|
| 1 | +# (C) Datadog, Inc. 2025-present |
| 2 | +# All rights reserved |
| 3 | +# Licensed under a 3-clause BSD style license (see LICENSE) |
| 4 | +from requests.exceptions import ConnectionError, HTTPError, InvalidURL, JSONDecodeError, Timeout |
| 5 | + |
| 6 | +from datadog_checks.base import AgentCheck |
| 7 | +from datadog_checks.proxmox.config_models import ConfigMixin |
| 8 | + |
| 9 | +from .constants import ( |
| 10 | + NODE_RESOURCE, |
| 11 | + OK_STATUS, |
| 12 | + PERF_METRIC_NAME, |
| 13 | + RESOURCE_COUNT_METRICS, |
| 14 | + RESOURCE_METRIC_NAME, |
| 15 | + RESOURCE_TYPE_MAP, |
| 16 | + VM_RESOURCE, |
| 17 | +) |
| 18 | + |
| 19 | + |
| 20 | +class ProxmoxCheck(AgentCheck, ConfigMixin): |
| 21 | + __NAMESPACE__ = 'proxmox' |
| 22 | + |
| 23 | + def __init__(self, name, init_config, instances): |
| 24 | + super(ProxmoxCheck, self).__init__(name, init_config, instances) |
| 25 | + self.check_initializations.append(self._parse_config) |
| 26 | + self.all_resources = {} |
| 27 | + |
| 28 | + def _parse_config(self): |
| 29 | + self.base_tags = [f"proxmox_server:{self.config.proxmox_server}"] |
| 30 | + if self.config.tags: |
| 31 | + self.base_tags.extend(self.config.tags) |
| 32 | + |
| 33 | + def _submit_resource_metrics(self, resource, tags, hostname): |
| 34 | + for metric_name, metric_name_remapped in RESOURCE_METRIC_NAME.items(): |
| 35 | + metric_value = resource.get(metric_name) |
| 36 | + metric_method = self.count if metric_name in RESOURCE_COUNT_METRICS else self.gauge |
| 37 | + if metric_value is not None: |
| 38 | + metric_method(f'{metric_name_remapped}', metric_value, tags=tags, hostname=hostname) |
| 39 | + |
| 40 | + def _get_vm_hostname(self, vm_id, vm_name, node): |
| 41 | + try: |
| 42 | + url = f"{self.config.proxmox_server}/nodes/{node}/qemu/{vm_id}/agent/get-host-name" |
| 43 | + hostname_response = self.http.get(url) |
| 44 | + hostname_json = hostname_response.json() |
| 45 | + except (HTTPError, InvalidURL, ConnectionError, Timeout, JSONDecodeError) as e: |
| 46 | + self.log.info( |
| 47 | + "Failed to get hostname for vm %s on node %s; endpoint: %s; %s", |
| 48 | + vm_id, |
| 49 | + node, |
| 50 | + self.config.proxmox_server, |
| 51 | + e, |
| 52 | + ) |
| 53 | + hostname_json = {} |
| 54 | + hostname = hostname_json.get("data", {}).get("result", {}).get("host-name", vm_name) |
| 55 | + return hostname |
| 56 | + |
| 57 | + def _collect_performance_metrics(self): |
| 58 | + metrics_response = self.http.get(f"{self.config.proxmox_server}/cluster/metrics/export") |
| 59 | + metrics_response_json = metrics_response.json() |
| 60 | + metrics = metrics_response_json.get('data', {}).get('data', []) |
| 61 | + |
| 62 | + for metric in metrics: |
| 63 | + resource_id = metric.get('id') |
| 64 | + resource = self.all_resources.get(resource_id, {}) |
| 65 | + metric_value = metric.get('value') |
| 66 | + metric_name = metric.get('metric') |
| 67 | + metric_type = metric.get('type') |
| 68 | + metric_name_remapped = PERF_METRIC_NAME.get(metric_name) |
| 69 | + hostname = resource.get('hostname') |
| 70 | + tags = resource.get('tags', []) |
| 71 | + if not resource or metric_name_remapped is None: |
| 72 | + self.log.debug( |
| 73 | + "Invalid metric entry found; metric name: %s, resource id: %s", metric_name_remapped, resource_id |
| 74 | + ) |
| 75 | + continue |
| 76 | + |
| 77 | + metric_method = self.count if metric_type == 'derive' else self.gauge |
| 78 | + metric_method(metric_name_remapped, metric_value, tags=tags, hostname=hostname) |
| 79 | + |
| 80 | + def _collect_resource_metrics(self): |
| 81 | + resources_response = self.http.get(f"{self.config.proxmox_server}/cluster/resources") |
| 82 | + resources_response_json = resources_response.json() |
