login failed - pvecm expect

[matda59]

I run proxmox 7.1-10 on some mini micros M93Ps and HP elitebook USFF in my homelab. i have around 4 in a cluster but generally don't need all of them powered on at once, typically only 1 is powered on. What i've found recently is i get a login failed when i use the correct username and password to log into the node that is powered on. i have to SSH into that node and always run the command.

pvecm expect 1

or change the number to whatever number of nodes i have powered on. I guess its just a bit fustrating, and also i would think that it should not say login failed as that is not the issue.

also to note, running this command often fixes a number of other issues like migrating between hosts etc.

 

[sterzy]

Hi,

you need at least 3 nodes in a cluster actively working to be quorate i.e. to form a majority of 50+% on every decision. You can read more about that in the wiki [1].

As I see it you have essentially three options:

1. Keep enough nodes online to find a quorum. This means keeping at least 50% +1 nodes powered on. If you do not want to keep all these nodes running, you could swap out one with a QDevice [2].
2. Find a different solution that does not require clustering your nodes.
3. Keep using the pvecm command, although that is not recommended.

Hope I could help!

[1]: https://pve.proxmox.com/wiki/Cluster_Manager#_quorum
[2]: https://pve.proxmox.com/pve-docs/pve-admin-guide.html#_corosync_external_vote_support

 

[matda59]

Thanks for your response. I guess what I'm saying is:

1. That the login failed error is incorrect and misleading.
2. I'm logging directly into the web UI of the powered on node not another offline node.
3. The cluster knows which nodes are available based on some sort of health heart beat.

what would be good if there was some web UI that would let you change the pvem expect # without having to SSH into a node and changing that value.

 

[sterzy]

1. That the login failed error is incorrect and misleading.

The error message occurs because your cluster takes a vote on whether you are allowed to log in. Let's say your cluster consists of 4 nodes, but only one is online and can participate in the quorum. Since no majority of at least 3 nodes can be found that agree on you being allowed to log in, the log in fails.

2. I'm logging directly into the web UI of the powered on node not another offline node.

Yes, see my answer to your previous point.

3. The cluster knows which nodes are available based on some sort of health heart beat.

The command you posted tells the cluster how many nodes it should expect to be present. From the clusters perspective there is no difference in nodes not participating in a quorum (be that due to congestion, the node crashing or any other reason) and actually being powered off.

 

[matda59]

Thanks, Still i do not agree with a login failed error message.

 

[K1LLUM1N471]

Well, I understand that I need a minimum of 3 nodes for HA, but I classify this as a bug when I even can not login in a typical web UI.

I think a minimal set of functionality should still be available in the web UI. Moreover the cluster admin should have the ability to decide how to use the cluster even when it means to use it without HA functionality.

Kind regards,
K1LLUM1N471