|
Hello Ben,
> Very well. Basically we need to know within a reasonable time frame,
> say 5 minutes at the outside, that the servers is down via
> email/pager/some other notification. Failover should obviously be
> performed during this time frame. Of course 5 minutes is a "LONG"
> time in the land of 9x5 guarantee's, but lets just say that is the
> Maximum allowed time.
This should be no problem unless the second server is down too. After
having reread the thread I'm not sure if I get your idea. Because if I
get it, I don't see any reason for using LVS and load balancing. Let's
first try to start from the same assumptions (I'm a little slow
sometimes, bear with me. I just couldn't figure things out reading the
sketch you provided since it was a little bit mangled):
o you have two RS with public IP addresses each
o both webservers are W2K
o the RS are geographically distributed or at least not on the same
physical segment as the outgoing director interface
o you don't want to load balance, you just want to reroute stuff once
a server is down
> What you are proposing I had already been thinking of trying.
> Although I was only going to write the script that would flush the
> table and create a redirect to the secondary server. I thought of
> using
But if I understand you correctly this would be a sole routing issue.
NAT and TUN are not going to work for you as you already found out and
LVS_DR can only work if RS are on the same physical segment. But you can
use policy routing or something similar. Write a script that does the
health checking and in case of a failure you switch the routing tables
by deleting the rule responsible for the first RS and point a new rule
to the routing table for the second RS.
> heartbeat to check the status of the real servers, and then should RS1
> fail, then it would fun an "alert"/routing script that would redirect
> requests to the remote FOS. Do you think this would work in the
> situation? Can you think of any other solutions that might be
> plausible?
The questions are:
o do I understand your setup?
o are the RS's geographically separated on public IP's
o is your director the main router for those public services?
HTH and never give up, there is always a solution (even if we have to
route packets through a cipe tunnel [works with w2k], set one service to
quiesced mode and fiddle with the routing to get a happy customer),
Roberto Nibali, ratz
|
