table
#   OCF_RESKEY_test_user
#   OCF_RESKEY_test_passwd
#   OCF_RESKEY_enable_creation
#   OCF_RESKEY_additional_parameters
#   OCF_RESKEY_log
#   OCF_RESKEY_pid
#   OCF_RESKEY_socket
#   OCF_RESKEY_replication_user
#   OCF_RESKEY_replication_passwd
#   OCF_RESKEY_replication_port
#######################################################################
# Initialization:

OCF_RESKEY_node_list_default=""

: ${OCF_FUNCTIONS_DIR=${OCF_ROOT}/lib/heartbeat}
. ${OCF_FUNCTIONS_DIR}/ocf-shellfuncs
. ${OCF_FUNCTIONS_DIR}/mysql-common.sh
#######################################################################

usage() {
  cat <<UEND
usage: $0 (start|stop|validate-all|meta-data|monitor|promote|demote|notify)

$0 manages a MariaDB Database as an HA resource.

The 'start' operation starts the database.
The 'stop' operation stops the database.
The 'status' operation reports whether the database is running
The 'monitor' operation reports whether the database seems to be working
The 'promote' operation makes this mysql server run as promoted
The 'demote' operation makes this mysql server run as unpromoted
The 'validate-all' operation reports whether the parameters are valid

UEND
}

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

<longdesc lang="en">
Resource script for MariaDB.

Manages a complete promotable replication setup with GTID, for simpler
uses look at the mysql resource agent which supports older replication
forms which mysql and mariadb have in common.

The resource must be setup to use notifications. Set 'notify=true' in the metadata
attributes when defining a MariaDB promotable instance.

The default behavior is to use uname -n values in the change promoted to command.
Other IPs can be specified manually by adding a node attribute
\${INSTANCE_ATTR_NAME}_mysql_master_IP giving the IP to use for replication.
For example, if the mariadb primitive you are using is p_mariadb, the
attribute to set will be p_mariadb_mysql_master_IP.
</longdesc>
<shortdesc lang="en">Manages a MariaDB promotable instance</shortdesc>
<parameters>

<parameter name="binary" unique="0" required="0">
<longdesc lang="en">
Location of the MariaDB server binary
</longdesc>
<shortdesc lang="en">MariaDB server binary</shortdesc>
<content type="string" default="${OCF_RESKEY_binary_default}" />
</parameter>

<parameter name="client_binary" unique="0" required="0">
<longdesc lang="en">
Location of the MariaDB client binary
</longdesc>
<shortdesc lang="en">MariaDB client binary</shortdesc>
<content type="string" default="${OCF_RESKEY_client_binary_default}" />
</parameter>

<parameter name="config" unique="0" required="0">
<longdesc lang="en">
Configuration file
</longdesc>
<shortdesc lang="en">MariaDB config</shortdesc>
<content type="string" default="${OCF_RESKEY_config_default}" />
</parameter>

<parameter name="datadir" unique="0" required="0">
<longdesc lang="en">
Directory containing databases
</longdesc>
<shortdesc lang="en">MariaDB datadir</shortdesc>
<content type="string" default="${OCF_RESKEY_datadir_default}" />
</parameter>

<parameter name="user" unique="0" required="0">
<longdesc lang="en">
User running MariaDB daemon
</longdesc>
<shortdesc lang="en">MariaDB user</shortdesc>
<content type="string" default="${OCF_RESKEY_user_default}" />
</parameter>

<parameter name="group" unique="0" required="0">
<longdesc lang="en">
Group running MariaDB daemon (for logfile and directory permissions)
</longdesc>
<shortdesc lang="en">MariaDB group</shortdesc>
<content type="string" default="${OCF_RESKEY_group_default}"/>
</parameter>

<parameter name="log" unique="0" required="0">
<longdesc lang="en">
The logfile to be used for mysqld.
</longdesc>
<shortdesc lang="en">MariaDB log file</shortdesc>
<content type="string" default="${OCF_RESKEY_log_default}"/>
</parameter>

<parameter name="node_list" unique="0" required="1">
<longdesc lang="en">
All node names of nodes that will execute mariadb.
Please separate each node name with a space.
This is required for the promoted selection to function.
</longdesc>
<shortdesc lang="en">node list</shortdesc>
<content type="string" default="${OCF_RESKEY_node_list_default}" />
</parameter>

