Fix bug 1258: metrics not properly collected in some Openstacks
[osm/MON.git] / osm_mon / collector / vnf_collectors / openstack.py
index ba7097e..df0424b 100644 (file)
 # For those usages not covered by the Apache License, Version 2.0 please
 # contact: bdiaz@whitestack.com or glavado@whitestack.com
 ##
-import datetime
-import json
+from enum import Enum
 import logging
+import time
 from typing import List
 
+from ceilometerclient import client as ceilometer_client
+from ceilometerclient.exc import HTTPException
 import gnocchiclient.exceptions
 from gnocchiclient.v1 import client as gnocchi_client
-from keystoneauth1 import session
-from keystoneauth1.identity import v3
+from keystoneauth1.exceptions.catalog import EndpointNotFound
+from keystoneclient.v3 import client as keystone_client
+from neutronclient.v2_0 import client as neutron_client
 
 from osm_mon.collector.metric import Metric
+from osm_mon.collector.utils.openstack import OpenstackUtils
 from osm_mon.collector.vnf_collectors.base_vim import BaseVimCollector
 from osm_mon.collector.vnf_metric import VnfMetric
-from osm_mon.core.auth import AuthManager
 from osm_mon.core.common_db import CommonDbClient
 from osm_mon.core.config import Config
 
+
 log = logging.getLogger(__name__)
 
 METRIC_MAPPINGS = {
     "average_memory_utilization": "memory.usage",
-    "disk_read_ops": "disk.read.requests",
-    "disk_write_ops": "disk.write.requests",
-    "disk_read_bytes": "disk.read.bytes",
-    "disk_write_bytes": "disk.write.bytes",
-    "packets_dropped": "interface.if_dropped",
-    "packets_received": "interface.if_packets",
-    "packets_sent": "interface.if_packets",
-    "cpu_utilization": "cpu_util",
+    "disk_read_ops": "disk.read.requests.rate",
+    "disk_write_ops": "disk.write.requests.rate",
+    "disk_read_bytes": "disk.read.bytes.rate",
+    "disk_write_bytes": "disk.write.bytes.rate",
+    "packets_in_dropped": "network.outgoing.packets.drop",
+    "packets_out_dropped": "network.incoming.packets.drop",
+    "packets_received": "network.incoming.packets.rate",
+    "packets_sent": "network.outgoing.packets.rate",
+    "cpu_utilization": "cpu",
+}
+
+METRIC_MULTIPLIERS = {
+    "cpu": 0.0000001
 }
 
+METRIC_AGGREGATORS = {
+    "cpu": "rate:mean"
+}
+
+INTERFACE_METRICS = ['packets_in_dropped', 'packets_out_dropped', 'packets_received', 'packets_sent']
+
+
+class MetricType(Enum):
+    INSTANCE = 'instance'
+    INTERFACE_ALL = 'interface_all'
+    INTERFACE_ONE = 'interface_one'
+
 
 class OpenstackCollector(BaseVimCollector):
     def __init__(self, config: Config, vim_account_id: str):
         super().__init__(config, vim_account_id)
-        self.conf = config
         self.common_db = CommonDbClient(config)
-        self.auth_manager = AuthManager(config)
-        self.granularity = self._get_granularity(vim_account_id)
-        self.gnocchi_client = self._build_gnocchi_client(vim_account_id)
+        vim_account = self.common_db.get_vim_account(vim_account_id)
+        self.backend = self._get_backend(vim_account)
+
+    def _build_keystone_client(self, vim_account: dict) -> keystone_client.Client:
+        sess = OpenstackUtils.get_session(vim_account)
+        return keystone_client.Client(session=sess)
 
-    def _get_resource_uuid(self, nsr_id, vnf_member_index, vdur_name) -> str:
+    def _get_resource_uuid(self, nsr_id: str, vnf_member_index: str, vdur_name: str) -> str:
         vdur = self.common_db.get_vdur(nsr_id, vnf_member_index, vdur_name)
         return vdur['vim-id']
 
