#!/bin/bash
#
#
# MySQL
#
# Description: Manages a MySQL 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
# Jervin Real, Kenny Gryp: Booth Compatibility Improvements
#
# Support: linux-ha@lists.linux-ha.org
# License: GNU General Public License (GPL)
#
# (c) 2002-2005 International Business Machines, Inc.
# 2005-2010 Linux-HA contributors
#
# An example usage in /etc/ha.d/haresources:
# node1 10.0.0.170 mysql
#
# Version: 20141112131457
#
# See usage() function below for more details...
#
# OCF instance parameters:
# OCF_RESKEY_binary
# OCF_RESKEY_binary_prefix
# OCF_RESKEY_client_binary
# OCF_RESKEY_config
# OCF_RESKEY_datadir
# OCF_RESKEY_user
# OCF_RESKEY_group
# 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
# OCF_RESKEY_replication_options
# OCF_RESKEY_max_slave_lag
# OCF_RESKEY_evict_outdated_slaves
# OCF_RESKEY_reader_attribute
# OCF_RESKEY_reader_failcount
# OCF_RESKEY_backup_lockfile
# OCF_RESKEY_geo_remote_IP
# OCF_RESKEY_geo_remote_user
# OCF_RESKEY_geo_remote_opts
# OCF_RESKEY_booth_master_ticket
# OCF_RESKEY_post_promote_script
# OCF_RESKEY_prm_binlog_parser_path
# OCF_RESKEY_try_restart_crashed_master
#######################################################################
# Initialization:
: ${OCF_FUNCTIONS_DIR=${OCF_ROOT}/lib/heartbeat}
. ${OCF_FUNCTIONS_DIR}/ocf-shellfuncs
#######################################################################
# Fill in some defaults if no values are specified
HOSTOS=`uname`
if [ "X${HOSTOS}" = "XOpenBSD" ];then
OCF_RESKEY_binary_default="/usr/local/bin/mysqld"
OCF_RESKEY_config_default="/etc/my.cnf"
OCF_RESKEY_datadir_default="/var/mysql"
OCF_RESKEY_user_default="_mysql"
OCF_RESKEY_group_default="_mysql"
OCF_RESKEY_log_default="/var/log/mysqld.log"
OCF_RESKEY_pid_default="/var/mysql/mysqld.pid"
OCF_RESKEY_socket_default="/var/run/mysql/mysql.sock"
else
OCF_RESKEY_binary_default="/usr/sbin/mysqld"
OCF_RESKEY_config_default="/etc/my.cnf"
OCF_RESKEY_datadir_default="/var/lib/mysql"
OCF_RESKEY_user_default="mysql"
OCF_RESKEY_group_default="mysql"
OCF_RESKEY_log_default="/var/log/mysqld.log"
OCF_RESKEY_pid_default="/var/run/mysql/mysqld.pid"
OCF_RESKEY_socket_default="/var/lib/mysql/mysql.sock"
fi
OCF_RESKEY_client_binary_default="mysql"
OCF_RESKEY_binary_prefix_default=""
OCF_RESKEY_test_user_default="root"
OCF_RESKEY_test_table_default="mysql.user"
OCF_RESKEY_test_passwd_default=""
OCF_RESKEY_enable_creation_default=0
OCF_RESKEY_additional_parameters_default=""
OCF_RESKEY_replication_port_default="3306"
OCF_RESKEY_max_slave_lag_default="3600"
OCF_RESKEY_evict_outdated_slaves_default="false"
OCF_RESKEY_reader_attribute_default="readable"
OCF_RESKEY_reader_failcount_default="1"
OCF_RESKEY_backup_lockfile_default="/var/lock/innobackupex"
OCF_RESKEY_booth_master_ticket_default="ticketMaster"
OCF_RESKEY_async_stop_default=0
OCF_RESKEY_try_restart_crashed_master_default=1
: ${OCF_RESKEY_binary=${OCF_RESKEY_binary_default}}
MYSQL_SBINDIR=`dirname ${OCF_RESKEY_binary}`
: ${OCF_RESKEY_binary_prefix=${OCF_RESKEY_binary_prefix_default}}
: ${OCF_RESKEY_client_binary=${OCF_RESKEY_client_binary_default}}
: ${OCF_RESKEY_config=${OCF_RESKEY_config_default}}
: ${OCF_RESKEY_datadir=${OCF_RESKEY_datadir_default}}
: ${OCF_RESKEY_user=${OCF_RESKEY_user_default}}
: ${OCF_RESKEY_group=${OCF_RESKEY_group_default}}
: ${OCF_RESKEY_log=${OCF_RESKEY_log_default}}
: ${OCF_RESKEY_pid=${OCF_RESKEY_pid_default}}
: ${OCF_RESKEY_socket=${OCF_RESKEY_socket_default}}
: ${OCF_RESKEY_test_user=${OCF_RESKEY_test_user_default}}
: ${OCF_RESKEY_test_table=${OCF_RESKEY_test_table_default}}
: ${OCF_RESKEY_test_passwd=${OCF_RESKEY_test_passwd_default}}
: ${OCF_RESKEY_enable_creation=${OCF_RESKEY_enable_creation_default}}
: ${OCF_RESKEY_additional_parameters=${OCF_RESKEY_additional_parameters_default}}
: ${OCF_RESKEY_replication_user=${OCF_RESKEY_replication_user_default}}
: ${OCF_RESKEY_replication_passwd=${OCF_RESKEY_replication_passwd_default}}
: ${OCF_RESKEY_replication_port=${OCF_RESKEY_replication_port_default}}
: ${OCF_RESKEY_replication_options=${OCF_RESKEY_replication_options_default}}
: ${OCF_RESKEY_max_slave_lag=${OCF_RESKEY_max_slave_lag_default}}
: ${OCF_RESKEY_evict_outdated_slaves=${OCF_RESKEY_evict_outdated_slaves_default}}
: ${OCF_RESKEY_reader_attribute=${OCF_RESKEY_reader_attribute_default}}
: ${OCF_RESKEY_reader_failcount=${OCF_RESKEY_reader_failcount_default}}
: ${OCF_RESKEY_backup_lockfile=${OCF_RESKEY_backup_lockfile_default}}
: ${OCF_RESKEY_geo_remote_IP}=""
: ${OCF_RESKEY_geo_remote_user}="root"
: ${OCF_RESKEY_geo_remote_opts}=""
: ${OCF_RESKEY_booth_master_ticket}=${OCF_RESKEY_booth_master_ticket_default}
: ${OCF_RESKEY_post_promote_script}=""
: ${OCF_RESKEY_prm_binlog_parser_path=${OCF_RESKEY_prm_binlog_parser_path_default}}
: ${OCF_RESKEY_async_stop=${OCF_RESKEY_async_stop_default}}
: ${OCF_RESKEY_try_restart_crashed_master=${OCF_RESKEY_try_restart_crashed_master_default}}
#######################################################################
# Convenience variables
MYSQL=$OCF_RESKEY_client_binary
#Add a timeout to the mysql client, no commands should take more than 2s
if [ -x "/usr/bin/timeout" ]; then
TIMEOUT="/usr/bin/timeout 5 "
else
TIMEOUT=""
fi
MYSQL="$TIMEOUT $MYSQL"
MYSQLBINLOG=`which mysqlbinlog 2> /dev/null`
MYSQL_BINDIR=`dirname ${OCF_RESKEY_client_binary}`
MYSQL_OPTIONS_LOCAL="-A -S $OCF_RESKEY_socket --connect_timeout=10"
MYSQL_OPTIONS_REPL="$MYSQL_OPTIONS_LOCAL --user=$OCF_RESKEY_replication_user --password=$OCF_RESKEY_replication_passwd"
MYSQL_OPTIONS_TEST="$MYSQL_OPTIONS_LOCAL --user=$OCF_RESKEY_test_user --password=$OCF_RESKEY_test_passwd"
MYSQL_LAST_ERR=0
MYSQL_TOO_MANY_CONN_ERR=1040
CRM_MASTER="$TIMEOUT ${HA_SBIN_DIR}/crm_master -l reboot "
HOSTNAME=`uname -n`
CRM_ATTR="$TIMEOUT ${HA_SBIN_DIR}/crm_attribute -N $HOSTNAME -q"
INSTANCE_ATTR_NAME=`echo ${OCF_RESOURCE_INSTANCE}| awk -F : '{print $1}'`
CRM_ATTR_REPL_INFO="$TIMEOUT ${HA_SBIN_DIR}/crm_attribute --type crm_config --name ${INSTANCE_ATTR_NAME}_REPL_INFO -s mysql_replication -q "
CRM_ATTR_REPL_STATUS="$TIMEOUT ${HA_SBIN_DIR}/crm_attribute --type crm_config --name ${INSTANCE_ATTR_NAME}_REPL_STATUS -s mysql_replication -q "
CRM_ATTR_LAST_TRX="$TIMEOUT ${HA_SBIN_DIR}/crm_attribute --type crm_config --name ${INSTANCE_ATTR_NAME}_NEW_MASTER_LAST_TRX -s mysql_replication -q "
CRM_ATTR_MASTER_CRASHED_TS="$CRM_ATTR -l reboot --name ${INSTANCE_ATTR_NAME}_master_crashed"
CRM_ATTR_NODE_LAST_TRX_MD5="$CRM_ATTR -l reboot --name ${INSTANCE_ATTR_NAME}_last_trx_md5"
CRM_RES="$TIMEOUT ${HA_SBIN_DIR}/crm_resource"
CRM_TICKET="$TIMEOUT ${HA_SBIN_DIR}/crm_ticket"
SSH="$TIMEOUT /usr/bin/ssh "
MAX_BINLOG_SIZE_CACHE="${HA_RSCTMP}/max_binlog_size_cache"
ASYNC_STOP_WITNESS_FILE="${HA_RSCTMP}/stop_${INSTANCE_ATTR_NAME}"
OCF_STOPPING=100 #custom error code for async_stop
#######################################################################
usage() {
cat <<UEND
usage: $0 (start|stop|validate-all|meta-data|monitor|promote|demote|notify)
$0 manages a MySQL Database as an HA resource.
The 'start' operation starts the database.
The 'stop' operation stops the database.
