ha/stx-ocf-scripts/src/ocf/nova-conductor

436 lines
17 KiB
Bash

#!/bin/sh
#
#
# OpenStack Conductor Service (nova-conductor)
#
# Description: Manages an OpenStack Conductor Service (nova-conductor) process as an HA resource
#
# Authors: Sébastien Han
# Mainly inspired by the Glance API resource agent written by Martin Gerhard Loschwitz from Hastexo: http://goo.gl/whLpr
#
# Support: openstack@lists.launchpad.net
# License: Apache Software License (ASL) 2.0
#
# Copyright (c) 2014 Wind River Systems, Inc.
#
# SPDX-License-Identifier: Apache-2.0
#
#
# See usage() function below for more details ...
#
# OCF instance parameters:
# OCF_RESKEY_binary
# OCF_RESKEY_config
# OCF_RESKEY_user
# OCF_RESKEY_pid
# OCF_RESKEY_monitor_binary
# OCF_RESKEY_database_server_port
# OCF_RESKEY_amqp_server_port
# OCF_RESKEY_zeromq
# OCF_RESKEY_additional_parameters
#######################################################################
# Initialization:
: ${OCF_FUNCTIONS_DIR=${OCF_ROOT}/lib/heartbeat}
. ${OCF_FUNCTIONS_DIR}/ocf-shellfuncs
#######################################################################
# Fill in some defaults if no values are specified
OCF_RESKEY_binary_default="nova-conductor"
OCF_RESKEY_config_default="/etc/nova/nova.conf"
OCF_RESKEY_user_default="nova"
OCF_RESKEY_pid_default="$HA_RSCTMP/$OCF_RESOURCE_INSTANCE.pid"
OCF_RESKEY_database_server_port_default="3306"
OCF_RESKEY_amqp_server_port_default="5672"
OCF_RESKEY_zeromq_default="false"
: ${OCF_RESKEY_binary=${OCF_RESKEY_binary_default}}
: ${OCF_RESKEY_config=${OCF_RESKEY_config_default}}
: ${OCF_RESKEY_user=${OCF_RESKEY_user_default}}
: ${OCF_RESKEY_pid=${OCF_RESKEY_pid_default}}
: ${OCF_RESKEY_database_server_port=${OCF_RESKEY_database_server_port_default}}
: ${OCF_RESKEY_amqp_server_port=${OCF_RESKEY_amqp_server_port_default}}
: ${OCF_RESKEY_zeromq=${OCF_RESKEY_zeromq_default}}
#######################################################################
usage() {
cat <<UEND
usage: $0 (start|stop|validate-all|meta-data|status|monitor)
$0 manages an OpenStack ConductorService (nova-conductor) process as an HA resource
The 'start' operation starts the conductor service.
The 'stop' operation stops the conductor service.
The 'validate-all' operation reports whether the parameters are valid
The 'meta-data' operation reports this RA's meta-data information
The 'status' operation reports whether the conductor service is running
The 'monitor' operation reports whether the conductor service seems to be working
UEND
}
meta_data() {
cat <<END
<?xml version="1.0"?>
<!DOCTYPE resource-agent SYSTEM "ra-api-1.dtd">
<resource-agent name="nova-conductor">
<version>1.0</version>
<longdesc lang="en">
Resource agent for the OpenStack Nova Conductor Service (nova-conductor)
May manage a nova-conductor instance or a clone set that
creates a distributed nova-conductor cluster.
