When vSAN Performance Service is enabled, all vSAN nodes in the same cluster are periodically sending latest metrics to the current vSAN master host. However under some circumstances, TCP RESET packets are being sent to other vSAN nodes using a wrong physical network uplink, which do still have the Source-IP from the vmkernel adapters used by vSAN. (For example: vSAN's vmk3 is only assigned to vmnic3. However, these packets from vmk3 are then unexpectedly leaving vmnic0 used solely for Management.)
These rogue IP TCP packets lead to Cisco ACI "learning" the vSAN-IPs on a wrong physical uplink and make unexpected routing decisions, and possibly sending vSAN data network traffic along a wrong path. This can affect the overall vSAN stability due to network connectivity issues between vSAN nodes across routed subnets.
There is currently no available resolution.
The issue has been identified and is addressed in a future release.
Workaround