<parameter name="pid" unique="0" required="0">
<longdesc lang="en">
The pidfile to be used for mysqld.
</longdesc>
<shortdesc lang="en">MariaDB pid file</shortdesc>
<content type="string" default="${OCF_RESKEY_pid_default}"/>
</parameter>

<parameter name="socket" unique="0" required="0">
<longdesc lang="en">
The socket to be used for mysqld.
</longdesc>
<shortdesc lang="en">MariaDB socket</shortdesc>
<content type="string" default="${OCF_RESKEY_socket_default}"/>
</parameter>

<parameter name="test_table" unique="0" required="0">
<longdesc lang="en">
Table to be tested in monitor statement (in database.table notation)
</longdesc>
<shortdesc lang="en">MariaDB test table</shortdesc>
<content type="string" default="${OCF_RESKEY_test_table_default}" />
</parameter>

<parameter name="test_user" unique="0" required="0">
<longdesc lang="en">
MariaDB test user, must have select privilege on test_table
</longdesc>
<shortdesc lang="en">MariaDB test user</shortdesc>
<content type="string" default="${OCF_RESKEY_test_user_default}" />
</parameter>

<parameter name="test_passwd" unique="0" required="0">
<longdesc lang="en">
MariaDB test user password
</longdesc>
<shortdesc lang="en">MariaDB test user password</shortdesc>
<content type="string" default="${OCF_RESKEY_test_passwd_default}" />
</parameter>

<parameter name="enable_creation" unique="0" required="0">
<longdesc lang="en">
If the MariaDB database does not exist, it will be created
</longdesc>
<shortdesc lang="en">Create the database if it does not exist</shortdesc>
<content type="boolean" default="${OCF_RESKEY_enable_creation_default}"/>
</parameter>

<parameter name="additional_parameters" unique="0" required="0">
<longdesc lang="en">
Additional parameters which are passed to the mysqld on startup.
(e.g. --skip-external-locking or --skip-grant-tables)
</longdesc>
<shortdesc lang="en">Additional parameters to pass to mysqld</shortdesc>
<content type="string" default="${OCF_RESKEY_additional_parameters_default}"/>
</parameter>

<parameter name="replication_user" unique="0" required="0">
<longdesc lang="en">
MariaDB replication user. This user is used for starting and stopping
MariaDB replication, for setting and resetting the promoted host, and for
setting and unsetting read-only mode. Because of that, this user must
have SUPER, REPLICATION SLAVE, REPLICATION CLIENT, PROCESS and RELOAD
privileges on all nodes within the cluster. Mandatory if you define a
promotable resource.
</longdesc>
<shortdesc lang="en">MariaDB replication user</shortdesc>
<content type="string" default="${OCF_RESKEY_replication_user_default}" />
</parameter>

<parameter name="replication_passwd" unique="0" required="0">
<longdesc lang="en">
MariaDB replication password. Used for replication client and unpromoted.
Mandatory if you define a promotable resource.
</longdesc>
<shortdesc lang="en">MariaDB replication user password</shortdesc>
<content type="string" default="${OCF_RESKEY_replication_passwd_default}" />
</parameter>

<parameter name="replication_port" unique="0" required="0">
<longdesc lang="en">
The port on which the Promoted MariaDB instance is listening.
</longdesc>
<shortdesc lang="en">MariaDB replication port</shortdesc>
<content type="string" default="${OCF_RESKEY_replication_port_default}" />
</parameter>

</parameters>

<actions>
<action name="start" timeout="120s" />
<action name="stop" timeout="120s" />
<action name="status" timeout="60s" />
<action name="monitor" depth="0" timeout="30s" interval="20s" />
<action name="monitor" role="Promoted" depth="0" timeout="30s" interval="10s" />
<action name="monitor" role="Unpromoted" depth="0" timeout="30s" interval="30s" />
<action name="promote" timeout="120s" />
<action name="demote" timeout="120s" />
<action name="notify" timeout="90s" />
<action name="validate-all" timeout="5s" />
<action name="meta-data" timeout="5s" />
</actions>
</resource-agent>
END
}

