Merge "Add cpu.delay meter"

ceilometer/compute/pollsters/instance_stats.py

@@ -38,6 +38,17 @@ class CPUPollster(InstanceStatsPollster):
         return {'cpu_number': c_data.cpu_number}
 
 
+class CPUDelayPollster(InstanceStatsPollster):
+    sample_name = 'cpu.delay'
+    sample_unit = 'ns'
+    sample_stats_key = 'cpu_delay'
+    sample_type = sample.TYPE_CUMULATIVE
+
+    @staticmethod
+    def get_additional_metadata(instance, c_data):
+        return {'cpu_number': c_data.cpu_number}
+
+
 class VCPUsPollster(InstanceStatsPollster):
     sample_name = 'vcpus'
     sample_unit = 'vcpu'

ceilometer/compute/virt/inspector.py

@@ -43,6 +43,7 @@ class InstanceStats:
         'power_state',       # the power state of the domain
         'cpu_number',        # number: number of CPUs
         'cpu_time',          # time: cumulative CPU time
+        'cpu_delay',         # delay: cumulative CPU steal time
         'memory_actual',     # actual: Amount of allocated memory
         'memory_available',  # available: Amount of usable memory
         'memory_usage',      # usage: Amount of memory used

ceilometer/compute/virt/libvirt/inspector.py

@@ -246,7 +246,7 @@ class LibvirtInspector(virt_inspector.Inspector):
         # TODO(sileht): stats also have the disk/vnic info
         # we could use that instead of the old method for Queen
         stats = self.connection.domainListGetStats([domain], 0)[0][1]
-        cpu_time = 0
+        cpu_time = cpu_delay = 0
         current_cpus = stats.get('vcpu.current')
         # Iterate over the maximum number of CPUs here, and count the
         # actual number encountered, since the vcpu.x structure can
@@ -257,6 +257,7 @@ class LibvirtInspector(virt_inspector.Inspector):
             try:
                 cpu_time += (stats.get('vcpu.%s.time' % vcpu) +
                              stats.get('vcpu.%s.wait' % vcpu))
+                cpu_delay += stats.get('vcpu.%s.delay' % vcpu)
                 current_cpus -= 1
             except TypeError:
                 # pass here, if there are too many holes, the cpu count will
@@ -266,11 +267,13 @@ class LibvirtInspector(virt_inspector.Inspector):
         if current_cpus:
             # There wasn't enough data, so fall back
             cpu_time = stats.get('cpu.time')
+            cpu_delay = 0
 
         return virt_inspector.InstanceStats(
             power_state=domain.info()[0],
             cpu_number=stats.get('vcpu.current'),
             cpu_time=cpu_time,
+            cpu_delay=cpu_delay,
             memory_actual=memory_actual,
             memory_available=memory_available,
             memory_usage=memory_used,

ceilometer/publisher/data/gnocchi_resources.yaml

@@ -86,6 +86,8 @@ resources:
       power.state:
       cpu:
         archive_policy_name: ceilometer-low-rate
+      cpu.delay:
+        archive_policy_name: ceilometer-low-rate
       disk.root.size:
       disk.ephemeral.size:
       disk.latency:

ceilometer/tests/unit/compute/pollsters/test_cpu.py

@@ -104,6 +104,37 @@ class TestCPUPollster(base.TestPollsterBase):
                          samples[0].resource_metadata['instance_type'])
 
 
+class TestCPUDelayPollster(base.TestPollsterBase):
+
+    def test_get_samples(self):
+        self._mock_inspect_instance(
+            virt_inspector.InstanceStats(
+                cpu_delay=1 * (10 ** 6), cpu_number=2),
+            virt_inspector.InstanceStats(
+                cpu_delay=3 * (10 ** 6), cpu_number=2),
+            # cpu_delay resets on instance restart
+            virt_inspector.InstanceStats(
+                cpu_delay=2 * (10 ** 6), cpu_number=2),
+        )
+
+        mgr = manager.AgentManager(0, self.CONF)
+        pollster = instance_stats.CPUDelayPollster(self.CONF)
+
+        def _verify_cpu_metering(expected_time):
+            cache = {}
+            samples = list(pollster.get_samples(mgr, cache, [self.instance]))
+            self.assertEqual(1, len(samples))
+            self.assertEqual({'cpu.delay'}, {s.name for s in samples})
+            self.assertEqual(expected_time, samples[0].volume)
+            self.assertEqual(2, samples[0].resource_metadata.get('cpu_number'))
+            # ensure elapsed time between polling cycles is non-zero
+            time.sleep(0.001)
+
+        _verify_cpu_metering(1 * (10 ** 6))
+        _verify_cpu_metering(3 * (10 ** 6))
+        _verify_cpu_metering(2 * (10 ** 6))
+
+
 class TestVCPUsPollster(base.TestPollsterBase):
 
