[SOLVED] Elasticsearch cluster over vxlan

[Astaoth]

Hi,

I've encountered a weird behaviour when building an Elasticsearch cluster over a Proxmox cluster :
I have a 2 node cluster (with an additional corosync node running on a rpi for the HA), on which I run only LXC. On this, I have an intra-cluster vxlan, in order to have a common internal LAN for my LXC. I also have on each node a bridge for giving external access to my LXC if necessary.

Then, on this cluster, I have deployed an Elasticsearch cluster with :
- on each proxmox node, a master/data node
- an additionnal master node which can be migrated between each proxmox node, for having a kind of HA which is enough for my setup

When the ElasticSearch nodes listen on the external bridge interfaces, the cluster can work fine, each nodes are visible.
However, when this nodes listen on the vxlan interfaces, only the two which are on the same Proxmox node are visible, the one on the other promox node can't be connected to the ElasticSearch cluster. But with a telnet or nc command I can reach the 9200 and 9300 of each ElasticSearch nodes from every ElasticSearch LXC.

Do you have an idea of how to correct this strange behaviour over the vxlan ?

 

[spirit]

maybe is it mtu related ? if your physical network mtu is 1500, do you have reduce mtu to 1450 in your vm ?

 

[Astaoth]

Thank you, that was this. Changing the MTU to 1450 made it work.