Windows logs that stay on the host are lost the moment an attacker clears them, a disk is encrypted or a busy Security log rolls over. Windows Event Forwarding (WEF) is the built-in, agent-free way to copy selected events from every Windows client and server to central Windows Event Collector (WEC) servers, using WinRM and Kerberos that are already in the domain. It costs no licences, but it is easy to deploy badly: subscriptions that forward everything, collectors that silently drop sources, and Security logs that overwrite before forwarding. This guide covers architecture, subscription and XPath design, Group Policy, certificate-based forwarding for non-domain hosts, scaling, health monitoring and the identities WEF depends on. It is for Windows engineers and detection teams; which event IDs matter is covered in the Windows Event Log Guide, so this page focuses on getting them off the host reliably.
This is an independent guide from NHI Mgmt Group, based on Microsoft's public documentation as of November 2026. It is not sponsored or reviewed by Microsoft. Product features change, so confirm settings against Microsoft's current documentation.
Key takeaways
- Source-initiated push scales; collector-initiated pull does not. In source-initiated mode clients learn the collector from Group Policy and the subscription ACL lists machine accounts, so there is no list of hosts to maintain and no remote-read credential to manage.
- The local log is the only buffer. Microsoft states that if events are overwritten before they are forwarded, the collector gets no notification and no gap indicator, so size Security logs for the longest plausible outage.
- Collectors have real limits. Microsoft plans for about 3,000 events per second per commodity WEC server, and warns that more than 100,000 lifetime sources can make the collector registry unreadable.
- Know the Server 2016 and 2019 trap. When WinRM and the Event Collector service run in separate svchost processes, the default HTTP URL ACL blocks the collector and clients log event 105; Microsoft documents a
netsh httpfix. - WEC servers are Tier 0 when they collect domain controller logs. They hold the evidence of every privileged action, so protect them like the systems they watch.
How WEF works
Each Windows machine has a built-in forwarding client. It connects over WS-Management (WinRM) to a subscription manager on the collector, downloads the subscriptions it is allowed to use, runs their XPath queries against local event channels and sends matching events, keeping a bookmark so that it can resume after an outage. On the collector, the Windows Event Collector service (Wecsvc) writes received events into the ForwardedEvents log or a custom channel, from where a SIEM agent or connector picks them up.
| Choice | Source-initiated (push) | Collector-initiated (pull) |
|---|---|---|
| How clients are targeted | Group Policy points clients at the collector; the subscription ACL allows machine accounts or groups of machine accounts | The collector holds an explicit list of computers |
| Credentials | Machine Kerberos (or client certificates outside the domain) | An account that can read each client's logs, typically added to Event Log Readers on every client |
| Scale | Fleet-wide | Small, fixed sets of servers |
| Typical use | Baseline subscription for all devices | Rarely needed; sometimes for a few appliances or isolated servers |
A subscription is either push or pull, never both. Microsoft's own deployment model uses a baseline subscription for every device and a suspect (targeted) subscription for hosts under investigation, which collects noisier events such as network logons and is changed without touching the baseline. Set "read existing events" on the suspect subscription, or you only get events created after the host receives it.
Subscription and query design
Delivery options
| Option | Mode | Batching (Microsoft defaults) | Use for |
|---|---|---|---|
| Normal | Pull | 5 items, 15-minute batch timeout | Reliable bulk collection where latency is unimportant |
| Minimize Bandwidth | Push | 6-hour batch timeout and heartbeat | Low-value telemetry over constrained links |
| Minimize Latency | Push | 30-second batch timeout | Security events that drive alerts |
| Custom | Set with wecutil ss <name> /cm:Custom | /dmi (max items) and /dmlt (max latency in milliseconds) | Tuning a specific subscription |
- Content format: the default
RenderedTextadds the human-readable message and roughly doubles or triples event size. Switching to binary XML withwecutil ss <name> /cf:Eventscan more than double the volume one collector handles, provided your SIEM parser does not need the rendered message. - Destination channel: everything lands in ForwardedEvents by default. Large deployments create custom event channels (an instrumentation manifest compiled into a DLL on the collector, as the Palantir WEF project does) so each subscription writes to its own log with its own size and retention.
