Add support for VM property metrics (#14787)

* Add base VM property metrics

* Add network ip

* Add new properties

* Report only VM values and use host tags

* Clean up code

* Add more tests and add to cache

* Add more tests

* use vm hostname

* Clean up code and add tests for all property metrics

* Add submit_property_metric function

* Add type hints and add helper methods

* Add new metrics and update metadata.csv

* Only log if in debug mode

* Add unneeded types and log

* Remove duplicated code, fix metadata.csv

* Move mocks to common file

* Sync models

* update metadata and docs to reflect interval

* Further refactor, standardize metric types, and add another test

* Add tests for logging

* Add cache time test

* Add tools version status metric

* Address review comments

* use new version of models

vsphere/assets/configuration/spec.yaml

@@ -310,6 +310,18 @@ files:
         value:
           type: boolean
           example: false
+      - name: collect_property_metrics
+        description: |
+          If true, the vSphere integration will collect additional metrics about the resource types
+          you have configured to monitor. These are configuration properties such as disk information
+          in a VM and DRS config status in clusters.
+
+          Note: These metrics are collected at a frequency of `refresh_infrastructure_cache_interval`.
+          Warning: Depending on the size of the vSphere environment, property metric collection can be slow,
+          very CPU intensive and put pressure on the vCenter Server.
+        value:
+          type: boolean
+          example: false
       - name: attributes_prefix
         description: |
           Custom attributes attached to vSphere resources will be prefixed with this prefix when collected.

vsphere/datadog_checks/vsphere/api.py

@@ -12,7 +12,12 @@
 
 from datadog_checks.base.log import CheckLoggingAdapter  # noqa: F401
 from datadog_checks.vsphere.config import VSphereConfig  # noqa: F401
-from datadog_checks.vsphere.constants import ALL_RESOURCES, MAX_QUERY_METRICS_OPTION, UNLIMITED_HIST_METRICS_PER_QUERY
+from datadog_checks.vsphere.constants import (
+    ALL_RESOURCES,
+    MAX_QUERY_METRICS_OPTION,
+    UNLIMITED_HIST_METRICS_PER_QUERY,
+    VM_PROPERTIES,
+)
 from datadog_checks.vsphere.event import ALLOWED_EVENTS
 from datadog_checks.vsphere.types import InfrastructureData
 
@@ -184,6 +189,10 @@ def _get_raw_infrastructure(self):
                 property_spec.pathSet.append("runtime.powerState")
                 property_spec.pathSet.append("runtime.host")
                 property_spec.pathSet.append("guest.hostName")
+                if self.config.collect_property_metrics:
+                    for vm_property in VM_PROPERTIES:
+                        property_spec.pathSet.append(vm_property)
+
             property_specs.append(property_spec)
 
         # Specify the attribute of the root object to traverse to obtain all the attributes
@@ -267,13 +276,21 @@ def get_infrastructure(self):
         root_folder = self._conn.content.rootFolder
         infrastructure_data[root_folder] = {"name": root_folder.name, "parent": None}
 
-        if self.config.should_collect_attributes:
+        if self.config.should_collect_attributes or self.config.collect_property_metrics:
             # Clean up attributes in infrastructure_data,
             # at this point they are custom pyvmomi objects and the attribute keys are not resolved.
 
             attribute_keys = {x.key: x.name for x in self._fetch_all_attributes()}
             for props in itervalues(infrastructure_data):
                 mor_attributes = []
+                if self.config.collect_property_metrics:
+                    all_properties = {}
+                    for attribute_name in VM_PROPERTIES:
+                        attribute_val = props.pop(attribute_name, None)
+                        if attribute_val is not None:
+                            all_properties[attribute_name] = attribute_val
+                    props['properties'] = all_properties
+
                 if 'customValue' not in props:
                     continue
                 for attribute in props.pop('customValue'):

vsphere/datadog_checks/vsphere/cache.py

@@ -148,3 +148,9 @@ def set_mor_props(self, mor, mor_data):
         if mor_type not in self._mors:
             self._mors[mor_type] = {}
         self._mors[mor_type][mor] = mor_data
+
+    def clear_properties(self):
+        # type: () -> None
+        for _, mors in self._mors.items():
+            for _, mor_props in mors.items():
+                mor_props.pop('properties', None)

vsphere/datadog_checks/vsphere/config.py

@@ -79,6 +79,7 @@ def __init__(self, instance, init_config, log):
         self.should_collect_tags = is_affirmative(instance.get("collect_tags", False))
         self.tags_prefix = instance.get("tags_prefix", DEFAULT_VSPHERE_TAG_PREFIX)
         self.should_collect_attributes = is_affirmative(instance.get("collect_attributes", False))
+        self.collect_property_metrics = is_affirmative(instance.get("collect_property_metrics", False))
         self.attr_prefix = instance.get("attributes_prefix", DEFAULT_VSPHERE_ATTR_PREFIX)
         self.excluded_host_tags = instance.get("excluded_host_tags", [])
         self.base_tags = instance.get("tags", []) + ["vcenter_server:{}".format(self.hostname)]

vsphere/datadog_checks/vsphere/config_models/defaults.py

@@ -28,6 +28,10 @@ def instance_collect_events_only():
     return False
 
 
+def instance_collect_property_metrics():
+    return False
+
+
 def instance_collect_tags():
     return False
 

