Updating vCPU (from 4 to 8) on an SEG through an API on 22.1.x controller will enable GRO unless explicitly disabled in segroup

Products

• VMware

Issue/Introduction

Symptoms:
1.Enabling GRO on versions prior to 22.1.5-2p1 may result in issue such as SE crash and high CPU utilization on SEs:
- The reason for SE crash is due to a rare scenario where the connection is closed while the data packet is in the GRO queue.
2.Enabling GRO on versions prior to 22.1.2-2p7 may result in issues such as connection drops:
- Connections may be dropped at the SE when the TCP timestamp option is not present in TCP data.

Cause

Upgrading from a prior NSX ALB version (21.1.x) will carry forward the GRO configuration for a SE group, however:
- If an SE group is newly created in 22.1.1, GRO config is in auto-mode.
- If the SE group has SEs with >=8 vcpus, GRO will be enabled.

Resolution

· SE Crash issue is scheduled to be fixed in 22.1.5-2p1.
· Connections drop issue when the TCP timestamp option is not present in TCP data is fixed in 22.1.3 and above.

Workaround:
To disable GRO from SEG config knob, below is an example:

[admin:cntrl]: > configure serviceenginegroup Default-Group
[admin:cntrl]: serviceenginegroup> disable_gro
Overwriting the previously entered value for disable_gro
[admin:cntrl]: serviceenginegroup> save

 

As GRO is a SEG property, it’s not affected with a SE reboot (for an example if GRO is explicitly disabled on SEG with 8vCPU it will not get turned on automatically after a SE reboot) and existing flows are not impacted