- Expiry: the subscription XML supports an
Expireselement. Leave it unset for baseline subscriptions; a past date silently stops collection.
Writing XPath queries
Subscriptions use a subset of XPath 1.0. Selectors must select event nodes (starting with * or Event), only the child and attribute axes are supported, and the only functions are band (bitwise AND) and timediff. Use a structured QueryList when a query has more than 20 subexpressions or reads several channels, and use Suppress to drop noise: an event matched by both a selector and a suppressor is excluded.
- Select by event ID per channel: one
Selectper channel with explicit IDs, for example*[System[(EventID=1102 or EventID=4719)]]on the Security channel, rather than everything at level 4. - Suppress known noise: for example share access to
IPC$in event 5140, or failed logon type 2 in event 4625, which Microsoft's baseline excludes. - Enable the channels first: many useful operational channels (CAPI2, DriverFrameworks-UserMode) are disabled by default. Microsoft's example enables them with
wevtutil sl <channel> /e:true, sets sizes with/ms:and, for CAPI2, grants Event Log Readers read access with a channel SDDL. A subscription cannot forward events that are never written. - Start from published lists: Microsoft's baseline and suspect appendices, the NSA Event Forwarding Guidance repository and the Palantir WEF repository provide maintained subscription XML. Review each against your audit policy rather than importing all of it. See the Sysmon Guide for adding Sysmon's channel.
Deploying WEF step by step
- Design collectors: plan at least two WEC servers per environment, with dedicated fast disks for the EVTX files (Microsoft identifies disk I/O as the bottleneck), and separate collectors for Tier 0 sources such as domain controllers.
- Prepare the collector: run
winrm qc -qandwecutil qc /q. On Windows Server 2016 and 2019, checknetsh http show urlacl; if thehttp://+:5985/wsman/andhttps://+:5986/wsman/entries list only the WinRM service SID, add the Wecsvc service SID as Microsoft's troubleshooting article describes. - Create subscriptions: write them as XML in version control and load them with
wecutil cs <file>.xml. Keep the defaultAllowedSourceDomainComputersSDDL (Domain Computers and Network Service) or restrict it to a specific group of machine accounts. - Configure clients by GPO: under Computer Configuration, Administrative Templates, Windows Components, Event Forwarding, enable Configure the server address, refresh interval, and issuer certificate authority of a target Subscription Manager with a value such as
Server=http://wec01.example.local:5985/wsman/SubscriptionManager/WEC,Refresh=120, one entry per collector. Make sure the WinRM service starts, and add NETWORK SERVICE to the local Event Log Readers group so the forwarder can read the Security log. - Set audit policy and log sizes in the same GPO set: forwarding only moves what is audited. Size the Security log on busy servers so that it holds more than your longest expected collector outage. See the Group Policy Security Guide.
- Pilot: apply the GPO to a pilot OU, then on a client confirm events 104 (connected to the subscription manager) and 100 (subscription created) in
Microsoft-Windows-Eventlog-ForwardingPlugin/Operational. On the collector, runwecutil gr <subscription>and check the runtime status shows the source as active. - Test end to end: generate a known event on a pilot host and time its arrival in the SIEM. Then block the collector for an hour and confirm the backlog arrives when it returns.
- Expand in rings: servers, then a share of workstations, then the rest, watching events per second on each collector.
- Rollback: unlink or disable the Subscription Manager GPO setting and the clients stop forwarding at the next refresh; disable a subscription with
wecutil ss <name> /e:false. Neither changes local logging.
Non-domain and DMZ hosts
Where Kerberos is not available, WEF uses HTTPS on port 5986 with mutual certificate authentication. Microsoft's procedure: a server authentication certificate on the collector, a client authentication certificate on each source (subject matching its FQDN), winrm set winrm/config/service/auth '@{Certificate="true"}', an HTTPS listener bound to the collector certificate's thumbprint, a WinRM certificate mapping for the issuing CA, read access for NETWORK SERVICE on the client certificate's private key, and a client policy string of the form Server=HTTPS://<collector FQDN>:5986/wsman/SubscriptionManager/WEC,Refresh=<seconds>,IssuerCA=<issuing CA thumbprint>. Both sides must be able to check certificate revocation. See the Machine Identity, PKI and Certificate Lifecycle Guide.
