Questions about New HA Features in vsphere 5

[mhdganji]

Hi !

I have some questions related to new ha functions in vsphere 5

as I found, there is a master ha in the cluster which monitors other hosts and protected vms through two type of heartbeats (network and datastore) and move vms if necessary. if that host fails, an election is held and new ha master is determined by vcenter server.

ok ! what happens if we install vCenter Server as a virtual machine inside the hosts which is the master ha in a cluster. if this host fails (something like a power out) its agent is also going away and also the vcenter is destroyed. So, who is gonna hold the election and introduce the new ha while the old ha and vCenter are out of play.

I know that ha agents are not fully dependent on vCenter and runs if the host holding vcenter is away (vms including vcenter or moved to another host are restarted on another one) but i think this one is different because of new master role

my next question is about different kinds of events which can happen to slave hosts and making them partitioned or isolated. I will ask them later (after this question is answered

thanks to you all

[weinstein5]

If the HA agents are properly functioning they  are not dependent at all on Virual Center - the remianinf agents will nominate one of the remaining hosts as the primary.

[mhdganji]

Thanks

the remianing agents will nominate one of the remaining hosts as the primary.

you mean master election (which needs to know host object id in the cluster and number of connected datastores and ...) is done by agents themselves ? and it needs no intervention of vCenter ?

and another thing. when the host with vCenter machine on it and master role fails, how the other agents find that the master is away? is it not the job of vcenter to notice and tell the others when master is not functioning? what i doubt is that: if slave hosts loose connectivity with master, they think they are themselves faulty and should introduce themselves as isolated or partitioned and it is the duty of the vcenter to find out when the master is not functioning ! can you please correct me here

[mcowger · Accepted solution]

you mean master election (which needs to know host object id in the cluster and number of connected datastores and ...) is done by agents themselves ? and it needs no intervention of vCenter ?

Correct - no intervention of vCenter for a master election.

and another thing. when the host with vCenter machine on it and master role fails, how the other agents find that the master is away? is it not the job of vcenter to notice and tell the others when master is not functioning?

Its not vCenter's job - thats the job of the master and the hearbeating process amongst the slaves.

what i doubt is that: if slave hosts loose connectivity with master, they think they are themselves faulty and should introduce themselves as isolated or partitioned and it is the duty of the vcenter to find out when the master is not functioning ! can you please correct me here

See above - its not vCenter's job to determine which node is failed or isolated - thats the job of the FDM agents.

You should read Duncan's book: http://www.amazon.com/dp/B005C1SARM/ref=as_li_tf_til?tag=yellowbricks-20&camp=0&creative=0&linkCode=...

It covers all of this in depth.