</longdesc>
<shortdesc lang="en">Manages the OpenStack Conductor Service (nova-conductor)</shortdesc>
<parameters>
<parameter name="binary" unique="0" required="0">
<longdesc lang="en">
Location of the OpenStack Nova Conductor server binary (nova-conductor)
</longdesc>
<shortdesc lang="en">OpenStack Nova Conductor server binary (nova-conductor)</shortdesc>
<content type="string" default="${OCF_RESKEY_binary_default}" />
</parameter>
<parameter name="config" unique="0" required="0">
<longdesc lang="en">
Location of the OpenStack Conductor Service (nova-conductor) configuration file
</longdesc>
<shortdesc lang="en">OpenStack Nova Conductor (nova-conductor) config file</shortdesc>
<content type="string" default="${OCF_RESKEY_config_default}" />
</parameter>
<parameter name="user" unique="0" required="0">
<longdesc lang="en">
User running OpenStack Conductor Service (nova-conductor)
</longdesc>
<shortdesc lang="en">OpenStack Conductor Service (nova-conductor) user</shortdesc>
<content type="string" default="${OCF_RESKEY_user_default}" />
</parameter>
<parameter name="pid" unique="0" required="0">
<longdesc lang="en">
The pid file to use for this OpenStack Conductor Service (nova-conductor) instance
</longdesc>
<shortdesc lang="en">OpenStack Conductor Service (nova-conductor) pid file</shortdesc>
<content type="string" default="${OCF_RESKEY_pid_default}" />
</parameter>
<parameter name="database_server_port" unique="0" required="0">
<longdesc lang="en">
The listening port number of the database server. Use for monitoring purposes
</longdesc>
<shortdesc lang="en">Database listening port</shortdesc>
<content type="integer" default="${OCF_RESKEY_database_server_port_default}" />
</parameter>
<parameter name="amqp_server_port" unique="0" required="0">
<longdesc lang="en">
The listening port number of the AMQP server. Use for monitoring purposes
</longdesc>
<shortdesc lang="en">AMQP listening port</shortdesc>
<content type="integer" default="${OCF_RESKEY_amqp_server_port_default}" />
</parameter>
<parameter name="zeromq" unique="0" required="0">
<longdesc lang="en">
If zeromq is used, this will disable the connection test to the AMQP server. Use for monitoring purposes
</longdesc>
<shortdesc lang="en">Zero-MQ usage</shortdesc>
<content type="boolean" default="${OCF_RESKEY_zeromq_default}" />
</parameter>
<parameter name="additional_parameters" unique="0" required="0">
<longdesc lang="en">
Additional parameters to pass on to the OpenStack Conductor Service (nova-conductor)
</longdesc>
<shortdesc lang="en">Additional parameters for nova-conductor</shortdesc>
<content type="string" />
</parameter>
</parameters>
<actions>
<action name="start" timeout="20" />
<action name="stop" timeout="20" />
<action name="status" timeout="20" />
<action name="monitor" timeout="30" interval="20" />
<action name="validate-all" timeout="5" />
<action name="meta-data" timeout="5" />
</actions>
</resource-agent>
END
}
#######################################################################
# Functions invoked by resource manager actions
nova_conductor_check_port() {
# This function has been taken from the squid RA and improved a bit
# The length of the integer must be 4
# Examples of valid port: "1080", "0080"
# Examples of invalid port: "1080bad", "0", "0000", ""
local int
local cnt
int="$1"
cnt=${#int}
echo $int |egrep -qx '[0-9]+(:[0-9]+)?(,[0-9]+(:[0-9]+)?)*'
if [ $? -ne 0 ] || [ $cnt -ne 4 ]; then
ocf_log err "Invalid port number: $1"
exit $OCF_ERR_CONFIGURED
fi
}
nova_conductor_validate() {
local rc
check_binary $OCF_RESKEY_binary
check_binary netstat
nova_conductor_check_port $OCF_RESKEY_database_server_port
nova_conductor_check_port $OCF_RESKEY_amqp_server_port
# A config file on shared storage that is not available
# during probes is OK.
if [ ! -f $OCF_RESKEY_config ]; then
if ! ocf_is_probe; then
ocf_log err "Config $OCF_RESKEY_config doesn't exist"
return $OCF_ERR_INSTALLED
fi
ocf_log_warn "Config $OCF_RESKEY_config not available during a probe"
fi
getent passwd $OCF_RESKEY_user >/dev/null 2>&1
rc=$?
