def __init__(self, message=None, **kwargs):
self.kwargs = kwargs
if 'code' not in self.kwargs:
try:
self.kwargs['code'] = self.code
except AttributeError:
pass
if not message:
try:
message = self.msg_fmt % kwargs
except Exception:
# kwargs doesn't match a variable in msg_fmt
# log the issue and the kwargs
LOG.exception(_LE('Exception in string format operation'))
for name, value in kwargs.items():
LOG.error(_LE("%(name)s: %(value)s"),
{'name': name, 'value': value})
if CONF.fatal_exception_format_errors:
raise
else:
# at least get the core msg_fmt out if something happened
message = self.msg_fmt
super(WatcherException, self).__init__(message)
def __init__(self, message=None, **kwargs):
self.kwargs = kwargs
if 'code' not in self.kwargs:
try:
self.kwargs['code'] = self.code
except AttributeError:
pass
if not message:
try:
message = self.msg_fmt % kwargs
except Exception as e:
# kwargs doesn't match a variable in msg_fmt
# log the issue and the kwargs
LOG.exception(_LE('Exception in string format operation'))
for name, value in kwargs.items():
LOG.error(_LE("%(name)s: %(value)s"),
{'name': name, 'value': value})
if CONF.fatal_exception_format_errors:
raise e
else:
# at least get the core msg_fmt out if something happened
message = self.msg_fmt
super(WatcherException, self).__init__(message)
def execute(self, original_model):
LOG.debug("Initializing Outlet temperature strategy")
if original_model is None:
raise wexc.ClusterStateNotDefined()
current_model = original_model
hosts_need_release, hosts_target = self.group_hosts_by_outlet_temp(
current_model)
if len(hosts_need_release) == 0:
# TODO(zhenzanz): return something right if there's no hot servers
LOG.debug("No hosts require optimization")
return self.solution
if len(hosts_target) == 0:
LOG.warning(_LE("No hosts under outlet temp threshold found"))
return self.solution
# choose the server with highest outlet t
hosts_need_release = sorted(hosts_need_release,
reverse=True,
key=lambda x: (x["outlet_temp"]))
vm_to_migrate = self.choose_vm_to_migrate(current_model,
hosts_need_release)
# calculate the vm's cpu cores,memory,disk needs
if vm_to_migrate is None:
return self.solution
mig_src_hypervisor, vm_src = vm_to_migrate
dest_servers = self.filter_dest_servers(current_model, hosts_target,
vm_src)
# sort the filtered result by outlet temp
# pick up the lowest one as dest server
if len(dest_servers) == 0:
# TODO(zhenzanz): maybe to warn that there's no resource
# for instance.
LOG.info(_LE("No proper target host could be found"))
return self.solution
dest_servers = sorted(dest_servers, key=lambda x: (x["outlet_temp"]))
# always use the host with lowerest outlet temperature
mig_dst_hypervisor = dest_servers[0]['hv']
# generate solution to migrate the vm to the dest server,
if current_model.get_mapping().migrate_vm(vm_src, mig_src_hypervisor,
mig_dst_hypervisor):
parameters = {
'migration_type': 'live',
'src_hypervisor': mig_src_hypervisor.uuid,
'dst_hypervisor': mig_dst_hypervisor.uuid
}
self.solution.add_action(action_type=self.MIGRATION,
resource_id=vm_src.uuid,
input_parameters=parameters)
self.solution.model = current_model
return self.solution
def execute(self, original_model):
LOG.debug("Initializing Outlet temperature strategy")
if original_model is None:
raise wexc.ClusterStateNotDefined()
current_model = original_model
hosts_need_release, hosts_target = self.group_hosts_by_outlet_temp(
current_model)
if len(hosts_need_release) == 0:
# TODO(zhenzanz): return something right if there's no hot servers
LOG.debug("No hosts require optimization")
return self.solution
if len(hosts_target) == 0:
LOG.warning(_LE("No hosts under outlet temp threshold found"))
return self.solution
# choose the server with highest outlet t
hosts_need_release = sorted(hosts_need_release,
reverse=True,
key=lambda x: (x["outlet_temp"]))
vm_to_migrate = self.choose_vm_to_migrate(current_model,
hosts_need_release)
# calculate the vm's cpu cores,memory,disk needs
if vm_to_migrate is None:
return self.solution
mig_src_hypervisor, vm_src = vm_to_migrate
dest_servers = self.filter_dest_servers(current_model,
hosts_target,
vm_src)
# sort the filtered result by outlet temp
# pick up the lowest one as dest server
if len(dest_servers) == 0:
# TODO(zhenzanz): maybe to warn that there's no resource
# for instance.
LOG.info(_LE("No proper target host could be found"))
return self.solution
dest_servers = sorted(dest_servers, key=lambda x: (x["outlet_temp"]))
# always use the host with lowerest outlet temperature
mig_dst_hypervisor = dest_servers[0]['hv']
# generate solution to migrate the vm to the dest server,
if current_model.get_mapping().migrate_vm(vm_src,
mig_src_hypervisor,
mig_dst_hypervisor):
parameters = {'migration_type': 'live',
'src_hypervisor': mig_src_hypervisor.uuid,
'dst_hypervisor': mig_dst_hypervisor.uuid}
self.solution.add_action(action_type=self.MIGRATION,
resource_id=vm_src.uuid,
input_parameters=parameters)
self.solution.model = current_model
return self.solution
def choose_instance_to_migrate(self, hosts):
"""Pick up an active instance instance to migrate from provided hosts
:param hosts: the array of dict which contains node object
"""
instances_tobe_migrate = []
for nodemap in hosts:
source_node = nodemap['node']
source_instances = self.compute_model.mapping.get_node_instances(
source_node)
if source_instances:
inlet_t = self.ceilometer.statistic_aggregation(
resource_id=source_node.uuid,
meter_name=self.meter_name_inlet_t,
period=self._period,
aggregate='avg')
power = self.ceilometer.statistic_aggregation(
resource_id=source_node.uuid,
meter_name=self.meter_name_power,
period=self._period,
aggregate='avg')
if (power < self.threshold_power and
inlet_t < self.threshold_inlet_t):
# hardware issue, migrate all instances from this node
for instance_id in source_instances:
try:
instance = (self.compute_model.
get_instance_from_id(instance_id))
instances_tobe_migrate.append(instance)
except wexc.InstanceNotFound:
LOG.error(_LE("Instance not found; error: %s"),
instance_id)
return source_node, instances_tobe_migrate
else:
# migrate the first active instance
for instance_id in source_instances:
try:
instance = (self.compute_model.
get_instance_from_id(instance_id))
if (instance.state !=
element.InstanceState.ACTIVE.value):
LOG.info(
_LI("Instance not active, skipped: %s"),
instance.uuid)
continue
instances_tobe_migrate.append(instance)
return source_node, instances_tobe_migrate
except wexc.InstanceNotFound:
LOG.error(_LE("Instance not found; error: %s"),
instance_id)
else:
LOG.info(_LI("Instance not found on node: %s"),
source_node.uuid)
def choose_instance_to_migrate(self, hosts):
"""Pick up an active instance instance to migrate from provided hosts
:param hosts: the array of dict which contains node object
"""
instances_tobe_migrate = []
for nodemap in hosts:
source_node = nodemap['node']
source_instances = self.compute_model.mapping.get_node_instances(
source_node)
if source_instances:
inlet_t = self.ceilometer.statistic_aggregation(
resource_id=source_node.uuid,
meter_name=self.meter_name_inlet_t,
period=self._period,
aggregate='avg')
power = self.ceilometer.statistic_aggregation(
resource_id=source_node.uuid,
meter_name=self.meter_name_power,
period=self._period,
aggregate='avg')
if (power < self.threshold_power and
inlet_t < self.threshold_inlet_t):
# hardware issue, migrate all instances from this node
for instance_id in source_instances:
try:
instance = (self.compute_model.