# Convenience functions

greater_than_equal_long()
{
    # there are values we need to compare in this script
    # that are too large for shell -gt to process
    local true=$(echo "$1 > $2" | bc)
    if [ "$true" -eq "1" ]; then
        return 0
    else
        return 1
    fi
}

greater_than_gtid()
{
    local gtid1_transaction_id=$(echo $1 | cut -d - -f 3)
    local gtid2_transaction_id=$(echo $2 | cut -d - -f 3)

    greater_than_equal_long $gtid1_transaction_id $gtid2_transaction_id
    return $?
}

set_gtid() {
    # Sets the GTID in CIB using attrd_updater for this node.

    local gtid=$($MYSQL $MYSQL_OPTIONS_REPL \
                        -s -N -e "show global variables like 'gtid_current_pos'" | cut -f 2)

    # Ensure that we got somethine like a valid GTID
    if ! echo $gtid | grep -q '-'; then
        ocf_exit_reason "Unable to read GTID from MariaDB"
        ocf_log err "Unable to read GTID from MariaDB"
        return $OCF_ERR_GENERIC
    fi

    ${HA_SBIN_DIR}/attrd_updater -p -n ${OCF_RESOURCE_INSTANCE}-gtid -U $gtid
}

read_gtid() {
    local node=$1
    local query_result
    local name
    local host
    local value

    # This produces output of the form 'name="var-name" host="node2" value="val"'.
    # This should be set at this point, because we have store our own GTID previously.
    if ! query_result=$(${HA_SBIN_DIR}/attrd_updater -p -N $node -n ${OCF_RESOURCE_INSTANCE}-gtid -Q); then
        ocf_exit_reason "Unable to read GTID from attrd"
        ocf_log err "Unable to read GTID from attrd"
        echo ""
        return
    fi

    # Evaluate the query result to place the variables in the local scope.
    eval ${query_result}

    echo ${value}
}

clear_all_gtid() {
    for node in $OCF_RESKEY_node_list; do
        ${HA_SBIN_DIR}/attrd_updater -n ${OCF_RESOURCE_INSTANCE}-gtid -N $node -D
    done
}

set_waiting_for_first_master() {
    ${HA_SBIN_DIR}/attrd_updater -p -n ${OCF_RESOURCE_INSTANCE}-waiting-for-first-master -U true
}

waiting_for_first_master() {
    local query_result
    local name
    local host
    local value

    if ! query_result=$(${HA_SBIN_DIR}/attrd_updater -p -n ${OCF_RESOURCE_INSTANCE}-waiting-for-first-master -Q); then
        ocf_exit_reason "Unable to read waiting-for-first-master from attrd"
        ocf_log err "Unable to read waiting-for-first-master from attrd"
        return 1
    fi

    # Evaluate the query result to place the variables in the local scope.
    eval ${query_result}

    if [ "$value" = "true" ]; then
        return 0
    else
        return 1
    fi
}

clear_waiting_for_first_master() {
    attrd_updater -n ${OCF_RESOURCE_INSTANCE}-waiting-for-first-master -D
}

have_master_with_priority() {
    # Go through each node and validate that at least one has
    # a set priority. Because we unset the priority on reboot
    # a lack of priority indicates that we need to select a
    # new master.
    for node in $OCF_RESKEY_node_list; do
        ocf_promotion_score -G -N $node >/dev/null 2>&1
        rc=$?
        if [ $rc -eq 0 ]; then
            return 0
        fi
    done
    return 1
}

attempt_to_set_master() {

    ocf_log info "Attempting to set master"
    
    local expected_node_count
    if waiting_for_first_master; then
        # Wait for all nodes to come online
        expected_node_count=$OCF_RESKEY_CRM_meta_clone_max
    else
        # We accept one node being down. This is not arbitrary,
        # synchronous replication requires acknowledgement from
        # at least one host, which means only two nodes must have
        # the latest GTID. So a set of n - 1 ensures that we do
        # not lose any writes.
        expected_node_count=$(($OCF_RESKEY_CRM_meta_clone_max-1))
    fi

    # Set the gtid for this node, making it available to other nodes
    set_gtid
    
    local node_count=0
    local highest_gtid=0
    local master_candidate=""
    for node in $OCF_RESKEY_node_list; do

        local node_gtid=$(read_gtid $node)
        if [ -z "$node_gtid" ]; then
            continue
        fi

