Windows: event logs, counters, WMI, Active Directory and DNS
The event logs you choose, the counters and WMI classes of your software, and two Windows Server roles watched out of the box - Active Directory and DNS.
The Windows agent (installation) runs as a service under the system account: it reads every log, Security included, and the host's performance counters. Everything below is read-only.
Event logs
By default the agent ships the System and Application logs. The logs.windows_eventlog list takes channel names, or objects with filters:
logs:
windows_eventlog:
- System
- channel: Application
exclude_providers: ["Microsoft-Windows-Security-SPP"] # sources left out (glob patterns)
- channel: Security
event_ids: [4625, 4740, "4720-4726", 4728, 4732, 4756] # IDs or ranges
service: windows-security # instead of the source name
tags: {role: dc}
- channel: Microsoft-Windows-TaskScheduler/Operational
levels: [critical, error, warning]
| Setting | Effect |
|---|---|
levels | critical, error, warning, information, verbose, audit_success, audit_failure; empty = all |
event_ids, exclude_event_ids | IDs (4625) or ranges (4624-4634) kept or left out |
providers, exclude_providers | sources (providers) kept or left out, case-insensitive, * and ? patterns allowed |
service | HYPHA service of the channel's lines (default: the source name) |
tags | tags added to the channel's lines |
Kept levels and IDs go into the XPath query sent to Windows: a domain controller's Security log is not read in full. A filtered-out event is not shipped, but the read position moves on: nothing is read again after a restart. Security events sit at level 0; filter them with audit_success and audit_failure, not information. A named channel (…/Operational, …/Admin) is read like any other; wevtutil el lists them.
Each line has the eventlog source and the attributes @eventlog.channel, .event_id, .provider, .record_id, .level (0 to 5), .level_name (the levels names above), .audit (success or failure), .task, .opcode, .keywords, the labels rendered by Windows in its own language (.level_text, .task_text, .opcode_text, .keywords_text), .user_sid, .pid, .computer and the event data (@eventlog.data.TargetUserName…). Search example: source:eventlog @eventlog.event_id:4625.
The windows.eventlog.events metric counts collected events per channel (channel) and status (status). Monitor templates (Windows family of the catalogue): critical Windows event, unexpected shutdown (6008), logon failures (4625, add the Security channel), locked-out account (4740), errors per log. Ready dashboard: "Journaux d'événements Windows".
Performance counters
The counters of your software, as you choose, in integrations.perf_counters: one set per piece of software, read together.
integrations:
perf_counters:
- name: sql-locks # instance_name tag
namespace: sqlserver # metric prefix (default: the name)
interval: 30s # 10 s to 1 h
timeout: 10s
max_series: 500
counters:
- path: '\SQLServer:Locks(_Total)\Lock Waits/sec'
metric: locks.waits # → sqlserver.locks.waits
unit: operation_per_second
- path: '\PhysicalDisk(*)\Avg. Disk sec/Read'
instance_tag: disk # one series per instance, tagged disk
exclude_instances: [_Total]
Paths use the English names of objects and counters, whatever the language of Windows (PowerShell shows them translated). The English list: (Get-ItemProperty 'HKLM:\SOFTWARE\Microsoft\Windows NT\CurrentVersion\Perflib\009').Counter. (*) reads every instance (one series each, filtered by instances and exclude_instances, glob patterns); a named instance ((_Total)) gives a single series. Without metric, the name comes from the counter: % Processor Time → processor_time_pct, Lock Waits/sec → lock_waits_per_s. "/sec" counters are already rates.
Safeguards: beyond max_series, further series are dropped; a reading that exceeds timeout is abandoned, and no other one starts until it returns. The perf_counters.can_collect check turns to warning when a counter does not exist on the host or series are dropped, to critical when nothing can be read. Reading metrics: perf_counters.series, .series.dropped, .duration.
WMI queries
For what counters do not tell (states, sizes, software-specific classes), integrations.wmi runs read-only WQL queries:
integrations:
wmi:
- name: os
class: Win32_OperatingSystem
properties: [FreePhysicalMemory, NumberOfProcesses]
- name: physical-disks
wmi_namespace: root\Microsoft\Windows\Storage # root\cimv2 by default
class: MSFT_PhysicalDisk
properties: [{property: HealthStatus, metric: health}]
tag_by: [FriendlyName] # text properties turned into tags
- name: stopped-services
class: Win32_Service
properties: [ExitCode]
filter: "StartMode = 'Auto' AND State = 'Stopped'"
tag_by: [Name]
interval: 5m
Each numeric property of each row becomes a <namespace>.<metric> series (default: the property name, FreePhysicalMemory → free_physical_memory), tagged by tag_by (FriendlyName → friendly_name). 64-bit integers, returned as text by WMI, are converted; a non-numeric property is reported (put it in tag_by). Interval from 15 s to 1 h (1 min by default), same timeout and max_series safeguards as counters. The wmi.can_query check turns to warning when the query returns no row, to critical when it fails. Metrics: wmi.rows, wmi.series, wmi.series.dropped, wmi.duration. For a rate, prefer the performance counter: a Win32_PerfFormattedData_* class read now and then gives unreliable averages.
perf_counters and wmi lists from several conf.d files add up. Monitor templates (Collecte générique family): unreadable performance counters, failing WMI query. Ready dashboard: "Compteurs de performance et WMI".
