Emit cpu_util from transformer instead of pollster

The emission of the cpu_util counter from the compute CPUUtilPollster
can be superceeded by the RateOfChangeTransformer (the intended
target usercase for the new transformer).

The caching of the CPUStats is now no longer required.

Note that the pipeline.yaml must be updated in deployments, otherwise
the cpu_util meter will no longer be collected.

Change-Id: I79ff047595b960de7a2acb2ec81c357fe2ee21e5
This commit is contained in:
Eoghan Glynn 2013-07-12 15:52:44 +00:00
parent ac2b21b722
commit 74c1b0d7aa
4 changed files with 32 additions and 150 deletions

View File

@ -18,9 +18,6 @@
# License for the specific language governing permissions and limitations
# under the License.
import abc
import datetime
from ceilometer import counter
from ceilometer.compute import plugin
from ceilometer.compute.pollsters import util
@ -29,91 +26,25 @@ from ceilometer.openstack.common import log
LOG = log.getLogger(__name__)
class _Base(plugin.ComputePollster):
CACHE_KEY_CPU = 'cpu'
def _get_cpu_info(self, inspector, instance_name, cache):
i_cache = cache.setdefault(self.CACHE_KEY_CPU, {})
if instance_name not in i_cache:
i_cache[instance_name] = inspector.inspect_cpus(instance_name)
return i_cache[instance_name]
@abc.abstractmethod
def _get_counter(instance, instance_name, cpu_info):
"""Return one Counter."""
class CPUPollster(plugin.ComputePollster):
def get_counters(self, manager, cache, instance):
LOG.info('checking instance %s', instance.id)
instance_name = util.instance_name(instance)
try:
cpu_info = self._get_cpu_info(
manager.inspector,
instance_name,
cache,
)
yield self._get_counter(
cpu_info = manager.inspector.inspect_cpus(instance_name)
LOG.info("CPUTIME USAGE: %s %d",
instance.__dict__, cpu_info.time)
cpu_num = {'cpu_number': cpu_info.number}
yield util.make_counter_from_instance(
instance,
instance_name,
cpu_info,
name='cpu',
type=counter.TYPE_CUMULATIVE,
unit='ns',
volume=cpu_info.time,
additional_metadata=cpu_num,
)
except Exception as err:
LOG.error('could not get CPU time for %s: %s',
instance.id, err)
LOG.exception(err)
class CPUPollster(_Base):
@staticmethod
def _get_counter(instance, instance_name, cpu_info):
LOG.info("CPUTIME USAGE: %s %d",
instance.__dict__, cpu_info.time)
cpu_num = {'cpu_number': cpu_info.number}
return util.make_counter_from_instance(
instance,
name='cpu',
type=counter.TYPE_CUMULATIVE,
unit='ns',
volume=cpu_info.time,
additional_metadata=cpu_num,
)
class CPUUtilPollster(_Base):
# FIXME(eglynn): once we have a way of configuring which measures
# are published to each sink, we should by default
# disable publishing this derived measure to the
# metering store, only publishing to those sinks
# that specifically need it
utilization_map = {}
def _get_cpu_util(self, instance, cpu_info):
prev_times = self.utilization_map.get(instance.id)
self.utilization_map[instance.id] = (cpu_info.time,
datetime.datetime.now())
cpu_util = 0.0
if prev_times:
prev_cpu = prev_times[0]
prev_timestamp = prev_times[1]
delta = self.utilization_map[instance.id][1] - prev_timestamp
elapsed = (delta.seconds * (10 ** 6) + delta.microseconds) * 1000
cores_fraction = 1.0 / cpu_info.number
# account for cpu_time being reset when the instance is restarted
time_used = (cpu_info.time - prev_cpu
if prev_cpu <= cpu_info.time else cpu_info.time)
cpu_util = 100 * cores_fraction * time_used / elapsed
return cpu_util
def _get_counter(self, instance, instance_name, cpu_info):
cpu_util = self._get_cpu_util(instance, cpu_info)
LOG.info("CPU UTILIZATION %%: %s %0.2f",
instance.__dict__, cpu_util)
return util.make_counter_from_instance(
instance,
name='cpu_util',
type=counter.TYPE_GAUGE,
unit='%',
volume=cpu_util,
)

View File

@ -7,3 +7,19 @@
transformers:
publishers:
- rpc://
-
name: cpu_pipeline
interval: 600
counters:
- "cpu"
transformers:
- name: "rate_of_change"
parameters:
target:
name: "cpu_util"
unit: "%"
type: "gauge"
scale: "100.0 / (10**9 * (resource_metadata.cpu_number or 1))"
replace: False
publishers:
- rpc://

