This article will help to avoid some users logging to DPU in order to change vmknic with esxcli and find the configuration missing after failover.
When network offloads are enabled on the VDS and the NSX transport node is created on VDS, vmknics created on mirror/ops network stack will be pushed to DPU. The vmknics on mirror/ops network stack is not visible with esxcli on esx side but if users use the sshdpu command and login to DPU, the vmknics are visible with esxcli on DPU side. In a dual DPU environment, if user login to the active DPU and uses esxcli(esxcli network ip interface) on DPU to operate the vmknics, the configuration will take effect on the active DPU, but the configurations will be missing after the active DPU is down and the vmknics failover to standby DPU.
For example, if vmk3 is on an active DPU and connected to a mirror network stack with an MTU of 1500. Using "esxcli network ip interface set -m 1600 -i vmk3"
to change the MTU from 1500 to 1600 can take effect on active DPU. However, if the active DPU is down, vmk3 will automatically a failover to standby DPU. Because the source of truth of the vmk3 is on ESX's configstore, its MTU is still 1500 in configstore, so after DPU failover and vmk3 created on standby DPU, the MTU is back to 1500.
VMware vSphere ESXi 8.0 Update 3
This is because ESX is the source of truth of vmknics
, users are not expected to use the sshdpu
command to login to DPU and using esxcli
to operate the vmknics
.
Currently there is no resolution.
Users should not use esxcli
to operate vmknics
on DPU. Users should only operate vmknics
on DPU from VC UI.