        # Got a valid gtid, increment node count
        node_count=$(($node_count+1))

        # Check if this is a good master candidate
        if greater_than_gtid $node_gtid $highest_gtid; then
            master_candidate=$node
            highest_gtid=$node_gtid
        fi
    done

    # If we managed to query a sufficient number of nodes
    # then set a master
    if [ $node_count -ge $expected_node_count ]; then
        ocf_log info "Promoting $master_candidate to master, highest gtid $highest_gtid, queried $node_count nodes."
        ocf_promotion_score -v 100 -N $master_candidate
    else
        ocf_log info "Not enough nodes ($node_count) contributed to select a master, need $expected_node_count nodes."
    fi
}

set_read_only() {
    # Sets or unsets read-only mode. Accepts one boolean as its
    # optional argument. If invoked without any arguments, defaults to
    # enabling read only mode. Should only be set in master/slave
    # setups.
    # Returns $OCF_SUCCESS if the operation succeeds, or
    # $OCF_ERR_GENERIC if it fails.
    local ro_val
    if ocf_is_true $1; then
        ro_val="on"
    else
        ro_val="off"
    fi
    ocf_run $MYSQL $MYSQL_OPTIONS_REPL \
        -e "SET GLOBAL read_only=${ro_val}"
}

get_read_only() {
    # Check if read-only is set
    local read_only_state

    read_only_state=$($MYSQL $MYSQL_OPTIONS_REPL \
        -e "SHOW VARIABLES" | grep -w read_only | awk '{print $2}')

    if [ "$read_only_state" = "ON" ]; then
        return 0
    else
        return 1
    fi
}

is_slave() {
    # Determine whether the machine is currently running as a MariaDB
    # slave, as determined per SHOW SLAVE STATUS. Returns 1 if SHOW
    # SLAVE STATUS creates an empty result set, 0 otherwise.
    local rc

    # Check whether this machine should be slave
    if ! get_read_only; then
        return 1
    fi
   
    if get_slave_info; then
       # show slave status is not empty
       # Is the slave sql thread running, then we are a slave!
       if [ "$slave_sql" == 'Yes' ]; then
          return 0
       else
          return 1
       fi
    else
       # "SHOW SLAVE STATUS" returns an empty set if instance is not a
       # replication slave
       return 1
    fi
}

parse_slave_info() {
    # Extracts field $1 from result of "SHOW SLAVE STATUS\G" from file $2
    sed -ne "s/^.* $1: \(.*\)$/\1/p" < $2
}

get_slave_info() {
            
    if [ "$master_log_file" -a "$master_host" ]; then
        # variables are already defined, get_slave_info has been run before
        return $OCF_SUCCESS
    else
        local tmpfile=$(mktemp ${HA_RSCTMP}/check_slave.${OCF_RESOURCE_INSTANCE}.XXXXXX)

        $MYSQL $MYSQL_OPTIONS_REPL \
        -e 'SHOW SLAVE STATUS\G' > $tmpfile

        if [ -s $tmpfile ]; then
            master_host=$(parse_slave_info Master_Host $tmpfile)
            master_user=$(parse_slave_info Master_User $tmpfile)
            master_port=$(parse_slave_info Master_Port $tmpfile)
            master_using_gtid=$(parse_slave_info Using_Gtid $tmpfile)
            master_log_file=$(parse_slave_info Master_Log_File $tmpfile)
            slave_sql=$(parse_slave_info Slave_SQL_Running $tmpfile)
            slave_io=$(parse_slave_info Slave_IO_Running $tmpfile)
            last_errno=$(parse_slave_info Last_Errno $tmpfile)
            last_error=$(parse_slave_info Last_Error $tmpfile)
            secs_behind=$(parse_slave_info Seconds_Behind_Master $tmpfile)
            last_io_errno=$(parse_slave_info Last_IO_Errno $tmpfile)
            last_io_error=$(parse_slave_info Last_IO_Error $tmpfile)
            ocf_log debug "MariaDB instance running as a replication slave"
            rm "$tmpfile"
        else
            # Instance produced an empty "SHOW SLAVE STATUS" output --
            # instance is not a slave
            rm "$tmpfile"
            return $OCF_ERR_GENERIC
        fi

        return $OCF_SUCCESS
    fi
}

check_slave() {
    # Checks slave status
    local rc new_master

    get_slave_info
    rc=$?

