Move SNMP E2E device-tag assertions to the metadata payload

snmp/tests/common.py

@@ -196,6 +196,189 @@
 
 EXCLUDED_E2E_TAG_KEYS = ['agent_version']
 
+# Tag keys the SNMP corecheck used to attach to every per-device metric but, since
+# DataDog/datadog-agent#49822, ships only on the `network-devices-metadata`
+# event-platform payload when `collect_device_metadata` is enabled (the default).
+# The backend re-joins these onto metrics at ingest from the metadata, so E2E
+# tests that ran against the raw `agent check --json` aggregator state must
+# assert them against the metadata payload (see test_e2e_core_metadata.py)
+# rather than against per-metric tags.
+DEVICE_METADATA_TAG_KEYS = (
+    # Device identity (always emitted from the instance config)
+    'agent_host',
+    'device_id',
+    'device_ip',
+    'device_namespace',
+    'snmp_device',
+    # Profile-level globals derived from sysObjectID / sysName / sysDescr
+    'device_hostname',
+    'device_vendor',
+    'snmp_host',
+    'snmp_profile',
+    # Common profile metadata.device.fields
+    'firmware_version',
+    'model',
+    'os_name',
+    'os_version',
+    'product_name',
+    'serial_num',
+    'ups_name',
+    'version',
+    # Profile-specific metadata.device.fields. Each entry below is a tag key
+    # that one or more profiles declare as a global metric_tag whose value
+    # comes from a metadata.device.fields lookup. They're stripped by the
+    # agent the same way as the common identity fields above.
+    'agent_inventory_machine_model',
+    'agent_inventory_software_version',
+    'avaya_cmg_active_controller_address',
+    'avaya_cmg_hw_type',
+    'avaya_cmg_model_number',
+    'avaya_cmg_serial_number',
+    'avaya_cmg_voip_admin_state',
+    'avaya_cmg_voip_current_ip_address',
+    'avaya_gen_cpu_utilization_enable_monitoring',
+    'avaya_s5_chas_ser_num',
+    'avaya_s5_chas_ver',
+    'ax_sys_a_fle_x_engine_version',
+    'ax_sys_firmware_version',
+    'ax_sys_serial_number',
+    'bcn_sys_id_os_release',
+    'bcn_sys_id_platform',
+    'bcn_sys_id_product',
+    'br_info_serial_number',
+    'cppm_cluster_node_type',
+    'cppm_num_cluster_nodes',
+    'cppm_nw_data_port_ip_address',
+    'cppm_nw_data_port_mac_address',
+    'cppm_nw_mgmt_port_ip_address',
+    'cppm_nw_mgmt_port_mac_address',
+    'cppm_system_num_cp_us',
+    'cppm_system_version',
+    'cppm_zone_name',
+    'e_pdu_ident_model_number',
+    'e_pdu_ident_name',
+    'e_pdu_ident_serial_number',
+    'envir_ident_location',
+    'envir_ident_name',
+    'fe_hardware_model',
+    'fe_serial_number',
+    'filer_bios_version',
+    'filer_cpu_arch',
+    'filer_cpu_model',
+    'filer_identifier',
+    'filer_package_format',
+    'filer_platform_name',
+    'filer_platform_type',
+    'filer_serial_number',
+    'filer_version',
+    'fs_sys_serial',
+    'fs_sys_version',
+    'hawk_i2_inv_fw_revision',
+    'hawk_i2_inv_hw_revision',
+    'hawk_i2_inv_serial_num',
+    'hawk_i2_ip_temp_scale_flag',
+    'hpe_fibrechannel_cpq_si_product_name',
+    'hpe_fibrechannel_cpq_si_sys_product_id',
+    'hpe_fibrechannel_cpq_si_sys_serial_num',
+    'huawei_hw_entity_system_model',
+    'ib_hardware_type',
+    'ib_nios_version',
+    'ib_serial_number',
+    'ibm_imm_machine_level_product_name',
+    'ibm_imm_machine_level_serial_number',
+    'ironport_cache_admin',
+    'ironport_cache_software',
+    'ironport_cache_version',
+    'ironport_http_ports',
