Auto host power on/off - how should this work?

Posted on

So I have three little microservers each with 16GB ram. I have an APC Switched rack PDU and the apc_snmp fencing agent defined for each host.

All running latest RHEV-H 4.1, subscribed and yum updated.

I have enabled a 'powersave' scheduling policy on my cluster with the following settings
HighUtilization 80
LowUtilization 10
EnableAutomaticHostPowerManagement true
CPUOverCommitDurationMinutes 2

So this is what I expect and want to happen.
1. vms migrate to one host and 2 under-utilised hosts shutdown (and fence stop)
2. When CPU utilisation of 1 host goes above 80% for 2 minutes, fence start +1 host and migrate vms to it
3. When utilisation drops down again goto 1.

What actually happens is
1. The hosts migrate OK and the 2 hosts 'soft' powerdown (tick)

When I apply load to get the utilisation up above 80% for 2 mins there appears to be no attempt to start another host.

When I fence start a second host - as soon as it is up it detects it as under-utilised and immediately shuts it down.

The only way I could get the cluster to 'attempt' to start up a second host was to set the 'HostsInReserve' to 2.
However this attempt failed to soft start (was hoping it would wake on lan). It then marks the host as 'Not Responding' but doesnt attempt to 'fence' start it.

After a while I power managed 'restart' the host from RHVM and once it is up it then migrated the machines to get the over utilised back under the threshold. However because I now have hosts in reserve it will not shut down the host once I get the load back down.

How can I get this to work as expected?
Am I misunderstanding something?