2 node cluster won't stay clustered

[fonefoo]

This is a new install. I'm using 2 Cisco UCS M4 blades (maybe 3 if all goes well).

I performed a default install via the ISO (latest version, 4.3 I believe). Everything goes well on both nodes.

Build cluster on first node, join 2nd node to cluster. Everything holds together still.

It appears that when I make any sort of change one of the cluster nodes shows offline, doing a pvecm status the quorum shows actively blocking. Changes I'm talking about are like trying to create a vm, or adding storage.

If I perform a 'service networking restart' on both nodes that cluster will resume for a time, but it seems like after a few minutes it'll disconnect again.

Note: If I run a node solo one of these blades I don't have any trouble, I can add iscsi storage, build vms, etc. I only begin to have trouble after clustering the nodes.

Any tips on things I should look for here?

 

[iptgeek]

2 node *cluster* is a very bad idea and will cause you all types of issues.

 

[fonefoo]

Is there a method proxmox offers to share storage and allow live migration in case of host failure using just two hosts?

 

[nibblerrick]

Just clustering two nodes shouldn't be a problem AFAIK, HA isn't good in that case. I just made a two node cluster also to have the management a little bit centralised.
One node won't stay online either, I read somewhere that setting the network interface in promiscousmode should help and since then I did that it worked. I used the ompingmethods from the wiki to find out which node hat the problem.

The storagequestion I have, too.
For just two nodes it could be the easiest to make NFS-shares on both nodes so it is possible to migrate and do backups. Of course it is not distributed storage.

For a shared nothing cluster with shared/distributed storage it would be interesting what is the recommended method? Sheepdog is just preview and DRBD9, too. CEPH I think is a bit overkill (for two nodes not usable, even for three nodes a bit overkill).
So what is a good solution for a two node not-HA or three node cluster?

 

[fonefoo]

Ok so it seems like the architecture is just a little different than what I'm used to or just familiar with.

I primarily manage windows 2012R2 hyper-v clusters. You cluster the hosts, and the storage is in use with one host but shared by all in the cluster.

The solution I'm looking for doesn't have to be a "cluster", I'm just looking to have the VMs migrate to another node should their hosting node fail.

btw, I tried the adding the promiscuous option to my bridge interfaces but it didn't seem to help. cluster eventually dies after a few minutes. (note omping host to host works, so it doesn't appear to be an mcast issue).

In short- if a two node cluster is not idea, what would be the ideal proxmox solution to allow live migration to another known host?
thanks.

 

[lilCDNnrg]

HA is only available for 3 or mode nodes as you need to have a way to break a tie vote if a node fails. I have a 2 node cluster running but it's more so that I can control 2 nodes under one web GUI. All the stores are on nfs shares and though I can migrate the vm between nodes it is not automatic.

 

[nibblerrick]

What Microsoft does with the HyperV-stuff isn't a bad thing and with the replicamechisms and so on they're doing many things right. Reminds me, that I wanted to play around with storage spaces direct, last time I did that was in TP3 or so, there wasn't everything running smooth.

Eventually the pve-zsync is another option for the two-way-thing.

But as said I am also interested which is the best method to distribute/mirror storage over two nodes the easiest way to make a manual failover.