| 83 | + resources = resources_response_json.get("data", []) |
| 84 | + |
| 85 | + external_tags = [] |
| 86 | + all_resources = {} |
| 87 | + |
| 88 | + for resource in resources: |
| 89 | + resource_type = resource.get('type') |
| 90 | + node = resource.get('node') |
| 91 | + resource_type_remapped = RESOURCE_TYPE_MAP.get(resource_type, resource_type) |
| 92 | + resource_id = resource.get('id') |
| 93 | + resource_name = resource.get('name') |
| 94 | + if resource_name is None: |
| 95 | + # some resource types don't have a name attribute |
| 96 | + resource_name = resource.get(resource.get('type', '')) |
| 97 | + |
| 98 | + resource_tags = { |
| 99 | + f'proxmox_type:{resource_type_remapped}', |
| 100 | + f'proxmox_name:{resource_name}', |
| 101 | + f'proxmox_id:{resource_id}', |
| 102 | + } |
| 103 | + |
| 104 | + proxmox_tags = resource.get('tags') |
| 105 | + if proxmox_tags: |
| 106 | + proxmox_tags = proxmox_tags.split(';') |
| 107 | + resource_tags.update(proxmox_tags) |
| 108 | + |
| 109 | + if node and resource_type_remapped != 'node': |
| 110 | + resource_tags.add(f'proxmox_node:{node}') |
| 111 | + |
| 112 | + pool = resource.get('pool') |
| 113 | + if pool and resource_type_remapped != 'pool': |
| 114 | + resource_tags.add(f'proxmox_pool:{pool}') |
| 115 | + |
| 116 | + self.gauge( |
| 117 | + f'{resource_type_remapped}.count', |
| 118 | + 1, |
| 119 | + tags=self.base_tags + list(resource_tags), |
| 120 | + ) |
| 121 | + |
| 122 | + status = resource.get("status") |
| 123 | + status = 1 if status in OK_STATUS else 0 |
| 124 | + |
| 125 | + hostname = None |
| 126 | + |
| 127 | + if (resource_type_remapped == VM_RESOURCE or resource_type_remapped == NODE_RESOURCE) and status == 0: |
| 128 | + # don't collect information about powered off VMs and nodes |
| 129 | + continue |
| 130 | + elif resource_type_remapped == VM_RESOURCE and status == 1: |
| 131 | + vm_id = resource.get('vmid') |
| 132 | + hostname = self._get_vm_hostname(vm_id, resource_name, node) |
| 133 | + elif resource_type_remapped == NODE_RESOURCE: |
| 134 | + hostname = node |
| 135 | + |
| 136 | + tags = [] |
| 137 | + if hostname is None: |
| 138 | + tags = self.base_tags + list(resource_tags) |
| 139 | + else: |
| 140 | + external_tags.append((hostname, {self.__NAMESPACE__: self.base_tags + list(resource_tags)})) |
| 141 | + |
| 142 | + all_resources[resource_id] = {'resource_type': resource_type_remapped, 'tags': tags, 'hostname': hostname} |
| 143 | + |
| 144 | + if resource_type_remapped != "pool": |
| 145 | + # pools don't have a status attribute |
| 146 | + self.gauge( |
| 147 | + f'{resource_type_remapped}.up', |
| 148 | + status, |
| 149 | + tags, |
| 150 | + hostname=hostname, |
| 151 | + ) |
| 152 | + self._submit_resource_metrics(resource, tags, hostname) |
| 153 | + |
| 154 | + self.all_resources = all_resources |
| 155 | + self.set_external_tags(external_tags) |
| 156 | + |
| 157 | + def check(self, _): |
| 158 | + try: |
| 159 | + response = self.http.get(f"{self.config.proxmox_server}/version") |
| 160 | + response.raise_for_status() |
| 161 | + |
| 162 | + response_json = response.json() |
| 163 | + version = response_json.get("data", {}).get("version") |
| 164 | + self.set_metadata('version', version) |
| 165 | + self.gauge("api.up", 1, tags=self.base_tags + ['proxmox_status:up']) |
| 166 | + |
| 167 | + except (HTTPError, InvalidURL, ConnectionError, Timeout, JSONDecodeError) as e: |
| 168 | + self.log.error( |
| 169 | + "Encountered an Exception when hitting the Proxmox API %s: %s", self.config.proxmox_server, e |
| 170 | + ) |
| 171 | + self.gauge("api.up", 0, tags=self.base_tags + ['proxmox_status:down']) |
| 172 | + raise |
| 173 | + |
| 174 | + self._collect_resource_metrics() |
| 175 | + self._collect_performance_metrics() |
0 commit comments