Performing a "Reconfigure for VMware HA" operation on the primary ESXi node in an HA cluster triggers false "unexpected virtual machine failover" alerts to be generated

Products

• VMware vCenter Server

Issue/Introduction

• When performing a "Reconfigure for VMware HA" operation on the primary node in an HA cluster, an unexpected virtual machine failover alert is triggered for the virtual machines running on that primary node.
• The vCenter Server events tab displays messages similar to:
  
  vCenter Server is disconnected from a master HA agent running on host <primary hostname> in HA_DRS_Cluster in Datacenter - vSphere HA agent on <primary hostname> in cluster HA_DRS_Cluster in Datacenter is disabled

The vSphere HA availability state of the host <primary hostname> in cluster in HA_DRS_Cluster in Datacenter has changed to Uninitialized

The vSphere HA availability state of the host <secondary hostname> in cluster in HA_DRS_Cluster in Datacenter has changed to Election

vSphere HA unsuccessfully failed over <virtual machine> on <secondary hostname> in cluster HA_DRS_Cluster in Datacenter. vSphere HA will retry if the maximum number of attempts has not been exceeded. Reason: The operation is not allowed in the current state.

Environment

VMware vCenter Server

Cause

When the primary HA host is manually reconfigured for HA, it causes the remaining secondary hosts to enter an election to find a new primary host.

The newly elected primary host places the virtual machines running on the old primary host in an unknown power state and waits up to 10 seconds for notification that the virtual machines on the old primary host are powered on and running.

If the old primary host does not become secondary within that 10-second interval, the new primary host assumes that the virtual machines are down and attempts to restart them. This causes a false failover event to occur, and consequently, the failover task fails because the virtual machines were never powered off. The virtual machines remain unaffected in this scenario.

Resolution

This behavior is an expected cosmetic issue. The alerts can be ignored and cleared once the HA configuration has finished resetting.

To avoid generating alerts, increase the monitor period (but this is usually unnecessary).

Note: Starting with vCenter Server 7.0 Update 1, the property name for fdm.policy.unknownStateMonitorPeriod has changed to fdm.unknownStateMonitorPeriod.

 

When completed, the das.config can be prefixed to these properties, which can apply to all the hosts in the cluster.

1. In vCenter, right-click the cluster and select Edit Settings.

2. Click vSphere HA and then Advanced Options.

3. Add a new option (if not already present).

• Default Option is 10:

vCenter version	Option	Value
8.0 Update 2 onwards	das.config.fdm.policy.unknownStateMonitorPeriod	10
7.0 Update 1 to 8.0 Update1	das.config.fdm.unknownStateMonitorPeriod	10
Pre-7.0 Update 1	das.config.fdm.policy.unknownStateMonitorPeriod	10

• • For this issue, change the value from 10 to 30:

vCenter version	Option	Value
8.0 Update 2 onwards	das.config.fdm.policy.unknownStateMonitorPeriod	30
7.0 Update 1 to 8.0 Update1	das.config.fdm.unknownStateMonitorPeriod	30
Pre-7.0 Update 1	das.config.fdm.policy.unknownStateMonitorPeriod	30

 

4. Disable and re-enable the HA settings of the cluster.

Additional Information

プライマリノードで VMware HA 操作の再構築を実行すると、予期しない仮想マシンのフェイルオーバーが発生する

在主节点上执行重新配置 VMware HA 操作引发意外虚拟机故障切换

Impact/Risks:
Increasing the monitor period also increases the time to start virtual machine failovers by the same amount (in this case, by 20 seconds) when a primary node stops during an actual HA failure.