Possible multicast problems with Proxmox cluster

[Mor H.]

Dear Proxmox community,

We've been having crazy problems for the past two days with our VPS node cluster.

What seems to happen is that at some point, all nodes show up with a red X through the management panel except for 1 node that has the green checkmark.

The nodes that have the red X can't be managed at all but its VPS continue to run and be accessible.

The node that has the green checkmark has all of its VPS down.

Looking at the logs, we've seen things like:

kernel: [ 8355.928291] nf_conntrack: table full, dropping packet

We've reviewed this section in detail:
https://pve.proxmox.com/wiki/Multicast_notes#Troubleshooting

Could these errors and the symptoms we're describing be caused by IGMP snooping being disabled?

Could anything like this be caused by IGMP snooping being enabled?

If that's the case, does it make sense that the cluster works fine for hours and then all of a sudden the above-mentioned symptoms happen and the cluster goes crazy?

We disabled IGMP snooping and are worried to turn it on now thinking it might be causing this issue.

Basically, we're not sure if IGMP snooping should be enabled or disabled in our environment and we also don't know how to ensure the querier is set up properly.

Thanks in advance.

 

[wolfgang]

Hi

The conntrack error can come for many reasons. DDOS attack wrong/too many Firewall rules or to many connections.
You can try to increase the nf_conntrack_max and reduce the timeout.