Clustering EFA Servers

[Socal_Steve]

I'm looking to "Cluster" 3 EFA servers and round robin the dns so that i can spread the load of spam filtering and create some redundancy. One way i could choose a master server and cron copy all the relevant files eg: transport settings, outbound settings, users, ect. The other way i could make the servers use the same database and config files hosted in a central location like on the master server.

Anyone ever done this? Have any advice for me? Ideas? Smarthost better?

Thanks

[Widmo]

I think it's hard to do.

But it should look like:

Haproxy --- ( round robin to 3 or more EFA ) --- EFA Servers with the same database, and one NFS server on backend, with shared spool, and other files used by EFA, postfix...

Easiest way is to buy more powerfull server

[shawniverson]

I know that there has been a RPC issue in MailWatch preventing this. I think the bug is now fixed and will be included in 3.0.0.5.

[darky83]

I am working on a side project setup that clusters my setup using:
- centralized MYSQL server (Mysql clustered)
- loadbalancer (zen loadbalancer in cluster).
- centralized management (mailwatch + puppet)
- 2 or more mailscanner servers

This setup requires at least 7 (virtual) machine's to be configured and high available shared storage, but should be scallable to handle milions of mails.

But this won't be included in E.F.A. as its a completely different setup (maybe under a different name..)

For now if you want clustering I would suggest creating your own setup and don't use E.F.A.
E.F.A. is meant to be a easy to use single appliance for 'normal' traffic, if you have to handle millions of mails each day then you should use a enterprise ready system or build your own.

[drewmorris]

I am interested in that centralized mysql db "clustered" setup and I have some ideas that may simplify the approach. I am actually contracting out a similar project as we speak so it may be a good idea to get a discussion going offline.

If you are interested... email me directly at drewmorris @ gmail.com

[ressel]

Hello,
Any news on this side project with cluster?

[darky83]

Not yet,

Still a work in progress, got a setup running but creating it into a easy to install configured 'platform' is a bit harder

[datatek]

Hi darky83,

would you please outline how you got your cluster running?

[datatek]

BUMP?

[ZimboKraut]

As I am also very interested in a clustering solution,
here is another BUMP

[darky83]

Still work in progress (and low prio

making the system clusterable/failover and keeping all EFA items (the config etc) seems to be an bad idea.
Will have to start from scratch, currenly testing and working out an setup based on Centos 7

[Mcoulianos]

Any reason why you don't just use MX records and set the priority the same for both servers? This is a simple way to load balance through DNS. Either that or set one server to a lower priority for a failover setup. I currently use this method without issue.

It does add some work keeping the blacklist/whitelist entries the same between both servers but I haven't really found that to be too troublesome. I rarely touch EFA anymore now that i've added a few more RBL's to the list. Oddly there is only one RBL by default and it just isn't enough IMO.

[darky83]

Which one's are you also using ?

[ZimboKraut]

How are you syncing the lists between the servers?
I have four EFA servers running at the moment in various locations around the world.
The can communicate via VPN.
As they all have multiple Domains with various mx prios, it would be very helpful to sync them.
Due to the differnt locations, a centralized DB is not ideal and could cause potential delay/failure issues.
has anyone got any experience with this?

[drewmorris]

I setup a clustering-like solution using a policy service and some basic scripting. Basically every node is standalone and the policy service works using a central control source (which can also be clustered or distributed). In my case I have 2 policy checks... 1 at rcpt and 1 at end-of-message. The end-of-message response includes the filter telling the message what machine to route to for its final destination.

The configs are all pulled from a central location so I only need to change one thing and all servers change.

I did have to build my own shared service for quarantine and things but otherwise it works well for me.

[z3us]

I'm setting up two nodes connected via IPSEC vpn with SQL master-master replication so each change in SQL gets immediately replicated to the other node.
I've enabled RPC in the Mailwatch configuration, so it's possible to release mail from node 1 on quarantine of node 2.
I've setup e two mx-records (1 of each node) for incoming mail traffic.

I'm still looking into this setup, but I can't see any upcoming problems yet.

[wilbourne]

I think with pacemaker you can do this and supported postfix mysql apache

[shawniverson]

EFA Clustering/Distributed setup is being developed for v4

[Socal_Steve]

Awesome looking forward to it.