Improve handling to use hostname

NodeInstance.py

@@ -36,7 +36,7 @@ class NodeInstance():
         self.inactive_node_list = []
         self.s_domain = s_domain
         self.domain_list = []
-        self.ipmiaddress = ''
+        self.ipmi_hostname = self.config['ipmi_hostname']
 
         # Zookeeper handlers for changed states
         @zk.DataWatch('/nodes/{}/state'.format(self.name))
@@ -195,12 +195,9 @@ class NodeInstance():
         # Close the Libvirt connection
         conn.close()
 
-        # Get IPMI address
-        self.ipmiaddress = fencenode.getIPMIAddress()
-
         # Display node information to the terminal
         ansiiprint.echo('{}{} keepalive{}'.format(ansiiprint.purple(), self.name, ansiiprint.end()), '', 't')
-        ansiiprint.echo('{0}CPUs:{1} {2}  {0}Free memory:{1} {3}  {0}Load:{1} {4}  {0}IPMI Address:{1} {5}'.format(ansiiprint.bold(), ansiiprint.end(), self.cpucount, self.memfree, self.cpuload, self.ipmiaddress), '', 'c')
+        ansiiprint.echo('{0}CPUs:{1} {2}  {0}Free memory:{1} {3}  {0}Load:{1} {4}  {0}IPMI Address:{1} {5}'.format(ansiiprint.bold(), ansiiprint.end(), self.cpucount, self.memfree, self.cpuload, self.ipmi_hostname), '', 'c')
         ansiiprint.echo('{}Active domains:{} {}'.format(ansiiprint.bold(), ansiiprint.end(), ' '.join(self.domain_list)), '', 'c')
 
         # Update our local node lists
@@ -213,8 +210,9 @@ class NodeInstance():
                 node_keepalive = 0
 
             # Handle deadtime and fencng if needed
-            # (A node is considered dead when its keepalive timer is >30s out-of-date while in 'start' state)
+            # (A node is considered dead when its keepalive timer is >6*keepalive_interval seconds
-            node_deadtime = int(time.time()) - 30
+            # out-of-date while in 'start' state)
+            node_deadtime = int(time.time()) - ( int(self.config['keepalive_interval']) * 6 )
             if node_keepalive < node_deadtime and node_state == 'start':
                 ansiiprint.echo('Node {} is dead - performing fence operation in 3 seconds'.format(node_name), '', 'w')
                 self.zk.set('/nodes/{}/state'.format(node_name), 'dead'.encode('ascii'))

fencenode.py

@@ -29,10 +29,10 @@ def fence(node_name, zk):
     time.sleep(3)
     ansiiprint.echo('Fencing node "{}" via IPMI reboot signal'.format(node_name), '', 'w')
 
-    ipmi_address = zk.get('/nodes/{}/ipmiaddress'.format(node_name))[0].decode('ascii')
+    ipmi_hostname = zk.get('/nodes/{}/ipmihostname'.format(node_name))[0].decode('ascii')
     ipmi_username = zk.get('/nodes/{}/ipmiusername'.format(node_name))[0].decode('ascii')
     ipmi_password = zk.get('/nodes/{}/ipmipassword'.format(node_name))[0].decode('ascii')
-    rebootViaIPMI(ipmi_address, ipmi_user, ipmi_password)
+    rebootViaIPMI(ipmi_hostname, ipmi_user, ipmi_password)
 
     ansiiprint.echo('Moving VMs from dead hypervisor "{}" to new hosts'.format(node_name), '', 'i')
     dead_node_running_domains = zk.get('/nodes/{}/runningdomains'.format(node_name))[0].decode('ascii').split()
@@ -57,17 +57,17 @@ def fence(node_name, zk):
         transaction.set_data('/domains/{}/lasthypervisor'.format(dom_uuid), current_hypervisor.encode('ascii'))
         transaction.commit()
 
-def getIPMIAddress():
+#def getIPMIAddress():
-    ipmi_command = ['bash', '/fakeipmi.sh']
+#    ipmi_command = ['bash', '/fakeipmi.sh']
+#
+#    # Get the IPMI address
+#    ipmi_lan_output = subprocess.run(ipmi_command, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
+#    ipmi_lan_parsed = ipmi_lan_output.stdout.decode('ascii').split('\n')
+#    ipmi_lan_address = [s for s in ipmi_lan_parsed if re.search('IP Address[ ]*:', s)][0].split(':')[-1].strip()
+#    return ipmi_lan_address
 
-    # Get the IPMI address
+def rebootViaIPMI(ipmi_hostname, ipmi_user, ipmi_password):
-    ipmi_lan_output = subprocess.run(ipmi_command, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
+    ipmi_command = ['ipmitool', '-H', ipmi_hostname, '-U', ipmi_user, '-P', ipmi_password, 'chassis', 'power', 'reset']
-    ipmi_lan_parsed = ipmi_lan_output.stdout.decode('ascii').split('\n')
-    ipmi_lan_address = [s for s in ipmi_lan_parsed if re.search('IP Address[ ]*:', s)][0].split(':')[-1].strip()
-    return ipmi_lan_address
-
-def rebootViaIPMI(ipmi_address, ipmi_user, ipmi_password):
-    ipmi_command = ['ipmitool', '-H', ipmi_address, '-U', ipmi_user, '-P', ipmi_password, 'chassis', 'power', 'reset']
     ipmi_command_output = subprocess.run(ipmi_command, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
     if ipmi_command_output == 0:
         ansiiprint.echo('Successfully rebooted dead node', '', 'o')

pvcd.py

@@ -106,9 +106,9 @@ else:
     zk.create('/domains/{}/runningdomains'.format(myhostname), ''.encode('ascii'))
     # Keepalives and fencing information
     zk.create('/domains/{}/keepalive'.format(myhostname), str(keepalive_time).encode('ascii'))
-    zk.create('/domains/{}/ipmiaddress'.format(myhostname), ''.encode('ascii'))
+    zk.create('/domains/{}/ipmihostname'.format(config['ipmi_hostname']), ''.encode('ascii'))
-    zk.create('/domains/{}/ipmiusername'.format(myhostname), ''.encode('ascii'))
+    zk.create('/domains/{}/ipmiusername'.format(config['ipmi_username']), ''.encode('ascii'))
-    zk.create('/domains/{}/ipmipassword'.format(myhostname), ''.encode('ascii'))
+    zk.create('/domains/{}/ipmipassword'.format(config['ipmi_password']), ''.encode('ascii'))
 
 t_node = dict()
 s_domain = dict()
@@ -144,7 +144,7 @@ update_zookeeper = this_node.update_zookeeper
 
 # Create timer to update this node in Zookeeper
 update_timer = apscheduler.schedulers.background.BackgroundScheduler()
-update_timer.add_job(update_zookeeper, 'interval', seconds=config['keepalive_interval'])
+update_timer.add_job(update_zookeeper, 'interval', seconds=int(config['keepalive_interval']))
 update_timer.start()
 
 # Tick loop