Fix more.

files/MonitorNeutron

@@ -98,8 +98,9 @@ END
 
 MonitorNeutron_start() {
     ocf_log info "MonitorNeutron_start"
-    su ${OCF_RESKEY_user} -s /bin/sh -c "python ${OCF_RESKEY_binary} $OCF_RESKEY_additional_parameters" \
+    su ${OCF_RESKEY_user} -s /bin/sh -c "python ${OCF_RESKEY_binary} \
-    ' >> /dev/null 2>&1 & echo $!' > $OCF_RESKEY_pid
+    $OCF_RESKEY_additional_parameters"' \
+     >> /dev/null 2>&1 & echo $!' > $OCF_RESKEY_pid
 
     if [ $? =  $OCF_SUCCESS ]; then
         return $OCF_SUCCESS

files/monitor.py

@@ -20,9 +20,7 @@ import sys
 import time
 
 from oslo.config import cfg
-from neutron.openstack.common import log as logging
+import logging as LOG
-
-LOG = logging.getLogger(__name__)
 
 
 class Daemon(object):
@@ -58,18 +56,16 @@ class Daemon(object):
         # fork second time
         self._fork()
 
-        # redirect standard file descriptors
         # redirect standard file descriptors
         sys.stdout.flush()
         sys.stderr.flush()
         stdin = open(self.stdin, 'r')
         stdout = open(self.stdout, 'a+')
         stderr = open(self.stderr, 'a+', 0)
-        #os.dup2(stdin.fileno(), sys.stdin.fileno())
+        os.dup2(stdin.fileno(), sys.stdin.fileno())
-        #os.dup2(stdout.fileno(), sys.stdout.fileno())
+        os.dup2(stdout.fileno(), sys.stdout.fileno())
-        #os.dup2(stderr.fileno(), sys.stderr.fileno())
+        os.dup2(stderr.fileno(), sys.stderr.fileno())
 
-        #atexit.register(self.delete_pid)
         signal.signal(signal.SIGTERM, self.handle_sigterm)
 
     def handle_sigterm(self, signum, frame):
@@ -93,19 +89,26 @@ class MonitorNeutronAgentsDaemon(Daemon):
         super(MonitorNeutronAgentsDaemon, self).__init__()
         self.check_interval = check_interval
         LOG.info('Monitor Neutron Agent Loop Init')
+        self.env = {}
 
     def get_env(self):
-        env = {}
+        envrc_f = '/etc/legacy_ha_envrc'
-        env_data = '/etc/legacy_ha_env_data'
+        envrc_f_m = False
-        if os.path.isfile(env_data):
+        if os.path.isfile(envrc_f):
-            with open(env_data, 'r') as f:
+            ctime = time.ctime(os.stat(envrc_f).st_ctime)
-                line = f.readline()
+            mtime = time.ctime(os.stat(envrc_f).st_mtime)
-                data = line.split('=').strip()
+            if ctime != mtime:
-                if data and data[0] and data[1]:
+                envrc_f_m = True
-                    env[data[0]] = env[data[1]]
+
-                else:
+            if not self.env or envrc_f_m:
-                    raise Exception("OpenStack env data uncomplete.")
+                with open(envrc_f, 'r') as f:
-        return env
+                    for line in f:
+                        data = line.strip().split('=')
+                        if data and data[0] and data[1]:
+                            self.env[data[0]] = data[1]
+                        else:
+                            raise Exception("OpenStack env data uncomplete.")
+        return self.env
 
     def reassign_agent_resources(self):
         ''' Use agent scheduler API to detect down agents and re-schedule '''
@@ -130,10 +133,6 @@ class MonitorNeutronAgentsDaemon(Daemon):
                                 region_name=env['region'])
 
         partner_gateways = []
-        #partner_gateways = [unit_private_ip().split('.')[0]]
-        #for partner_gateway in relations_of_type(reltype='cluster'):
-        #    gateway_hostname = get_hostname(partner_gateway['private-address'])
-        #    partner_gateways.append(gateway_hostname.partition('.')[0])
 
         agents = quantum.list_agents(agent_type=DHCP_AGENT)
         dhcp_agents = []
@@ -147,8 +146,7 @@ class MonitorNeutronAgentsDaemon(Daemon):
                             agent['id'])['networks']:
                     networks[network['id']] = agent['id']
             else:
-                if agent['host'].partition('.')[0] in partner_gateways:
+                dhcp_agents.append(agent['id'])
-                    dhcp_agents.append(agent['id'])
 
         agents = quantum.list_agents(agent_type=L3_AGENT)
         routers = {}
@@ -160,8 +158,7 @@ class MonitorNeutronAgentsDaemon(Daemon):
                             agent['id'])['routers']:
                     routers[router['id']] = agent['id']
             else:
-                if agent['host'].split('.')[0] in partner_gateways:
+                l3_agents.append(agent['id'])
-                    l3_agents.append(agent['id'])
 
         if len(dhcp_agents) == 0 or len(l3_agents) == 0:
             LOG.info('Unable to relocate resources, there are %s dhcp_agents '
@@ -194,11 +191,9 @@ class MonitorNeutronAgentsDaemon(Daemon):
     def run(self):
         while True:
             LOG.info('Monitor Neutron Agent Loop Start')
-            print "Monitor Neutron Agent Loop Start"
+            LOG.info("Start : %s" % time.ctime())
-            print "Start : %s" % time.ctime()
+            time.sleep(self.check_interval)
-            #time.sleep(self.check_interval)
+            LOG.info("End : %s" % time.ctime())
-            time.sleep( 15 )
-            print "End : %s" % time.ctime()
             self.reassign_agent_resources()
 