The 'status' operation reports whether the database is running
The 'monitor' operation reports whether the database seems to be working
The 'promote' operation makes this mysql server run as master
The 'demote' operation makes this mysql server run as slave
The 'validate-all' operation reports whether the parameters are valid
UEND
}
meta_data() {
cat <<END
<?xml version="1.0"?>
<!DOCTYPE resource-agent SYSTEM "ra-api-1.dtd">
<resource-agent name="mysql">
<version>1.0</version>
<longdesc lang="en">
Resource script for MySQL.
May manage a standalone MySQL database, a clone set with externally
managed replication, or a complete master/slave replication setup.
While managing replication, the default behavior is to use uname -n
values in the change master to command. Other IPs can be specified
manually by adding a node attribute \${INSTANCE_ATTR_NAME}_mysql_master_IP
giving the IP to use for replication. For example, if the mysql primitive
you are using is p_mysql, the attribute to set will be
p_mysql_mysql_master_IP.
</longdesc>
<shortdesc lang="en">Manages a MySQL database instance</shortdesc>
<parameters>
<parameter name="binary" unique="0" required="0">
<longdesc lang="en">
Location of the MySQL server binary
</longdesc>
<shortdesc lang="en">MySQL server binary</shortdesc>
<content type="string" default="${OCF_RESKEY_binary_default}" />
</parameter>
<parameter name="binary_prefix" unique="0" required="0">
<longdesc lang="en">
A prefix to the MySQL server binary. I could be for example a LD_PRELOAD or
a call to numactl.
</longdesc>
<shortdesc lang="en">MySQL server binary prefix</shortdesc>
<content type="string" default="${OCF_RESKEY_binary_prefix_default}" />
</parameter>
<parameter name="client_binary" unique="0" required="0">
<longdesc lang="en">
Location of the MySQL client binary
</longdesc>
<shortdesc lang="en">MySQL client binary</shortdesc>
<content type="string" default="${OCF_RESKEY_client_binary_default}" />
</parameter>
<parameter name="config" unique="0" required="0">
<longdesc lang="en">
Configuration file
</longdesc>
<shortdesc lang="en">MySQL config</shortdesc>
<content type="string" default="${OCF_RESKEY_config_default}" />
</parameter>
<parameter name="datadir" unique="0" required="0">
<longdesc lang="en">
Directory containing databases
</longdesc>
<shortdesc lang="en">MySQL datadir</shortdesc>
<content type="string" default="${OCF_RESKEY_datadir_default}" />
</parameter>
<parameter name="user" unique="0" required="0">
<longdesc lang="en">
User running MySQL daemon
</longdesc>
<shortdesc lang="en">MySQL user</shortdesc>
<content type="string" default="${OCF_RESKEY_user_default}" />
</parameter>
<parameter name="group" unique="0" required="0">
<longdesc lang="en">
Group running MySQL daemon (for logfile and directory permissions)
</longdesc>
<shortdesc lang="en">MySQL group</shortdesc>
<content type="string" default="${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">MySQL log file</shortdesc>
<content type="string" default="${OCF_RESKEY_log_default}"/>
</parameter>
<parameter name="pid" unique="0" required="0">
<longdesc lang="en">
The pidfile to be used for mysqld.
</longdesc>
<shortdesc lang="en">MySQL pid file</shortdesc>
<content type="string" default="${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">MySQL socket</shortdesc>
<content type="string" default="${OCF_RESKEY_socket_default}"/>
</parameter>
<parameter name="test_table" unique="0" required="0">
<longdesc lang="en">
Table to be tested in monitor statement (in database.table notation)
</longdesc>
<shortdesc lang="en">MySQL test table</shortdesc>
<content type="string" default="${OCF_RESKEY_test_table_default}" />
</parameter>
<parameter name="test_user" unique="0" required="0">
<longdesc lang="en">
MySQL test user, must have select privilege on test_table
</longdesc>
<shortdesc lang="en">MySQL test user</shortdesc>
<content type="string" default="${OCF_RESKEY_test_user_default}" />
</parameter>
<parameter name="test_passwd" unique="0" required="0">
<longdesc lang="en">
MySQL test user password
</longdesc>
<shortdesc lang="en">MySQL test user password</shortdesc>
<content type="string" default="${OCF_RESKEY_test_passwd_default}" />
</parameter>
<parameter name="enable_creation" unique="0" required="0">
<longdesc lang="en">
If the MySQL 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">
MySQL replication user. This user is used for starting and stopping
MySQL replication, for setting and resetting the master host, and for
setting and unsetting read-only mode. Because of that, this user must
have SUPER, REPLICATION SLAVE, REPLICATION CLIENT, and PROCESS
privileges on all nodes within the cluster. Mandatory if you define
a master-slave resource.
</longdesc>
<shortdesc lang="en">MySQL replication user</shortdesc>
<content type="string" default="${OCF_RESKEY_replication_user_default}" />
</parameter>
<parameter name="replication_passwd" unique="0" required="0">
<longdesc lang="en">
MySQL replication password. Used for replication client and slave.
Mandatory if you define a master-slave resource.
</longdesc>
<shortdesc lang="en">MySQL replication user password</shortdesc>
<content type="string" default="${OCF_RESKEY_replication_passwd_default}" />
</parameter>
<parameter name="replication_port" unique="0" required="0">
<longdesc lang="en">
The port on which the Master MySQL instance is listening.
</longdesc>
<shortdesc lang="en">MySQL replication port</shortdesc>
<content type="string" default="${OCF_RESKEY_replication_port_default}" />
</parameter>
<parameter name="replication_options" unique="0" required="0">
<longdesc lang="en">
Extra options to pass to CHANGE MASTER, be sure to pass a preceeding comma. Handy for SSL, for example:
replication_options=", MASTER_SSL=1, MASTER_SSL_CA='/path/to/ca.crt'"
</longdesc>
<shortdesc lang="en">MySQL replication options</shortdesc>
<content type="string" default="${OCF_RESKEY_replication_options_default}" />
</parameter>
<parameter name="max_slave_lag" unique="0" required="0">
<longdesc lang="en">
The maximum number of seconds a replication slave is allowed to lag
behind its master. Do not set this to zero. What the cluster manager
does in case a slave exceeds this maximum lag is determined by the
evict_outdated_slaves parameter.
</longdesc>
<shortdesc lang="en">Maximum time (seconds) a MySQL slave is allowed
to lag behind a master</shortdesc>
<content type="integer" default="${OCF_RESKEY_max_slave_lag_default}"/>
</parameter>
<parameter name="evict_outdated_slaves" unique="0" required="0">
<longdesc lang="en">
If set to true, any slave which is more than max_slave_lag seconds
behind the master has its MySQL instance shut down. If this parameter
is set to false in a primitive or clone resource, it is simply
ignored. If set to false in a master/slave resource, then exceeding
the maximum slave lag will merely push down the master preference so
the lagging slave is never promoted to the new master.
</longdesc>
<shortdesc lang="en">Determines whether to shut down badly lagging
slaves</shortdesc>
<content type="boolean" default="${OCF_RESKEY_evict_outdated_slaves_default}" />
</parameter>
<parameter name="reader_attribute" unique="1" required="0">
<longdesc lang="en">
An attribute that the RA can manage to specify whether a node
can be read from. This node attribute will be 1 if it's fine to
read from the node, and 0 otherwise (for example, when a slave
has lagged too far behind the master).
A typical example for the use of this attribute would be to tie
a set of IP addresses to MySQL slaves that can be read from.
This parameter is only meaningful in master/slave set configurations.
</longdesc>
<shortdesc lang="en">Sets the node attribute that determines
whether a node is usable for clients to read from.</shortdesc>
<content type="string" default="${OCF_RESKEY_reader_attribute_default}" />
</parameter>
<parameter name="reader_failcount" unique="1" required="0">
<longdesc lang="en">
The number of times a monitor operation can find the slave
to be unsuitable for reader VIP before failing. Useful if
there are short intermittent issues like clock adjustments in VMs.
</longdesc>
<shortdesc lang="en">Allowed failcount for reader</shortdesc>
<content type="integer" default="${OCF_RESKEY_reader_failcount_default}" />
</parameter>
<parameter name="backup_lockfile" unique="1" required="0">
<longdesc lang="en">
The path to a file that will be exclusively locked by any backup
process. The lockfile serves to provide a reliable way of determining
whether to restart the slave process or not. If a thirdparty process
locks this file, the agent will fail to lock the file and will not
start the slave. When this agent is able to lock the file, it is
assumed that backups are finished and the slave thread should be
running and will start it.
A typical cron command example would be like:
flock -xn /var/lock/innobackupex innobackupex --safe-slave-backup /tmp/mysqlbackup
This example will use innobackupex's ability to stop the slave when necessary
to ensure backup consistency. During this time the agent will not start the
slave. Once the backup is complete, the lock will automatically expire and
the agent can start the slave if it isn't already.
</longdesc>
<shortdesc lang="en">Path to backup lockfile</shortdesc>
<content type="string" default="${OCF_RESKEY_backup_lockfile_default}" />
</parameter>
<parameter name="geo_remote_IP" unique="1" required="0">
<longdesc lang="en">
In case multiple Geo redundant sites are connected with the booth protocol
this is the IP to use to connect to the remote cluster to query replication info.
Normally this would be the writer VIP on the remote cluster. Also, ssh is used
for communication so make sure keys are exchanged and that ssh options are set
in a way that connection doesn't take many seconds. If empty, the booth type
behavior is not triggered.
</longdesc>
<shortdesc lang="en">IP of the remote cluster</shortdesc>
<content type="string" default="" />
</parameter>
<parameter name="geo_remote_user" unique="1" required="0">
<longdesc lang="en">
When geo_remote_IP is not empty, this will be the custom SSH user that can be
used to connect to remote sites.
</longdesc>
<shortdesc lang="en">SSH user for remote-enabled sites</shortdesc>
<content type="string" default="root" />
</parameter>
<parameter name="geo_remote_opts" unique="1" required="0">
<longdesc lang="en">
When geo_remote_IP is not empty, this will be the custom SSH options that can be
used to connect to remote sites.
</longdesc>
<shortdesc lang="en">SSH options for remote-enabled sites</shortdesc>
<content type="string" default="" />
</parameter>
<parameter name="booth_master_ticket" unique="1" required="0">
<longdesc lang="en">
In case multiple Geo redundant sites are connected with the booth protocol
this is the name of the ticket used to identify the master side.
</longdesc>
<shortdesc lang="en">Booth ticket name</shortdesc>
<content type="string" default="${OCF_RESKEY_booth_master_ticket_default}" />
</parameter>
<parameter name="post_promote_script" unique="1" required="0">
<longdesc lang="en">
Allows to run custom code following a promotion. An application of this is to
prevent fail-back of the master role after an initial failover.
</longdesc>
<shortdesc lang="en">Post promote script</shortdesc>
<content type="string" default="" />
</parameter>
<parameter name="prm_binlog_parser_path" unique="1" required="0">
<longdesc lang="en">
Path to the prm_binlog_parser tool that is used to publish the last trx of a new
master after a hard crash of the previous master. The tool can be downloaded from
https://github.com/percona/percona-pacemaker-agents/tree/master/tools/ybinlogp
</longdesc>
<shortdesc lang="en">Path to the prm_binlog_parser tool</shortdesc>
<content type="string" default="" />
</parameter>
<parameter name="async_stop" unique="0" required="0">
<longdesc lang="en">
If set to true, PRM will not wait for MySQL to stop after sending the
SIGTERM signal. This can be useful to speed up failover when a server has a
large number of dirty pages and takes a long time to shutdown, or worse, receives
a SIGKILL after the stop timeout. The main drawback is that if PRM wants to restart
MySQL before it completed its shutdown, the operation will error out.
</longdesc>
<shortdesc lang="en">Asynchronous stop of MySQL</shortdesc>
<content type="boolean" default="${OCF_RESKEY_async_stop_default}" />
</parameter>
<parameter name="try_restart_crashed_master" unique="0" required="0">
<longdesc lang="en">
If set to true, PRM will try to restart a failed master in place instead of promoting
another node. This can help recover untransmitted binary logs. However, if you have
a large database that takes a long time to recovery, this may not be a good option
for you.
</longdesc>
<shortdesc lang="en">Try restarting a crashed master</shortdesc>
<content type="boolean" default="${OCF_RESKEY_try_restart_crashed_master_default}" />
</parameter>
</parameters>
<actions>
<action name="start" timeout="120" />
<action name="stop" timeout="120" />
<action name="status" timeout="60" />
<action name="monitor" depth="0" timeout="30" interval="20" />
<action name="monitor" role="Master" depth="0" timeout="30" interval="10" />
<action name="monitor" role="Slave" depth="0" timeout="30" interval="30" />
<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>
END
}
# Convenience functions
# get the binlog references and md5 of the payload. Currently unable to report
# the 1st trx.
get_last_binlog() {
local binlog_files
local maxEntries
local lastNSec
local Ts Year Month Day Time
local startAt nlines
local bltempfile
binlog_files=$1
maxEntries=$2
#lastNSec=$3
# for debugging
cp "$binlog_files" /tmp
let nlines=maxEntries*3+6 # below return 3 lines per Xid + need to skip last 2
startAt=`$MYSQLBINLOG -vvv --base64-output=DECODE-ROWS $1 | \
grep 'Xid =' -A2 | grep -v '\-\-' | tail -n $nlines | \
egrep 'Xid|\# at [0-9]{2,10}' | tac | grep -A2 Xid | \
grep '# at ' | tail -n 1 | rev | cut -d' ' -f1 | rev`
md5=''
queryok=0
bltempfile=`mktemp ${HA_RSCTMP}/${OCF_RESOURCE_INSTANCE}.XXXXXX`
( for blf in $1; do
blf_no_path=`echo "$blf" | rev | cut -d'/' -f1 | rev`
echo "$blf_no_path@"
$MYSQLBINLOG --base64-output=DECODE-ROWS -vvv --start-position="$startAt" $blf | \
( while read line; do
if [ "$queryok" -eq 1 ]; then
if [[ $line =~ ^\# ]]; then
if [[ $line =~ .*end_log_pos\ ([0-9]*).*Xid\ =\ [0-9]*$ ]]; then
echo "${BASH_REMATCH[1]},`md5sum $bltempfile | cut -d' ' -f1`"
echo -n '' > $bltempfile
else
if [[ $line =~ ^\#\#\# ]]; then
echo "${line}" >> $bltempfile
fi
fi
else
echo "${line}" >> $bltempfile
fi
else
if [[ $line =~ ^BEGIN ]]; then
queryok=1
fi
fi
done )
done ) | tail -n $maxEntries
echo "@"
rm -f $bltempfile
}
# Set the master score, the presence of the master_crashed_ts attribute
# presents the update.
set_master_score() {
local master_crashed_ts
master_crashed_ts=`$CRM_ATTR_MASTER_CRASHED_TS --query --default=0`
if [ "$master_crashed_ts" -eq "0" ]; then
$CRM_MASTER -v $1
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
mysql_run -Q -sw $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_run -Q -sw -O $MYSQL -N $MYSQL_OPTIONS_REPL \
-e "SHOW VARIABLES like '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 MySQL
# slave, as determined per SHOW SLAVE STATUS. Returns 1 if SHOW
# SLAVE STATUS creates an empty result set, 0 otherwise.
local rc
local tmpfile
# Check whether this machine should be slave
if ! ocf_is_ms || ! get_read_only; then
return 1
fi
get_slave_info
rc=$?
if [ $rc -eq 0 ]; then
# show slave status is not empty
# Is there a master_log_file defined? (master_log_file is deleted
# by reset slave
if [ "$master_log_file" ]; 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 the current max_binlog_size. Since this value rarely change, we cache
# it for 1h
get_max_binlog_size() {
if [ -e $MAX_BINLOG_SIZE_CACHE ]; then
cat $MAX_BINLOG_SIZE_CACHE
if [ `date +%s` -gt "$((`stat -c %Z $MAX_BINLOG_SIZE_CACHE`+3600))" ]; then
rm $MAX_BINLOG_SIZE_CACHE
fi
else
mysql_run -Q -sw -O $MYSQL -N $MYSQL_OPTIONS_REPL \
-e "Show global variables like 'max_binlog_size';" | \
awk '{ print $2 }' > $MAX_BINLOG_SIZE_CACHE
cat $MAX_BINLOG_SIZE_CACHE
fi
}
get_slave_info() {
local mysql_options tmpfile
if [ "$master_log_file" -a "$master_host" ]; then
# variables are already defined, get_slave_info has been run before
return $OCF_SUCCESS
else
tmpfile=`mktemp ${HA_RSCTMP}/check_slave.${OCF_RESOURCE_INSTANCE}.XXXXXX`
mysql_run -Q -sw -O $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_log_file=`parse_slave_info Master_Log_File $tmpfile`
relay_log_file=`parse_slave_info Relay_Log_File $tmpfile`
master_log_pos=`parse_slave_info Read_Master_Log_Pos $tmpfile`
slave_sql=`parse_slave_info Slave_SQL_Running $tmpfile`
slave_io=`parse_slave_info Slave_IO_Running $tmpfile`
slave_io_state=`parse_slave_info Slave_IO_State $tmpfile`
last_errno=`parse_slave_info Last_Errno $tmpfile`
secs_behind=`parse_slave_info Seconds_Behind_Master $tmpfile`
ocf_log debug "MySQL instance has a non empty slave status"
else
# Instance produced an empty "SHOW SLAVE STATUS" output --
# instance is not a slave
ocf_log err "check_slave invoked on an instance that is not a replication slave."
rm -f $tmpfile
return $OCF_ERR_GENERIC
fi
rm -f $tmpfile
return $OCF_SUCCESS
fi
}
check_slave() {
# Checks slave status
local rc new_master
get_slave_info
rc=$?
if [ $rc -eq 0 ]; then
# Did we receive an error other than max_connections?
if [ $last_errno -ne 0 -a $last_errno -ne "$MYSQL_TOO_MANY_CONN_ERR" ]; then
# Whoa. Replication ran into an error. This slave has
# diverged from its master. Make sure this resource
# doesn't restart in place.
ocf_log err "MySQL instance configured for replication, but replication has failed."
# Just pull the reader VIP away, killing MySQL here would be pretty evil
# on a loaded server
set_reader_attr 0
#Since replication is broken, not suitable to be a master
set_master_score -INF
exit $OCF_SUCCESS
fi
# If we got max_connections, let's only remove the vip
if [ $last_errno -eq "$MYSQL_TOO_MANY_CONN_ERR" ]; then
set_reader_attr 0
exit $OCF_SUCCESS
fi
if [ "$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 "MySQL Slave IO threads currently not running."
# Sanity check, are we at least on the right master
if [ "$master_host" != "$glb_cib_master" ]; then
# Not pointing to the right master
# Is this a recent master failover on the remote side
if [ "${#glb_remote_info}" -gt "0" -a "$slave_sql" = 'Yes' ]; then
# looks like, the sql thread is still running, no need
# to remove the vip, doing nothing
:
else
set_reader_attr 0
fi
# try setting up the slave with the new master
set_master
exit $OCF_SUCCESS
elif [ "$slave_sql" == 'Yes' ]; then
# If the slq thread is running, it is an issue with the io thread
# let's try to restart it
if [ "$slave_io_state" != "" ]; then
# The io thread is running but is not connected, let's restart it.
mysql_run -Q -sw $MYSQL $MYSQL_OPTIONS_REPL \
-e "STOP SLAVE IO_THREAD"
fi
# At this point, the io_thread should be stopped.
# let's try to start it again.
mysql_run -Q -sw $MYSQL $MYSQL_OPTIONS_REPL \
-e "START SLAVE IO_THREAD"
# We give some time to connect
sleep 2
get_slave_info
rc=$?
if [ $rc -eq 0 -a "$slave_io" == 'Yes' ]; then
ocf_log info "MySQL Slave IO thread started succesfully."
else
ocf_log warn "We could not start the MySQL Slave IO thread."
fi
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 MySQL restart.
ocf_log err "MySQL Slave SQL threads currently not running."