-    def _build_gnocchi_client(self, vim_account_id: str) -> gnocchi_client.Client:
-        creds = self.auth_manager.get_credentials(vim_account_id)
-        verify_ssl = self.auth_manager.is_verify_ssl(vim_account_id)
-        auth = v3.Password(auth_url=creds.url,
-                           username=creds.user,
-                           password=creds.password,
-                           project_name=creds.tenant_name,
-                           project_domain_id='default',
-                           user_domain_id='default')
-        sess = session.Session(auth=auth, verify=verify_ssl)
-        return gnocchi_client.Client(session=sess)
-
-    def _get_granularity(self, vim_account_id: str):
-        creds = self.auth_manager.get_credentials(vim_account_id)
-        vim_config = json.loads(creds.config)
-        if 'granularity' in vim_config:
-            return int(vim_config['granularity'])
-        else:
-            return int(self.conf.get('openstack', 'default_granularity'))
-
     def collect(self, vnfr: dict) -> List[Metric]:
         nsr_id = vnfr['nsr-id-ref']
         vnf_member_index = vnfr['member-vnf-index-ref']
         vnfd = self.common_db.get_vnfd(vnfr['vnfd-id'])
+
+        # Populate extra tags for metrics
+        tags = {}
+        tags['ns_name'] = self.common_db.get_nsr(nsr_id)['name']
+        if vnfr['_admin']['projects_read']:
+            tags['project_id'] = vnfr['_admin']['projects_read'][0]
+        else:
+            tags['project_id'] = ''
+
         metrics = []
         for vdur in vnfr['vdur']:
             # This avoids errors when vdur records have not been completely filled
@@ -99,19 +111,177 @@ class OpenstackCollector(BaseVimCollector):
             if 'monitoring-param' in vdu:
                 for param in vdu['monitoring-param']:
                     metric_name = param['nfvi-metric']
-                    gnocchi_metric_name = METRIC_MAPPINGS[metric_name]
-                    delta = 10 * self.granularity
-                    start_date = datetime.datetime.now() - datetime.timedelta(seconds=delta)
-                    resource_id = self._get_resource_uuid(nsr_id, vnf_member_index, vdur['name'])
+                    interface_name = param['interface-name-ref'] if 'interface-name-ref' in param else None
+                    openstack_metric_name = METRIC_MAPPINGS[metric_name]
+                    metric_type = self._get_metric_type(metric_name, interface_name)
                     try:
-                        measures = self.gnocchi_client.metric.get_measures(gnocchi_metric_name,
-                                                                           start=start_date,
-                                                                           resource_id=resource_id,
-                                                                           granularity=self.granularity)
-                        if len(measures):
-                            metric = VnfMetric(nsr_id, vnf_member_index, vdur['name'], metric_name, measures[-1][2])
+                        resource_id = self._get_resource_uuid(nsr_id, vnf_member_index, vdur['name'])
+                    except ValueError:
+                        log.warning(
+                            "Could not find resource_uuid for vdur %s, vnf_member_index %s, nsr_id %s. "
+                            "Was it recently deleted?",
+                            vdur['name'], vnf_member_index, nsr_id)
+                        continue
+                    try:
+                        log.info("Collecting metric type: %s and metric_name: %s and resource_id %s and "
+                                 "interface_name: %s", metric_type, metric_name, resource_id, interface_name)
+                        value = self.backend.collect_metric(metric_type, openstack_metric_name, resource_id,
+                                                            interface_name)
+                        if value is not None:
+                            log.info("value: %s", value)
+                            if interface_name:
+                                tags['interface'] = interface_name
+                            metric = VnfMetric(nsr_id, vnf_member_index, vdur['name'], metric_name, value, tags)
                             metrics.append(metric)
-                    except gnocchiclient.exceptions.NotFound as e:
-                        log.debug("No metric found: %s", e)
-                        pass
+                        else:
+                            log.info("metric value is empty")
+                    except Exception as e:
+                        log.exception("Error collecting metric %s for vdu %s" % (metric_name, vdur['name']))
+                        log.info("Error in metric collection: %s" % e)
         return metrics
+
+    def _get_backend(self, vim_account: dict):
+        try:
+            ceilometer = CeilometerBackend(vim_account)
+            ceilometer.client.capabilities.get()
+            log.info("Using ceilometer backend to collect metric")
+            return ceilometer
+        except (HTTPException, EndpointNotFound):
+            gnocchi = GnocchiBackend(vim_account)
+            gnocchi.client.metric.list(limit=1)
+            log.info("Using gnocchi backend to collect metric")
+            return gnocchi
+
+    def _get_metric_type(self, metric_name: str, interface_name: str) -> MetricType:
+        if metric_name not in INTERFACE_METRICS:
+            return MetricType.INSTANCE
+        else:
+            if interface_name:
+                return MetricType.INTERFACE_ONE
+            return MetricType.INTERFACE_ALL
+
+
+class OpenstackBackend:
+    def collect_metric(self, metric_type: MetricType, metric_name: str, resource_id: str, interface_name: str):
+        pass
+
+
+class GnocchiBackend(OpenstackBackend):
+
+    def __init__(self, vim_account: dict):
+        self.client = self._build_gnocchi_client(vim_account)
+        self.neutron = self._build_neutron_client(vim_account)
+
+    def _build_gnocchi_client(self, vim_account: dict) -> gnocchi_client.Client:
+        sess = OpenstackUtils.get_session(vim_account)
