> > This seem to be very strange and quite not acceptable since my primary
> > director machine may go down while a realserver is down;
>
> You cannot setup an LVS with machines that aren't working.
> you configure the LVS with working machines. After you've
> configured them, then you add a second layer to monitor the machines
> and handle the failures
>
> Joe
>
Hmm, this sounds a bit awkward. Today, I already have a MON alert script
that removes/adds hosts on the fly to/from the NAT table when needed using
the iptables command.
This means that my alert script must perform the following additional tasks
each time a realserver goes up or down:
1. Parse my LVS config file and remove/add the failing host's name from the
SERVICE variable.
2. Execute 'configure' on my LVS script.
3. Copy the resulting file 'rc.lvs_nat' to the '/etc/init.d/' directory in
order for it to be executed during restart of the OS.
Is there a "smoother" way to accomplish the above?
Btw, what's the reason for LVS to ping all hosts (defined in the SERVICE
variable) during execution of the 'rc.lvs_nat' script? Why does it hang upon
ping failures? Is it possible to use a time-out to avoid the hang issue?
Thanks in advance.
/Tommy
|