# Remove reader vip
set_reader_attr 0
# If sql is not running, can't be a master
set_master_score -INF
# Check that the flock tool exists first
if type flock &>/dev/null; then
(
flock -xn 8
if [ $? -eq 0 ]; then
mysql_run -Q -sw $MYSQL $MYSQL_OPTIONS_REPL \
-e "START SLAVE"
else
ocf_log info "Unable to lock $OCF_RESKEY_backup_lockfile. Not starting slave."
fi
) 8>$OCF_RESKEY_backup_lockfile
else
# try to restart slave
mysql_run -Q -sw $MYSQL $MYSQL_OPTIONS_REPL \
-e "START SLAVE"
fi
# Return success to prevent a restart
exit $OCF_SUCCESS
fi
if ocf_is_true $OCF_RESKEY_evict_outdated_slaves; then
# We're supposed to bail out if we lag too far
# behind. Let's check our lag.
if [ $secs_behind -gt $OCF_RESKEY_max_slave_lag ]; then
ocf_log err "MySQL Slave is $secs_behind seconds behind master (allowed maximum: $OCF_RESKEY_max_slave_lag)."
# Remove reader vip
set_reader_attr 0
exit $OCF_ERR_INSTALLED
fi
elif ocf_is_ms; then
# Even if we're not set to evict lagging slaves, we can
# still use the seconds behind master value to set our
# master preference.
local master_pref
test $secs_behind -eq 0 2>/dev/null
if [ $? -eq 2 ]; then
# SBM is undefined or not an integer
master_pref=0
set_reader_attr 0
else
master_pref=$((${OCF_RESKEY_max_slave_lag}-${secs_behind}))
if [ $master_pref -lt 0 ]; then
# Sanitize a below-zero preference to just zero
master_pref=0
fi
if [ $secs_behind -gt $OCF_RESKEY_max_slave_lag ]; then
set_reader_attr 0
else
set_reader_attr 1
fi
#Edge case verification, check if on the right master
set_master nologging
fi
# Is the datadir almost full
if check_datadir_state; then
set_master_score $master_pref
else
# full so not good for a master
set_master_score -2147483640
fi
fi
ocf_log debug "MySQL 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
if [ $MYSQL_LAST_ERR -eq "$MYSQL_TOO_MANY_CONN_ERR" ]; then
# Remove the vip
set_reader_attr 0
return $OCF_SUCCESS
fi
# An empty status could happen when a master is demote in a
# geo DR setup, let's check
if [ $MYSQL_LAST_ERR -eq 0 -a $glb_master_exists -eq 1 ]; then
# This is not the master side, let's try to setup the slave
# No need to unset the master since slave status is empty
set_reader_attr 0
set_master
return $OCF_SUCCESS
fi
ocf_log err "check_slave invoked on an instance that is not a replication slave."
exit $OCF_ERR_GENERIC
fi
}
set_master() {
local new_master master_log_file master_log_pos new_master_info
local master_params new_master_log_file new_master_log_pos
if [ "$glb_master_exists" ]; then
if [ "${#glb_remote_info}" -gt "0" ]; then
# geo_remote_IP is defined, let's do the booth part
if [ $glb_master_side -ne 0 ]; then
# this is _not_ the side with the token
new_master_info=`echo $glb_remote_info | awk '{ print $1 }'`
new_master=`echo $new_master_info | cut -d'|' -f1`
new_master_log_file=`echo $new_master_info | cut -d'|' -f2`
new_master_log_pos=`echo $new_master_info | cut -d'|' -f3`
fi
fi
if [ "${#new_master_info}" -eq "0" ]; then
new_master=`echo $glb_local_info | cut -d'|' -f1`
new_master_log_file=`echo $glb_local_info | cut -d'|' -f2`
new_master_log_pos=`echo $glb_local_info | cut -d'|' -f3`
fi
# Keep replication position
get_slave_info
if [ "$master_log_file" -a "$new_master" = "$master_host" ]; then
# master_params=", MASTER_LOG_FILE='$master_log_file', \
# MASTER_LOG_POS=$master_log_pos"
if [ "$1" = "nologging" ]; then
:
else
ocf_log info "Kept master pos for $master_host : $master_log_file:$master_log_pos"
fi
return
else
if [ -n "$new_master_log_file" -a -n "$new_master_log_pos" ]; then
master_params=", MASTER_LOG_FILE='$new_master_log_file', \
MASTER_LOG_POS=$new_master_log_pos"
ocf_log info "Restored master pos for $new_master : $new_master_log_file:$new_master_log_pos"
fi
fi
# Informs the MySQL 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.
mysql_run -Q -sw $MYSQL $MYSQL_OPTIONS_REPL \
-e "STOP SLAVE;CHANGE MASTER TO MASTER_HOST='$new_master', \
MASTER_PORT=$OCF_RESKEY_replication_port, \
MASTER_USER='$OCF_RESKEY_replication_user', \
MASTER_PASSWORD='$OCF_RESKEY_replication_passwd' \
$OCF_RESKEY_replication_options $master_params;START SLAVE;"
fi
}
unset_master(){
# Instructs the MySQL server to stop replicating from a master
# host.
# If we're currently not configured to be replicating from any
# host, then there's nothing to do. But we do log a warning as
# no-one but the CRM should be touching the MySQL master/slave
# configuration.
is_slave
rc=$?
if [ $rc -ne 0 ]; 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
local tmpfile
tmpfile=`mktemp ${HA_RSCTMP}/unset_master.${OCF_RESOURCE_INSTANCE}.XXXXXX`
# At this point, the master is read only so there should not be much binlogs to transfer
# Let's wait for the last bits
while true; do
get_slave_info
rc=$?
# Is the slave_io thread running?
if [ "$slave_io" != 'Yes' ]; then
ocf_log info "Slave IO thread not running, master likely dead or stopped"
break;
fi
mysql_run -Q -sw -O $MYSQL $MYSQL_OPTIONS_REPL \
-e 'SHOW PROCESSLIST\G' > $tmpfile
if grep -i 'Master has sent all binlog to slave' $tmpfile >/dev/null; then
ocf_log info "MySQL slave has finished reading master binary log"
break
fi
if grep -i 'Waiting for master to send event' $tmpfile >/dev/null; then
ocf_log info "MySQL slave has finished reading master binary log"
break
fi
if grep -i 'Waiting to reconnect after' $tmpfile >/dev/null; then
ocf_log info "Master is down, no more binary logs to come"
break
fi
if grep -i 'Reconnecting after a failed' $tmpfile >/dev/null; then
ocf_log info "Master is down, no more binary logs to come"
break
fi
if grep -i 'Connecting to master' $tmpfile >/dev/null; then
ocf_log info "Master is down, no more binary logs to come"
break
fi
if ! grep 'system user' $tmpfile >/dev/null; then
ocf_log info "Slave is not running - not waiting to finish"
break
fi
sleep 1
done
# Now, stop the slave I/O thread and wait for relay log
# processing to complete
mysql_run -Q -sw $MYSQL $MYSQL_OPTIONS_REPL \
-e "STOP SLAVE IO_THREAD"
if [ $? -gt 0 ]; then
ocf_log err "Error stopping slave IO thread"
rm -f $tmpfile
exit $OCF_ERR_GENERIC
fi
while true; do
mysql_run -Q -sw -O $MYSQL $MYSQL_OPTIONS_REPL \
-e 'SHOW PROCESSLIST\G' > $tmpfile
# Of course, slave messages changed over MySQL versions...
if grep -i 'Has read all relay log' $tmpfile >/dev/null; then
ocf_log info "MySQL 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 MySQL slave to finish processing relay log"
sleep 1
done
rm -f $tmpfile
# Now, stop all slave activity and unset the master host
mysql_run -Q -sw $MYSQL $MYSQL_OPTIONS_REPL \
-e "STOP SLAVE"
if [ $? -gt 0 ]; then
ocf_log err "Error stopping rest slave threads"
exit $OCF_ERR_GENERIC
fi
# a last get_slave_info to save the status in variables may be needed
# by pre-promote notification
unset master_host # need to unset for get_slave_info to run
get_slave_info
# First, where are the relay logs? That will be easier when the using_multi_config
# branch will be merged. If the path is not defined, the output will be "."
relaylog_path=`${OCF_RESKEY_binary} --defaults-file=$OCF_RESKEY_config --verbose --help --user=$OCF_RESKEY_user | grep -e '^relay-log ' | awk '{ print $2 }'`
relaylog_path=`dirname $relaylog_path`
if [ "$relaylog_path" == "." ]; then
relaylog_path=$OCF_RESKEY_datadir
fi
last_relaylog_file="$relay_log_file"
cp ${relaylog_path}/${last_relaylog_file} /tmp/${last_relaylog_file}
#ok now we need to find the md5 of the last trx
last_trx_md5=$(get_last_binlog /tmp/${last_relaylog_file} 1 60)
last_trx_md5=`echo $last_trx_md5 | cut -d',' -f2`
if [ ! -z "$last_trx_md5" ]; then
$CRM_ATTR_NODE_LAST_TRX_MD5 -v $last_trx_md5
fi
# Let's establish the master score based on the following
# rule.
# score = ((file number diff)*master_max_binlog_size
# + fileposdiff/10 + constante
#
# All events are at least 10 bytes so dividing by 10 doesn't reduce
# the resolution and increases the span.
#
# Since the master publishes its status only once per few
# seconds, the fileposdiff is likely positive.
# We'll cap all values to int signed range et target
# 1B as the value if a slave is fully in sync with the master
# Master-score is normally calculated from slave-lag but it is better to use
# binlog offset position to pick the most up to date slave
local master_status_attr
local Iter=5
while [ "$Iter" -gt 0 ]; do
master_status_attr=`$CRM_ATTR_REPL_STATUS --query`
if [ "$?" -eq 0 ]; then
break;
else
sleep 1
let Iter-=1
fi
done
if [ $? -eq 0 ]; then
# There's a master status entry although we don't know if it is
# a valid one
local last_reported_master_file last_reported_master_pos master_max_binlog_size
local last_reported_master_file_number master_log_file_number master_score
last_reported_master_file=`echo $master_status_attr | cut -d'|' -f1`
last_reported_master_pos=`echo $master_status_attr | cut -d'|' -f2`
master_max_binlog_size=`echo $master_status_attr | cut -d'|' -f3`
last_reported_master_file_number=`echo $last_reported_master_file | cut -d'.' -f2 | sed -ne "s/^0*\([1-9][0-9]*\)$/\1/p"`
master_log_file_number=`echo $master_log_file | cut -d'.' -f2 | sed -ne "s/^0*\([1-9][0-9]*\)$/\1/p"`
master_score=$((1000000000+(\
($master_log_file_number-$last_reported_master_file_number)*\
$master_max_binlog_size+$master_log_pos-$last_reported_master_pos)/100))
