Shutdown Standby Node in Cluster, Any Impact?

[aychprox]

I am running 9 nodes cluster with 5 nodes for ceph and 4 nodes for VM.
Since the capacity is not fully utilized, and for the purpose of saving electricity, I shutdown 1 of the compute node and use as standby node.

May I know in term of HA and corosync, will it be any impact in long run especially the moment power up the standby node for the purpose of maintenance, etc?
Will it fence and bring down other nodes?

Hope to learn from the community.
Thanks.

 

[Alwin]

May I know in term of HA and corosync, will it be any impact in long run especially the moment power up the standby node for the purpose of maintenance, etc?
Will it fence and bring down other nodes?

It depends on your setup. If corosync is on a separated NIC port not sharing traffic and there are multiple links, then it should be no problem. Besides it will always show that corosync has an offline member.

If this node isn't needed on the cluster, why not just remove it completely? In a cluster all VM/CT can be migrated to other nodes and updates can be done a node at a time.

 

[aychprox]

thanks Alwin.

sorry, it is 10 nodes in total, 5 nodes for ceph and 5 nodes for VM.

corosync is on different vLan and 2 ring on 2 different switches.
reason behind to keep this node is just for emergency spike of other nodes, so can easily boot up this spare node and migrate over. This especially during public holiday. In normal day, we don't need to use this spare compute node at all.

here are the pvecm status

Votequorum information
----------------------
Expected votes: 10
Highest expected: 10
Total votes: 9
Quorum: 6
Flags: Quorate

 

[Alwin]

corosync is on different vLan and 2 ring on 2 different switches.

I can't stress enough, a VLAN is not a separate physical NIC port. Most important, to provide a stable quorum, the corosync traffic must not share its link.

reason behind to keep this node is just for emergency spike of other nodes, so can easily boot up this spare node and migrate over. This especially during public holiday. In normal day, we don't need to use this spare compute node at all.

This sound to me, like the cluster is operating already in the >80% capacity range. On a node failure that capacity becomes even smaller. I'd recommend against the "cold-standby" node.

Aside from the above, it should be doable. Though keep in mind to keep to node on the same update level as the remaining cluster. And that it doesn't participate in any HA groups.