get_instance_by_uuid(instance_id))
instances_tobe_migrate.append(instance)
except wexc.InstanceNotFound:
LOG.error(_LE("Instance not found; error: %s"),
instance_id)
return source_node, instances_tobe_migrate
else:
# migrate the first active instance
for instance_id in source_instances:
try:
instance = (self.compute_model.
get_instance_by_uuid(instance_id))
if (instance.state !=
element.InstanceState.ACTIVE.value):
LOG.info(
_LI("Instance not active, skipped: %s"),
instance.uuid)
continue
instances_tobe_migrate.append(instance)
return source_node, instances_tobe_migrate
except wexc.InstanceNotFound:
LOG.error(_LE("Instance not found; error: %s"),
instance_id)
else:
LOG.info(_LI("Instance not found on node: %s"),
source_node.uuid)
def calculate_score_instance(self, instance):
"""Calculate Score of virtual machine
:param instance: the virtual machine
:return: score
"""
instance_cpu_utilization = self.ceilometer. \
statistic_aggregation(
resource_id=instance.uuid,
meter_name=self.INSTANCE_CPU_USAGE_METRIC_NAME,
period="7200",
aggregate='avg'
)
if instance_cpu_utilization is None:
LOG.error(
_LE("No values returned by %(resource_id)s "
"for %(metric_name)s") %
dict(resource_id=instance.uuid,
metric_name=self.INSTANCE_CPU_USAGE_METRIC_NAME))
instance_cpu_utilization = 100
cpu_capacity = self.compute_model.get_resource_by_uuid(
element.ResourceType.cpu_cores).get_capacity(instance)
total_cores_used = cpu_capacity * (instance_cpu_utilization / 100.0)
return self.calculate_weight(instance, total_cores_used, 0, 0)
def calculate_score_node(self, node):
"""Calculate the score that represent the utilization level
:param node: :py:class:`~.ComputeNode` instance
:return: Score for the given compute node
:rtype: float
"""
resource_id = "%s_%s" % (node.uuid, node.hostname)
host_avg_cpu_util = self.ceilometer.statistic_aggregation(
resource_id=resource_id,
meter_name=self.HOST_CPU_USAGE_METRIC_NAME,
period="7200",
aggregate='avg')
if host_avg_cpu_util is None:
LOG.error(
_LE("No values returned by %(resource_id)s "
"for %(metric_name)s") %
dict(resource_id=resource_id,
metric_name=self.HOST_CPU_USAGE_METRIC_NAME))
host_avg_cpu_util = 100
cpu_capacity = self.compute_model.get_resource_by_uuid(
element.ResourceType.cpu_cores).get_capacity(node)
total_cores_used = cpu_capacity * (host_avg_cpu_util / 100.0)
return self.calculate_weight(node, total_cores_used, 0, 0)
def get_vm_load(self, vm_uuid):
"""
Gather vm load through ceilometer statistic.
:param vm_uuid: vm for which statistic is gathered.
:return: dict
"""
LOG.warning('get_vm_load started')
flavor_id = self.nova.servers.get(vm_uuid).flavor['id']
vm_vcpus = self.nova.flavors.get(flavor_id).vcpus
vm_load = {'uuid': vm_uuid, 'vcpus': vm_vcpus}
for meter in self.metrics:
avg_meter = self.ceilometer.statistic_aggregation(
resource_id=vm_uuid,
meter_name=meter,
period="120",
aggregate='avg'
)
if avg_meter is None:
LOG.error(
_LE("No values returned by %(resource_id)s "
"for %(metric_name)s"),
resource_id=vm_uuid,
metric_name=avg_meter,
)
vm_load[meter] = avg_meter if avg_meter else 0
return vm_load
def calculate_score_node(self, hypervisor, model):
"""calculate the score that represent the utilization level
:param hypervisor:
:param model:
:return:
"""
resource_id = "%s_%s" % (hypervisor.uuid, hypervisor.hostname)
vm_avg_cpu_util = self.ceilometer. \
statistic_aggregation(resource_id=resource_id,
meter_name=self.HOST_CPU_USAGE_METRIC_NAME,
period="7200",
aggregate='avg'
)
if vm_avg_cpu_util is None:
LOG.error(
_LE("No values returned by %(resource_id)s "
"for %(metric_name)s"),
resource_id=resource_id,
metric_name=self.HOST_CPU_USAGE_METRIC_NAME,
)
vm_avg_cpu_util = 100
cpu_capacity = model.get_resource_from_id(
resource.ResourceType.cpu_cores).get_capacity(hypervisor)
total_cores_used = cpu_capacity * (vm_avg_cpu_util / 100)
return self.calculate_weight(model, hypervisor, total_cores_used, 0, 0)
def calculate_score_vm(self, vm, cluster_data_model):
"""Calculate Score of virtual machine
:param vm: the virtual machine
:param cluster_data_model: the cluster model
:return: score
"""
if cluster_data_model is None:
raise exception.ClusterStateNotDefined()
vm_cpu_utilization = self.ceilometer. \
statistic_aggregation(
resource_id=vm.uuid,
meter_name=self.INSTANCE_CPU_USAGE_METRIC_NAME,
period="7200",
aggregate='avg'
)
if vm_cpu_utilization is None:
LOG.error(
_LE("No values returned by %(resource_id)s "
"for %(metric_name)s"),
resource_id=vm.uuid,
metric_name=self.INSTANCE_CPU_USAGE_METRIC_NAME,
)
vm_cpu_utilization = 100
cpu_capacity = cluster_data_model.get_resource_from_id(
resource.ResourceType.cpu_cores).get_capacity(vm)
total_cores_used = cpu_capacity * (vm_cpu_utilization / 100.0)
return self.calculate_weight(cluster_data_model, vm, total_cores_used,
0, 0)
def group_hosts_by_outlet_temp(self, cluster_data_model):
"""Group hosts based on outlet temp meters"""
hypervisors = cluster_data_model.get_all_hypervisors()
size_cluster = len(hypervisors)
if size_cluster == 0:
raise wexc.ClusterEmpty()
hosts_need_release = []
hosts_target = []
for hypervisor_id in hypervisors:
hypervisor = cluster_data_model.get_hypervisor_from_id(
hypervisor_id)
resource_id = hypervisor.uuid
outlet_temp = self.ceilometer.statistic_aggregation(
resource_id=resource_id,
meter_name=self._meter,
period="30",
aggregate='avg')
# some hosts may not have outlet temp meters, remove from target
if outlet_temp is None:
LOG.warning(_LE("%s: no outlet temp data"), resource_id)
continue
LOG.debug("%s: outlet temperature %f" % (resource_id, outlet_temp))
hvmap = {'hv': hypervisor, 'outlet_temp': outlet_temp}
if outlet_temp >= self.threshold:
# mark the hypervisor to release resources
hosts_need_release.append(hvmap)
else:
hosts_target.append(hvmap)
return hosts_need_release, hosts_target
def calculate_score_node(self, node):
"""Calculate the score that represent the utilization level
:param node: :py:class:`~.ComputeNode` instance
:return: Score for the given compute node
:rtype: float
"""
resource_id = "%s_%s" % (node.uuid, node.hostname)
host_avg_cpu_util = self.ceilometer. \
statistic_aggregation(resource_id=resource_id,
meter_name=self.HOST_CPU_USAGE_METRIC_NAME,
period="7200",
aggregate='avg')
if host_avg_cpu_util is None:
LOG.error(
_LE("No values returned by %(resource_id)s "
"for %(metric_name)s"),
resource_id=resource_id,
metric_name=self.HOST_CPU_USAGE_METRIC_NAME,
)
host_avg_cpu_util = 100
cpu_capacity = self.compute_model.get_resource_from_id(
element.ResourceType.cpu_cores).get_capacity(node)
total_cores_used = cpu_capacity * (host_avg_cpu_util / 100)