# now, the caps, the upper cap is unlikely
if [ $master_score -gt 2147483647 ]; then
master_score=2147483647
fi
# the lower cap could happened if a slave lags behind by
# more then 30GB of binlog. In that case... do we really care
# if we floor the value
if [ $master_score -lt -2147483647 ]; then
master_score=-2147483647
fi
set_master_score $master_score
# Next, we need a reminder that the master crashed and when, that will be
# used to publish the last trx in the promote event if we are picket
# as the new master.
$CRM_ATTR_MASTER_CRASHED_TS -v `date +%s`
# Let's give pacemaker some time to realize there are new scores
sleep 1
fi
mysql_run -Q -sw $MYSQL $MYSQL_OPTIONS_REPL \
-e "RESET SLAVE /*!50516 ALL */;"
if [ $? -gt 0 ]; then
ocf_log err "Failed to reset slave"
exit $OCF_ERR_GENERIC
fi
}
# Start replication as slave
start_slave() {
mysql_run -Q -sw $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 [ "$1" -eq "0" ]; then
if [ "$curr_attr_value" -gt "0" ]; then
curr_attr_value=$((${curr_attr_value}-1))
$CRM_ATTR -l reboot --name ${OCF_RESKEY_reader_attribute} -v $curr_attr_value
else
$CRM_ATTR -l reboot --name ${OCF_RESKEY_reader_attribute} -v 0
fi
else
if [ "$curr_attr_value" -ne "$OCF_RESKEY_reader_failcount" ]; then
$CRM_ATTR -l reboot --name ${OCF_RESKEY_reader_attribute} -v $OCF_RESKEY_reader_failcount
fi
fi
}
is_master_side() {
#Returns true (0) if this cluster has a grant for the booth ticket OCF_RESKEY_booth_master_ticket
local ticket crmTicketRet
if [ "${#OCF_RESKEY_geo_remote_IP}" -gt "0" ]; then
#Try the new format
crmTicketRet=`file $CRM_TICKET | grep -c 'Bourne-Again shell script'`
if [ "$crmTicketRet" -eq "1" ]; then
# got an error, we assume the old format
ticket=`$CRM_TICKET -t $OCF_RESKEY_booth_master_ticket -Q | grep -c 'true'`
else
ticket=`$CRM_TICKET --info | grep $OCF_RESKEY_booth_master_ticket | awk '{ print $2 }' | grep -c granted`
fi
if [ "$ticket" -eq "1" ]; then
return 0
else
return 1
fi
else
return 0
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 $attr_value
else
echo -1
fi
}
# Stores data for MASTER STATUS from MySQL
update_data_master_status() {
master_status_file="${HA_RSCTMP}/master_status.${OCF_RESOURCE_INSTANCE}"
mysql_run -Q -sw -O $MYSQL $MYSQL_OPTIONS_REPL -e "SHOW MASTER STATUS\G" > $master_status_file
}
# Returns the specified value from the stored copy of SHOW MASTER STATUS.
# should be call after update_data_master_status for tmpfile
# Arguments:
# $1 The value to get.
get_master_status() {
awk -v var="$1" '$1 == var ":" {print substr($0, index($0, ":") + 2)}' "$master_status_file"
}
# 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=10.2.2.1611
# 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`
if [ ! $? -eq 0 ]; then
uname -n
else
echo $IP
fi
}
# Determine if the datadir is full or almost full, the threshold is 97%
check_datadir_state() {
# Get the free space of the binlogdir
FREE_SPC_PCT=`/bin/df -P $OCF_RESKEY_datadir | /bin/grep -v Filesystem \
| /bin/sed -e 's/ \+/ /g' | /usr/bin/cut -d' ' -f 5 \
| /usr/bin/tr -d '%'`
if [ "$FREE_SPC_PCT" -ge "97" ]; then
ocf_log warn "Partition $OCF_RESKEY_datadir usage is at " \
"or more than 97, " \
"unsuitable for master..."
return 1
fi
return 0
}
#######################################################################
# Functions invoked by resource manager actions
mysql_validate() {
check_binary $OCF_RESKEY_binary
check_binary $OCF_RESKEY_client_binary
if [ ! -f $OCF_RESKEY_config ]; then
ocf_log err "Config $OCF_RESKEY_config doesn't exist";
return $OCF_ERR_INSTALLED;
fi
if [ ! -d $OCF_RESKEY_datadir ]; then
ocf_log err "Datadir $OCF_RESKEY_datadir doesn't exist";
return $OCF_ERR_INSTALLED;
fi
getent passwd $OCF_RESKEY_user >/dev/null 2>&1
if [ ! $? -eq 0 ]; then
ocf_log err "User $OCF_RESKEY_user doesn't exit";
return $OCF_ERR_INSTALLED;
fi
getent group $OCF_RESKEY_group >/dev/null 2>&1
if [ ! $? -eq 0 ]; then
ocf_log err "Group $OCF_RESKEY_group doesn't exist";
return $OCF_ERR_INSTALLED;
fi
true
}
# Return the status of mysqld
# $1 the loglevel to use (mandatory)
# $2 Override async_stop if 1, default to 0
mysql_status() {
local last_restart_ts
local kill_exit_code
local witness_pid
local override_async_stop
override_async_stop=0
test $2 -eq 0 2>/dev/null
if [ $? -ne 2 ]; then
override_async_stop=$2
fi
if [ ! -e $OCF_RESKEY_pid ]; then
ocf_log $1 "MySQL is not running"
if [ -e $ASYNC_STOP_WITNESS_FILE ]; then
# MySQL is stopped and there's a witness file, cleanup
rm -f $ASYNC_STOP_WITNESS_FILE
fi
return $OCF_NOT_RUNNING;
fi
pid=`cat $OCF_RESKEY_pid`;
if [ -d /proc -a -d /proc/1 ]; then
[ "u$pid" != "u" -a -d /proc/$pid ]
else
kill -s 0 $pid >/dev/null 2>&1
fi
kill_exit_code=$?
if [ "$OCF_RESKEY_async_stop" -eq "1" -a \
-e $ASYNC_STOP_WITNESS_FILE ]; then
# Async stop seems to be in progress
witness_pid=`cat $ASYNC_STOP_WITNESS_FILE | grep pid | cut -d':' -f2`
if [ $kill_exit_code -eq 0 -a $witness_pid -eq $pid ]; then
# Should we lie or tell the truth
if [ "$override_async_stop" -eq "0" ]; then
# we lie
# still running but because of async, we report stopped
return $OCF_NOT_RUNNING
else
# we tell the truth
return $OCF_STOPPING #custom error code
fi
else
# That shouldn't happen execpt if SIGKILL, cleanup
rm -f $ASYNC_STOP_WITNESS_FILE
fi
fi
if [ $kill_exit_code -eq 0 ]; then
return $OCF_SUCCESS
else
ocf_log $1 "MySQL not running: removing old PID file"
rm -f $OCF_RESKEY_pid
# This is abnormal, is this host the master defined in the cib?
# Also confirm it succeed in starting with the socket file
if [ "$glb_master_exists" -eq "1" -a "$glb_cib_master" = $(get_local_ip) \
-a -e "$OCF_RESKEY_socket" ]; then
#This is a crashed master
if [ "$OCF_RESKEY_try_restart_crashed_master" -eq "1" ]; then
# This is the master, let's give it a change to restart
# that will allow the slaves a better chance to sync but we
# need to avoid letting it restart forever. Has it tried to
# restart within the last hour
last_crash_ts=`$CRM_ATTR -l reboot --name ${INSTANCE_ATTR_NAME}_last_crash --query -q --default=0`
if [ "$last_crash_ts" -ne "0" ]; then
if [ `date +%s` -lt "$((${last_crash_ts}+3600))" ]; then
# too soon, multiple crash, let's error out
set_master_score 0
return $OCF_NOT_RUNNING
fi
fi
$CRM_ATTR -l reboot --name ${INSTANCE_ATTR_NAME}_last_crash -v `date +%s`
mysql_start_low
rc=$?
if [ "$rc" -eq "0" ]; then
set_read_only OFF
fi
return $rc
else
$CRM_ATTR -l reboot --name ${INSTANCE_ATTR_NAME}_last_crash -v `date +%s`
# OCF_ERR_ARGS is a hard error, won't wait for restart
set_master_score 0
return $OCF_ERR_ARGS
fi
fi
return $OCF_NOT_RUNNING
fi
}
mysql_monitor() {
local rc
local status_loglevel="err"
local master_resource
local master_status_attr
local new_master_status_attr
: ${OCF_RESKEY_CRM_meta_interval=0}
# Set loglevel to info during probe
if ocf_is_probe; then
status_loglevel="info"
fi
mysql_status $status_loglevel
rc=$?
# TODO: check max connections error
# If status returned an error, return that immediately
if [ $rc -ne $OCF_SUCCESS ]; then
return $rc
fi
if [ $OCF_CHECK_LEVEL -gt 0 -a -n "$OCF_RESKEY_test_table" ]; then
# Check if this instance is configured as a slave, and if so
# check slave status
# Are we currently having a master?
if [ "$glb_master_exists" -ne "0" ]; then
is_slave
rc=$?
if [ $rc -eq 0 -o "$OCF_RESKEY_CRM_meta_role" = "Slave" ]; then
check_slave
else
update_data_master_status
master_status_attr=`$CRM_ATTR_REPL_STATUS --query -q`
new_master_status_attr="$(get_master_status File)|$(get_master_status Position)|$(get_max_binlog_size)"
rm -f $master_status_file
if [ "$master_status_attr" != "$new_master_status_attr" ]; then
# Doing in bg, no need to wait and that can hang if a node is lost at the same time
$CRM_ATTR_REPL_STATUS -v "$new_master_status_attr" &
fi
# Is this following a recent master crash?
master_crashed_ts=`$CRM_ATTR_MASTER_CRASHED_TS --query --default=0`
if [ "$master_crashed_ts" -gt "0" ]; then
if [ `date +%s` -gt "$((${master_crashed_ts}+3600))" ]; then
#Let's cleanup the cib
$CRM_ATTR_MASTER_CRASHED_TS -D
$CRM_ATTR_LAST_TRX -D
fi
fi
fi
else
is_slave
rc=$?