Decision guidance
- WEF or an agent: WEF is free, native and needs no software on clients, which suits domain-joined fleets and teams without endpoint agent coverage. A SIEM or EDR agent gives richer parsing, file and non-Windows sources, and vendor support. Many organisations use WEF to collect and one agent on each WEC server to ship to the SIEM.
- Small estates: one or two collectors, one baseline subscription with Minimize Latency for security events, and the ForwardedEvents log are enough.
- Large estates: plan collectors from Microsoft's 3,000 events per second figure, use binary XML content, custom channels per subscription, and manage subscriptions with
wecutil(Microsoft recommends it over Event Viewer above 50,000 lifetime sources). - Mobile and VPN clients: Microsoft states forwarding reconnects over VPN, RAS and DirectAccess and sends the backlog; laptops off the network for weeks need larger local logs or an internet-reachable HTTPS collector.
Scaling and resilience
| Constraint | Microsoft guidance | Design response |
|---|---|---|
| Throughput | About 3,000 events per second average per commodity collector | Measure events per second in the pilot and size collector count from it |
| Disk | EVTX write speed is the bottleneck | Dedicated, fast volumes for collector logs |
| Connections | Limited by available TCP ports on the collector | Spread large fleets across collectors by OU |
| Lifetime sources | Over 1,000: Event Viewer subscription node slows; over 50,000: use wecutil; over 100,000: registry may become unreadable | Rebuild or rotate collectors in very large or high-churn estates; avoid pointing VDI pools at long-lived collectors |
| Outages | Client log is the buffer; overwritten events are lost silently | Larger Security logs on servers, two collectors per client, alerting on missing heartbeats |
| Event size | RenderedText roughly doubles or triples size | Binary XML (/cf:Events) where the SIEM supports it |
Where identity fits
WEF depends on several non-human identities, and each one is either a control or an attack path.
- Machine accounts: in source-initiated mode the subscription ACL decides which computer accounts may forward. Restricting it to a group of known machines prevents rogue domain-joined hosts from flooding a collector or injecting noise.
- NETWORK SERVICE in Event Log Readers: this is what lets the forwarder read the Security log. It is a deliberate, scoped grant; do not replace it with broader rights.
- Collector-initiated credentials: a pull subscription needs an account in Event Log Readers on every client. That is a standing service account with fleet-wide read access to security logs, so prefer push. If pull is unavoidable, use a group managed service account and alert on its interactive use. See the Group Managed Service Accounts (gMSA) Guide.
- Certificate mapping user: the HTTPS procedure creates a local administrator on the collector for the certificate mapping. Microsoft notes that WEF never uses it to impersonate the client and that it can be deleted; remove it unless another certificate-authenticated WinRM use needs it.
- Collector administrators: whoever administers a WEC server can delete or alter forwarded domain controller events. Put those collectors in Tier 0 and restrict admin rights accordingly. See the Tiered Administration and Enterprise Access Model Guide.
WinRM itself is also used for lateral movement (ATT&CK T1021.006 Windows Remote Management). Enabling the service on clients for WEF does not require opening inbound WinRM on workstations for push subscriptions, so keep host firewall rules for 5985 and 5986 restricted to what each role needs. See the Windows Remote Management and WinRM/PowerShell Remoting Security Guide.
Monitoring WEF health and measuring success
| Signal | Where | Meaning |
|---|---|---|
| Event 104 and 100 | Client: Microsoft-Windows-Eventlog-ForwardingPlugin/Operational | Client reached the subscription manager and created the subscription |
| Event 105 with error 2150859027 | Client forwarding operational log | The collector URL is not available, often the Server 2016 or 2019 URL ACL issue |
| Runtime status and last heartbeat | Collector: wecutil gr <subscription> | Sources that have stopped sending |
| Events 1102 and 104 | Forwarded Security and System logs | Security or System log cleared (T1685.005 Clear Windows Event Logs) |
| Event 4719 | Forwarded Security log | Audit policy changed, a common way to stop events being written (T1685.001 Disable or Modify Windows Event Log) |
| Collector service and disk | Collector System log, performance counters | Wecsvc stopped, disk full or queue growth |
Metrics
- Source coverage: percentage of active computer accounts that forwarded at least one event in the last 24 hours (target 100 percent for domain controllers and servers).
