|
I'm trying to design a truly redundant architecture to make sure that
if power goes out in any node (be it LVS, SQL, webserver, or even the
switch) or hardware fails or anything - it will have a secondary
failover path built-in.
Here's some basic diagrams I sketched up in Visio.
http://mikehost.com/~mike/tmp/servers.htm
I'm wondering which of these things LVS could possibly manage. We've
talked before about MySQL, and obviously it can handle web. It can
also heartbeat/self-check and failover itself. I'm wondering if I can
apply that to these other things.
I suppose that MySQL could use LVS no problem as well. On fail of the
master, I'd have to do something to tell the slave it is now the
master, so that it understands it needs to replicate back to the other
server once it is up (unless I can setup a dual-master setup... that's
another thing to look at.)
But if I rely on LVS for too much, especially in my internal network,
that will increase my reliance on LVS and use a lot more resources to
basically be routing all external traffic through LVS to load balance
the web traffic, and then internally for MySQL, NFS? and the other
services I require.
Also does anyone see a problem with the same physical LVS machines
doing the external web traffic AND internal traffic? I have seen some
threads (haven't had the chance to mess around with an LVS setup yet
myself) mentioning that you cannot reach internal addresses or
something from the LVS machine itself (which I would think then
creates a problem with my thinking so far...)
Any thoughts, examples, help is appreciated. Again, just trying to
sanity check myself here.
Also, the internal network will all be gigabit ethernet.
Thanks,
mike
|
