Quick answer
Disable TCP offloading (Large Send Offload, TCP Checksum Offload) and RSS on the cluster network adapters, then reboot the affected nodes. If that doesn't stick, update the NIC driver to a version from late 2022 or newer.
What's actually happening here
This error appears in the System event log on Windows Server 2016, 2019, or 2022 nodes in a Failover Cluster. The cluster service tries to send a heartbeat or replication chunk to another node, but the TCP stack only transmits part of the buffer before the network card decides it's done. The kernel returns STATUS_PARTIAL_COPY, which the cluster code maps to this error code.
The root cause? Almost always the NIC driver or its offloading features. Modern network cards try to be clever — they offload checksum calculation, segmentation, and receive-side scaling to hardware. When a driver mishandles a partial offload request, the TCP connection gets left in a weird state. The cluster service sees the incomplete send, logs the error, and may evict the node.
This isn't a cluster configuration problem. Don't waste time tweaking quorum or witness settings. The network stack is failing at layer 2 or 3.
Fix steps
- Identify the cluster NICs — Open Failover Cluster Manager, go to Networks, find the network used for cluster communication (usually private, no default gateway). Note the adapter name.
- Open adapter properties — In Network Connections, right-click each cluster NIC, select Properties, then Configure.
- Disable TCP offloading features — On the Advanced tab, set these to Disabled:
- Large Send Offload (LSO) — both IPv4 and IPv6
- TCP Checksum Offload (IPv4 and IPv6)
- UDP Checksum Offload (IPv4 and IPv6)
- Also disable RSS — Receive Side Scaling. Set to Disabled.
- Apply and reboot the node — This is non-disruptive to the cluster as long as you move roles first. Or schedule a maintenance window.
- Repeat on all nodes — The error happens on the sending node, but the misconfiguration lives on every node.
Alternative fixes if step 3 doesn't help
Update the NIC driver. Download the latest driver directly from the chipset vendor (Intel, Broadcom, Mellanox). The version bundled with Windows Update is often a year behind. For Intel X710/XL710 adapters, anything before version 1.14.131.0 has a known bug where LSO with specific MTU sizes causes partial sends. Update to 1.14.131.0 or later.
Check for jumbo frames mismatch. If you have jumbo frames enabled on the switch but not on the node (or vice versa), the TCP stack can fragment in odd ways. Set all cluster NICs to 1500 MTU. Run netsh interface ipv4 set subinterface "Cluster" mtu=1500 store=persistent on each node.
Disable VMQ if you're on Hyper-V hosts. Virtual Machine Queue can cause similar partial-send symptoms on physical nodes hosting clustered VMs. Run Set-NetAdapterVmq -Name "ClusterNIC" -Enabled $false in PowerShell.
Prevention tip
Standardize your cluster network hardware. Don't mix Intel and Broadcom adapters for the same cluster network. Use the exact same driver version on all nodes. Before applying a NIC driver update, test it on one node with Get-ClusterLog -Destination C:\ClusterLogs, run for 24 hours, then check for this error. The cluster is brutally honest about network problems — a clean event log is your signal that the fix works.