+    'ironport_license_expiration',
+    'juniper_ive_esap_version',
+    'juniper_ive_ive_max_concurrent_users_license_capacity',
+    'juniper_ive_product_name',
+    'juniper_ive_product_version',
+    'kcprt_general_model_name',
+    'kcprt_serial_number',
+    'legacy_pdu_macaddress',
+    'legacy_pdu_model',
+    'legacy_pdu_name',
+    'legacy_pdu_version',
+    'lgp_agent_ident_model',
+    'lgp_agent_ident_serial_number',
+    'mcafee_mwg_k_build_number',
+    'mcafee_mwg_k_company_name',
+    'mcafee_mwg_k_custom_version',
+    'mcafee_mwg_k_hotfix_version',
+    'mcafee_mwg_k_major_version',
+    'mcafee_mwg_k_micro_version',
+    'mcafee_mwg_k_minor_version',
+    'mcafee_mwg_k_product_name',
+    'mcafee_mwg_k_product_version',
+    'mcafee_mwg_k_revision',
+    'mcafee_mwg_p_am_engine_version',
+    'mcafee_mwg_p_am_proactive_version',
+    'mcafee_mwg_p_mfe_engine_version',
+    'mcafee_mwg_p_mfedat_version',
+    'mcafee_mwg_p_tsdb_version',
+    'netscaler_sdx_system_bios_version',
+    'netscaler_sdx_system_gateway',
+    'netscaler_sdx_system_gateway_type',
+    'netscaler_sdx_system_netmask',
+    'netscaler_sdx_system_netmask_type',
+    'netscaler_sdx_system_network_interface',
+    'netscaler_sdx_system_product',
+    'netscaler_sdx_system_svm_ip_address',
+    'netscaler_sdx_system_svm_ip_address_type',
+    'netscaler_sdx_system_xen_ip_address',
+    'netscaler_sdx_system_xen_ip_address_type',
+    'peplink_bal_firmware',
+    'peplink_bal_serial_number',
+    'peplink_device_firmware_version',
+    'peplink_device_model',
+    'peplink_device_serial_number',
+    'powernet_r_pdu_ident_firmware_rev',
+    'powernet_r_pdu_ident_hardware_rev',
+    'powernet_r_pdu_ident_model_number',
+    'powernet_r_pdu_ident_name',
+    'powernet_r_pdu_ident_serial_number',
+    'riverbed_interceptor_model',
+    'riverbed_interceptor_serial_number',
+    'ruckus_unleashed_system_licensed_aps',
+    'ruckus_unleashed_system_model',
+    'ruckus_unleashed_system_serial_number',
+    'servertech_sentry3_system_nic_serial_number',
+    'servertech_sentry3_system_version',
+    'sfos_device_fw_version',
+    'sfos_device_type',
+    'sfos_ips_version',
+    'synology_model_name',
+    'synology_serial_number',
+    'synology_version',
+    'tl_ups_ident_id',
+    'tl_ups_ident_serial_num',
+    'tl_ups_location',
+    'tl_ups_snmp_card_serial_num',
+    'ups_ident_manufacturer',
+    'ups_ident_model',
+    'ups_ident_name',
+    'vertiv_product_friendly_name',
+    'vertiv_product_mac_address',
+    'vertiv_product_title',
+    'vertiv_product_version',
+    'wdmycloudex2_agent_ver',
+    'wdmycloudex2_ftp_server',
+    'wdmycloudex2_host_name',
+    'wdmycloudex2_net_type',
+    'wdmycloudex2_software_version',
+    'wlsx_model_name',
+    'wlsx_switch_license_serial_number',
+    'wlsx_switch_role',
+    'wlsx_sys_ext_hw_version',
+    'wlsx_sys_ext_sw_version',
+    'zyxel_sys_product_model',
+    'zyxel_sys_product_serial_number',
+    'zyxel_sys_sw_platform',
+    'zyxel_sys_sw_version_string',
+)
+
 snmp_listener_only = pytest.mark.skipif(SNMP_LISTENER_ENV != 'true', reason='Agent snmp lister tests only')
 snmp_integration_only = pytest.mark.skipif(SNMP_LISTENER_ENV != 'false', reason='Normal tests')
 py3_plus_only = pytest.mark.skipif(sys.version_info[0] < 3, reason='Run test with Python 3+ only')
@@ -279,43 +462,74 @@ def create_check(instance):
     return SnmpCheck('snmp', {}, [instance])
 
