HA and Linux heartbeats

[ranjitcool]

I have always got a little confused with this.

I know HA is for hosts or nodes as we call it. If a host dies, HA detects that and starts all the vms that were on that host on another host.

What if a vm in a HA enabled cluster dies, for some odd reason. HA will restart that VM. Does HA detect this from heart beats between vms?

If Yes, then What about linux vms that don't have vm tools installed.

Please advice.

RJ

[Troy_Clavell]

VMM (virtual machine monitoring) is different from HA (Host monitoring).  You can enable VMM which uses the VMware guest tools as a heartbeat.  But if VMM is not enabled a failure of a guest will not be noticed or monitored within HA

for single vCPU guests, you may have better success using FT

...and if no VMware Guest tools, VMM won't work either.

[ranjitcool]

Thanks Troy,

VMM however is part of the HA package right?

Failure of vm = failure to recive heartbeat from vm tools which kicks in response from vmm/HA right?

Does the heartbeat in the vmm scenario also have a 1 sec interval with a 15 second response time?

Thanks

[Troy_Clavell]

VMM however is part of the HA package right?

yes, you need HA to enable VMM, but they are two different feature sets

Failure of vm = failure to recive heartbeat from vm tools which kicks in response from vmm/HA right?

yes

Does the heartbeat in the vmm scenario also have a 1 sec interval with a 15 second response time?

default is 30 seconds, but can be chaned to as high as 2 minutes

[depping]

VMM is part of the HA configuration but not part of the HA agent. VMM lives in "VPXD" and uses the VMware tools heartbeat!

<shameless self promotion> most of this info can be found on yellow-bricks.com and of course in my book</end>

Duncan (VCDX)

Available now on Amazon: vSphere 4.1 HA and DRS technical deepdive

[mcowger]

You can also adjust the sensitivty of the hearbeat detection.

[EdWilts]

We've had nothing but problems with VMM and Linux guests and I believe that there's a KB article outlining this but I can't find one right now for a current version of ESX (I did find the one for ESX 3.5).  If the guest is up and running but hearthbeat doesn't get delivered, vmm will reboot the guest causing you an outage where one isn't required.

We've turned off VMM cluster-wide.  We've never had an issue with Windows guests.

[DSTAVERT]

It is a matter of adjusting the sensitivity and timeout values. Once you get it right for your environment it works well.