Shared storage and dedicated NIC

[SysAx]

Hi,

I have a 3 nodes cluster with Proxmox 4.2.

Each network card (eth0) of the hosts is in a dedicated VLAN used to sync the cluster. I also have a shared storage through NFS on a FreeNAS for the Promox backup on the same card and VLAN. VMs are on other cards and other VLANs.

I have a question about the requirements that I have read in the documentation.
"If shared storage is used, a dedicated NIC for the traffic is needed".

I don't use HA and would like to know if it's dangerous to keep this configuration? (A node has stopped to work in the GUI last week and I had to move VMs manually).

Should I use another card and another VLAN to connect to the NAS ?

Thanks,

Christian

 

[t.lamprecht]

Hi,

I have a 3 nodes cluster with Proxmox 4.2.

Each network card (eth0) of the hosts is in a dedicated VLAN used to sync the cluster. I also have a shared storage through NFS on a FreeNAS for the Promox backup on the same card and VLAN. VMs are on other cards and other VLANs.

I have a question about the requirements that I have read in the documentation.
"If shared storage is used, a dedicated NIC for the traffic is needed".

So you have the cluster network (corosync) on the same card as the Storage? That's not ideal, cluster traffic is more sensitive about delays and network congestion and storage traffic is often huge, so not the best combination.

I don't use HA and would like to know if it's dangerous to keep this configuration? (A node has stopped to work in the GUI last week and I had to move VMs manually).

Should I use another card and another VLAN to connect to the NAS ?

Thanks,

Christian

If possible then yes, would be better to do so, but not strictly necessary.

But I cannot say yet if that was the cause for your GUI which stopped working problem. If it does that again can you look if the node is still quorate (pvecm status) and whats happening in the logs?

A restart of pve-cluster/pveproxy may also resolve it.

If you have them available you could attach the log output from that time here, journalctl has parameters to limit its output to a time span, e.g.:

journalctl --since "2016-06-07 18:00" --until "2016-06-10"

see:
man journalctl
for more details if you do not know them already and are interested.

 

[SysAx]

I will put the NFS storage on the same card than the VMs (not with corosync) to avoid problems in the future.

I restarted pve-cluster/pveproxy, but unfortunately, the node is always showned in the Proxmox web console.

It seems that it's not completely deleted from the cluster and I can see it in the file /etc/pve/.members

{
"nodename": "ax-prox-r710",
"version": 6,
"cluster": { "name": "AX-PROX4", "version": 3, "nodes": 3, "quorate": 1 },
"nodelist": {
"ax-prox-2950": { "id": 2, "online": 0},
"ax-prox-r710": { "id": 1, "online": 1, "ip": "192.168.100.105"},
"ax-prox-r730": { "id": 3, "online": 1, "ip": "192.168.100.104"}
}
}

For now, it's not a major problem because I have to re-install the cluster with new IPs in another VLAN.

Thanks for your help, Thomas.

Best regards,
Christian

 

[t.lamprecht]

I restarted pve-cluster/pveproxy, but unfortunately, the node is always showned in the Proxmox web console.

It seems that it's not completely deleted from the cluster and I can see it in the file /etc/pve/.members

Ok now I'm a little confused. Did you delete the node and it still shows up? If yes use

pvecm delnode NODENAME

If it should work but shows up dead then please post the output from:

journalctl -b -u corosync -u pve-cluster.service

 

[SysAx]

pvecm delnode NODENAME worked. Everything seems to be fine now. Might've missed something the first time.

I'm almost sure I did this command....or perhaps I forgot with all the other problems.

Thanks again.

Christian