Re: Hardware failover webserver cluster

[Dan Ritter <dsr@randomstring.org>]

On Thu, Mar 27, 2014 at 04:47:15PM +0100, basti wrote:
> Hello,
> perhaps that's a bit off topic here but can someone explain what I need
> to build a hardware failover nginx cluster?
> 
> The unclear is:
> How does the client know that the server 1 is down and use the other one?
> DNS failover has a delay because of the design I think.
> 
> client ---> SRV1
>       |---> SRV2 (failover if SRV1 is down and load balancing)
> 
> how did it work on google or somethink else?

Here's what you need to do a failover cluster:

- two or more sets of hardware
- individual addresses for each machine
- a shared address that both will answer to when needed
- a system for sharing needed resources
- a system for choosing which hardware will answer the shared
  address


DNS has two methods for providing alternate resources: you can
use a RR (round robin), in which case one name will
be resolved into several IPs, and the client may or may not
choose appropriately; or you can use a short TTL and update the
A or CNAME record to match your desired server.

Neither of those methods is much good, and in practice they are
only used as the first stage of a more complicated load
balancing system.


Please read up on corosync, pacemaker and wackamole.

-dsr-