vsphere/datadog_checks/vsphere/config_models/instance.py

@@ -127,6 +127,7 @@ class InstanceConfig(BaseModel):
     collect_events: Optional[bool] = None
     collect_events_only: Optional[bool] = None
     collect_per_instance_filters: Optional[CollectPerInstanceFilters] = None
+    collect_property_metrics: Optional[bool] = None
     collect_tags: Optional[bool] = None
     collection_level: Optional[int] = None
     collection_type: Optional[str] = None

vsphere/datadog_checks/vsphere/constants.py

@@ -57,3 +57,33 @@
 
 DEFAULT_VSPHERE_TAG_PREFIX = ""
 DEFAULT_VSPHERE_ATTR_PREFIX = ""
+
+PROPERTY_COUNT_METRICS = [
+    "guest.net",
+    "guest.ipStack.ipRoute",
+    "guest.net.ipConfig.address",
+    "guest.toolsRunningStatus",
+    "guest.toolsVersionStatus2",
+    "guest.toolsVersion",
+    "guest.guestFullName",
+]
+VM_OBJECT_PROPERTIES = ["guest.disk", "guest.net", "guest.ipStack"]
+
+VM_SIMPLE_PROPERTIES = [
+    "guest.toolsRunningStatus",
+    "guest.toolsVersionStatus2",
+    "guest.toolsVersion",
+    "config.hardware.numCoresPerSocket",
+    "config.cpuAllocation.limit",
+    "config.cpuAllocation.overheadLimit",
+    "config.memoryAllocation.limit",
+    "config.memoryAllocation.overheadLimit",
+    "summary.config.numCpu",
+    "summary.config.memorySizeMB",
+    "summary.config.numEthernetCards",
+    "summary.config.numVirtualDisks",
+    "summary.quickStats.uptimeSeconds",
+    "guest.guestFullName",
+]
+
+VM_PROPERTIES = VM_OBJECT_PROPERTIES + VM_SIMPLE_PROPERTIES

vsphere/datadog_checks/vsphere/data/conf.yaml.example

@@ -279,6 +279,17 @@ instances:
     #
     # collect_attributes: false
 
+    ## @param collect_property_metrics - boolean - optional - default: false
+    ## If true, the vSphere integration will collect additional metrics about the resource types
+    ## you have configured to monitor. These are configuration properties such as disk information
+    ## in a VM and DRS config status in clusters.
+    ##
+    ## Note: These metrics are collected at a frequency of `refresh_infrastructure_cache_interval`.
+    ## Warning: Depending on the size of the vSphere environment, property metric collection can be slow,
+    ## very CPU intensive and put pressure on the vCenter Server.
+    #
+    # collect_property_metrics: false
+
     ## @param attributes_prefix - string - optional
     ## Custom attributes attached to vSphere resources will be prefixed with this prefix when collected.
     ## Example use cases:

vsphere/datadog_checks/vsphere/types.py

@@ -5,7 +5,7 @@
 from typing import Any, Dict, List, Optional, Pattern, Type, TypedDict
 
 # CONFIG ALIASES
-from pyVmomi import vim
+from pyVmomi import VmomiSupport, vim
 
 ResourceFilterConfig = TypedDict(
     'ResourceFilterConfig', {'resource': str, 'property': str, 'type': str, 'patterns': List[str]}
@@ -52,13 +52,28 @@
 MetricName = str
 CounterId = int
 
+VmomiObject = VmomiSupport.Object
+
 InfrastructureDataItem = TypedDict(
     'InfrastructureDataItem',
     {
         'name': str,
         'runtime.host': vim.ManagedEntity,
         'guest.hostName': str,
         'runtime.powerState': str,
+        'summary.config.numCpu': int,
+        'summary.config.memorySizeMB': int,
+        'summary.config.numEthernetCards': int,
+        'summary.config.numVirtualDisks': int,
+        'summary.quickStats.uptimeSeconds': int,
+        'guest.guestFullName': str,
+        'guest.disk': List[VmomiObject],
+        'guest.net': List[VmomiObject],
+        'guest.ipStack': List[VmomiObject],
+        'guest.toolsRunningStatus': str,
+        'guest.toolsVersionStatus2': str,
+        'guest.toolsVersion': str,
+        'config.hardware.numCoresPerSocket': str,
         'parent': Optional[vim.ManagedEntity],
         'attributes': List[str],
     },

vsphere/datadog_checks/vsphere/utils.py

@@ -1,7 +1,7 @@
 # (C) Datadog, Inc. 2019-present
 # All rights reserved
 # Licensed under Simplified BSD License (see LICENSE)
-from typing import List, Optional, Type  # noqa: F401
+from typing import Any, Dict, List, Optional, Type  # noqa: F401
 
 from pyVmomi import vim
 from six import iteritems
@@ -153,3 +153,10 @@ def get_mapped_instance_tag(metric_name):
         if metric_name.startswith(prefix):
             return tag_key
     return 'instance'
+
+
+def add_additional_tags(tags, additional_tags):
+    # type: (List[str], Dict[str, Optional[Any]]) -> List[str]
+    for tag_name, tag_value in additional_tags.items():
+        if tag_value is not None:
+            tags.append("{}:{}".format(tag_name, tag_value))