    if [ $rc -eq 0 ]; then

        # Check normal errors
        if [ $last_errno -ne 0 ]; then
            ocf_exit_reason "MariaDB slave replication has failed ($last_errno): $last_error"

            exit $OCF_ERR_GENERIC
        fi

        # Check IO Errors, ignore 2003 which indicates a connection failure to the master
        if [ $last_io_errno -ne 0 ] && [ $last_io_errno -ne 2003 ]; then
            ocf_exit_reason "MariaDB slave io has failed ($last_io_errno): $last_io_error"

            exit $OCF_ERR_GENERIC
        fi

        if [ $last_io_errno -eq 2003 ]; then
            ocf_log warn "MariaDB master not reachable from slave"
        fi
        
        if [ "$slave_io" != 'Yes' ]; then
            # Not necessarily a bad thing. The master may have
            # temporarily shut down, and the slave may just be
            # reconnecting. A warning can't hurt, though.
            ocf_log warn "MariaDB Slave IO threads currently not running."

            # Sanity check, are we at least on the right master
            new_master=$($CRM_ATTR_REPL_INFO --query  -q)

            if [ "$master_host" != "$new_master" ]; then
               # Not pointing to the right master, not good, removing the VIPs
               set_reader_attr 0

               exit $OCF_SUCCESS
            fi

        fi

        if [ "$slave_sql" != 'Yes' ]; then
            # We don't have a replication SQL thread running. Not a
            # good thing. Try to recoved by restarting the SQL thread
            # and remove reader vip.  Prevent MariaDB restart.
            ocf_exit_reason "MariaDB Slave SQL threads currently not running."

            # Remove reader vip
            set_reader_attr 0

            # try to restart slave
            ocf_run $MYSQL $MYSQL_OPTIONS_REPL \
                -e "START SLAVE"

            # Return success to prevent a restart
            exit $OCF_SUCCESS
        fi

        ocf_log debug "MariaDB instance running as a replication slave"
    else
        # Instance produced an empty "SHOW SLAVE STATUS" output --
        # instance is not a slave
        # TODO: Needs to handle when get_slave_info will return too many connections error
        ocf_exit_reason "check_slave invoked on an instance that is not a replication slave."
        exit $OCF_ERR_GENERIC
    fi
}

set_master() {
    local new_master=$($CRM_ATTR_REPL_INFO --query  -q)

    # Informs the MariaDB server of the master to replicate
    # from. Accepts one mandatory argument which must contain the host
    # name of the new master host. The master must either be unchanged
    # from the laste master the slave replicated from, or freshly
    # reset with RESET MASTER.
    ocf_log info "Changing MariaDB configuration to replicate from $new_master."

    ocf_run $MYSQL $MYSQL_OPTIONS_REPL \
        -e "CHANGE MASTER TO MASTER_HOST='$new_master', \
        MASTER_PORT=$OCF_RESKEY_replication_port, \
        MASTER_USER='$OCF_RESKEY_replication_user', \
        MASTER_PASSWORD='$OCF_RESKEY_replication_passwd', \
        MASTER_USE_GTID=current_pos";
}

unset_master(){
    # Instructs the MariaDB server to stop replicating from a master
    # host.

    # If we're currently not configured to be replicating from any
    # host, then there's nothing to do. But we do log a warning as
    # no-one but the CRM should be touching the MariaDB master/slave
    # configuration.
    if ! is_slave; then
        ocf_log warn "Attempted to unset the replication master on an instance that is not configured as a replication slave"
        return $OCF_SUCCESS
    fi

    # Stop the slave I/O thread and wait for relay log
    # processing to complete
    ocf_run $MYSQL $MYSQL_OPTIONS_REPL \
        -e "STOP SLAVE IO_THREAD"
    if [ $? -gt 0 ]; then
        ocf_exit_reason "Error stopping slave IO thread"
        exit $OCF_ERR_GENERIC
    fi

    local tmpfile=$(mktemp ${HA_RSCTMP}/threads.${OCF_RESOURCE_INSTANCE}.XXXXXX)
    while true; do
        $MYSQL $MYSQL_OPTIONS_REPL \
            -e 'SHOW PROCESSLIST\G' > $tmpfile
        if grep -i 'Has read all relay log' $tmpfile >/dev/null; then
            ocf_log info "MariaDB slave has finished processing relay log"
            break
        fi
        if ! grep -q 'system user' $tmpfile; then
            ocf_log info "Slave not runnig - not waiting to finish"
            break
        fi
        ocf_log info "Waiting for MariaDB slave to finish processing relay log"
        sleep 1
    done
    rm -f $tmpfile

