From 5ad49d239ef9e20ad47b2c1ad4b95dd3c2a3f63e Mon Sep 17 00:00:00 2001
From: Joshua Boniface <joshua@boniface.me>
Date: Mon, 11 Jun 2018 20:12:11 -0400
Subject: [PATCH] Add some tweaks to prevent clobbering state changes happening
 at startup and during each keepalive cycle

---
 NodeInstance.py |  8 ++++++--
 VMInstance.py   | 24 +++++++++++++++++-------
 2 files changed, 23 insertions(+), 9 deletions(-)

diff --git a/NodeInstance.py b/NodeInstance.py
index d2a5dc6b..5c7e621e 100644
--- a/NodeInstance.py
+++ b/NodeInstance.py
@@ -198,10 +198,12 @@ class NodeInstance():
         else:
             self.daemon_state = 'start'
 
-        # Toggle state management of all VMs and remove any non-running VMs
+        # Toggle state management of dead VMs
         for domain, instance in self.s_domain.items():
             if instance.inshutdown == False and domain in self.domain_list:
-                instance.manage_vm_state()
+                if instance.getstate() == 'start' and instance.gethypervisor() == self.name and instance.dom == None:
+                    instance.manage_vm_state()
+
                 if instance.dom == None:
                     try:
                         self.domain_list.remove(domain)
@@ -218,6 +220,8 @@ class NodeInstance():
                             self.domain_list.remove(domain)
                         except:
                             pass
+                    else:
+                        instance.setnoclobber()
 
         # toggle state management of this node
         if self.domain_state == 'flush':
diff --git a/VMInstance.py b/VMInstance.py
index 744067b5..9bc55d2a 100644
--- a/VMInstance.py
+++ b/VMInstance.py
@@ -40,17 +40,24 @@ class VMInstance:
         self.inmigrate = False
         self.inreceive = False
 
-        self.dom = self.lookupByUUID(self.domuuid)
+        # These stop a weird clobber at startup
+        self.noclobber = False
 
-        # Watch for changes to the hypervisor field in Zookeeper
-        @zk.DataWatch('/domains/{}/hypervisor'.format(self.domuuid))
-        def watch_hypervisor(data, stat, event=""):
-            self.manage_vm_state()
+        self.dom = self.lookupByUUID(self.domuuid)
 
         # Watch for changes to the state field in Zookeeper
         @zk.DataWatch('/domains/{}/state'.format(self.domuuid))
         def watch_state(data, stat, event=""):
-            self.manage_vm_state()
+            if self.noclobber == False:
+                self.noclobber = True
+                self.manage_vm_state()
+
+        # Watch for changes to the hypervisor field in Zookeeper
+        @zk.DataWatch('/domains/{}/hypervisor'.format(self.domuuid))
+        def watch_hypervisor(data, stat, event=""):
+            if self.noclobber == False:
+                self.noclobber = True
+                self.manage_vm_state()
 
     # Get data functions
     def getstate(self):
@@ -59,6 +66,10 @@ class VMInstance:
     def gethypervisor(self):
         return self.hypervisor
 
+    # Allow the node to set the noclobber status once it performs its first keepalive
+    def setnoclobber(self):
+        self.noclobber = False
+
     # Start up the VM
     def start_vm(self, xmlconfig):
         ansiiprint.echo('Starting VM', '{}:'.format(self.domuuid), 'i')
@@ -286,7 +297,6 @@ class VMInstance:
             domxml = str(domxml.decode('ascii'))
             self.start_vm(domxml)
 
-
     # This function is a wrapper for libvirt.lookupByUUID which fixes some problems
     # 1. Takes a text UUID and handles converting it to bytes
     # 2. Try's it and returns a sensible value if not