Windows Event Log Forwarding with WEF: Collector Node and Subscription Tuning

Windows Event Forwarding turns a set of independent Windows servers into a system with one audit trail. Instead of logging on to each host to read the Security log — and trusting that an attacker has not cleared it — each source machine pushes selected events to a central collector that the sources cannot write to or delete from. This is the native, agentless mechanism for that, and it is built into every supported Windows Server version, but the default configuration is almost never appropriate for a production estate: the default subscription pulls enormous volumes of noise, the default delivery mode can lose events under load, and the default collector sizing assumption is wildly optimistic.

This guide covers the collector node as a piece of infrastructure rather than a one-off configuration: how to size it, how to structure subscriptions so that one noisy source cannot starve the others, the subscription XML that expresses the filtering, and the GPO fields that actually matter. It assumes a domain environment, since WEF’s source-initiated model depends on Kerberos authentication. If you are planning the host for the collector, our Windows hosting comparison covers which configurations give you the disk throughput this role needs.

Collector vs Forwarder: The Two Halves

The forwarding side is the WinRM client configuration on each source: the service must be running, the listener must exist, and the client must be pointed at the collector with the right authentication scheme. The collecting side is the Event Collector service on the collector node, plus a subscription definition that specifies which events to request and which machines to request them from. Configuration is delivered to sources by Group Policy in almost every real deployment, because touching WinRM settings on hundreds of machines by hand is not a maintainable plan.

The direction of trust matters. In a source-initiated subscription, the source connects to the collector — which means the collector needs an inbound rule for WinRM over HTTPS on 5986 (or 5985 if you accept unencrypted transport inside a trusted network, which you should not). The source authenticates with its computer account, and the collector’s Windows Event Collector service authorizes it by group membership. That group membership is the access control for the entire audit pipeline.

Sizing the Collector Before You Point Sources at It

WEF writes forwarded events into ForwardedEvents, a single log with a configurable maximum size. The sizing question is the product of event rate and retention. A typical server produces a few hundred Security events per hour at moderate audit settings; a domain controller with detailed auditing produces tens of thousands. Multiply by the number of sources, decide how many days of retention you need on the collector before your SIEM or archive picks them up, and size the log accordingly — 8 GB to 16 GB is common for a few hundred servers.

Disk throughput matters more than capacity. Forwarded events arrive in bursts, and a collector whose volume cannot absorb the write burst drops events in ways that are not always visible in the log. Keep ForwardedEvents on a volume separate from the operating system, and monitor the log for the “events dropped” error state that the Event Collector surfaces in its own operational log. Retention also interacts with the source-side buffer: sources spool events locally when the collector is unreachable, and if the local log rolls over before delivery, the events are gone.

Subscription XML: The Minimum You Need

A subscription is an XML document with three meaningful sections: the delivery settings, the query (QueryList), and the source selection. The Delivery element controls the batch size and frequency and the important ContentFormat attribute, which should be RenderedText for readability and SIEM ingestion — Events is smaller on the wire but requires the consumer to resolve message resources from the source, which fails for offline sources.

A minimal, workable pattern for a Security-audit subscription looks like this: a <Query> with Path>Security</Path>, a <Select Path="Security"> filter using System/EventID on the event IDs you actually alert on — 4624, 4625, 4648, 4672, 4720, 4726, 4732, 1102 — and a <Suppress> clause to drop the highest-volume low-value events such as 4688 for expected administrative binaries. Set MaxEventsPerSecond explicitly rather than letting it default, and split subscriptions by purpose so that a logon-only subscription can be tuned without touching the process-creation stream.

GPO Fields That Actually Matter

Four Group Policy paths do the work. Under Computer Configuration > Policies > Administrative Templates > Windows Components > Event Forwarding, enable Configure target Subscription Manager and specify the collector server with Server=http://collector.domain:5985/wsman/SubscriptionManager/WEC,Refresh=60 using the HTTPS URL in production. Under Windows Components > Windows Remote Management, enable the WinRM service and set Allow Basic Authentication to Disabled while keeping Kerberos enabled.

The remaining two are on the collector side: AdmPwd-style group membership for the Event Log Readers group on each source, and the collector’s own authorization group for incoming sources. The refresh interval in the subscription manager URL controls how often sources re-read their subscription configuration; 60 minutes is normal for stable estates, and shorter intervals during initial rollout make the staging process much easier to debug. Also set the Security log size on sources through policy — a source log that rolls over in two hours cannot spool events during a collector outage.

Source-Side Prerequisites in Practice

Each source needs the WinRM service started and set to automatic, a listener configured for HTTPS, and outbound connectivity to the collector on 5986. The computer account must be a member of the collector’s authorization group, and the collector must be able to resolve the source’s fully qualified name — Kerberos does not tolerate short names or IP addresses in the subscription manager URI. A source that shows as “running” but never delivers anything is almost always failing at one of these three checks.

Test with wecutil rs on the source to list subscriptions and their status, and wecutil gr <SubscriptionName> to query runtime status including the number of events collected and the last error. If the status reports a WinRM error 0x80338029, the source cannot reach or authenticate to the collector; if it reports a subscription error, the name does not match what the collector publishes. These two codes distinguish the majority of configuration failures.

Backlog Behavior and What Happens at the Limit

WEF’s backlog is not unlimited and its semantics are subtle. Sources buffer events locally when delivery fails, but the buffer is bounded by the local log size and by the subscription’s MaxItems and MaxLatencyTime settings. When the backlog exceeds the limit, the oldest events are discarded — silently from the collector’s perspective, because they were never sent. This is the mechanism by which an audit pipeline fails without failing: the collector is healthy, the sources are healthy, and the gaps are invisible.

Monitor for it directly. Count events per source per hour and alert on drops below a rolling baseline rather than only on absolute zero. A source that normally sends 8,000 events per hour and sends 900 is telling you something, and it is the same signal an attacker generating cover traffic would produce. Pair that with alerting on event 1102 (audit log cleared) from any source, which is the one event you must never tune away.

Retention, Archive, and the Collector as a Target

The collector concentrates the audit trail, which makes it a target in its own right. Harden it like a domain controller: no interactive logons from general staff, no application roles, firewall rules that accept WinRM only from known source ranges, and an explicit rule blocking attempts to clear or modify ForwardedEvents. Back up the subscription definitions and the collector’s configuration alongside the log itself, because the configuration is what makes the log interpretable.

For retention beyond the collector’s log size, forward from the collector to a SIEM, a long-term archive, or a write-once object store. The architectural principle is that each hop reduces the ability of a compromised host to erase its own history, so the chain is collector, then archive, then offline or immutable copy. If your estate is still being built, the infrastructure guide on our main site covers separating the collector onto its own instance from day one, which is materially easier than extracting it from an application server later.

Leave a Comment