     def test_get_samples(self):

ceilometer/tests/unit/compute/virt/libvirt/test_inspector.py

@@ -64,10 +64,12 @@ class TestLibvirtInspection(base.BaseTestCase):
             'cpu.time': 999999,
             'vcpu.maximum': 4,
             'vcpu.current': 2,
-            'vcpu.0.time': 10000,
-            'vcpu.0.wait': 10000,
+            'vcpu.0.time': 20000,
+            'vcpu.0.wait': 2000,
+            'vcpu.0.delay': 100,
             'vcpu.2.time': 10000,
-            'vcpu.2.wait': 10000,
+            'vcpu.2.wait': 1000,
+            'vcpu.2.delay': 200,
             'perf.cpu_cycles': 7259361,
             'perf.instructions': 8815623,
             'perf.cache_references': 74184,
@@ -78,7 +80,8 @@ class TestLibvirtInspection(base.BaseTestCase):
             stats = self.inspector.inspect_instance(self.instance, None)
             self.assertEqual(0, stats.power_state)
             self.assertEqual(2, stats.cpu_number)
-            self.assertEqual(40000, stats.cpu_time)
+            self.assertEqual(33000, stats.cpu_time)
+            self.assertEqual(300, stats.cpu_delay)
             self.assertEqual(54400 / units.Ki, stats.memory_actual)
             self.assertEqual(51200 / units.Ki, stats.memory_available)
             self.assertEqual(25600 / units.Ki, stats.memory_usage)
@@ -110,6 +113,7 @@ class TestLibvirtInspection(base.BaseTestCase):
             stats = self.inspector.inspect_instance(self.instance)
             self.assertEqual(2, stats.cpu_number)
             self.assertEqual(999999, stats.cpu_time)
+            self.assertEqual(0, stats.cpu_delay)
 
     def test_inspect_instance_with_domain_shutoff(self):
         domain = mock.Mock()
@@ -138,6 +142,7 @@ class TestLibvirtInspection(base.BaseTestCase):
             self.assertEqual(2, stats.cpu_number)
             self.assertEqual(2097152 / units.Ki, stats.memory_actual)
             self.assertIsNone(stats.cpu_time)
+            self.assertIsNone(stats.cpu_delay)
             self.assertIsNone(stats.memory_available)
             self.assertIsNone(stats.memory_usage)
             self.assertIsNone(stats.memory_resident)

doc/source/admin/telemetry-measurements.rst

@@ -236,6 +236,11 @@ The following meters are collected for OpenStack Compute.
 |           |       |      |          |          |         | from within the  |
 |           |       |      |          |          |         | instance         |
 +-----------+-------+------+----------+----------+---------+------------------+
+| **Meters added in the Indri release**                                       |
++-----------+-------+------+----------+----------+---------+------------------+
+| cpu.delay | Cumu\ | ns   | instance | Pollster | Libvirt | CPU steal time   |
+|           | lative|      | ID       |          |         |                  |
++-----------+-------+------+----------+----------+---------+------------------+
 
 .. note::
 

pyproject.toml

@@ -103,6 +103,7 @@ local_node = "ceilometer.polling.discovery.localnode:LocalNodeDiscovery"
 "disk.device.write.latency" = "ceilometer.compute.pollsters.disk:PerDeviceDiskWriteLatencyPollster"
 "power.state" = "ceilometer.compute.pollsters.instance_stats:PowerStatePollster"
 "cpu" = "ceilometer.compute.pollsters.instance_stats:CPUPollster"
+"cpu.delay" = "ceilometer.compute.pollsters.instance_stats:CPUDelayPollster"
 "vcpus" = "ceilometer.compute.pollsters.instance_stats:VCPUsPollster"
 "network.incoming.bytes" = "ceilometer.compute.pollsters.net:IncomingBytesPollster"
 "network.incoming.packets" = "ceilometer.compute.pollsters.net:IncomingPacketsPollster"

releasenotes/notes/cpu_delay-54bd14eee60880e4.yaml

@@ -0,0 +1,5 @@
+---
+features:
+  - |
+    The new ``cpu.delay`` meter has been added. This meter shows total CPU
+    steal time observed per instance.