    # Now, stop all slave activity and unset the master host
    ocf_run $MYSQL $MYSQL_OPTIONS_REPL \
        -e "STOP SLAVE"
    if [ $? -gt 0 ]; then
        ocf_exit_reason "Error stopping rest slave threads"
        exit $OCF_ERR_GENERIC
    fi

    ocf_run $MYSQL $MYSQL_OPTIONS_REPL \
        -e "RESET SLAVE /*!50516 ALL */;"
    if [ $? -gt 0 ]; then
        ocf_exit_reason "Failed to reset slave"
        exit $OCF_ERR_GENERIC
    fi
}

# Start replication as slave
start_slave() {
    ocf_run $MYSQL $MYSQL_OPTIONS_REPL \
        -e "START SLAVE"
}

# Set the attribute controlling the readers VIP
set_reader_attr() {
    local curr_attr_value

    curr_attr_value=$(get_reader_attr)

    if [ "$curr_attr_value" -ne "$1" ]; then
        $CRM_ATTR -l reboot --name ${OCF_RESKEY_reader_attribute} -v $1
    fi

}

# get the attribute controlling the readers VIP
get_reader_attr() {
    local attr_value
    local rc

    attr_value=$($CRM_ATTR -l reboot --name ${OCF_RESKEY_reader_attribute} --query -q)
    rc=$?
    if [ "$rc" -eq "0" ]; then
        echo $attr_value
    else
        echo -1
    fi
}

# Determines what IP address is attached to the current host.  The output of the
# crm_attribute command looks like this:
# scope=nodes  name=IP value=10.2.2.161
# If the ${INSTANCE_ATTR_NAME}_MYSQL_MASTER_IP node attribute is not defined, fallback is to uname -n
# The ${INSTANCE_ATTR_NAME}_MYSQL_MASTER_IP is the IP address that will be used for the 
# change master to command.
get_local_ip() {
   local IP
   IP=$($CRM_ATTR -l forever -n ${INSTANCE_ATTR_NAME}_mysql_master_IP -q -G 2>/dev/null)
   if [ ! $? -eq 0 ]; then
      uname -n
   else
      echo $IP
   fi
}

#######################################################################

# Functions invoked by resource manager actions

mysql_monitor() {
    local rc
    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 status returned an error, return that immediately
    if [ $rc -ne $OCF_SUCCESS ]; then
        return $rc
    fi

    # Check if this instance is configured as a slave, and if so
    # check slave status
    if is_slave; then
        if ! check_slave; then
            return $OCF_ERR_GENERIC
        fi
    fi
    
    if [ -n "$OCF_RESKEY_test_table" ]; then

        # Check for test table
        ocf_run -q $MYSQL $MYSQL_OPTIONS_TEST \
            -e "SELECT COUNT(*) FROM $OCF_RESKEY_test_table"
        rc=$?

        if [ $rc -ne 0 ]; then
            ocf_exit_reason "Failed to select from $test_table";
            return $OCF_ERR_GENERIC;
        fi
    fi

    # Check if we are in read-only mode and there is no master
    # with priority then we attempt to select a master
    if get_read_only && ! have_master_with_priority; then
        attempt_to_set_master
    fi
    
    if ! get_read_only; then
        ocf_log debug "MariaDB monitor succeeded (master)";
        return $OCF_RUNNING_MASTER
    else
        ocf_log debug "MariaDB monitor succeeded";
        return $OCF_SUCCESS
    fi
}

mysql_start() {
    local rc
    
    if ! ocf_is_ms; then
        ocf_exit_reason "Resource is not configured as master/slave"
        return $OCF_ERR_GENERIC
    fi
    
