Question about clustering

[Alessandro 123]

Let's assume a simple 3 node cluster.
I'm referring, for semplicity, to resource utilization on each server as number of VMs running (in example, a server running 40 VMs, is used at 40%)

If server1, server2, and server3 are used at 30% each and a node (let's say server2) fails, all VMs are automatically moved on one of the remaining host ?

This movement, is automated ?
All VMs are moved on the same hosts or splitted ? In example, 15 VMs on node1, 15VMs on node3, ....
Any "warning" in PVE GUI saying that a node can't be migrated in case of failure, due to too many resource ? In other words, if I create some VMs on server2 and usage grow up from 30% to 51%, node2 can't be protected as other nodes doesn't have 51% of resource available.

 

[dcsapak]

This movement, is automated ?

only if you enable ha for those vms

All VMs are moved on the same hosts or splitted ? In example, 15 VMs on node1, 15VMs on node3, ....

the ha services will be one after another tried to restarted on one of the remaining hosts, AFAIK the crm tries to equally distribute the services

there are options on how often it tries to restart the services, and on how many nodes it tries to start them

this is described in detail in our reference documentation (just click the "? help" button when in datacenter -> ha) or https://pve.proxmox.com/wiki/High_Availability in the wiki

 

[Alessandro 123]

only if you enable ha for those vms

Well, that's obvious

the ha services will be one after another tried to restarted on one of the remaining hosts, AFAIK the crm tries to equally distribute the services

there are options on how often it tries to restart the services, and on how many nodes it tries to start them

this is described in detail in our reference documentation (just click the "? help" button when in datacenter -> ha) or https://pve.proxmox.com/wiki/High_Availability in the wiki

I've looked this weekend at HA page you linked here but i've not found anything useful for this question.

Anyway, can I create a new cluster with a production host with running VMs or is better to create an empty cluster from scratch ?
In other words, I have here a new host that will be (in following months) part of a new cluster when we add more servers.
Should I create an empty cluster now, or is safe to start some VMs on this standalone host and then create the cluster only when needed ?

 

[dcsapak]

I've looked this weekend at HA page you linked here but i've not found anything useful for this question.

details for the restart/node options are here: https://pve.proxmox.com/wiki/High_Availability#ha_manager_start_failure_policy

Anyway, can I create a new cluster with a production host with running VMs or is better to create an empty cluster from scratch ?
In other words, I have here a new host that will be (in following months) part of a new cluster when we add more servers.
Should I create an empty cluster now, or is safe to start some VMs on this standalone host and then create the cluster only when needed ?

you can always create the cluster later, but beware you can only add empty nodes to a cluster (because it takes the configuration from the existing cluster)

 

[Alessandro 123]

you can always create the cluster later, but beware you can only add empty nodes to a cluster (because it takes the configuration from the existing cluster)

The second node would be empty. What is full is the first node, that i'll use to create the cluster. Is this safe ?

 

[dcsapak]

The second node would be empty. What is full is the first node, that i'll use to create the cluster. Is this safe ?

yes

 

[Alessandro 123]

details for the restart/node options are here: https://pve.proxmox.com/wiki/High_Availability#ha_manager_start_failure_policy

Clear, but this doesn't tell if corosync tries to allocate failed VMs to the maximum number of nodes or only to one.

In example, if node3 is able to start all failed VM, corosync will start all failed VMs on node3 or still try to balance starting 50% on node1 and 50% on node3 and so on ?

In other words, usign as much nodes as possible is better than starting everything on a single node.