View File

@ -56,7 +56,6 @@ ceilometer.poll.compute =
disk.read.bytes = ceilometer.compute.pollsters.disk:ReadBytesPollster
disk.write.bytes = ceilometer.compute.pollsters.disk:WriteBytesPollster
cpu = ceilometer.compute.pollsters.cpu:CPUPollster
cpu_util = ceilometer.compute.pollsters.cpu:CPUUtilPollster
network.incoming.bytes = ceilometer.compute.pollsters.net:IncomingBytesPollster
network.incoming.packets = ceilometer.compute.pollsters.net:IncomingPacketsPollster
network.outgoing.bytes = ceilometer.compute.pollsters.net:OutgoingBytesPollster

View File

@ -55,8 +55,6 @@ class TestCPUPollster(base.TestPollsterBase):
self.assertEqual(set([c.name for c in counters]),
set(['cpu']))
assert counters[0].volume == expected_time
assert pollster.CACHE_KEY_CPU in cache
assert self.instance.name in cache[pollster.CACHE_KEY_CPU]
self.assertEquals(counters[0].resource_metadata.get('cpu_number'),
2)
# ensure elapsed time between polling cycles is non-zero
@ -67,78 +65,16 @@ class TestCPUPollster(base.TestPollsterBase):
_verify_cpu_metering(2 * (10 ** 6))
@mock.patch('ceilometer.pipeline.setup_pipeline', mock.MagicMock())
def test_get_counters_cache(self):
def test_get_counters_no_caching(self):
self.inspector.inspect_cpus(self.instance.name).AndReturn(
virt_inspector.CPUStats(time=1 * (10 ** 6), number=2))
self.mox.ReplayAll()
mgr = manager.AgentManager()
pollster = cpu.CPUPollster()
cache = {
pollster.CACHE_KEY_CPU: {
self.instance.name: virt_inspector.CPUStats(
time=10 ** 6,
number=2,
),
},
}
cache = {}
counters = list(pollster.get_counters(mgr, cache, self.instance))
self.assertEquals(len(counters), 1)
self.assertEquals(counters[0].volume, 10 ** 6)
class TestCPUUtilPollster(base.TestPollsterBase):
def setUp(self):
super(TestCPUUtilPollster, self).setUp()
@mock.patch('ceilometer.pipeline.setup_pipeline', mock.MagicMock())
def test_get_counters(self):
self.inspector.inspect_cpus(self.instance.name).AndReturn(
virt_inspector.CPUStats(time=1 * (10 ** 6), number=2))
self.inspector.inspect_cpus(self.instance.name).AndReturn(
virt_inspector.CPUStats(time=3 * (10 ** 6), number=2))
# cpu_time resets on instance restart
self.inspector.inspect_cpus(self.instance.name).AndReturn(
virt_inspector.CPUStats(time=2 * (10 ** 6), number=2))
self.mox.ReplayAll()
mgr = manager.AgentManager()
pollster = cpu.CPUUtilPollster()
pollster.utilization_map = {} # clear the internal cache
def _verify_cpu_metering(zero):
cache = {}
counters = list(pollster.get_counters(mgr, cache, self.instance))
self.assertEquals(len(counters), 1)
self.assertEqual(set([c.name for c in counters]),
set(['cpu_util']))
assert (counters[0].volume == 0.0 if zero else
counters[0].volume > 0.0)
assert pollster.CACHE_KEY_CPU in cache
assert self.instance.name in cache[pollster.CACHE_KEY_CPU]
# ensure elapsed time between polling cycles is non-zero
time.sleep(0.001)
_verify_cpu_metering(True)
_verify_cpu_metering(False)
_verify_cpu_metering(False)
@mock.patch('ceilometer.pipeline.setup_pipeline', mock.MagicMock())
def test_get_counters_cache(self):
self.mox.ReplayAll()
mgr = manager.AgentManager()
pollster = cpu.CPUUtilPollster()
pollster.utilization_map = {} # clear the internal cache
cache = {
pollster.CACHE_KEY_CPU: {
self.instance.name: virt_inspector.CPUStats(
time=10 ** 6,
number=2,
),
},
}
counters = list(pollster.get_counters(mgr, cache, self.instance))
self.assertEquals(len(counters), 1)
self.assertEquals(counters[0].volume, 0)
self.assertEquals(len(cache), 0)