Reword some of the stuff and handle updates better

NodeInstance.py

@@ -155,7 +155,7 @@ class NodeInstance():
                 transaction.set_data('/domains/{}/state'.format(dom_uuid), 'migrate'.encode('ascii'))
                 transaction.set_data('/domains/{}/hypervisor'.format(dom_uuid), target_hypervisor.encode('ascii'))
                 transaction.set_data('/domains/{}/lasthypervisor'.format(dom_uuid), current_hypervisor.encode('ascii'))
-                result = transaction.commit()
+                transaction.commit()
 
             # Wait 2s between migrations
             time.sleep(2)
@@ -177,7 +177,7 @@ class NodeInstance():
             transaction.set_data('/domains/{}/state'.format(dom_uuid), 'migrate'.encode('ascii'))
             transaction.set_data('/domains/{}/hypervisor'.format(dom_uuid), self.name.encode('ascii'))
             transaction.set_data('/domains/{}/lasthypervisor'.format(dom_uuid), ''.encode('ascii'))
-            result = transaction.commit()
+            transaction.commit()
 
             # Wait 2s between migrations
             time.sleep(2)
@@ -196,7 +196,7 @@ class NodeInstance():
         past_state = self.zk.get('/nodes/{}/daemonstate'.format(self.name))[0].decode('ascii')
         if past_state != 'start':
             self.daemon_state = 'start'
-            self.zk.set('/nodes/{}/daemonstate'.format(self.name), 'start'.encode('ascii'))
+            self.zk.set('/nodes/{}/daemonstate'.format(self.name), 'run'.encode('ascii'))
         else:
             self.daemon_state = 'start'
 
@@ -221,13 +221,15 @@ class NodeInstance():
         self.domains_count = len(conn.listDomainsID())
         keepalive_time = int(time.time())
         try:
-            self.zk.set('/nodes/{}/cpucount'.format(self.name), str(self.cpucount).encode('ascii'))
+            transaction = self.zk.transaction()
-            self.zk.set('/nodes/{}/memused'.format(self.name), str(self.memused).encode('ascii'))
+            transaction.set_data('/nodes/{}/cpucount'.format(self.name), str(self.cpucount).encode('ascii'))
-            self.zk.set('/nodes/{}/memfree'.format(self.name), str(self.memfree).encode('ascii'))
+            transaction.set_data('/nodes/{}/memused'.format(self.name), str(self.memused).encode('ascii'))
-            self.zk.set('/nodes/{}/cpuload'.format(self.name), str(self.cpuload).encode('ascii'))
+            transaction.set_data('/nodes/{}/memfree'.format(self.name), str(self.memfree).encode('ascii'))
-            self.zk.set('/nodes/{}/runningdomains'.format(self.name), ' '.join(self.domain_list).encode('ascii'))
+            transaction.set_data('/nodes/{}/cpuload'.format(self.name), str(self.cpuload).encode('ascii'))
-            self.zk.set('/nodes/{}/domainscount'.format(self.name), str(self.domains_count).encode('ascii'))
+            transaction.set_data('/nodes/{}/runningdomains'.format(self.name), ' '.join(self.domain_list).encode('ascii'))
-            self.zk.set('/nodes/{}/keepalive'.format(self.name), str(keepalive_time).encode('ascii'))
+            transaction.set_data('/nodes/{}/domainscount'.format(self.name), str(self.domains_count).encode('ascii'))
+            transaction.set_data('/nodes/{}/keepalive'.format(self.name), str(keepalive_time).encode('ascii'))
+            transaction.commit()
         except:
             return
 

fencenode.py

@@ -26,8 +26,22 @@ import os, sys, libvirt, uuid, kazoo.client, time, subprocess, re, ansiiprint
 # Trigger function
 #
 def fence(node_name, zk):
-    time.sleep(3)
+    failcount = 0
-    ansiiprint.echo('Fencing node "{}" via IPMI reboot signal'.format(node_name), '', 'w')
+    while failcount < 3
+        # Wait 5 seconds
+        time.sleep(5)
+        # Get the state
+        node_daemon_state = self.zk.get('/nodes/{}/daemonstate'.format(node_name))[0].decode('ascii')
+        # Is it still 'dead'
+        if node_daemon_state == 'dead':
+            failcount += 1
+            ansiiprint.echo('Node "{}" failed {} saving throw'.format(node_name, failcount), '', 'w')
+        # It changed back to something else so it must be alive
+        else:
+            ansiiprint.echo('Node "{}" passed a saving throw; canceling fence'.format(node_name), '', 'o')
+            return
+
+    ansiiprint.echo('Fencing node "{}" via IPMI reboot signal'.format(node_name), '', 'e')
 
     ipmi_hostname = zk.get('/nodes/{}/ipmihostname'.format(node_name))[0].decode('ascii')
     ipmi_username = zk.get('/nodes/{}/ipmiusername'.format(node_name))[0].decode('ascii')

pvc.py

@@ -225,12 +225,16 @@ RAM (MiB): {node_mem_total: <6} {node_mem_used: <6} {node_mem_free: <6} {node_me
             
     # Format the string (elements)
     for node_name in node_list:
-        if node_daemon_state[node_name] == 'start':
+        if node_daemon_state[node_name] == 'run':
             daemon_state_colour = ansiiprint.green()
         elif node_daemon_state[node_name] == 'stop':
             daemon_state_colour = ansiiprint.red()
-        else:
+        elif node_daemon_state[node_name] == 'init':
             daemon_state_colour = ansiiprint.yellow()
+        elif node_daemon_state[node_name] == 'dead':
+            daemon_state_colour = ansiiprint.red() + ansiiprint.bold()
+        else
+            daemon_state_colour = ansiiprint.blue()
 
         if node_domain_state[node_name] == 'ready':
             domain_state_colour = ansiiprint.green()