if [ $rc -ne 0 ]; then
ocf_log err "User $OCF_RESKEY_user doesn't exist"
return $OCF_ERR_INSTALLED
fi
true
}
nova_conductor_status() {
local pid
local rc
if [ ! -f $OCF_RESKEY_pid ]; then
ocf_log info "OpenStack Nova Conductor (nova-conductor) is not running"
return $OCF_NOT_RUNNING
else
pid=`cat $OCF_RESKEY_pid`
fi
ocf_run -warn kill -s 0 $pid
rc=$?
if [ $rc -eq 0 ]; then
return $OCF_SUCCESS
else
ocf_log info "Old PID file found, but OpenStack Nova Conductor (nova-conductor) is not running"
rm -f $OCF_RESKEY_pid
return $OCF_NOT_RUNNING
fi
}
check_port() {
local port=$1
local pid=$2
netstat -punt | grep -s "$port" | grep -s "$pid" | grep -qs "ESTABLISHED"
rc=$?
if [ $rc -eq 0 ]; then
return 0
else
return 1
fi
}
nova_conductor_monitor() {
local rc
local pid
local rc_db
local rc_amqp
local conductor_db_check
local conductor_amqp_check
nova_conductor_status
rc=$?
# If status returned anything but success, return that immediately
if [ $rc -ne $OCF_SUCCESS ]; then
return $rc
fi
# Check the connections according to the PID.
# We are sure to hit the conductor process and not other nova process with the same connection behavior (for example nova-cert)
if ocf_is_true "$OCF_RESKEY_zeromq"; then
pid=`cat $OCF_RESKEY_pid`
rc_db=`check_port $OCF_RESKEY_database_server_port $pid`
if [ $rc_db -ne 0 ]; then
ocf_log err "Nova Conductor is not connected to the database server: $rc_db"
return $OCF_NOT_RUNNING
fi
else
pid=`cat $OCF_RESKEY_pid`
rc_db=`check_port $OCF_RESKEY_database_server_port $pid`
rc_amqp=`check_port $OCF_RESKEY_amqp_server_port $pid`
if [ $rc_amqp -ne 0 ] || [ $rc_db -ne 0 ]; then
# may have multiple workers, in which case $pid is the parent and we want to check the children
# If there are no children or at least one child is not connected to both DB and AMQP then we fail.
KIDPIDS=`pgrep -P $pid -f nova-conductor`
if [ ! -z "$KIDPIDS" ]; then
for pid in $KIDPIDS
do
rc_db=`check_port $OCF_RESKEY_database_server_port $pid`
rc_amqp=`check_port $OCF_RESKEY_amqp_server_port $pid`
if [ $rc_amqp -ne 0 ] || [ $rc_db -ne 0 ]; then
ocf_log err "Nova Conductor pid $pid is not connected to the AMQP server and/or the database server: AMQP connection test returned $rc_amqp and database connection test returned $rc_db"
return $OCF_NOT_RUNNING
fi
done
else
ocf_log err "Nova Conductor pid $pid is not connected to the AMQP server and/or the database server: AMQP connection test returned $rc_amqp and database connection test returned $rc_db"
return $OCF_NOT_RUNNING
fi
fi
fi
ocf_log debug "OpenStack Nova Conductor (nova-conductor) monitor succeeded"
return $OCF_SUCCESS
}
nova_conductor_start() {
local rc
nova_conductor_status
rc=$?
if [ $rc -eq $OCF_SUCCESS ]; then
ocf_log info "OpenStack Nova Conductor (nova-conductor) already running"
return $OCF_SUCCESS
fi
# Change the working dir to /, to be sure it's accesible
cd /
# Run the pre-start hooks. This can be used to trigger a nova database sync, for example.
/usr/bin/nova-controller-runhooks
# run the actual nova-conductor daemon. Don't use ocf_run as we're sending the tool's output
# straight to /dev/null anyway and using ocf_run would break stdout-redirection here.
su ${OCF_RESKEY_user} -s /bin/sh -c "${OCF_RESKEY_binary} --config-file=$OCF_RESKEY_config \
$OCF_RESKEY_additional_parameters"' >> /dev/null 2>&1 & echo $!' > $OCF_RESKEY_pid
# Spin waiting for the server to come up.
while true; do
nova_conductor_monitor
rc=$?
[ $rc -eq $OCF_SUCCESS ] && break
if [ $rc -ne $OCF_NOT_RUNNING ]; then
ocf_log err "OpenStack Nova Conductor (nova-conductor) start failed"
exit $OCF_ERR_GENERIC
fi
sleep 1
done
ocf_log info "OpenStack Nova Conductor (nova-conductor) started"
return $OCF_SUCCESS
}
nova_conductor_confirm_stop() {
local my_bin
local my_processes
my_binary=`which ${OCF_RESKEY_binary}`
my_processes=`pgrep -l -f "^(python|/usr/bin/python|/usr/bin/python2) ${my_binary}([^\w-]|$)"`
if [ -n "${my_processes}" ]
then
ocf_log info "About to SIGKILL the following: ${my_processes}"
pkill -KILL -f "^(python|/usr/bin/python|/usr/bin/python2) ${my_binary}([^\w-]|$)"
fi
}
nova_conductor_stop() {
local rc
local pid
nova_conductor_status
rc=$?
if [ $rc -eq $OCF_NOT_RUNNING ]; then
ocf_log info "OpenStack Nova Conductor (nova-conductor) already stopped"
nova_conductor_confirm_stop
return $OCF_SUCCESS
fi
# Try SIGTERM
pid=`cat $OCF_RESKEY_pid`
ocf_run kill -s TERM $pid
rc=$?
if [ $rc -ne 0 ]; then
ocf_log err "OpenStack Nova Conductor (nova-conductor) couldn't be stopped"
nova_conductor_confirm_stop
exit $OCF_ERR_GENERIC
fi
# stop waiting
shutdown_timeout=15
if [ -n "$OCF_RESKEY_CRM_meta_timeout" ]; then
shutdown_timeout=$((($OCF_RESKEY_CRM_meta_timeout/1000)-5))
fi
count=0
while [ $count -lt $shutdown_timeout ]; do
nova_conductor_status
rc=$?
if [ $rc -eq $OCF_NOT_RUNNING ]; then
break
fi
count=`expr $count + 1`
sleep 1
ocf_log debug "OpenStack Nova Conductor (nova-conductor) still hasn't stopped yet. Waiting ..."
done
nova_conductor_status
rc=$?
if [ $rc -ne $OCF_NOT_RUNNING ]; then
# SIGTERM didn't help either, try SIGKILL
ocf_log info "OpenStack Nova Conductor (nova-conductor) failed to stop after ${shutdown_timeout}s \
using SIGTERM. Trying SIGKILL ..."
ocf_run kill -s KILL $pid
fi
nova_conductor_confirm_stop
ocf_log info "OpenStack Nova Conductor (nova-conductor) stopped"
rm -f $OCF_RESKEY_pid
return $OCF_SUCCESS
}
#######################################################################
case "$1" in
meta-data) meta_data
exit $OCF_SUCCESS;;
usage|help) usage
exit $OCF_SUCCESS;;
esac
# Anything except meta-data and help must pass validation
nova_conductor_validate || exit $?
# What kind of method was invoked?
case "$1" in
start) nova_conductor_start;;
stop) nova_conductor_stop;;
status) nova_conductor_status;;
monitor) nova_conductor_monitor;;
validate-all) ;;
*) usage
exit $OCF_ERR_UNIMPLEMENTED;;
esac