# Need to cover for crashed master... if so, unset_master will set
# the last trx md5 in the cib ...
if [ $rc -eq 0 -o "$OCF_RESKEY_CRM_meta_role" = "Slave" ]; then
# If there no quorum, we will not reset master
# This is optimistic setting as the original master could be part of
# majority. If there is no quorum, this node will restart in any case
if [ "$cluster_has_quorum" -ne "0" ]; then
ocf_log info "Cluster has quorum, resetting replication"
unset_master
fi
set_reader_attr 0
fi
fi
# Check for test table
mysql_run -Q -sw $MYSQL $MYSQL_OPTIONS_TEST \
-e "SELECT COUNT(*) FROM $OCF_RESKEY_test_table"
if [ $MYSQL_LAST_ERR -ne "$MYSQL_TOO_MANY_CONN_ERR" ]; then
if [ $MYSQL_LAST_ERR -ne 0 ]; then
ocf_log err "Failed to select from $OCF_RESKEY_test_table";
return $OCF_ERR_GENERIC;
fi
else
ocf_log info "Master hit max_connections"
fi
fi
if ocf_is_ms && ! get_read_only; then
ocf_log debug "MySQL monitor succeeded (master)";
if [ "$OCF_RESKEY_CRM_meta_interval" -eq "0" ]; then
# this is a probe and this server is a master so need to set master_score
set_master_score 2147483647
set_reader_attr 1
fi
if ! check_datadir_state; then
set_master_score -2147483640
fi
return $OCF_RUNNING_MASTER
else
ocf_log debug "MySQL monitor succeeded";
return $OCF_SUCCESS
fi
}
# Start MySQL in the master-slave context
mysql_start() {
local current_status
if ocf_is_ms; then
# Initialize the ReaderVIP attribute, monitor will enable it
set_reader_attr 0
# set master_score to 0 in case mysql crashes on startup
set_master_score 0
fi
mysql_status info 1 # Adding 2nd param here to get the true state in case
# async_stop is used
current_status=$?
if [ "$current_status" = "$OCF_SUCCESS" ]; then
ocf_log info "MySQL already running"
return $OCF_SUCCESS
fi
# Is MySQL still stopping, OCF_STOPPING is a custom error code
if [ "$current_status" = "$OCF_STOPPING" ]; then
# Waiting 5s less than the op timeout
shutdown_timeout=$((($OCF_RESKEY_CRM_meta_timeout/1000)-5))
count=0
while [ $count -lt $shutdown_timeout ]
do
sleep 1
mysql_status info 1
if [ "$?" != "$OCF_STOPPING" ]; then
break
fi
count=$(($count+1))
ocf_log debug "MySQL hasn't stopped yet (async_stop). Waiting before trying to start."
done
# Has MySQL stopped in time
if [ $count -eq $shutdown_timeout ]; then
# No, the start request was too soon
ocf_log err "MySQL asked to start while still stopping"
return $OCF_ERR_GENERIC
fi
fi
mysql_start_low
rc=$?
if [ $rc != $OCF_SUCCESS ]; then
ocf_log err "Wasn't able to start MySQL, stopping 'start'."
return $rc
fi
if ocf_is_ms; then
# We're configured as a stateful resource. We must start as
# slave by default. At this point we don't know if the CRM has
# already promoted a master. So, we simply start in read only
# mode. Should already be from command line.
set_read_only on
# Now, let's see whether there is a master. We might be a new
# node that is just joining the cluster, and the CRM may have
# promoted a master before.
if [ "$glb_master_exists" -ne 0 -a "$glb_cib_master" != $(get_local_ip) ]; then
# In case of network issues, we need to make sure not to reset replication
# because REPL_INFO stored on CIBADMIN can be outdated, if that happens
# we will break replication.
# First, since this is a slave, let's check for current replication
# info, if the master host matches current master IP, we should just
# resume replication, otherwise we should reset.
get_slave_info
rc=$?
if [ $rc -eq 0 -a "$glb_cib_master" == "$master_host" ]; then
ocf_log info "Current Master_Host matches current cluster master, starting slave"
else
ocf_log info "Changing MySQL configuration to replicate from $master_host."
set_master
fi
start_slave
if [ $? -ne 0 ]; then
ocf_log err "Failed to start slave"
return $OCF_ERR_GENERIC
fi
else
ocf_log info "No MySQL master present - clearing replication state"
unset_master
fi
# We also need to set a master preference, otherwise Pacemaker
# won't ever promote us in the absence of any explicit
# preference set by the administrator. We choose a low
# greater-than-zero preference.
set_master_score 1
fi
# Initial monitor action
if [ -n "$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_log err "Failed initial monitor action"
return $rc
fi
ocf_log info "MySQL started"
return $OCF_SUCCESS
}
# low level MySQL start
mysql_start_low() {
touch $OCF_RESKEY_log
chown $OCF_RESKEY_user:$OCF_RESKEY_group $OCF_RESKEY_log
chmod 0640 $OCF_RESKEY_log
[ -x /sbin/restorecon ] && /sbin/restorecon $OCF_RESKEY_log
if ocf_is_true "$OCF_RESKEY_enable_creation" && [ ! -d $OCF_RESKEY_datadir/mysql ] ; then
ocf_log info "Initializing MySQL database: "
$MYSQL_SBINDIR/mysql_install_db --datadir=$OCF_RESKEY_datadir
rc=$?
if [ $rc -ne 0 ] ; then
ocf_log err "Initialization failed: $rc";
exit $OCF_ERR_GENERIC
fi
chown -R $OCF_RESKEY_user:$OCF_RESKEY_group $OCF_RESKEY_datadir
fi
pid_dir=`dirname $OCF_RESKEY_pid`
if [ ! -d $pid_dir ] ; then
ocf_log info "Creating PID dir: $pid_dir"
mkdir -p $pid_dir
chown $OCF_RESKEY_user:$OCF_RESKEY_group $pid_dir
fi
socket_dir=`dirname $OCF_RESKEY_socket`
if [ ! -d $socket_dir ] ; then
ocf_log info "Creating socket dir: $socket_dir"
mkdir -p $socket_dir
chown $OCF_RESKEY_user:$OCF_RESKEY_group $socket_dir
fi
# Regardless of whether we just created the directory or it
# already existed, check whether it is writable by the configured
# user
for dir in $pid_dir $socket_dir; do
# needed to wrap around su a bit, sssd causing issue
if [ `su - $OCF_RESKEY_user -s /bin/bash -c "if test -w $dir; then echo yes; else echo no; fi" 2> /dev/null` != "yes" ]; then
ocf_log err "Directory $dir is not writable by $OCF_RESKEY_user"
exit $OCF_ERR_PERM;
fi
done
# Uncomment to perform permission clensing
# - not convinced this should be enabled by default
#
#chmod 0755 $OCF_RESKEY_datadir
#chown -R $OCF_RESKEY_user $OCF_RESKEY_datadir
#chgrp -R $OCF_RESKEY_group $OCF_RESKEY_datadir
mysql_extra_params=
if ocf_is_ms; then
mysql_extra_params="$mysql_extra_params --skip-slave-start --read-only"
fi
${OCF_RESKEY_binary_prefix} ${OCF_RESKEY_binary} --defaults-file=$OCF_RESKEY_config \
--pid-file=$OCF_RESKEY_pid \
--socket=$OCF_RESKEY_socket \
--datadir=$OCF_RESKEY_datadir \
--user=$OCF_RESKEY_user $OCF_RESKEY_additional_parameters \
$mysql_extra_params >/dev/null 2>&1 &
rc=$?
# we also get the process id from $! because the PID file is only
# created by mysql as soon as mysql is fully up and running
# for example, when recovery is busy, the pid file does not exist yet
# this part already creates the PID file as the mysql user
# so that other PRM checks know
# When recovery happens, the PID file does not exist yet.
process_pid=$!
# mysql_status expects that if the pid is there and it's running
# that mysql is completely active
#su $OCF_RESKEY_user -c "echo '$process_pid' > $OCF_RESKEY_pid"
echo "$process_pid" > ${OCF_RESKEY_pid}.starting
if [ $rc != 0 ]; then
ocf_log err "MySQL start command failed: $rc"
return $rc
fi
# Spin waiting for the server to come up.
# Let the CRM/LRM time us out if required.
start_wait=1
while [ $start_wait = 1 ]; do
mysql_status info
rc=$?
if [ $rc = $OCF_SUCCESS ]; then
start_wait=0
elif [ $rc != $OCF_NOT_RUNNING ]; then
ocf_log info "MySQL start failed: $rc"
return $rc
fi
# if mysql died in the meantime, we shall not wait
# until the timeout is reached.
kill -s 0 $process_pid > /dev/null
mysqld_pid_status=$?
if [ "$mysqld_pid_status" -ne "0" ]; then
ocf_log err "MySQL daemon died during start, giving up."
return $OCF_ERR_GENERIC
fi
sleep 2
done
return $OCF_SUCCESS
}
mysql_stop() {
if ocf_is_ms; then
# clear preference for becoming master
$CRM_MASTER -D
# Remove VIP capability
set_reader_attr 0
fi
# we rely only on ${OCF_RESKEY_pid}.starting
# as this certainly contains the file we need with the PID
if [ ! -f ${OCF_RESKEY_pid}.starting ]; then
ocf_log info "MySQL is not running"
return $OCF_SUCCESS
fi
pid=`cat ${OCF_RESKEY_pid}.starting 2> /dev/null`
/bin/kill $pid > /dev/null
rc=$?
if [ $rc != 0 ]; then
ocf_log err "MySQL couldn't be stopped"
return $OCF_ERR_GENERIC
fi
if [ "$OCF_RESKEY_async_stop" -eq "1" ]; then
#Ok, MySQL is stopping and the async_stop option is set, just put the
#pid and a timestamp in the witness file and return
echo "pid:$pid" > $ASYNC_STOP_WITNESS_FILE
echo "ts:`date +%s`" >> $ASYNC_STOP_WITNESS_FILE
#Don't know yet why the ts, just seems useful for debugging for now
ocf_log info "MySQL async stopped";
return $OCF_SUCCESS
fi
# stop waiting
shutdown_timeout=15
if [ -n "$OCF_RESKEY_CRM_meta_timeout" ]; then
shutdown_timeout=$((($OCF_RESKEY_CRM_meta_timeout/1000)-5))
fi
#Normally, pacemaker handles timeout but here we want to be able to do
#a SIGKILL (-9) before the timeout occurs.
count=0
while [ $count -lt $shutdown_timeout ]
do
kill -s 0 $pid
rc=$?
if [ $rc -ne 0 ]; then
break
fi
count=`expr $count + 1`
sleep 1
ocf_log debug "MySQL still hasn't stopped yet. Waiting..."
done
kill -s 0 $pid
if [ $? -eq 0 ]; then
ocf_log info "MySQL failed to stop after ${shutdown_timeout}s using SIGTERM. Trying SIGKILL..."