+        return gnocchi_client.Client(session=sess)
+
+    def _build_neutron_client(self, vim_account: dict) -> neutron_client.Client:
+        sess = OpenstackUtils.get_session(vim_account)
+        return neutron_client.Client(session=sess)
+
+    def collect_metric(self, metric_type: MetricType, metric_name: str, resource_id: str, interface_name: str):
+        if metric_type == MetricType.INTERFACE_ONE:
+            return self._collect_interface_one_metric(metric_name, resource_id, interface_name)
+
+        if metric_type == MetricType.INTERFACE_ALL:
+            return self._collect_interface_all_metric(metric_name, resource_id)
+
+        elif metric_type == MetricType.INSTANCE:
+            return self._collect_instance_metric(metric_name, resource_id)
+
+        else:
+            raise Exception('Unknown metric type %s' % metric_type.value)
+
+    def _collect_interface_one_metric(self, metric_name, resource_id, interface_name):
+        ports = self.neutron.list_ports(name=interface_name, device_id=resource_id)
+        if not ports or not ports['ports']:
+            raise Exception(
+                'Port not found for interface %s on instance %s' % (interface_name, resource_id))
+        port = ports['ports'][0]
+        port_uuid = port['id'][:11]
+        tap_name = 'tap' + port_uuid
+        interfaces = self.client.resource.search(resource_type='instance_network_interface',
+                                                 query={'=': {'name': tap_name}})
+        measures = self.client.metric.get_measures(metric_name,
+                                                   resource_id=interfaces[0]['id'],
+                                                   limit=1)
+        return measures[-1][2] if measures else None
+
+    def _collect_interface_all_metric(self, openstack_metric_name, resource_id):
+        total_measure = None
+        interfaces = self.client.resource.search(resource_type='instance_network_interface',
+                                                 query={'=': {'instance_id': resource_id}})
+        for interface in interfaces:
+            try:
+                measures = self.client.metric.get_measures(openstack_metric_name,
+                                                           resource_id=interface['id'],
+                                                           limit=1)
+                if measures:
+                    if not total_measure:
+                        total_measure = 0.0
+                    total_measure += measures[-1][2]
+
+            except gnocchiclient.exceptions.NotFound as e:
+                log.debug("No metric %s found for interface %s: %s", openstack_metric_name,
+                          interface['id'], e)
+        return total_measure
+
+    def _collect_instance_metric(self, openstack_metric_name, resource_id):
+        value = None
+        try:
+            aggregation = METRIC_AGGREGATORS.get(openstack_metric_name)
+
+            try:
+                measures = self.client.metric.get_measures(openstack_metric_name,
+                                                           aggregation=aggregation,
+                                                           start=time.time() - 1200,
+                                                           resource_id=resource_id)
+                if measures:
+                    value = measures[-1][2]
+            except gnocchiclient.exceptions.NotFound as e:
+                # CPU metric in previous Openstack versions do not support rate:mean aggregation method
+                if openstack_metric_name == "cpu":
+                    log.debug("No metric %s found for instance %s: %s", openstack_metric_name, resource_id, e)
+                    log.debug("Retrying to get metric %s for instance %s without aggregation",
+                              openstack_metric_name, resource_id)
+                    measures = self.client.metric.get_measures(openstack_metric_name,
+                                                               resource_id=resource_id,
+                                                               limit=1)
+                else:
+                    raise e
+                # measures[-1] is the last measure
+                # measures[-2] is the previous measure
+                # measures[x][2] is the value of the metric
+                if measures and len(measures) >= 2:
+                    value = measures[-1][2] - measures[-2][2]
+            if value:
+                # measures[-1][0] is the time of the reporting interval
+                # measures[-1][1] is the duration of the reporting interval
+                if aggregation:
+                    # If this is an aggregate, we need to divide the total over the reported time period.
+                    # Even if the aggregation method is not supported by Openstack, the code will execute it
+                    # because aggregation is specified in METRIC_AGGREGATORS
+                    value = value / measures[-1][1]
+                if openstack_metric_name in METRIC_MULTIPLIERS:
+                    value = value * METRIC_MULTIPLIERS[openstack_metric_name]
+        except gnocchiclient.exceptions.NotFound as e:
+            log.debug("No metric %s found for instance %s: %s", openstack_metric_name, resource_id,
+                      e)
+        return value
+
+
+class CeilometerBackend(OpenstackBackend):
+    def __init__(self, vim_account: dict):
+        self.client = self._build_ceilometer_client(vim_account)
+
+    def _build_ceilometer_client(self, vim_account: dict) -> ceilometer_client.Client:
+        sess = OpenstackUtils.get_session(vim_account)
+        return ceilometer_client.Client("2", session=sess)
+
+    def collect_metric(self, metric_type: MetricType, metric_name: str, resource_id: str, interface_name: str):
+        if metric_type != MetricType.INSTANCE:
+            raise NotImplementedError('Ceilometer backend only support instance metrics')
+        measures = self.client.samples.list(meter_name=metric_name, limit=1, q=[
+            {'field': 'resource_id', 'op': 'eq', 'value': resource_id}])
+        return measures[0].counter_volume if measures else None