VM connected to a VLAN segment intermittently starts choosing wrong uplink to send the traffic despite the named teaming policy configured

Products

• VMware NSX

Issue/Introduction

• A VMs connected to a VLAN segment intermittently starts choosing wrong uplink to send the traffic despite the named teaming policy configured.
  
  For an instance, the default teaming policy applicable to overlay traffic is configured for 4 uplinks in the below snippet, whereas the named teaming policy is configured for 2 uplinks each:
   

The uplink mappings are as below:

uplink-1: vmnic1
uplink-2: vmnic2
uplink-3: vmnic3
uplink-4: vmnic4

• When the issue occurs, the VMs connected to the VLAN segment with "VLAN" named teaming policy mapped, will start choosing uplink-1 (vmnic1) instead of uplink-3 (vmnic3) or uplink-4 (vmnic4), thus impacting or possibly blackholing the traffic.
• Same can be verified with nsxdp-cli vswitch instance list to confirm the uplink being chosen for sending the traffic out.

 

Environment

VMware NSX

Cause

This issue can occur due to the race condition between MP and CCP. When setting the vlan based logical switch teaming policy, the port level teaming policy name gets to DP before the teaming policy, so the default policy is used even after the expected teaming policy is pushed from MP later.

Resolution

This issue is resolved in VMware NSX 4.2

Workaround:

• Disconnecting and re-connecting the network adapter of the VM
• vMotion of the VM to another host

Additional Information

Impact/Risks:

Wrong teaming policy will be used, which might impact the network connectivity.