Active Directory
On a domain controller (NTDS service present) the agent turns itself on: integrations.active_directory: {enabled: auto} (false to turn it off).
- Every 30 s, counters: LDAP sessions, searches, writes and binds and their duration (
active_directory.ldap.*), directory reads, writes and searches (.ds.*), inbound and outbound replication and pending operations (.dra.*), ATQ queue, Kerberos and NTLM authentications (.kerberos.*,.ntlm.authentications), Netlogon semaphores, the NTLM authentication bottleneck (.netlogon.*). - Every 30 s,
active_directory.servicecheck of the role's services present: NTDS, Netlogon, Kdc, ADWS, DFSR, IsmServ, W32Time, LanmanServer, DNS (a disabled service is ignored);active_directory.can_connectcritical when NTDS is stopped or its counters unreadable. - Every 5 min, through PowerShell (ActiveDirectory module, served by ADWS): inbound replication partners,
active_directory.replicationcheck per partner and partition (warning on failure, critical with no success for 24 h or ever), age of the last success (.replication.age), FSMO roles held (.fsmo_roles, and in thecan_connectmessage),active_directory.sysvolcheck (SYSVOL and NETLOGON shares). - Directory Service and DFS Replication logs added to collection (
logs: falseto leave them out; an entry for the same channel inlogs.windows_eventlogwins, with its filters).
Monitor templates (Active Directory family): domain controller down, role service stopped, replication failing, SYSVOL or NETLOGON not shared, slow LDAP binds, Netlogon bottleneck (NTLM). Ready dashboard: "Active Directory". For logon failures and locked-out accounts, add the Security channel (see above).
Microsoft DNS
On a Windows DNS server (DNS service present), same principle: integrations.windows_dns.
integrations:
windows_dns:
exclude_zones: ["*.in-addr.arpa"] # zones neither checked nor measured (glob patterns)
test_zone: example.com # zone of the SOA question (default: the first primary zone)
address: 127.0.0.1:53 # the server's address if it does not listen on loopback
- Every 30 s: queries received (UDP, TCP, dropped), responses, recursive queries and their failures and timeouts, dynamic updates (
windows_dns.*); rejected updates, failed secure updates, successful and failed zone transfers, counted between two readings; cache and database memory. - Every 30 s, the agent asks the server a real SOA question:
windows_dns.can_connectcritical when the service is stopped, does not answer, answers with an error (REFUSED, SERVFAIL…) or without authority; response timewindows_dns.response_time. - Every 5 min, through PowerShell (DnsServer module):
windows_dns.zonecheck per zone (warning when paused or when a secondary zone's last transfer failed, critical when shut down or expired), serial number of primary zones (windows_dns.zone.serial), age of the last successful transfer of secondary zones. - DNS Server log added to collection. Record changes are in the
Microsoft-Windows-DNSServer/Auditchannel; add it tologs.windows_eventlogif you want to keep them (large with many dynamic updates).
Monitor templates (DNS Microsoft family): silent DNS server, failing zone, recursion failures. Ready dashboard: "DNS Microsoft".
WSUS
The probe is off by default: it only starts on an explicit decision, on the WSUS server itself (integrations.wsus.enabled: true). It reads the WSUS administration API (PowerShell, Microsoft.UpdateServices.Administration assembly, installed with the role console), read-only: nothing is approved, declined, synchronised or cleaned up.
integrations:
wsus:
enabled: true
stale_days: 30 # computers with no report for this many days
sync_max_age: 48h # last successful synchronisation older than this: warning
- Every 60 s: role services (
WsusService, IISW3SVC, internal databaseMSSQL$MICROSOFT##WID),wsus.servicecheck per service. - Every 5 min, every minute while a fault is seen: computers (
wsus.computers,.needing_updates,.with_errors,.stale), updates (wsus.updates.total,.approved,.declined,.expired,.not_approved,.critical_not_approved,.needing_files,.with_client_errors,.with_server_errors), synchronisation (wsus.sync.agesince the last successful one,wsus.sync.duration,wsus.sync.failed), SUSDB size (internal database) and free space of the content folder (wsus.database.size,wsus.content.free). - Checks:
wsus.can_connect(service stopped or API failing: CRIT),wsus.sync(last synchronisation failed: CRIT, with the WSUS message; none successful, older thansync_max_age, or cancelled: WARN; running: OK),wsus.app_pool(IISWsusPoolpool stopped: CRIT).
Monitor templates: WSUS injoignable, Synchronisation WSUS en échec, Service du rôle WSUS arrêté, Pool d'applications WsusPool arrêté, Ordinateurs en défaut dans WSUS, Place libre du contenu WSUS. Ready-made dashboard: WSUS.
Tested on a test WSUS server (Windows Server 2025 in French, WID internal database), seen in HYPHA: never synchronised (WARN), first synchronisation with Microsoft Update running (OK), WsusService and the WsusPool pool stopped (CRIT, then OK on return), synchronisation failure caused by a non-existent proxy (CRIT, with the WSUS message). Not yet seen on a real server: a successful synchronisation (the first category synchronisation took more than an hour on the test machine) and client computers reporting.
Compatibility
Windows Server 2016 and later, plus Windows 10 and 11 for logs, counters and WMI. Proven on Windows Server 2025 in French, sole domain controller of a domain with directory-integrated DNS: replication between several controllers and secondary zones are read but have not been proven on a real server yet.
Updated: 21 November 2026