svn commit: r1338303 - in /incubator/tashi/trunk/src/tashi: agents/primitive.py clustermanager/clustermanagerservice.py nodemanager/nodemanagerservice.py nodemanager/vmcontrol/qemu.py

[st...@apache.org]

Author: stroucki
Date: Mon May 14 17:31:45 2012
New Revision: 1338303

URL: http://svn.apache.org/viewvc?rev=1338303&view=rev
Log:
clustermanagerservice, qemu, nodemanagerservice, primitive: mark out plan to avoid overscheduling on a host due to old state information

Modified:
    incubator/tashi/trunk/src/tashi/agents/primitive.py
    incubator/tashi/trunk/src/tashi/clustermanager/clustermanagerservice.py
    incubator/tashi/trunk/src/tashi/nodemanager/nodemanagerservice.py
    incubator/tashi/trunk/src/tashi/nodemanager/vmcontrol/qemu.py

Modified: incubator/tashi/trunk/src/tashi/agents/primitive.py
URL: http://svn.apache.org/viewvc/incubator/tashi/trunk/src/tashi/agents/primitive.py?rev=1338303&r1=1338302&r2=1338303&view=diff
==============================================================================
--- incubator/tashi/trunk/src/tashi/agents/primitive.py (original)
+++ incubator/tashi/trunk/src/tashi/agents/primitive.py Mon May 14 17:31:45 2012
@@ -76,8 +76,9 @@ class Primitive(object):
 		# XXXstroucki put held machines behind pending ones
 		heldInstances = []
 		for i in instances.itervalues():
+			# Nonrunning VMs will have hostId of None, but
+			# so will Suspended VMs.
 			if (i.hostId or i.state == InstanceState.Pending):
-				# Nonrunning VMs will have hostId of None
 				load[i.hostId] = load[i.hostId] + [i.id]
 			elif (i.hostId is None and i.state == InstanceState.Held):
 				heldInstances = heldInstances + [i.id]
@@ -245,8 +246,21 @@ class Primitive(object):
 	def start(self):
 		oldInstances = {}
 
+		# XXXstroucki: scheduling races have been observed, where
+		# a vm is scheduled on a host that had not updated its
+		# capacity with the clustermanager, leading to overloaded
+		# hosts. I think the place to insure against this happening
+		# is in the nodemanager. This scheduler will keep an
+		# internal state of cluster loading, but that is best
+		# effort and will be refreshed from CM once the buffer
+		# of vms to be scheduled is exhausted.
+
 		while True:
 			try:
+				# XXXstroucki: to get a list of vms to be
+				# scheduled, it asks the CM for a full
+				# cluster state, and will look at those
+				# without a host.
 				self.__getState()
 				
 				# Check for VMs that have exited and call

Modified: incubator/tashi/trunk/src/tashi/clustermanager/clustermanagerservice.py
URL: http://svn.apache.org/viewvc/incubator/tashi/trunk/src/tashi/clustermanager/clustermanagerservice.py?rev=1338303&r1=1338302&r2=1338303&view=diff
==============================================================================
--- incubator/tashi/trunk/src/tashi/clustermanager/clustermanagerservice.py (original)
+++ incubator/tashi/trunk/src/tashi/clustermanager/clustermanagerservice.py Mon May 14 17:31:45 2012
@@ -675,6 +675,9 @@ class ClusterManagerService(object):
 		return "success"
 	
 	def activateVm(self, instanceId, host):
+		# XXXstroucki: check my idea of the host's capacity before
+		# trying.
+
 		dataHost = self.data.acquireHost(host.id)
 
 		if (dataHost.name != host.name):

Modified: incubator/tashi/trunk/src/tashi/nodemanager/nodemanagerservice.py
URL: http://svn.apache.org/viewvc/incubator/tashi/trunk/src/tashi/nodemanager/nodemanagerservice.py?rev=1338303&r1=1338302&r2=1338303&view=diff
==============================================================================
--- incubator/tashi/trunk/src/tashi/nodemanager/nodemanagerservice.py (original)
+++ incubator/tashi/trunk/src/tashi/nodemanager/nodemanagerservice.py Mon May 14 17:31:45 2012
@@ -286,6 +286,8 @@ class NodeManagerService(object):
 
 	# remote
 	def instantiateVm(self, instance):
+		# XXXstroucki: check my capacity before instantiating
+
 		self.__ACCOUNT("NM VM INSTANTIATE", instance=instance)
 		try:
 			vmId = self.vmm.instantiateVm(instance)

Modified: incubator/tashi/trunk/src/tashi/nodemanager/vmcontrol/qemu.py
URL: http://svn.apache.org/viewvc/incubator/tashi/trunk/src/tashi/nodemanager/vmcontrol/qemu.py?rev=1338303&r1=1338302&r2=1338303&view=diff
==============================================================================
--- incubator/tashi/trunk/src/tashi/nodemanager/vmcontrol/qemu.py (original)
+++ incubator/tashi/trunk/src/tashi/nodemanager/vmcontrol/qemu.py Mon May 14 17:31:45 2012
@@ -653,6 +653,8 @@ class Qemu(VmControlInterface):
 
 	# extern	
 	def instantiateVm(self, instance):
+		# XXXstroucki: check capacity before instantiating
+
 		try:
 			(vmId, cmd) = self.__startVm(instance, None)
 			child = self.__getChildFromPid(vmId)