Live migration of a Virtual Machine (VM), which is part an in-guest VCS cluster where LLT is configured over Ethernet, from one Windows Server 2012 Hyper-V host to another may result in inconsistent HAD state.

Article: 100028636
Last Published: 2013-02-13
Ratings: 0 0
Product(s): InfoScale & Storage Foundation

Problem

Consider the following scenario:
- Two or more Hyper-V hosts are set up, between which live migration can be performed.
- Two or more virtual machines (for example, VM1 and VM2) are configured as nodes of a VCS cluster within the guest, on one of the hosts.
- In the VCS cluster, LLT is configured over Ethernet.

Perform live migration of one of the virtual machines (say VM1, which may be the active node).
After live migration, the HAD state is reported as follows:
- VM1 shows the HAD state as RUNNING for both the nodes.
- VM2 shows the HAD state as FAULTED for VM1, but RUNNING for VM2.

Events such as the following may be seen in the System Event Viewer for VM1 (one event for each LLT link):

Error Message

LLT ERROR V-14-1-10085 LLT protocol is unbinding from link adapterID

Solution

Veritas recommends that LLT to be configured over UDP as this prevents the issue.

Should the problem occur, perform the following steps to work around this issue:

1. Forcibly stop the VCS High Availability Daemon (HAD) service on the migrated node using the following command:
taskkill /f /im had.exe

2. If the HAD service starts again, stop it using the following command:
hastop -local
 
3. Verify that the HAD service is in the stopped state.
 
4. Run the following commands sequentially on the migrated node:
net stop vcscomm
net stop gab
net stop llt
 
5. On the migrated node, restart the VCS cluster using the following command:
hastart
 
6. Verify that all the cluster nodes report a consistent state for the HAD service using the following command:
hasys -state
 

Applies To

Microsoft Windows Server 2012 Hyper-V

References

UMI : V-14-1-10085 Etrack : 3056450 Etrack : 3053241

Was this content helpful?