    # Initialize the ReaderVIP attribute, monitor will enable it
    set_reader_attr 0

    mysql_common_status info
    if [ $? = $OCF_SUCCESS ]; then
        ocf_log info "MariaDB already running"
        return $OCF_SUCCESS
    fi

    mysql_common_prepare_dirs

    mysql_common_start --skip-slave-start --log-slave-updates
    rc=$?
    if [ $rc != $OCF_SUCCESS ]; then
        return $rc
    fi

    # Enable semi-sync
    ocf_run -q $MYSQL $MYSQL_OPTIONS_TEST \
            -e "SET GLOBAL rpl_semi_sync_slave_enabled='ON', \
                           rpl_semi_sync_master_enabled='ON', \
                           rpl_semi_sync_master_wait_no_slave='OFF', \
                           rpl_semi_sync_master_wait_point='AFTER_SYNC', \
                           gtid_strict_mode='ON', \
                           sync_binlog=1, \
                           sync_master_info=1, \
                           innodb_flush_log_at_trx_commit=1;"
    rc=$?
    if [ $rc -ne 0 ]; then
        ocf_exit_reason "Failed to enable semi-sync and set variables";
        return $OCF_ERR_GENERIC;
    fi
    
    # We're configured as a stateful resource. We must start as
    # slave by default. At this point we don't know if the CRM has
    # already promoted a master. So, we simply start in read only
    # mode and make sure our old score is invalidated.
    set_read_only on
    ocf_promotion_score -D

    # Now, let's see whether there is a master. We might be a new
    # node that is just joining the cluster, and the CRM may have
    # promoted a master before.
    new_master_host=$(echo $OCF_RESKEY_CRM_meta_notify_master_uname|tr -d " ")
    if [ "$new_master_host" -a "$new_master_host" != ${NODENAME} ]; then
        set_master
        start_slave
        if [ $? -ne 0 ]; then
            ocf_exit_reason "Failed to start slave"
            return $OCF_ERR_GENERIC
        fi
    else
        ocf_log info "No MariaDB master present - clearing replication state, setting gtid in attrd, waiting for first master"
        unset_master
        set_waiting_for_first_master
    fi
    
    # Initial monitor action
    if [ -n "$OCF_RESKEY_test_table" -a -n "$OCF_RESKEY_test_user" -a -n "$OCF_RESKEY_test_passwd" ]; then
        OCF_CHECK_LEVEL=10
    fi

    mysql_monitor
    rc=$?
    if [ $rc != $OCF_SUCCESS -a $rc != $OCF_RUNNING_MASTER ]; then
        ocf_exit_reason "Failed initial monitor action"
        return $rc
    fi

    ocf_log info "MariaDB started"
    return $OCF_SUCCESS
}

mysql_stop() {
    # clear preference for becoming master
    ocf_promotion_score -D

    # Remove VIP capability
    set_reader_attr 0

    mysql_common_stop
}

mysql_promote() {
    local master_info

    if ( ! mysql_common_status err ); then
        return $OCF_NOT_RUNNING
    fi
    ocf_run $MYSQL $MYSQL_OPTIONS_REPL \
        -e "STOP SLAVE"

    set_read_only off || return $OCF_ERR_GENERIC
    # Force the master to wait for timeout period on slave disconnect
    ocf_run -q $MYSQL $MYSQL_OPTIONS_TEST \
            -e "SET GLOBAL rpl_semi_sync_master_wait_no_slave='ON';"
    
    # Set Master Info in CIB, cluster level attribute
    master_info="$(get_local_ip)"
    ${CRM_ATTR_REPL_INFO} -v "$master_info"
    
    # A master can accept reads
    set_reader_attr 1

    # Clear the gtids in attrd now that there is a master
    clear_all_gtid

    return $OCF_SUCCESS
}

mysql_demote() {
    if ! mysql_common_status err; then
        return $OCF_NOT_RUNNING
    fi

    # Return to default no wait setting.
    ocf_run -q $MYSQL $MYSQL_OPTIONS_TEST \
            -e "SET GLOBAL rpl_semi_sync_master_wait_no_slave='OFF';"

    # Return master preference to default, so the cluster manager gets
    # a chance to select a new master
    ocf_promotion_score -D
}

mysql_notify() {
    local type_op
    type_op="${OCF_RESKEY_CRM_meta_notify_type}-${OCF_RESKEY_CRM_meta_notify_operation}"

    ocf_log debug "Received $type_op notification."