return self.calculate_weight(node, total_cores_used, 0, 0)
def calculate_score_node(self, hypervisor):
"""Calculate the score that represent the utilization level
:param hypervisor:
:return:
"""
resource_id = "%s_%s" % (hypervisor.uuid, hypervisor.hostname)
host_avg_cpu_util = self.ceilometer. \
statistic_aggregation(resource_id=resource_id,
meter_name=self.HOST_CPU_USAGE_METRIC_NAME,
period="7200",
aggregate='avg')
if host_avg_cpu_util is None:
LOG.error(
_LE("No values returned by %(resource_id)s "
"for %(metric_name)s"),
resource_id=resource_id,
metric_name=self.HOST_CPU_USAGE_METRIC_NAME,
)
host_avg_cpu_util = 100
cpu_capacity = self.model.get_resource_from_id(
resource.ResourceType.cpu_cores).get_capacity(hypervisor)
total_cores_used = cpu_capacity * (host_avg_cpu_util / 100)
return self.calculate_weight(hypervisor, total_cores_used, 0, 0)
def choose_vm_to_migrate(self, hosts, avg_workload, workload_cache):
"""Pick up an active vm instance to migrate from provided hosts
:param hosts: the array of dict which contains hypervisor object
:param avg_workload: the average workload value of all hypervisors
:param workload_cache: the map contains vm to workload mapping
"""
for hvmap in hosts:
source_hypervisor = hvmap['hv']
source_vms = self.model.get_mapping().get_node_vms(
source_hypervisor)
if source_vms:
delta_workload = hvmap['workload'] - avg_workload
min_delta = 1000000
instance_id = None
for vm_id in source_vms:
try:
# select the first active VM to migrate
vm = self.model.get_vm_from_id(vm_id)
if vm.state != vm_state.VMState.ACTIVE.value:
LOG.debug("VM not active; skipped: %s",
vm.uuid)
continue
current_delta = delta_workload - workload_cache[vm_id]
if 0 <= current_delta < min_delta:
min_delta = current_delta
instance_id = vm_id
except wexc.InstanceNotFound:
LOG.error(_LE("VM not found; error: %s"), vm_id)
if instance_id:
return source_hypervisor, self.model.get_vm_from_id(
instance_id)
else:
LOG.info(_LI("VM not found on hypervisor: %s"),
source_hypervisor.uuid)
def obj_class_from_name(cls, objname, objver):
"""Returns a class from the registry based on a name and version."""
if objname not in cls._obj_classes:
LOG.error(
_LE('Unable to instantiate unregistered object type '
'%(objtype)s'), dict(objtype=objname))
raise exception.UnsupportedObjectError(objtype=objname)
latest = None
compatible_match = None
for objclass in cls._obj_classes[objname]:
if objclass.VERSION == objver:
return objclass
version_bits = tuple([int(x) for x in objclass.VERSION.split(".")])
if latest is None:
latest = version_bits
elif latest < version_bits:
latest = version_bits
if versionutils.is_compatible(objver, objclass.VERSION):
compatible_match = objclass
if compatible_match:
return compatible_match
latest_ver = '%i.%i' % latest
raise exception.IncompatibleObjectVersion(objname=objname,
objver=objver,
supported=latest_ver)
def group_hosts_by_cpu_util(self, model):
"""Calculate the workloads of each hypervisor
try to find out the hypervisors which have reached threshold
and the hypervisors which are under threshold.
and also calculate the average workload value of all hypervisors.
and also generate the VM workload map.
"""
hypervisors = model.get_all_hypervisors()
cluster_size = len(hypervisors)
if not hypervisors:
raise wexc.ClusterEmpty()
# get cpu cores capacity of hypervisors and vms
cap_cores = model.get_resource_from_id(resource.ResourceType.cpu_cores)
overload_hosts = []
nonoverload_hosts = []
# total workload of cluster
# it's the total core numbers being utilized in a cluster.
cluster_workload = 0.0
# use workload_cache to store the workload of VMs for reuse purpose
workload_cache = {}
for hypervisor_id in hypervisors:
hypervisor = model.get_hypervisor_from_id(hypervisor_id)
vms = model.get_mapping().get_node_vms(hypervisor)
hypervisor_workload = 0.0
for vm_id in vms:
vm = model.get_vm_from_id(vm_id)
try:
cpu_util = self.ceilometer.statistic_aggregation(
resource_id=vm_id,
meter_name=self._meter,
period=self._period,
aggregate='avg')
except Exception as e:
LOG.error(_LE("Can not get cpu_util: %s"), e.message)
continue
if cpu_util is None:
LOG.debug("%s: cpu_util is None", vm_id)
continue
vm_cores = cap_cores.get_capacity(vm)
workload_cache[vm_id] = cpu_util * vm_cores / 100
hypervisor_workload += workload_cache[vm_id]
LOG.debug("%s: cpu_util %f", vm_id, cpu_util)
hypervisor_cores = cap_cores.get_capacity(hypervisor)
hy_cpu_util = hypervisor_workload / hypervisor_cores * 100
cluster_workload += hypervisor_workload
hvmap = {'hv': hypervisor, "cpu_util": hy_cpu_util, 'workload':
hypervisor_workload}
if hy_cpu_util >= self.threshold:
# mark the hypervisor to release resources
overload_hosts.append(hvmap)
else:
nonoverload_hosts.append(hvmap)
avg_workload = cluster_workload / cluster_size
return overload_hosts, nonoverload_hosts, avg_workload, workload_cache
def calculate_score_vm(self, vm):
"""Calculate Score of virtual machine
:param vm: the virtual machine
:param self.model: the cluster model
:return: score
"""
vm_cpu_utilization = self.ceilometer. \
statistic_aggregation(
resource_id=vm.uuid,
meter_name=self.INSTANCE_CPU_USAGE_METRIC_NAME,
period="7200",
aggregate='avg'
)
if vm_cpu_utilization is None:
LOG.error(
_LE("No values returned by %(resource_id)s "
"for %(metric_name)s"),
resource_id=vm.uuid,
metric_name=self.INSTANCE_CPU_USAGE_METRIC_NAME,
)
vm_cpu_utilization = 100
cpu_capacity = self.model.get_resource_from_id(
resource.ResourceType.cpu_cores).get_capacity(vm)
total_cores_used = cpu_capacity * (vm_cpu_utilization / 100.0)
return self.calculate_weight(vm, total_cores_used, 0, 0)
def group_hosts_by_airflow(self):
"""Group hosts based on airflow meters"""
hypervisors = self.model.get_all_hypervisors()
if not hypervisors:
raise wexc.ClusterEmpty()
overload_hosts = []
nonoverload_hosts = []
for hypervisor_id in hypervisors:
hypervisor = self.model.get_hypervisor_from_id(hypervisor_id)
resource_id = hypervisor.uuid
airflow = self.ceilometer.statistic_aggregation(
resource_id=resource_id,
meter_name=self.meter_name_airflow,
period=self._period,
aggregate='avg')
# some hosts may not have airflow meter, remove from target
if airflow is None:
LOG.warning(_LE("%s: no airflow data"), resource_id)
continue
LOG.debug("%s: airflow %f" % (resource_id, airflow))
hvmap = {'hv': hypervisor, 'airflow': airflow}
if airflow >= self.threshold_airflow:
# mark the hypervisor to release resources
overload_hosts.append(hvmap)
else:
nonoverload_hosts.append(hvmap)
return overload_hosts, nonoverload_hosts
def obj_class_from_name(cls, objname, objver):
"""Returns a class from the registry based on a name and version."""
if objname not in cls._obj_classes:
LOG.error(_LE('Unable to instantiate unregistered object type '
'%(objtype)s'), dict(objtype=objname))
raise exception.UnsupportedObjectError(objtype=objname)
latest = None
compatible_match = None
for objclass in cls._obj_classes[objname]:
if objclass.VERSION == objver:
return objclass
version_bits = tuple([int(x) for x in objclass.VERSION.split(".")])
if latest is None:
latest = version_bits
elif latest < version_bits:
latest = version_bits
if versionutils.is_compatible(objver, objclass.VERSION):
compatible_match = objclass
if compatible_match:
return compatible_match
latest_ver = '%i.%i' % latest
raise exception.IncompatibleObjectVersion(objname=objname,
objver=objver,
supported=latest_ver)
def cleanup():
global TRANSPORT, NOTIFICATION_TRANSPORT, NOTIFIER
if NOTIFIER is None:
LOG.exception(_LE("RPC cleanup: NOTIFIER is None"))
TRANSPORT.cleanup()
NOTIFICATION_TRANSPORT.cleanup()
TRANSPORT = NOTIFICATION_TRANSPORT = NOTIFIER = None
def get_vm_load(self, vm_uuid):
"""
Gather vm load through ceilometer statistic.
:param vm_uuid: vm for which statistic is gathered.
:return: dict
"""
LOG.warning('get_vm_load started')
flavor_id = self.nova.servers.get(vm_uuid).flavor['id']
vm_vcpus = self.nova.flavors.get(flavor_id).vcpus
vm_load = {'uuid': vm_uuid, 'vcpus': vm_vcpus}
for meter in self.metrics:
avg_meter = self.ceilometer.statistic_aggregation(
resource_id=vm_uuid,
meter_name=meter,
period="120",
aggregate='avg')
if avg_meter is None:
LOG.error(
_LE("No values returned by %(resource_id)s "
"for %(metric_name)s"),
resource_id=vm_uuid,
metric_name=avg_meter,
)
vm_load[meter] = avg_meter if avg_meter else 0
return vm_load
def calculate_score_instance(self, instance):
"""Calculate Score of virtual machine
:param instance: the virtual machine
:return: score
"""
instance_cpu_utilization = self.ceilometer. \
statistic_aggregation(
resource_id=instance.uuid,
meter_name=self.INSTANCE_CPU_USAGE_METRIC_NAME,
period="7200",
aggregate='avg'
)
if instance_cpu_utilization is None:
LOG.error(
_LE("No values returned by %(resource_id)s "
"for %(metric_name)s"),
resource_id=instance.uuid,
metric_name=self.INSTANCE_CPU_USAGE_METRIC_NAME,
)
instance_cpu_utilization = 100
cpu_capacity = self.compute_model.get_resource_from_id(
element.ResourceType.cpu_cores).get_capacity(instance)
total_cores_used = cpu_capacity * (instance_cpu_utilization / 100.0)
return self.calculate_weight(instance, total_cores_used, 0, 0)
def group_hosts_by_outlet_temp(self, cluster_data_model):
"""Group hosts based on outlet temp meters"""
hypervisors = cluster_data_model.get_all_hypervisors()
size_cluster = len(hypervisors)
if size_cluster == 0:
raise wexc.ClusterEmpty()
hosts_need_release = []
hosts_target = []
for hypervisor_id in hypervisors:
hypervisor = cluster_data_model.get_hypervisor_from_id(
hypervisor_id)
resource_id = hypervisor.uuid
outlet_temp = self.ceilometer.statistic_aggregation(
resource_id=resource_id,
meter_name=self._meter,
period="30",
aggregate='avg')
# some hosts may not have outlet temp meters, remove from target
if outlet_temp is None:
LOG.warning(_LE("%s: no outlet temp data"), resource_id)
continue
LOG.debug("%s: outlet temperature %f" % (resource_id, outlet_temp))
hvmap = {'hv': hypervisor, 'outlet_temp': outlet_temp}
if outlet_temp >= self.threshold:
# mark the hypervisor to release resources
hosts_need_release.append(hvmap)
else:
hosts_target.append(hvmap)
return hosts_need_release, hosts_target
def choose_vm_to_migrate(self, hosts):
"""pick up an active vm instance to migrate from provided hosts
:param hosts: the array of dict which contains hypervisor object
"""
vms_tobe_migrate = []
for hvmap in hosts:
source_hypervisor = hvmap['hv']
source_vms = self.model.get_mapping().get_node_vms(
source_hypervisor)
if source_vms:
inlet_t = self.ceilometer.statistic_aggregation(
resource_id=source_hypervisor.uuid,
meter_name=self.meter_name_inlet_t,
period=self._period,
aggregate='avg')
power = self.ceilometer.statistic_aggregation(
resource_id=source_hypervisor.uuid,
meter_name=self.meter_name_power,
period=self._period,
aggregate='avg')
if (power < self.threshold_power and
inlet_t < self.threshold_inlet_t):
# hardware issue, migrate all vms from this hypervisor
for vm_id in source_vms:
try:
vm = self.model.get_vm_from_id(vm_id)
vms_tobe_migrate.append(vm)
except wexc.InstanceNotFound:
LOG.error(_LE("VM not found Error: %s"), vm_id)
return source_hypervisor, vms_tobe_migrate
else:
# migrate the first active vm
for vm_id in source_vms:
try:
vm = self.model.get_vm_from_id(vm_id)
if vm.state != vm_state.VMState.ACTIVE.value:
LOG.info(_LE("VM not active, skipped: %s"),
vm.uuid)
continue
vms_tobe_migrate.append(vm)
return source_hypervisor, vms_tobe_migrate
except wexc.InstanceNotFound:
LOG.error(_LE("VM not found Error: %s"), vm_id)
else:
LOG.info(_LI("VM not found from hypervisor: %s"),
source_hypervisor.uuid)
def get_vm_utilization(self, vm_uuid, model, period=3600, aggr='avg'):
"""Collect cpu, ram and disk utilization statistics of a VM.
:param vm_uuid: vm object
:param model: model_root object
:param period: seconds
:param aggr: string
:return: dict(cpu(number of vcpus used), ram(MB used), disk(B used))
"""
if vm_uuid in self.ceilometer_vm_data_cache.keys():
return self.ceilometer_vm_data_cache.get(vm_uuid)
cpu_util_metric = 'cpu_util'
ram_util_metric = 'memory.usage'
ram_alloc_metric = 'memory'
disk_alloc_metric = 'disk.root.size'
vm_cpu_util = self.ceilometer.statistic_aggregation(
resource_id=vm_uuid,
meter_name=cpu_util_metric,
period=period,
aggregate=aggr)
vm_cpu_cores = model.get_resource_from_id(
resource.ResourceType.cpu_cores).get_capacity(
model.get_vm_from_id(vm_uuid))
if vm_cpu_util:
total_cpu_utilization = vm_cpu_cores * (vm_cpu_util / 100.0)
else:
total_cpu_utilization = vm_cpu_cores
vm_ram_util = self.ceilometer.statistic_aggregation(
resource_id=vm_uuid,
meter_name=ram_util_metric,
period=period,
aggregate=aggr)
if not vm_ram_util:
vm_ram_util = self.ceilometer.statistic_aggregation(
resource_id=vm_uuid,
meter_name=ram_alloc_metric,
period=period,
aggregate=aggr)
vm_disk_util = self.ceilometer.statistic_aggregation(
resource_id=vm_uuid,
meter_name=disk_alloc_metric,
period=period,
aggregate=aggr)
if not vm_ram_util or not vm_disk_util:
LOG.error(_LE('No values returned by %(resource_id)s '
'for memory.usage or disk.root.size'),
resource_id=vm_uuid)
raise exception.NoDataFound
self.ceilometer_vm_data_cache[vm_uuid] = dict(
cpu=total_cpu_utilization, ram=vm_ram_util, disk=vm_disk_util)
return self.ceilometer_vm_data_cache.get(vm_uuid)
def setter(self, value, name=name, typefn=typefn):
self._changed_fields.add(name)
try:
return setattr(self, get_attrname(name), typefn(value))
except Exception:
attr = "%s.%s" % (self.obj_name(), name)
LOG.exception(_LE('Error setting %(attr)s'), {'attr': attr})
raise
def stop(self):
try:
self.rpcserver.stop()
self.rpcserver.wait()
except Exception as e:
LOG.exception(_LE('Service error occurred when stopping the '
'RPC server. Error: %s'), e)
try:
self.manager.del_host(deregister=self.deregister)
except Exception as e:
LOG.exception(_LE('Service error occurred when cleaning up '
'the RPC manager. Error: %s'), e)
super(RPCService, self).stop(graceful=True)
LOG.info(_LI('Stopped RPC server for service %(service)s on host '
'%(host)s.'),
{'service': self.topic, 'host': self.host})
def group_hosts_by_cpu_util(self):
"""Calculate the workloads of each node
try to find out the nodes which have reached threshold
and the nodes which are under threshold.
and also calculate the average workload value of all nodes.
and also generate the instance workload map.
"""
nodes = self.compute_model.get_all_compute_nodes()
cluster_size = len(nodes)
if not nodes:
raise wexc.ClusterEmpty()
overload_hosts = []
nonoverload_hosts = []
# total workload of cluster
cluster_workload = 0.0
# use workload_cache to store the workload of VMs for reuse purpose
workload_cache = {}
for node_id in nodes:
node = self.compute_model.get_node_by_uuid(node_id)
instances = self.compute_model.get_node_instances(node)
node_workload = 0.0
for instance in instances:
try:
cpu_util = self.ceilometer.statistic_aggregation(
resource_id=instance.uuid,
meter_name=self._meter,
period=self._period,
aggregate='avg')
except Exception as exc:
LOG.exception(exc)
LOG.error(_LE("Can not get cpu_util from Ceilometer"))
continue
if cpu_util is None:
LOG.debug("Instance (%s): cpu_util is None", instance.uuid)
continue
workload_cache[instance.uuid] = cpu_util * instance.vcpus / 100
node_workload += workload_cache[instance.uuid]
LOG.debug("VM (%s): cpu_util %f", instance.uuid, cpu_util)
node_cpu_util = node_workload / node.vcpus * 100
cluster_workload += node_workload
instance_data = {
'node': node,
"cpu_util": node_cpu_util,
'workload': node_workload
}
if node_cpu_util >= self.threshold:
# mark the node to release resources
overload_hosts.append(instance_data)
else:
nonoverload_hosts.append(instance_data)
avg_workload = cluster_workload / cluster_size
return overload_hosts, nonoverload_hosts, avg_workload, workload_cache
def get_instance_utilization(self, instance, period=3600, aggr='avg'):
"""Collect cpu, ram and disk utilization statistics of a VM.
:param instance: instance object
:param period: seconds
:param aggr: string
:return: dict(cpu(number of vcpus used), ram(MB used), disk(B used))
"""
if instance.uuid in self.ceilometer_instance_data_cache.keys():
return self.ceilometer_instance_data_cache.get(instance.uuid)
cpu_util_metric = 'cpu_util'
ram_util_metric = 'memory.usage'
ram_alloc_metric = 'memory'
disk_alloc_metric = 'disk.root.size'
instance_cpu_util = self.ceilometer.statistic_aggregation(
resource_id=instance.uuid,
meter_name=cpu_util_metric,
period=period,
aggregate=aggr)
if instance_cpu_util:
total_cpu_utilization = (instance.vcpus *
(instance_cpu_util / 100.0))
else:
total_cpu_utilization = instance.vcpus
instance_ram_util = self.ceilometer.statistic_aggregation(
resource_id=instance.uuid,
meter_name=ram_util_metric,
period=period,
aggregate=aggr)
if not instance_ram_util:
instance_ram_util = self.ceilometer.statistic_aggregation(
resource_id=instance.uuid,
meter_name=ram_alloc_metric,
period=period,
aggregate=aggr)
instance_disk_util = self.ceilometer.statistic_aggregation(
resource_id=instance.uuid,
meter_name=disk_alloc_metric,
period=period,
aggregate=aggr)
if not instance_ram_util or not instance_disk_util:
LOG.error(
_LE('No values returned by %s for memory.usage '
'or disk.root.size'), instance.uuid)
raise exception.NoDataFound
self.ceilometer_instance_data_cache[instance.uuid] = dict(
cpu=total_cpu_utilization,
ram=instance_ram_util,
disk=instance_disk_util)
return self.ceilometer_instance_data_cache.get(instance.uuid)
def setter(self, value, name=name, typefn=typefn):
self._changed_fields.add(name)
try:
return setattr(self, get_attrname(name), typefn(value))
except Exception:
attr = "%s.%s" % (self.obj_name(), name)
LOG.exception(_LE('Error setting %(attr)s'),
{'attr': attr})
raise
def __call__(self, environ, start_response):
# Request for this state, modified by replace_start_response()
# and used when an error is being reported.
state = {}
def replacement_start_response(status, headers, exc_info=None):
"""Overrides the default response to make errors parsable."""
try:
status_code = int(status.split(' ')[0])
state['status_code'] = status_code
except (ValueError, TypeError): # pragma: nocover
raise Exception(_(
'ErrorDocumentMiddleware received an invalid '
'status %s') % status)
else:
if (state['status_code'] // 100) not in (2, 3):
# Remove some headers so we can replace them later
# when we have the full error message and can
# compute the length.
headers = [(h, v)
for (h, v) in headers
if h not in ('Content-Length', 'Content-Type')]
# Save the headers in case we need to modify them.
state['headers'] = headers
return start_response(status, headers, exc_info)
app_iter = self.app(environ, replacement_start_response)
if (state['status_code'] // 100) not in (2, 3):
req = webob.Request(environ)
if (
req.accept.best_match(
['application/json',
'application/xml']) == 'application/xml'
):
try:
# simple check xml is valid
body = [
et.ElementTree.tostring(
et.ElementTree.Element(
'error_message', text='\n'.join(app_iter)))]
except et.ElementTree.ParseError as err:
LOG.error(_LE('Error parsing HTTP response: %s'), err)
body = ['<error_message>%s'
'</error_message>' % state['status_code']]
state['headers'].append(('Content-Type', 'application/xml'))
else:
if six.PY3:
app_iter = [i.decode('utf-8') for i in app_iter]
body = [jsonutils.dumps(
{'error_message': '\n'.join(app_iter)})]
if six.PY3:
body = [item.encode('utf-8') for item in body]
state['headers'].append(('Content-Type', 'application/json'))
state['headers'].append(('Content-Length', str(len(body[0]))))
else:
body = app_iter
return body
def get_instance_utilization(self, instance_uuid, model,
period=3600, aggr='avg'):
"""Collect cpu, ram and disk utilization statistics of a VM.