/bin/kill -KILL $pid > /dev/null
fi
rm ${OCF_RESKEY_pid}.starting
ocf_log info "MySQL stopped";
rm -f /var/lock/subsys/mysqld
rm -f $OCF_RESKEY_socket
return $OCF_SUCCESS
}
mysql_promote() {
local master_info
local master_crashed_ts
local log_bin_path
local tmpfiletrx
local nb_trx
local last_binlog_number
if ( ! mysql_status err ); then
return $OCF_NOT_RUNNING
fi
unset_master
# Set Master Info in CIB, cluster level attribute
update_data_master_status
master_info="$(get_local_ip)|$(get_master_status File)|$(get_master_status Position)"
${CRM_ATTR_REPL_INFO} -v "$master_info"
rm -f $master_status_file
master_crashed_ts=`$CRM_ATTR_MASTER_CRASHED_TS --query`
if [ ! -z "$master_crashed_ts" ]; then
if [ `date +%s` -lt "$((${master_crashed_ts}+3600))" ]; then
# Master crashed less than 1h ago, let's publish the last trx
# First, where are the binlogs? That will be easier when the using_multi_config
# branch will be merged. If the path is not defined, the output will be "."
log_bin_path=`${OCF_RESKEY_binary} --defaults-file=$OCF_RESKEY_config --verbose --help --user=$OCF_RESKEY_user | grep -e '^log-bin ' | awk '{ print $2 }'`
log_bin_path=`dirname $log_bin_path`
if [ "$log_bin_path" == "." ]; then
log_bin_path=$OCF_RESKEY_datadir
fi
# Let's find the last binlog file
update_data_master_status
last_binlog_file="$(get_master_status File)"
tmpfiletrx=`mktemp ${HA_RSCTMP}/trx.${OCF_RESOURCE_INSTANCE}.XXXXXX`
# Are we at the beginning of a binlog file, if so mtime of the previous will
# be less than 1min
last_binlog_number=`echo $last_binlog_file | cut -d'.' -f2 | sed -ne "s/^0*\([1-9][0-9]*\)$/\1/p"`
prev_binlog_number=$(($last_binlog_number-1))
last_binlog_file_prefix="`echo $last_binlog_file | cut -d'.' -f1`"
if [ "$prev_binlog_number" -gt 0 ]; then
# re-adding the 0 padding
while [ "${#prev_binlog_number}" -lt "6" ]; do
prev_binlog_number="0${prev_binlog_number}"
done
prev_binlog_mtime=`stat -c %Y ${log_bin_path}/${last_binlog_file_prefix}.${prev_binlog_number}`
now=`date +%s`
tDiff=$(($now-$prev_binlog_mtime))
if [ "$tDiff" -lt 60 ]; then
get_last_binlog '${log_bin_path}/${last_binlog_file_prefix}.${prev_binlog_number} ${log_bin_path}/${last_binlog_file}' 3000 60 > $tmpfiletrx
else
get_last_binlog ${log_bin_path}/${last_binlog_file} 3000 60 > $tmpfiletrx
fi
else
get_last_binlog ${log_bin_path}/${last_binlog_file} 3000 60 > $tmpfiletrx
fi
#now we load all that to the cib so that it reaches the other nodes
$CRM_ATTR_LAST_TRX -v "`cat $tmpfiletrx | tr '\n' '|'`"
rm -f $tmpfiletrx
fi
fi
set_read_only off || return $OCF_ERR_GENERIC
# Existing master gets a higher-than-default master preference, so
# the cluster manager does not shuffle the master role around
# unnecessarily
$CRM_ATTR_MASTER_CRASHED_TS -D 2> /dev/null
$CRM_MASTER -v 2147483647
# A master can accept reads
set_reader_attr 1
if [ "${#OCF_RESKEY_post_promote_script}" -gt "0" -a \
-x "${OCF_RESKEY_post_promote_script}" -a \
! -L "${OCF_RESKEY_post_promote_script}" ]; then
${OCF_RESKEY_post_promote_script}
fi
return $OCF_SUCCESS
}
mysql_demote() {
if ! mysql_status err; then
set_master_score 0
exit $OCF_SUCCESS
else
# Return master preference to default, so the cluster manager gets
# a chance to select a new master
set_master_score 1
exit $OCF_SUCCESS
fi
}
mysql_notify() {
local master_crashed_ts
# If not configured as a Stateful resource, we make no sense of
# notifications.
if ! ocf_is_ms; then
ocf_log info "This agent makes no use of notifications unless running in master/slave mode."
return $OCF_SUCCESS
fi
local type_op
type_op="${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')
get_slave_info
if [ $? -eq "$OCF_SUCCESS" ]; then
# We'll be here only if the master crashed and there has been no
# monitor event since. In the event of
# a graceful demote, a post-demote notification event would have occurred.
# The post-demote include an unset-master that
# resets the slave after the completion of the IO and SQL
# threads. The post-demote doesn't run if the master host
# crashed.
# Unset master will publish the updated master_score and the md5 of the
# last trx. The update of the master_score will trigger another
# pre-promote notification
unset_master
else
# Apparently we sometimes need to help Pacemaker choose the best host.
# We'll get here after running pre-promote a first time because
# the previous will have run unset_master
master_crashed_ts=`$CRM_ATTR_MASTER_CRASHED_TS --query`
if [ ! -z "$master_crashed_ts" ]; then
if [ `date +%s` -lt "$((${master_crashed_ts}+3600))" ]; then
highestScore=-2147483647
thisNodeScore=0
for node in $OCF_RESKEY_CRM_meta_notify_slave_uname; do
score=`$CRM_MASTER -N $node -G -q`
if [ "$node" == "$HOSTNAME" ]; then
thisNodeScore=$score
fi
if [ "$highestScore" -lt "$score" ]; then
highestScore=$score
fi
done
if [ "$thisNodeScore" -eq "$highestScore" ]; then
if [ "$thisNodeScore" -ne "2147483647" ]; then
# This node should be the next master
$CRM_MASTER -v 2147483647
fi
else
# This node shouldn't be the next master
if [ "$thisNodeScore" -ne "0" ]; then
$CRM_MASTER -v 0
fi
fi
fi
fi
fi
;;
'post-promote')
# The master has completed its promotion. Now is a good
# time to check whether our replication slave is working
# correctly.
# Is the notification for our set
notify_resource=`echo $OCF_RESKEY_CRM_meta_notify_promote_resource|cut -d: -f1`
my_resource=`echo $OCF_RESOURCE_INSTANCE|cut -d: -f1`
if [ $notify_resource != ${my_resource} ]; then
ocf_log debug "Notification is not for us"
return $OCF_SUCCESS
fi
master_host=`echo $OCF_RESKEY_CRM_meta_notify_promote_uname|tr -d " "`
if [ "$master_host" = ${HOSTNAME} ]; then
ocf_log info "This will be the new master, ignoring post-promote notification."
else
# Is this following a recent master crash?
master_crashed_ts=`$CRM_ATTR_MASTER_CRASHED_TS --query`
if [ ! -z "$master_crashed_ts" ]; then
if [ `date +%s` -lt "$((${master_crashed_ts}+3600))" ]; then
# Master crashed less than 1h ago, let's see what our last trx was.
# Since the master crashed, we didn't have the post-demote notification
# so the slave may still configured if no monitor ops has run
# if a monitor ops occurred, it should be saved in the cib
last_trx_md5=`$CRM_ATTR_NODE_LAST_TRX_MD5 --query`
if [ -z "$last_trx_md5" ]; then
# no last_trx_md5 set in the cib, let's try to find it
# First, where are the relay logs? That will be easier when the using_multi_config
# branch will be merged. If the path is not defined, the output will be "."
relaylog_path=`${OCF_RESKEY_binary} --defaults-file=$OCF_RESKEY_config --verbose --help --user=$OCF_RESKEY_user | grep -e '^relay-log ' | awk '{ print $2 }'`
relaylog_path=`dirname $relaylog_path`
if [ "$log_bin_path" == "." ]; then
relaylog_path=$OCF_RESKEY_datadir
fi
# Let's find the last binlog file
get_slave_info
last_relaylog_file="$relay_log_file"
#ok now we need to find the md5 of the last trx
last_trx_md5=$(get_last_binlog ${relaylog_path}/${last_relaylog_file} 1 60)
last_trx_md5=`echo $last_trx_md5 | cut -d',' -f2`
fi
if [ ! -z "$last_trx_md5" ]; then
# now, let's try to find this md5 in the NEW_MASTER_LAST_TRX attribute
# There maybe up to 2 binlog files in the attribute
#1st file
binlog_file=`$CRM_ATTR_LAST_TRX --query | cut -d'@' -f1`
binlog_pos=`$CRM_ATTR_LAST_TRX --query | cut -d'@' -f2 | tr '|' '\n' | grep $last_trx_md5 | cut -d',' -f1`
#found?