 
@@ -207,11 +202,15 @@ if __name__ == '__main__':
         cfg.StrOpt('check_interval',
                    default=15,
                    help='Check Neutron Agents interval.'),
+        cfg.StrOpt('log_file',
+                   default='/var/log/monitor.log',
+                   help='log file'),
     ]
 
     cfg.CONF.register_cli_opts(opts)
     cfg.CONF(project='monitor_neutron_agents', default_config_files=[])
 
+    LOG.basicConfig(filename=cfg.CONF.log_file, level=LOG.INFO)
     monitor_daemon = MonitorNeutronAgentsDaemon(
         check_interval=cfg.CONF.check_interval)
     monitor_daemon.start()

hooks/quantum_hooks.py

@@ -192,6 +192,9 @@ def nm_changed():
         ca_crt = b64decode(relation_get('ca_cert'))
         install_ca_cert(ca_crt)
 
+    if config('ha-legacy-mode'):
+        cache_env_data()
+
 
 @hooks.hook("cluster-relation-departed")
 @restart_on_change(restart_map())
@@ -243,21 +246,20 @@ def ha_relation_joined():
                               'op monitor on-fail="restart" interval="10s"'
                               .format(external_agent=external_agent),
             'res_MonitorHA': 'op monitor interval="5s"',
-            'nees_connectivity': 'location res_MonitorHA '
-                                 'rule pingd: defined pingd'
-                              #'rule -inf: not_defined pingd or pingd lte 0'
         }
-
         clones = {
             'cl_PingCheck': 'res_PingCheck',
             'cl_ClusterMon': 'res_ClusterMon'
         }
+        constraints = {'location': 'nees_connectivity res_MonitorHA '
+                                   'rule pingd: defined pingd'}
 
         relation_set(corosync_bindiface=cluster_config['ha-bindiface'],
                      corosync_mcastport=cluster_config['ha-mcastport'],
                      resources=resources,
                      resource_params=resource_params,
-                     clones=clones)
+                     clones=clones,
+                     constraints=constraints)
 
 
 if __name__ == '__main__':

hooks/quantum_utils.py

@@ -473,11 +473,7 @@ def reassign_agent_resources():
                             auth_url=auth_url,
                             region_name=env['region'])
 
-    partner_gateways = [unit_private_ip().split('.')[0]]
+    partner_gateways = get_quantum_gateway_cluster_nodes()
-    for partner_gateway in relations_of_type(reltype='cluster'):
-        gateway_hostname = get_hostname(partner_gateway['private-address'])
-        partner_gateways.append(gateway_hostname.partition('.')[0])
-
     agents = quantum.list_agents(agent_type=DHCP_AGENT)
     dhcp_agents = []
     l3_agents = []
@@ -608,6 +604,14 @@ def get_dns_host():
     return ' '.join(dns_hosts)
 
 
+def get_quantum_gateway_cluster_nodes():
+    partner_gateways = get_hostname(unit_private_ip())
+    for partner_gateway in relations_of_type(reltype='cluster'):
+        gateway_hostname = get_hostname(partner_gateway['private-address'])
+        partner_gateways.append(gateway_hostname.partition('.')[0])
+    return partner_gateways
+
+
 def copy_file(source_dir, des_dir, f, f_mod=None, update=False):
     if not os.path.isdir(des_dir):
         mkdir(des_dir)
@@ -663,7 +667,7 @@ def install_legacy_ha_files(update=False):
     if config('ha-legacy-mode'):
         init_ocf_MonitorNeutron_f(update=update)
         init_external_agent_f(update=update)
-        #init_reassign_agent_services_binary()
+        # init_reassign_agent_services_binary()
         init_monitor_daemon(update=update)
 
 
@@ -679,6 +683,24 @@ def cache_env_data():
         log('Unable to get NetworkServiceContext at this time', level=ERROR)
         return
 
-    with open('/etc/legacy_ha_env_data', 'w') as f:
+    no_envrc = False
-        for k, v in env.items():
+    envrc_f = '/etc/legacy_ha_envrc'
-            f.write(''.join([k, '=', v, '\n']))
+    if os.path.isfile(envrc_f):
+        with open(envrc_f, 'r') as f:
+            data = f.read()
+        data = data.strip().split('\n')
+
+        diff = False
+        for line in data:
+            k = line.split('=')[0]
+            v = line.split('=')[1]
+            if k not in env or v != env[k]:
+                diff = True
+                break
+    else:
+        no_envrc = True
+
+    if no_envrc or diff:
+        with open(envrc_f, 'w') as f:
+            for k, v in env.items():
+                f.write(''.join([k, '=', v, '\n']))