[V2]clone ping and ClusterMon, when start OK monitor daemon is started.

This commit is contained in:
Hui Xiang 2014-12-15 12:24:16 +08:00
parent 7a24f44ad3
commit 601422cbe0
5 changed files with 57 additions and 33 deletions

View File

@ -189,8 +189,10 @@ class MonitorNeutronAgentsDaemon(Daemon):
index += 1
def run(self):
LOG.info('Monitor Neutron Agent Loop Start')
self.reassign_agent_resources()
while True:
LOG.info('Monitor Neutron Agent Loop Start')
time.sleep(15)
self.reassign_agent_resources()
if __name__ == '__main__':

View File

@ -11,36 +11,13 @@ logger "CRM_notify_node: $CRM_notify_node"
logger " ** "
logger " ** "
if [[ ${CRM_notify_task} == 'monitor' && $CRM_notify_rsc == 'res_PingCheck' ]]; then
if [[ ${CRM_notify_desc} == 'unknown error' || ${CRM_notify_desc} == 'Timed Out' ]]; then
if [[ ${CRM_notify_task} == 'start' && $CRM_notify_rsc == 'res_PingCheck' ]]; then
if [[ ${CRM_notify_desc} == 'OK' ]]; then
hostname=`hostname`
logger "monitor error hostname: $CRM_notify_node"
logger "hostname: $hostname"
if [[ $hostname == $CRM_notify_node ]]; then
logger "Cleaning up namespace and ovs on node $CRM_notify_node !"
for ns in $(ip netns list |grep 'qrouter-'); do ip netns delete $ns; done;
for ns in $(ip netns list |grep 'qdhcp-'); do ip netns delete $ns; done;
neutron-ovs-cleanup
logger "Cleaning done."
else
sudo crm_node -p
if [ $? -ne 0 ]; then
logger "Failed to executing command 'crm_node -p'."
exit
fi
nodes=`sudo crm_node -p`
logger "Cluster partition nodes: $nodes"
if [ ! -z "$nodes" ]; then
for node in $nodes
do
if [ "$node" != "$CRM_notify_node" ]; then
logger "Executing monitor to reschedule Neutron agents..."
sudo python /usr/local/bin/monitor.py
break
fi
done
fi
fi
logger "Executing monitor to reschedule Neutron agents..."
sudo python /usr/local/bin/monitor.py
fi
fi

46
files/ns_ovs_cleanup.sh.ori Executable file
View File

@ -0,0 +1,46 @@
#! /bin/bash
logger "Start running ns_ovs_cleanup.sh..."
logger " ** "
logger " ** "
logger "CRM_notify_task: $CRM_notify_task"
logger "CRM_notify_desc: $CRM_notify_desc"
logger "CRM_notify_rsc: $CRM_notify_rsc"
logger "CRM_notify_node: $CRM_notify_node"
logger " ** "
logger " ** "
if [[ ${CRM_notify_task} == 'monitor' && $CRM_notify_rsc == 'res_PingCheck' ]]; then
if [[ ${CRM_notify_desc} == 'unknown error' || ${CRM_notify_desc} == 'Timed Out' ]]; then
hostname=`hostname`
logger "monitor error hostname: $CRM_notify_node"
logger "hostname: $hostname"
if [[ $hostname == $CRM_notify_node ]]; then
logger "Cleaning up namespace and ovs on node $CRM_notify_node !"
for ns in $(ip netns list |grep 'qrouter-'); do ip netns delete $ns; done;
for ns in $(ip netns list |grep 'qdhcp-'); do ip netns delete $ns; done;
neutron-ovs-cleanup
logger "Cleaning done."
else
sudo crm_node -p
if [ $? -ne 0 ]; then
logger "Failed to executing command 'crm_node -p'."
exit
fi
nodes=`sudo crm_node -p`
logger "Cluster partition nodes: $nodes"
if [ ! -z "$nodes" ]; then
for node in $nodes
do
if [ "$node" != "$CRM_notify_node" ]; then
logger "Executing monitor to reschedule Neutron agents..."
sudo python /usr/local/bin/monitor.py
break
fi
done
fi
fi
fi
fi

View File

@ -235,11 +235,11 @@ def ha_relation_joined():
}
resource_params = {
'res_PingCheck': 'params host_list="{host}" dampen="5s" '
'debug={debug} multiplier="100" '
'failure_score="100" '
'debug={debug} multiplier="1000" '
'failure_score="10" '
'op monitor on-fail="restart" interval="10s" '
'timeout="60s" '.format(host=dns_hosts,
debug=debug),
debug=debug),
'res_ClusterMon': 'params user="root" update="30" '
'extra_options="-E {external_agent}" '
'op monitor on-fail="restart" interval="10s"'

View File

@ -595,7 +595,6 @@ def get_dns_host():
except Exception:
log('Failed to get nameserver from resolv.conf !', level=ERROR)
dns_hosts.append('8.8.8.8')
if config('dns_hosts'):
dnss = config('dns_hosts').split(' ')
for dns in dnss: