Cluster with two node on openvpn and with drbd

[siyanre]

Hello everyone!
I have two servers, with 2 nic on 1st and 2nd servers. Servers colacation on :
City 1 - server 1; ext.Ip
City 2- server 2; ext.Ip
Internet speed on colacation 20mb/20mb. Can i make cluster with it and HA. If i use openvpn i have problem with qourum. But if add node with ext.ip all ok!

How do I properly implement this scheme ?

 

[wolfgang]

Hi,

Cluster could be possible but really not recommended.
And for what propose?

Live mig will not work! Migration is very slow over WAM!

Why do you want to have a Cluster over WAM?

 

[siyanre]

Hmm... I have two fast servers
In different cities. City 1 and City 2 i need make drbd storage and if server 1 in down all user connect to other server. Right now maybe 2 server connect to vpn for 1 ext.ip.if 1 server down second server up and all user work.

 

[siyanre]

Provider can make speed 50/100Mb

 

[stacktrace]

Sorry, but forget it. We did something like that this one single time, because an customer wanted us to do so. The bandwith between servers was much better than your example, but we just got from one trouble to antoher. In theory it works, practically it ist not useable. And even with 100 MBit you can forget DRBD. Dependent on the ping between the cities you will have much fun with STONITH. Also: A two-node DRBD-WAN-Cluster is the best precondition for a split-brain scenario.

HA means "High Availability". This is not just a function, but a concept. If you really need and want HA, you will have to invest in redundancy (Networks, Storage, Nodes, Fencing). If you want HA over WAN, you also have to invest in bandwith. Think in terms of GBit. One or two servers more dont make a big difference then.

 

[siyanre]

Ok thank you for answer. If i just make drbd storage without cluster and HA, its ok? If just server in other city down i just run vm in another city and thats all and user conect to another server

 

[stacktrace]

If you let one node just be the spare for the other and just use it, if the other node fails, it could probably work. But Replication is slow and nobody would give you a guarantee that this will work reliable. Also, if the first node fails, you'd have to think about some kind of fencing anyhow.

 

[siyanre]

Ok i get it, thank you very much. May be you know another some scheme ?