- Latency: 95th percentile time from event creation to SIEM arrival for the alerting subscription (for example, under five minutes).
- Collector headroom: peak events per second per collector against the 3,000 planning figure.
- Buffer margin: hours of history in the Security log of the busiest servers compared with the longest collector outage in the last quarter.
- Silent sources: number of hosts with no heartbeat for more than a day, each with a ticket.
Framework mapping
| Framework | Reference | What WEF provides |
|---|---|---|
| NIST SP 800-53 Rev 5 | AU-6, AU-9(2) Store on Separate Physical Systems or Components, AU-4, AU-12 | Central review, off-host copies, capacity planning, generation through audit policy |
| NIST SP 800-53 Rev 5 | SC-8 Transmission Confidentiality and Integrity | Kerberos or TLS-protected transport |
| NIST CSF 2.0 | PR.PS-04, DE.CM-09 | Log records made available for continuous monitoring |
| CIS Controls v8.1 | Control 8 Audit Log Management (central collection and retention) | Collection layer for Windows |
| ISO/IEC 27001:2022 Annex A | A.8.15 Logging, A.8.16 Monitoring activities | Protected, centralised logs |
| MITRE ATT&CK | T1685.005, T1685.001 | Evidence survives log clearing and audit tampering |
See the CIS Control 8 Audit Log Management Guide and the Log Management and Retention Guide.
Common failure patterns
- Forwarding everything: a subscription for all Security events at RenderedText overloads collectors and the SIEM licence, and teams cut back by dropping the useful events. Select by ID and suppress noise.
- Silent collector rejection on Server 2016 and 2019: the collector looks healthy but receives nothing because of the URL ACL; clients log event 105 that nobody reads.
- Security log not readable: NETWORK SERVICE was never added to Event Log Readers, so every channel except Security forwards and the gap goes unnoticed.
- Single collector: one WEC server is a single point of failure, and during its outage the busiest servers overwrite events before they can be sent.
- No health monitoring: sources stop forwarding after rebuilds, renames or GPO changes, and coverage decays quietly over months.
- Collectors outside Tier 0: domain controller logs on a collector managed by the server team give a Tier 1 compromise the ability to remove evidence of a Tier 0 attack.
Practitioner checklist
- Use source-initiated subscriptions with an ACL limited to known machine account groups.
- Deploy at least two collectors per environment and separate collectors for domain controllers, managed as Tier 0.
- Check and fix the WinRM URL ACL on Windows Server 2016 and 2019 collectors.
- Configure the target Subscription Manager setting, the WinRM service and NETWORK SERVICE in Event Log Readers by GPO.
- Write subscriptions as XML in version control with explicit event IDs and suppressions, loaded with
wecutil cs. - Use Minimize Latency or custom batching for alerting events and binary XML where the SIEM supports it.
- Size Security logs on busy servers to cover the longest expected collector outage.
- Monitor events 104, 100 and 105 on clients, runtime status on collectors, and alert on 1102, 104 and 4719.
- Use HTTPS with certificate mapping for non-domain hosts and delete the mapping user if unused.
- Report source coverage, latency and collector headroom monthly.
Standards and references
- Microsoft Learn: Use Windows Event Forwarding to help with intrusion detection
- Microsoft Learn: Setting up a source initiated subscription
- Microsoft Learn: Windows Server event collector doesn't forward events
- Microsoft Learn: Consuming events (XPath query support)
- NSA Cybersecurity: Event Forwarding Guidance
- Palantir: Windows Event Forwarding Guidance
- NIST: SP 800-53 Rev. 5
- MITRE ATT&CK: Clear Windows Event Logs
Related NHI Mgmt Group resources: Windows Event Log Guide · Active Directory Security Monitoring and Event IDs Guide · SIEM Guide · Windows Server Hardening Guide