Asymmetric Scale out of Shared or Parent Child VirtualServices leads to traffic disruption.
Article ID: 376061
Updated On:
Products
Issue/Introduction
- Error: "Atleast one SE needed, cannot ScaleIn" occurs when attempting to scale in any Service Engine after the Virtual Service has been hosted on two different Service Engines.
- Traffic is being handled by the parent VS and sometimes by the child VS randomly.
Environment
NSX-T
Cause
In the event of a network disruption where Service Engines (SEs) get disconnected, the Controller attempts to scale out the Virtual Services (VS) to different Service Engines.
After the scale-out, the Controller initiates a scale-in for the older Service Engine. Before doing so, it checks if cloud programming is complete on all the SEs. However, if cloud programming is not completed for the newly added SE, it can prevent the scale-out from being successful.
If the old SE reconnects within a few seconds and the scale-in never occurs, it can result in a situation where the VIPs are asymmetrically distributed.
Resolution
Due to a race condition, when an SE disconnects and reconnects with the Controller some time later, the virtual service placed on that SE can end up in a state where it is placed on more SEs than the number of SEs requested configured in the SE Group. In the case of Parent-Child VS it can cause traffic issues if the child VS is disabled/enabled in this state, causing the Parent and Child VS to be placed on a different set of SEs. This can lead to the child FQDN being handled sometimes by the Parent VS and sometimes by the Child VS.
This bug is for both SNI VS and shared vip scenario.
Fix Version: 22.1.4, 22.1.6, 30.1.1, 30.2.1
Additional Information
Feedback