    case "$type_op" in
        'pre-promote')
            # A master is now being promoted, remove the waiting-for-first-master flag
            clear_waiting_for_first_master
        ;;
        'post-promote')
            # The master has completed its promotion. Now is a good
            # time to check whether our replication slave is working
            # correctly.
            new_master_host=$(echo $OCF_RESKEY_CRM_meta_notify_promote_uname|tr -d " ")
            if [ "$new_master_host" = ${NODENAME} ]; then
                ocf_log info "This will be the new master, ignoring post-promote notification."
            else
                ocf_log info "Resetting replication, uname of master: $new_master_host"
                unset_master
                if [ $? -ne 0 ]; then
                    return $OCF_ERR_GENERIC
                fi

                set_master
                if [ $? -ne 0 ]; then
                    return $OCF_ERR_GENERIC
                fi

                start_slave
                if [ $? -ne 0 ]; then
                    ocf_exit_reason "Failed to start slave"
                    return $OCF_ERR_GENERIC
                fi
            fi
            return $OCF_SUCCESS
        ;;
        'pre-demote')
            demote_host=$(echo $OCF_RESKEY_CRM_meta_notify_demote_uname|tr -d " ")
            if [ $demote_host = ${NODENAME} ]; then
                ocf_log info "pre-demote notification for $demote_host"
                set_read_only on
                if [ $? -ne 0 ]; then
                    ocf_exit_reason "Failed to set read-only";
                    return $OCF_ERR_GENERIC;
                fi

                # Must kill all existing user threads because they are still Read/write
                # in order for the slaves to complete the read of binlogs
                local tmpfile=$(mktemp ${HA_RSCTMP}/threads.${OCF_RESOURCE_INSTANCE}.XXXXXX)
                $MYSQL $MYSQL_OPTIONS_REPL -e "SHOW PROCESSLIST" > $tmpfile
                for thread in $(awk '$0 !~ /Binlog Dump|system user|event_scheduler|SHOW PROCESSLIST/ && $0 ~ /^[0-9]/ {print $1}' $tmpfile)
                do
                    ocf_run $MYSQL $MYSQL_OPTIONS_REPL \
                        -e "KILL ${thread}"
                done
                rm -f $tmpfile
            else
               ocf_log info "Ignoring post-demote notification execpt for my own demotion."
            fi
            return $OCF_SUCCESS
        ;;
        'post-demote')
            demote_host=$(echo $OCF_RESKEY_CRM_meta_notify_demote_uname|tr -d " ")
            if [ $demote_host = ${NODENAME} ]; then
                ocf_log info "Ignoring post-demote notification for my own demotion."
                return $OCF_SUCCESS
            fi
            ocf_log info "post-demote notification for $demote_host."
            # The former master has just been gracefully demoted.
            unset_master
        ;;
        *)
            return $OCF_SUCCESS
        ;;
    esac
}

mysql_validate() {
    check_binary bc
}

#######################################################################

case "$1" in
  meta-data)    meta_data
        exit $OCF_SUCCESS;;
  usage|help)   usage
        exit $OCF_SUCCESS;;
esac

mysql_common_validate
rc=$?
LSB_STATUS_STOPPED=3
if [ $rc -ne 0 ]; then
    case "$1" in
        stop) ;;
        monitor)
            mysql_common_status "info"
            if [ $? -eq $OCF_SUCCESS ]; then
                # if validatation fails and pid is active, always treat this as an error
                ocf_exit_reason "environment validation failed, active pid is in unknown state."
                exit $OCF_ERR_GENERIC
            fi
            # validation failed and pid is not active, it's safe to say this instance is inactive.
            exit $OCF_NOT_RUNNING;;

        status) exit $LSB_STATUS_STOPPED;;
        *) exit $rc;;
    esac
fi

# What kind of method was invoked?
case "$1" in
  start)    mysql_start;;
  stop)     mysql_stop;;
  status)   mysql_common_status err;;
  monitor)  mysql_monitor;;
  promote)  mysql_promote;;
  demote)   mysql_demote;;
  notify)   mysql_notify;;
  validate-all) mysql_validate;;

 *)     usage
        exit $OCF_ERR_UNIMPLEMENTED;;
esac

# vi:sw=4:ts=4:et: