openstack/nova
Code Issues Proposed changes

VMware: fix memory stats

Browse Source 
The total memory for the vCenter cluster managed by Nova
should be the aggregated sum of total memory of each ESX host in the
cluster. This is more accurate than using the available memory of the
resource pool associated to the cluster.

Partial-Bug: #1462957
Change-Id: I030cee9cebb0f030361aa6bbb612da5cd4202a7f
This commit is contained in:
Radoslav Gerganov 2017-10-31 11:22:20 +02:00
parent 32c8ac6b7d
commit 93bd310b91
5 changed files with 51 additions and 36 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
nova/tests/unit/virt/vmwareapi/fake.py
View File

@ -772,6 +772,7 @@ class HostSystem(ManagedObject):
self.set("capability.maxHostSupportedVcpus", 600)
self.set("summary.hardware", hardware)
self.set("summary.runtime", runtime)
self.set("summary.quickStats", quickstats)
self.set("config.network.pnic", net_info_pnic)
self.set("connected", connected)

4
nova/tests/unit/virt/vmwareapi/test_driver_api.py
View File

@ -2113,8 +2113,8 @@ class VMwareAPIVMTestCase(test.NoDBTestCase,
self.assertEqual(32, stats['vcpus'])
self.assertEqual(1024, stats['local_gb'])
self.assertEqual(1024 - 500, stats['local_gb_used'])
self.assertEqual(1000, stats['memory_mb'])
self.assertEqual(500, stats['memory_mb_used'])
self.assertEqual(2048, stats['memory_mb'])
self.assertEqual(1000, stats['memory_mb_used'])
self.assertEqual('VMware vCenter Server', stats['hypervisor_type'])
self.assertEqual(5001000, stats['hypervisor_version'])
self.assertEqual(self.node_name, stats['hypervisor_hostname'])

42
nova/tests/unit/virt/vmwareapi/test_vm_util.py
View File

@ -17,6 +17,7 @@
import collections
import mock
from oslo_utils import units
from oslo_utils import uuidutils
from oslo_vmware import exceptions as vexc
from oslo_vmware.objects import datastore as ds_obj
@ -68,21 +69,32 @@ class VMwareVMUtilTestCase(test.NoDBTestCase):
hardware.numCpuThreads = 16
hardware.vendor = "Intel"
hardware.cpuModel = "Intel(R) Xeon(R)"
hardware.memorySize = 4 * units.Gi
runtime_host_1 = fake.DataObject()
runtime_host_1.connectionState = "connected"
runtime_host_1.inMaintenanceMode = False
quickstats_1 = fake.DataObject()
quickstats_1.overallMemoryUsage = 512
quickstats_2 = fake.DataObject()
quickstats_2.overallMemoryUsage = 512
runtime_host_2 = fake.DataObject()
runtime_host_2.connectionState = connection_state
runtime_host_2.inMaintenanceMode = maintenance_mode
prop_list_host_1 = [fake.Prop(name="hardware_summary", val=hardware),
fake.Prop(name="runtime_summary",
val=runtime_host_1)]
prop_list_host_2 = [fake.Prop(name="hardware_summary", val=hardware),
fake.Prop(name="runtime_summary",
val=runtime_host_2)]
prop_list_host_1 = [fake.Prop(name="summary.hardware", val=hardware),
fake.Prop(name="summary.runtime",
val=runtime_host_1),
fake.Prop(name="summary.quickStats",
val=quickstats_1)]
prop_list_host_2 = [fake.Prop(name="summary.hardware", val=hardware),
fake.Prop(name="summary.runtime",
val=runtime_host_2),
fake.Prop(name="summary.quickStats",
val=quickstats_2)]
fake_objects = fake.FakeRetrieveResult()
fake_objects.add_object(fake.ObjectContent("prop_list_host1",
@ -90,29 +102,25 @@ class VMwareVMUtilTestCase(test.NoDBTestCase):
fake_objects.add_object(fake.ObjectContent("prop_list_host1",
prop_list_host_2))
respool_resource_usage = fake.DataObject()
respool_resource_usage.maxUsage = 5368709120
respool_resource_usage.overallUsage = 2147483648
def fake_call_method(*args):
if "get_object_properties_dict" in args:
return prop_dict
elif "get_properties_for_a_collection_of_objects" in args:
return fake_objects
else:
return respool_resource_usage
raise Exception('unexpected method call')
session = fake.FakeSession()
with mock.patch.object(session, '_call_method', fake_call_method):
result = vm_util.get_stats_from_cluster(session, "cluster1")
mem_info = {}
if connection_state == "connected" and not maintenance_mode:
vcpus = 32
num_hosts = 2
else:
vcpus = 16
mem_info['total'] = 5120
mem_info['free'] = 3072
expected_stats = {'vcpus': vcpus, 'mem': mem_info}
num_hosts = 1
expected_stats = {'vcpus': num_hosts * 16,
'mem': {'total': num_hosts * 4096,
'free': num_hosts * 4096 -
num_hosts * 512}}
self.assertEqual(expected_stats, result)
def test_get_stats_from_cluster_hosts_connected_and_active(self):

33
nova/virt/vmwareapi/vm_util.py
View File

@ -1159,7 +1159,8 @@ def get_vm_state(session, instance):
def get_stats_from_cluster(session, cluster):
"""Get the aggregate resource stats of a cluster."""
vcpus = 0
mem_info = {'total': 0, 'free': 0}
used_mem_mb = 0
total_mem_mb = 0
# Get the Host and Resource Pool Managed Object Refs
prop_dict = session._call_method(vutil,
"get_object_properties_dict",
@ -1172,27 +1173,25 @@ def get_stats_from_cluster(session, cluster):
result = session._call_method(vim_util,
"get_properties_for_a_collection_of_objects",
"HostSystem", host_mors,
["summary.hardware", "summary.runtime"])
["summary.hardware", "summary.runtime",
"summary.quickStats"])
for obj in result.objects:
hardware_summary = obj.propSet[0].val
runtime_summary = obj.propSet[1].val
host_props = propset_dict(obj.propSet)
hardware_summary = host_props['summary.hardware']
runtime_summary = host_props['summary.runtime']
stats_summary = host_props['summary.quickStats']
if (runtime_summary.inMaintenanceMode is False and
runtime_summary.connectionState == "connected"):
# Total vcpus is the sum of all pCPUs of individual hosts
# The overcommitment ratio is factored in by the scheduler
vcpus += hardware_summary.numCpuThreads
res_mor = prop_dict.get('resourcePool')
if res_mor:
res_usage = session._call_method(vutil, "get_object_property",
res_mor, "summary.runtime.memory")
if res_usage:
# maxUsage is the memory limit of the cluster available to VM's
mem_info['total'] = int(res_usage.maxUsage / units.Mi)
# overallUsage is the hypervisor's view of memory usage by VM's
consumed = int(res_usage.overallUsage / units.Mi)
mem_info['free'] = mem_info['total'] - consumed
stats = {'vcpus': vcpus, 'mem': mem_info}
threads = hardware_summary.numCpuThreads
vcpus += threads
used_mem_mb += stats_summary.overallMemoryUsage
mem_mb = hardware_summary.memorySize // units.Mi
total_mem_mb += mem_mb
stats = {'vcpus': vcpus,
'mem': {'total': total_mem_mb,
'free': total_mem_mb - used_mem_mb}}
return stats

7
releasenotes/notes/vmware-mem-stats-a9b6fac815d2bc57.yaml Normal file
View File

@ -0,0 +1,7 @@
---
fixes:
- |
Fixes how memory stats are reported for VMware. The total memory for the
vCenter cluster managed by Nova should be the aggregated sum of total
memory of each ESX host in the cluster. This is more accurate than using
the available memory of the resource pool associated to the cluster.