if [ -z "$binlog_pos" ]; then
#no, let's try if there's a 2nd file
binlog_file=`$CRM_ATTR_LAST_TRX --query | cut -d'@' -f3`
if [ -z "$binlog_file" ]; then
binlog_pos=`$CRM_ATTR_LAST_TRX --query | cut -d'@' -f4 | tr '|' '\n' | grep $last_trx_md5 | cut -d',' -f1`
fi
fi
# TODO: we could be at the end of the 2nd file so we should point to the first entry of the first
# file. Edge case, will deal with it later.
# have we found something?
if [ ! -z "$binlog_file" -a ! -z "$binlog_pos" ]; then
# Let's overwrite the glb_local_info variable
glb_local_info="`echo $glb_local_info | cut -d'|' -f1`|$binlog_file|$binlog_pos"
fi
$CRM_ATTR_NODE_LAST_TRX_MD5 -D
fi
fi
fi
ocf_log info "Resetting replication"
unset_master #Should be unset already execpt if master crashed
if [ $? -ne 0 ]; then
return $OCF_ERR_GENERIC
fi
ocf_log info "Changing MySQL configuration to replicate from $master_host"
set_master
if [ $? -ne 0 ]; then
return $OCF_ERR_GENERIC
fi
start_slave
if [ $? -ne 0 ]; then
ocf_log err "Failed to start slave"
return $OCF_ERR_GENERIC
fi
# Remove the master_crashed_ts attribute on this node since there has been a promotion
$CRM_ATTR_MASTER_CRASHED_TS -D 2> /dev/null
fi
return $OCF_SUCCESS
;;
'pre-demote')
# Is the notification for our set
notify_resource=`echo $OCF_RESKEY_CRM_meta_notify_demote_resource|cut -d: -f1`
my_resource=`echo $OCF_RESOURCE_INSTANCE|cut -d: -f1`
if [ $notify_resource != ${my_resource} ]; then
ocf_log debug "Notification is not for us"
return $OCF_SUCCESS
fi
demote_host=`echo $OCF_RESKEY_CRM_meta_notify_demote_uname|tr -d " "`
if [ $demote_host = ${HOSTNAME} ]; then
ocf_log info "post-demote notification for $demote_host"
while ! get_read_only
do
# This may timeout (5s) if there are long running queries (issue #44)
set_read_only on
# 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
tmpfile=`mktemp ${HA_RSCTMP}/threads.${OCF_RESOURCE_INSTANCE}.XXXXXX`
mysql_run -Q -sw -O $MYSQL $MYSQL_OPTIONS_REPL \
-e "SHOW PROCESSLIST" > $tmpfile
local killthreads=""
for thread in `awk '$0 !~ /Binlog Dump|system user|event_scheduler|SHOW PROCESSLIST/ && $0 ~ /^[0-9]/ {print $1}' $tmpfile`
do
killthreads="${killthreads} KILL ${thread};"
done
mysql_run -Q -sw $MYSQL $MYSQL_OPTIONS_REPL \
-e "${killthreads}"
rm -f $tmpfile
done
else
ocf_log info "Ignoring post-demote notification execpt for my own demotion."
fi
return $OCF_SUCCESS
;;
'post-demote')
# Is the notification for our set
notify_resource=`echo $OCF_RESKEY_CRM_meta_notify_demote_resource|cut -d: -f1`
my_resource=`echo $OCF_RESOURCE_INSTANCE|cut -d: -f1`
if [ $notify_resource != ${my_resource} ]; then
ocf_log debug "Notification is not for us"
return $OCF_SUCCESS
fi
demote_host=`echo $OCF_RESKEY_CRM_meta_notify_demote_uname|tr -d " "`
if [ $demote_host = ${HOSTNAME} ]; 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_run: Run a mysql command, log its output and return the proper error code.
# Usage: mysql_run [-Q] [-info|-warn|-err] [-O] [-sw] <command>
# -Q: don't log the output of the command if it succeeds
# -info|-warn|-err: log the output of the command at given
# severity if it fails (defaults to err)
# -O: echo the output of the command
# -sw: Suppress 5.6 client warning when password is used on the command line
# Adapted from ocf_run.
#
mysql_run() {
local rc
local output outputfile
local verbose=1
local returnoutput
local loglevel=err
local suppress_56_password_warning
local var
for var in 1 2 3 4
do
case "$1" in
"-Q")
verbose=""
shift 1;;
"-info"|"-warn"|"-err")
loglevel=`echo $1 | sed -e s/-//g`
shift 1;;
"-O")
returnoutput=1
shift 1;;
"-sw")
suppress_56_password_warning=1
shift 1;;
*)
;;
esac
done
outputfile=`mktemp ${HA_RSCTMP}/mysql_run.${OCF_RESOURCE_INSTANCE}.XXXXXX`
error=`"$@" 2>&1 1>$outputfile`
rc=$?
if [ "$suppress_56_password_warning" -eq 1 ]; then
error=`echo "$error" | egrep -v '^Warning: Using a password on the command line'`
fi
output=`cat $outputfile`
rm -f $outputfile
if [ $rc -eq 0 ]; then
if [ "$verbose" -a ! -z "$output" ]; then
ocf_log info "$output"
fi
if [ "$returnoutput" -a ! -z "$output" ]; then
echo "$output"
fi
MYSQL_LAST_ERR=$OCF_SUCCESS
return $OCF_SUCCESS
else
if [ ! -z "$error" ]; then
ocf_log $loglevel "$error"
regex='^ERROR ([[:digit:]]{4}).*'
if [[ $error =~ $regex ]]; then
mysql_code=${BASH_REMATCH[1]}
if [ -n "$mysql_code" ]; then
MYSQL_LAST_ERR=$mysql_code
return $rc
fi
fi
else
ocf_log $loglevel "command failed: $*"
fi
# No output to parse so return the standard exit code.
MYSQL_LAST_ERR=$rc
return $rc
fi
}
#######################################################################
##########################################################################
# If DEBUG_LOG is set, make this resource agent easy to debug: set up the
# debug log and direct all output to it. Otherwise, redirect to /dev/null.
# The log directory must be a directory owned by root, with permissions 0700,
# and the log must be writable and not a symlink.
##########################################################################
DEBUG_LOG="/tmp/mysql.ocf.ra.debug/log"
if [ "${DEBUG_LOG}" -a -w "${DEBUG_LOG}" -a ! -L "${DEBUG_LOG}" ]; then
DEBUG_LOG_DIR="${DEBUG_LOG%/*}"
if [ -d "${DEBUG_LOG_DIR}" ]; then
exec 9>>"$DEBUG_LOG"
exec 2>&9
date >&9
if [ "$OCF_RESKEY_CRM_meta_notify" == "true" ]; then
echo "$OCF_RESKEY_CRM_meta_notify_key_type - $OCF_RESKEY_CRM_meta_notify_key_operation" >&9
fi
echo "$*" >&9
env | grep OCF_ | sort >&9
set -x
else
exec 9>/dev/null
fi
fi
case "$1" in
meta-data) meta_data
exit $OCF_SUCCESS;;
usage|help) usage
exit $OCF_SUCCESS;;
esac
if [ "$#" -lt "1" ]; then
usage
exit $OCF_SUCCESS
fi
mysql_validate
rc=$?
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 $rc;;
esac
fi
#Global info missing from OCF_RESKEY
resources=`$CRM_RES --list`
# now we need the master-slave clone set name, need to walk around limitations
# of older pacemaker
if [[ "$OCF_RESKEY_crm_feature_set" > "3.0.1" ]]; then
glb_master_resource=`echo "$resources" | egrep "\[$INSTANCE_ATTR_NAME\]" | awk '{print $3}' | head -n 1`
else
# older versions of Pacemaker don't write the primitive name in the resources list
for msr in `echo "$resources" | grep 'Master/Slave' | awk '{print $3}'`; do
isThere=`$CRM_RES -q -r $msr | grep primitive | grep -c $INSTANCE_ATTR_NAME`
if [ "$isThere" -gt "0" ]; then
glb_master_resource=$msr
fi
done
fi
# We check if there is a location constraint against this node
# where $OCF_RESOURCE_INSTANCE should not be running here i.e.
# -INFINITY score, if so we ignore monitor call for this node
contrnt=$(timeout 5 cibadmin --query --xpath "/cib/configuration/constraints/rsc_location[@rsc='${glb_master_resource}']/rule[@score='-INFINITY']/expression[@attribute='#uname' and @value='${HOSTNAME}']" > /dev/null 2> /dev/null)
if [ "$?" -eq "0" ]; then
exit $OCF_SUCCESS
fi
is_master_side
glb_master_side=$?
if [ "${#OCF_RESKEY_geo_remote_IP}" -gt "0" -a $glb_master_side -ne 0 ]; then
# geo_remote_IP is defined, let's query the remote side
# the variable content will be like: pacemaker-1-1|binlog.000156|107 1
# The commands here are boolean tested so if any of them fails we will
# not get a MASTER_OK which indicates we got a useful value at least
ssh_cmd="$SSH $OCF_RESKEY_geo_remote_IP -l $OCF_RESKEY_geo_remote_user -o ConnectTimeout=3 -o PasswordAuthentication=no -q $OCF_RESKEY_geo_remote_opts"
ssh_query="$CRM_ATTR_REPL_INFO --query -q 2> /dev/null"
ssh_query="$ssh_query && $CRM_RES --list | egrep -A2 ' $glb_master_resource ' | egrep -c 'Master[^\/]'"
ssh_query="$ssh_query && echo 'MASTER_OK' || echo 'MASTER_NOT_FOUND'"
glb_remote_info=`$ssh_cmd "$ssh_query" | tr '\n' ' '`
echo "$glb_remote_info" | grep MASTER_OK > /dev/null 2>&1
if [ "$?" -ne "0" ]; then
glb_master_exists=0
else
glb_master_exists=`echo $glb_remote_info | awk '{ print $(NF-1) }'`
if [[ -z "$glb_master_exists" ]]; then
glb_master_exists=0
fi
fi
else
glb_master_exists=`echo "$resources" | grep -A2 " $glb_master_resource " | egrep -c 'Master[^\/]'`
fi
if [ "$glb_master_exists" -eq "1" ]; then
if [ "${#glb_remote_info}" -gt "0" ]; then
glb_cib_master=`echo $glb_remote_info | awk '{ print $1 }' | cut -d'|' -f1`
else
glb_local_info=`$CRM_ATTR_REPL_INFO --query -q`
glb_cib_master=`echo $glb_local_info | cut -d'|' -f1`
fi
fi
# Also check if the cluster has quorum, we may need to preserve some state
# If there was split brain and the cluster recovers
cluster_has_quorum=$(timeout 5 cibadmin --query|egrep -c 'have-quorum=\"1\"')
# What kind of method was invoked?
case "$1" in
start) mysql_start;;
stop) mysql_stop;;
status) mysql_status err;;
monitor) mysql_monitor;;
promote) mysql_promote;;
demote) mysql_demote;;
notify) mysql_notify;;
validate-all) exit $OCF_SUCCESS;;
*) usage
exit $OCF_ERR_UNIMPLEMENTED;;
esac