 
+def filter_metric_tags(tags: list[str] | None) -> list[str] | None:
+    """Drop device-level tags now joined from the NDM metadata payload at ingest.
+
+    Tests pass the device-level `common_tags` list to both metric and metadata
+    assertions for historical reasons. With the new agent tagging model the
+    per-metric assertions must only check tags the agent still attaches to
+    metrics; use this helper to project a `common_tags` list into the subset
+    that is valid for `aggregator.assert_metric(tags=...)` calls.
+    """
+    if tags is None:
+        return None
+    return remove_tags(tags, DEVICE_METADATA_TAG_KEYS)
+
+
 def assert_common_metrics(aggregator, tags=None, is_e2e=False, loader=None):
     assert_common_check_run_metrics(aggregator, tags, is_e2e, loader=loader)
     assert_common_device_metrics(aggregator, tags, is_e2e, loader=loader)
 
 
 def assert_common_check_run_metrics(aggregator, tags=None, is_e2e=False, loader=None):
+    # When running against the core loader, the agent strips device-level tags
+    # from per-device metrics; assert only what the agent still emits.
     if is_e2e and loader == 'core':
-        aggregator.assert_metric('snmp.device.reachable', metric_type=aggregator.GAUGE, tags=tags, at_least=1, value=1)
+        metric_tags = filter_metric_tags(tags)
         aggregator.assert_metric(
-            'snmp.device.unreachable', metric_type=aggregator.GAUGE, tags=tags, at_least=1, value=0
+            'snmp.device.reachable', metric_type=aggregator.GAUGE, tags=metric_tags, at_least=1, value=1
+        )
+        aggregator.assert_metric(
+            'snmp.device.unreachable', metric_type=aggregator.GAUGE, tags=metric_tags, at_least=1, value=0
+        )
+        aggregator.assert_metric(
+            'snmp.interface.status', metric_type=aggregator.GAUGE, tags=metric_tags, at_least=0, value=1
         )
-        aggregator.assert_metric('snmp.interface.status', metric_type=aggregator.GAUGE, tags=tags, at_least=0, value=1)
 
     monotonic_type = aggregator.MONOTONIC_COUNT
     if is_e2e:
         monotonic_type = aggregator.COUNT
     loader = loader or 'python'
-    if tags is not None:
-        tags = tags + ['loader:' + loader]
-    aggregator.assert_metric('datadog.snmp.check_duration', metric_type=aggregator.GAUGE, tags=tags)
-    aggregator.assert_metric('datadog.snmp.check_interval', metric_type=monotonic_type, tags=tags)
-    aggregator.assert_metric('datadog.snmp.submitted_metrics', metric_type=aggregator.GAUGE, tags=tags)
+    telemetry_tags = tags
+    if loader == 'core':
+        telemetry_tags = filter_metric_tags(tags)
+    if telemetry_tags is not None:
+        telemetry_tags = telemetry_tags + ['loader:' + loader]
+    aggregator.assert_metric('datadog.snmp.check_duration', metric_type=aggregator.GAUGE, tags=telemetry_tags)
+    aggregator.assert_metric('datadog.snmp.check_interval', metric_type=monotonic_type, tags=telemetry_tags)
+    aggregator.assert_metric('datadog.snmp.submitted_metrics', metric_type=aggregator.GAUGE, tags=telemetry_tags)
     if loader == 'core':
         # request_type tag can be get, getbulk, or getnext
         aggregator.assert_metric_has_tag_prefix('datadog.snmp.requests', tag_prefix='request_type:get')
-        for tag in tags:
+        for tag in telemetry_tags:
             aggregator.assert_metric_has_tag('datadog.snmp.requests', tag)
 
 
 def assert_common_device_metrics(
     aggregator, tags=None, is_e2e=False, count=None, devices_monitored_value=None, loader=None
 ):
     loader = loader or 'python'
-    if tags is not None:
-        tags = tags + ['loader:' + loader]
+    metric_tags = tags
+    if loader == 'core':
+        metric_tags = filter_metric_tags(tags)
+    if metric_tags is not None:
+        metric_tags = metric_tags + ['loader:' + loader]
     aggregator.assert_metric(
-        'snmp.devices_monitored', metric_type=aggregator.GAUGE, tags=tags, count=count, value=devices_monitored_value
+        'snmp.devices_monitored',
+        metric_type=aggregator.GAUGE,
+        tags=metric_tags,
+        count=count,
+        value=devices_monitored_value,
     )