:param instance_uuid: instance object
:param model: model_root object
:param period: seconds
:param aggr: string
:return: dict(cpu(number of vcpus used), ram(MB used), disk(B used))
"""
if instance_uuid in self.ceilometer_instance_data_cache.keys():
return self.ceilometer_instance_data_cache.get(instance_uuid)
cpu_util_metric = 'cpu_util'
ram_util_metric = 'memory.usage'
ram_alloc_metric = 'memory'
disk_alloc_metric = 'disk.root.size'
instance_cpu_util = self.ceilometer.statistic_aggregation(
resource_id=instance_uuid, meter_name=cpu_util_metric,
period=period, aggregate=aggr)
instance_cpu_cores = model.get_resource_from_id(
element.ResourceType.cpu_cores).get_capacity(
model.get_instance_from_id(instance_uuid))
if instance_cpu_util:
total_cpu_utilization = (
instance_cpu_cores * (instance_cpu_util / 100.0))
else:
total_cpu_utilization = instance_cpu_cores
instance_ram_util = self.ceilometer.statistic_aggregation(
resource_id=instance_uuid, meter_name=ram_util_metric,
period=period, aggregate=aggr)
if not instance_ram_util:
instance_ram_util = self.ceilometer.statistic_aggregation(
resource_id=instance_uuid, meter_name=ram_alloc_metric,
period=period, aggregate=aggr)
instance_disk_util = self.ceilometer.statistic_aggregation(
resource_id=instance_uuid, meter_name=disk_alloc_metric,
period=period, aggregate=aggr)
if not instance_ram_util or not instance_disk_util:
LOG.error(
_LE('No values returned by %(resource_id)s '
'for memory.usage or disk.root.size'),
resource_id=instance_uuid
)
raise exception.NoDataFound
self.ceilometer_instance_data_cache[instance_uuid] = dict(
cpu=total_cpu_utilization, ram=instance_ram_util,
disk=instance_disk_util)
return self.ceilometer_instance_data_cache.get(instance_uuid)
def __call__(self, environ, start_response):
# Request for this state, modified by replace_start_response()
# and used when an error is being reported.
state = {}
def replacement_start_response(status, headers, exc_info=None):
"""Overrides the default response to make errors parsable."""
try:
status_code = int(status.split(' ')[0])
state['status_code'] = status_code
except (ValueError, TypeError): # pragma: nocover
raise Exception(_(
'ErrorDocumentMiddleware received an invalid '
'status %s') % status)
else:
if (state['status_code'] // 100) not in (2, 3):
# Remove some headers so we can replace them later
# when we have the full error message and can
# compute the length.
headers = [(h, v)
for (h, v) in headers
if h not in ('Content-Length', 'Content-Type')
]
# Save the headers in case we need to modify them.
state['headers'] = headers
return start_response(status, headers, exc_info)
app_iter = self.app(environ, replacement_start_response)
if (state['status_code'] // 100) not in (2, 3):
req = webob.Request(environ)
if (req.accept.best_match(['application/json', 'application/xml']
) == 'application/xml'):
try:
# simple check xml is valid
body = [et.ElementTree.tostring(
et.ElementTree.Element('error_message',
text='\n'.join(app_iter)))]
except et.ElementTree.ParseError as err:
LOG.error(_LE('Error parsing HTTP response: %s'), err)
body = [et.ElementTree.tostring(
et.ElementTree.Element('error_message',
text=state['status_code']))]
state['headers'].append(('Content-Type', 'application/xml'))
else:
if six.PY3:
app_iter = [i.decode('utf-8') for i in app_iter]
body = [jsonutils.dumps(
{'error_message': '\n'.join(app_iter)})]
if six.PY3:
body = [item.encode('utf-8') for item in body]
state['headers'].append(('Content-Type', 'application/json'))
state['headers'].append(('Content-Length', str(len(body[0]))))
else:
body = app_iter
return body
def _soft_delete_removed_goals(self):
removed_goals = [
g for g in self.available_goals
if g.name not in self.discovered_map['goals']]
for removed_goal in removed_goals:
removed_goal.soft_delete()
filters = {"goal_id": removed_goal.id}
invalid_ats = objects.AuditTemplate.list(self.ctx, filters=filters)
for at in invalid_ats:
LOG.warning(
_LE("Audit Template '%(audit_template)s' references a "
"goal that does not exist"),
audit_template=at.uuid)
def execute(self, *args, **kwargs):
try:
LOG.debug("Running action: %s", self.name)
self.action.execute()
self.engine.notify(self._db_action, objects.action.State.SUCCEEDED)
except Exception as e:
LOG.exception(e)
LOG.error(_LE('The workflow engine has failed '
'to execute the action: %s'), self.name)
self.engine.notify(self._db_action, objects.action.State.FAILED)
raise
def execute(self, *args, **kwargs):
try:
LOG.debug("Running action: %s", self.name)
self.action.execute()
self.engine.notify(self._db_action,
obj_action.State.SUCCEEDED)
except Exception as e:
LOG.exception(e)
LOG.error(_LE('The workflow engine has failed '
'to execute the action: %s'), self.name)
self.engine.notify(self._db_action,
obj_action.State.FAILED)
raise
def get_state_str(self, state):
"""Get resource state in string format.
:param state: resource state of unknown type
"""
if isinstance(state, six.string_types):
return state
elif isinstance(state, (element.InstanceState, element.ServiceState)):
return state.value
else:
LOG.error(_LE('Unexpexted resource state type, '
'state=%(state)s, state_type=%(st)s.'),
state=state,
st=type(state))
raise exception.WatcherException
def get_state_str(self, state):
"""Get resource state in string format.
:param state: resource state of unknown type
"""
if isinstance(state, six.string_types):
return state
elif isinstance(state, (element.InstanceState, element.ServiceState)):
return state.value
else:
LOG.error(
_LE('Unexpexted resource state type, '
'state=%(state)s, state_type=%(st)s.') %
dict(state=state, st=type(state)))
raise exception.WatcherException
def get_state_str(self, state):
"""Get resource state in string format.
:param state: resource state of unknown type
"""
if isinstance(state, six.string_types):
return state
elif (type(state) == hyper_state.HypervisorState
or type(state) == vm_state.VMState):
return state.value
else:
LOG.error(_LE('Unexpexted resource state type, '
'state=%(state)s, state_type=%(st)s.'),
state=state,
st=type(state))
raise exception.WatcherException
def _configure(self):
try:
self.__transport = om.get_transport(CONF)
self.__notifier = self.build_notifier()
if len(self.__endpoints):
target = om.Target(
topic=self.topic_name,
# For compatibility, we can override it with 'host' opt
server=CONF.host or socket.getfqdn(),
version=self.__version,
)
self.__server = self.build_server(target)
else:
LOG.warning(_LW("No endpoint defined; can only publish events"))
except Exception as e:
LOG.exception(e)
LOG.error(_LE("Messaging configuration error"))
def execute(self, *args, **kwargs):
try:
LOG.debug("Running action %s", self.name)
# todo(jed) remove return (true or false) raise an Exception
result = self.action.execute()
if result is not True:
self.engine.notify(self._db_action, obj_action.State.FAILED)
else:
self.engine.notify(self._db_action, obj_action.State.SUCCEEDED)
except Exception as e:
LOG.exception(e)
LOG.error(
_LE('The WorkFlow Engine has failed '
'to execute the action %s'), self.name)
self.engine.notify(self._db_action, obj_action.State.FAILED)
raise
def _configure(self):
try:
self.__transport = om.get_transport(CONF)
self.__notifier = self.build_notifier()
if len(self.__endpoints):
target = om.Target(
topic=self.topic_name,
# For compatibility, we can override it with 'host' opt
server=CONF.host or socket.getfqdn(),
version=self.__version,
)
self.__server = self.build_server(target)
else:
LOG.warning(
_LW("No endpoint defined; can only publish events"))
except Exception as e:
LOG.exception(e)
LOG.error(_LE("Messaging configuration error"))
def calculate_score_instance(self, instance):
"""Calculate Score of virtual machine
:param instance: the virtual machine
:return: score
"""
instance_cpu_utilization = self.get_instance_cpu_usage(instance)
if instance_cpu_utilization is None:
LOG.error(
_LE("No values returned by %(resource_id)s "
"for %(metric_name)s") %
dict(resource_id=instance.uuid,
metric_name=self.METRIC_NAMES[self.config.datasource]
['instance_cpu_usage']))
instance_cpu_utilization = 100
total_cores_used = instance.vcpus * (instance_cpu_utilization / 100.0)
return self.calculate_weight(instance, total_cores_used, 0, 0)
def add_migration(self, vm_uuid, src_hypervisor, dst_hypervisor, model):
"""Add an action for VM migration into the solution.
:param vm_uuid: vm uuid
:param src_hypervisor: hypervisor object
:param dst_hypervisor: hypervisor object
:param model: model_root object
:return: None
"""
vm = model.get_vm_from_id(vm_uuid)
vm_state_str = self.get_state_str(vm.state)
if vm_state_str != vm_state.VMState.ACTIVE.value:
'''
Watcher curently only supports live VM migration and block live
VM migration which both requires migrated VM to be active.
When supported, the cold migration may be used as a fallback
migration mechanism to move non active VMs.
'''
LOG.error(_LE('Cannot live migrate: vm_uuid=%(vm_uuid)s, '
'state=%(vm_state)s.'),
vm_uuid=vm_uuid,
vm_state=vm_state_str)
raise exception.WatcherException
migration_type = 'live'
dst_hyper_state_str = self.get_state_str(dst_hypervisor.state)
if dst_hyper_state_str == hyper_state.HypervisorState.OFFLINE.value:
self.add_action_activate_hypervisor(dst_hypervisor)
model.get_mapping().unmap(src_hypervisor, vm)
model.get_mapping().map(dst_hypervisor, vm)
params = {
'migration_type': migration_type,
'src_hypervisor': src_hypervisor.uuid,
'dst_hypervisor': dst_hypervisor.uuid
}
self.solution.add_action(action_type='migrate',
resource_id=vm.uuid,
input_parameters=params)
self.number_of_migrations += 1
def execute(self, *args, **kwargs):
try:
LOG.debug("Running action %s", self.name)
# todo(jed) remove return (true or false) raise an Exception
result = self.action.execute()
if result is not True:
self.engine.notify(self._db_action,
obj_action.State.FAILED)
else:
self.engine.notify(self._db_action,
obj_action.State.SUCCEEDED)
except Exception as e:
LOG.exception(e)
LOG.error(_LE('The WorkFlow Engine has failed '
'to execute the action %s'), self.name)
self.engine.notify(self._db_action,
obj_action.State.FAILED)
raise
def add_migration(self, vm_uuid, src_hypervisor,
dst_hypervisor, model):
"""Add an action for VM migration into the solution.
:param vm_uuid: vm uuid
:param src_hypervisor: hypervisor object
:param dst_hypervisor: hypervisor object
:param model: model_root object
:return: None
"""
vm = model.get_vm_from_id(vm_uuid)
vm_state_str = self.get_state_str(vm.state)
if vm_state_str != vm_state.VMState.ACTIVE.value:
'''
Watcher curently only supports live VM migration and block live
VM migration which both requires migrated VM to be active.
When supported, the cold migration may be used as a fallback
migration mechanism to move non active VMs.
'''
LOG.error(_LE('Cannot live migrate: vm_uuid=%(vm_uuid)s, '
'state=%(vm_state)s.'),
vm_uuid=vm_uuid,
vm_state=vm_state_str)
raise exception.WatcherException
migration_type = 'live'
dst_hyper_state_str = self.get_state_str(dst_hypervisor.state)
if dst_hyper_state_str == hyper_state.HypervisorState.OFFLINE.value:
self.add_action_activate_hypervisor(dst_hypervisor)
model.get_mapping().unmap(src_hypervisor, vm)
model.get_mapping().map(dst_hypervisor, vm)
params = {'migration_type': migration_type,
'src_hypervisor': src_hypervisor.uuid,
'dst_hypervisor': dst_hypervisor.uuid}
self.solution.add_action(action_type='migrate',
resource_id=vm.uuid,
input_parameters=params)
self.number_of_migrations += 1
def choose_vm_to_migrate(self, hosts):
"""Pick up an active vm instance to migrate from provided hosts"""
for hvmap in hosts:
mig_src_hypervisor = hvmap['hv']
vms_of_src = self.model.get_mapping().get_node_vms(
mig_src_hypervisor)
if len(vms_of_src) > 0:
for vm_id in vms_of_src:
try:
# select the first active VM to migrate
vm = self.model.get_vm_from_id(vm_id)
if vm.state != vm_state.VMState.ACTIVE.value:
LOG.info(_LE("VM not active, skipped: %s"),
vm.uuid)
continue
return mig_src_hypervisor, vm
except wexc.InstanceNotFound as e:
LOG.exception(e)
LOG.info(_LI("VM not found"))
return None
def choose_instance_to_migrate(self, hosts, avg_workload, workload_cache):
"""Pick up an active instance instance to migrate from provided hosts
:param hosts: the array of dict which contains node object
:param avg_workload: the average workload value of all nodes
:param workload_cache: the map contains instance to workload mapping
"""
for instance_data in hosts:
source_node = instance_data['node']
source_instances = self.compute_model.mapping.get_node_instances(
source_node)
if source_instances:
delta_workload = instance_data['workload'] - avg_workload
min_delta = 1000000
instance_id = None
for inst_id in source_instances:
try:
# select the first active VM to migrate
instance = self.compute_model.get_instance_from_id(
inst_id)
if (instance.state !=
element.InstanceState.ACTIVE.value):
LOG.debug("Instance not active, skipped: %s",
instance.uuid)
continue
current_delta = (
delta_workload - workload_cache[inst_id])
if 0 <= current_delta < min_delta:
min_delta = current_delta
instance_id = inst_id
except wexc.InstanceNotFound:
LOG.error(_LE("Instance not found; error: %s"),
instance_id)
if instance_id:
return (source_node,
self.compute_model.get_instance_from_id(
instance_id))
else:
LOG.info(_LI("VM not found from node: %s"),
source_node.uuid)
def add_migration(self, instance_uuid, source_node,
destination_node, model):
"""Add an action for VM migration into the solution.
:param instance_uuid: instance uuid
:param source_node: node object
:param destination_node: node object
:param model: model_root object
:return: None
"""
instance = model.get_instance_from_id(instance_uuid)
instance_state_str = self.get_state_str(instance.state)
if instance_state_str != element.InstanceState.ACTIVE.value:
# Watcher curently only supports live VM migration and block live
# VM migration which both requires migrated VM to be active.
# When supported, the cold migration may be used as a fallback
# migration mechanism to move non active VMs.
LOG.error(
_LE('Cannot live migrate: instance_uuid=%(instance_uuid)s, '
'state=%(instance_state)s.'),
instance_uuid=instance_uuid,
instance_state=instance_state_str)
raise exception.WatcherException
migration_type = 'live'
destination_node_state_str = self.get_state_str(destination_node.state)
if destination_node_state_str == element.ServiceState.DISABLED.value:
self.add_action_enable_compute_node(destination_node)
model.mapping.unmap(source_node, instance)
model.mapping.map(destination_node, instance)
params = {'migration_type': migration_type,
'source_node': source_node.uuid,
'destination_node': destination_node.uuid}
self.solution.add_action(action_type='migrate',
resource_id=instance.uuid,
input_parameters=params)
self.number_of_migrations += 1
def choose_instance_to_migrate(self, hosts, avg_workload, workload_cache):
"""Pick up an active instance instance to migrate from provided hosts
:param hosts: the array of dict which contains node object
:param avg_workload: the average workload value of all nodes
:param workload_cache: the map contains instance to workload mapping
"""
for instance_data in hosts:
source_node = instance_data['node']
source_instances = self.compute_model.mapping.get_node_instances(
source_node)
if source_instances:
delta_workload = instance_data['workload'] - avg_workload
min_delta = 1000000
instance_id = None
for inst_id in source_instances:
try:
# select the first active VM to migrate
instance = self.compute_model.get_instance_by_uuid(
inst_id)
if (instance.state !=
element.InstanceState.ACTIVE.value):
LOG.debug("Instance not active, skipped: %s",
instance.uuid)
continue
current_delta = (delta_workload -
workload_cache[inst_id])
if 0 <= current_delta < min_delta:
min_delta = current_delta
instance_id = inst_id
except wexc.InstanceNotFound:
LOG.error(_LE("Instance not found; error: %s"),
instance_id)
if instance_id:
return (
source_node,
self.compute_model.get_instance_by_uuid(instance_id))
else:
LOG.info(_LI("VM not found from node: %s"), source_node.uuid)
def choose_vm_to_migrate(self, cluster_data_model, hosts):
"""pick up an active vm instance to migrate from provided hosts"""
for hvmap in hosts:
mig_src_hypervisor = hvmap['hv']
vms_of_src = cluster_data_model.get_mapping().get_node_vms(
mig_src_hypervisor)
if len(vms_of_src) > 0:
for vm_id in vms_of_src:
try:
# select the first active VM to migrate
vm = cluster_data_model.get_vm_from_id(vm_id)
if vm.state != vm_state.VMState.ACTIVE.value:
LOG.info(_LE("VM not active, skipped: %s"),
vm.uuid)
continue
return mig_src_hypervisor, vm
except wexc.InstanceNotFound as e:
LOG.info("VM not found Error: %s" % e.message)
pass
return None
def calculate_score_node(self, node):
"""Calculate the score that represent the utilization level
:param node: :py:class:`~.ComputeNode` instance
:return: Score for the given compute node
:rtype: float
"""
host_avg_cpu_util = self.get_node_cpu_usage(node)
if host_avg_cpu_util is None:
resource_id = "%s_%s" % (node.uuid, node.hostname)
LOG.error(
_LE("No values returned by %(resource_id)s "
"for %(metric_name)s") %
dict(resource_id=resource_id,
metric_name=self.METRIC_NAMES[self.config.datasource]
['host_cpu_usage']))
host_avg_cpu_util = 100
total_cores_used = node.vcpus * (host_avg_cpu_util / 100.0)
return self.calculate_weight(node, total_cores_used, 0, 0)
def add_migration(self, instance, source_node, destination_node):
"""Add an action for VM migration into the solution.
:param instance: instance object
:param source_node: node object
:param destination_node: node object
:return: None
"""
instance_state_str = self.get_state_str(instance.state)
if instance_state_str != element.InstanceState.ACTIVE.value:
# Watcher curently only supports live VM migration and block live
# VM migration which both requires migrated VM to be active.
# When supported, the cold migration may be used as a fallback
# migration mechanism to move non active VMs.
LOG.error(
_LE('Cannot live migrate: instance_uuid=%(instance_uuid)s, '
'state=%(instance_state)s.') %
dict(instance_uuid=instance.uuid,
instance_state=instance_state_str))
return
migration_type = 'live'
destination_node_state_str = self.get_state_str(destination_node.state)
if destination_node_state_str == element.ServiceState.DISABLED.value:
self.add_action_enable_compute_node(destination_node)
if self.compute_model.migrate_instance(instance, source_node,
destination_node):
params = {
'migration_type': migration_type,
'source_node': source_node.uuid,
'destination_node': destination_node.uuid
}
self.solution.add_action(action_type='migrate',
resource_id=instance.uuid,
input_parameters=params)
self.number_of_migrations += 1
def group_hosts_by_cpu_util(self):
"""Calculate the workloads of each node
try to find out the nodes which have reached threshold
and the nodes which are under threshold.
and also calculate the average workload value of all nodes.
and also generate the instance workload map.
"""
nodes = self.compute_model.get_all_compute_nodes()
cluster_size = len(nodes)
if not nodes:
raise wexc.ClusterEmpty()
# get cpu cores capacity of nodes and instances
cap_cores = self.compute_model.get_resource_from_id(
element.ResourceType.cpu_cores)
overload_hosts = []
nonoverload_hosts = []
# total workload of cluster
# it's the total core numbers being utilized in a cluster.
cluster_workload = 0.0
# use workload_cache to store the workload of VMs for reuse purpose
workload_cache = {}
for node_id in nodes:
node = self.compute_model.get_node_from_id(
node_id)
instances = self.compute_model.mapping.get_node_instances(node)
node_workload = 0.0
for instance_id in instances:
instance = self.compute_model.get_instance_from_id(instance_id)
try:
cpu_util = self.ceilometer.statistic_aggregation(
resource_id=instance_id,
meter_name=self._meter,
period=self._period,
aggregate='avg')
except Exception as exc:
LOG.exception(exc)
LOG.error(_LE("Can not get cpu_util from Ceilometer"))
continue
if cpu_util is None:
LOG.debug("Instance (%s): cpu_util is None", instance_id)
continue
instance_cores = cap_cores.get_capacity(instance)
workload_cache[instance_id] = cpu_util * instance_cores / 100
node_workload += workload_cache[instance_id]
LOG.debug("VM (%s): cpu_util %f", instance_id, cpu_util)
node_cores = cap_cores.get_capacity(node)
hy_cpu_util = node_workload / node_cores * 100
cluster_workload += node_workload
instance_data = {
'node': node, "cpu_util": hy_cpu_util,
'workload': node_workload}
if hy_cpu_util >= self.threshold:
# mark the node to release resources
overload_hosts.append(instance_data)
else:
nonoverload_hosts.append(instance_data)
avg_workload = cluster_workload / cluster_size
return overload_hosts, nonoverload_hosts, avg_workload, workload_cache