|
> Also, could someone explain how Piranha fits into this? On RH
> site, it says
> Piranha only supports NAT (is this outdated text)? Since it's
> based on LVS,
> why can't it support DR or TUN?
Piranha is Redhat, NOT LVS. My understanding is that Redhat created a rift
a while ago by sucking code and doing "their thing", aka branching. Please
see the pirahna mailing list for help regarding pirahna. (Hmm I suppose if I
were a company I'd want to have control of an important project like HA,
too?)
> > Cluster1
> > www1
> > www2
> > www3
> > /Cluster1
> >
> > Cluster2
> > www2
> > www4
> > /Cluster2
Not only possible, but trivial. I use a perl add-on for LVS called
ldirectord. To init config changes, I simply killall -HUP ldirectord.
Straight LVS simply requires the appropriate ipvsadm command.
> > adding/removing servers? Scheduling changes?
All this is possible in ipvsadm, although I'm only fairly sure on schedule
changes. Anyone?
I think pirahna is the same, but you'll have to ask Redhat..
If you ask me, I think you should try a straight LVS method. It is the
better long term option - you aren't reliant on a company _AND_ you get
quicker updates / direct connection to the developers). Remember that
you're using UNIX, not windows. Do you want flexibility?
> > balancer directors where "if one fails, the other
> transparently takes
> over"
> > not supported at this time? If it is, how is this handled?
There's a new section in the howto that describes how to use
lvs+heartbeat+ldirectord+mon to fasciliate this process. It was submitted
by me and isn't as thorough as it could be. It does however have excellent
reference links and should speed you on your way. At the very least you'd
avoid a few of the pitfalls I ran into.
There's 3 valid ways of doing HA in LVS.
1.) Redhat HA-pulse thing
2.) Heartbeat + LVS (+ldirectord recommended) + mon. This is what I use.
I'm glad I chose this method - not only is it stable, but I can apply the
heartbeat stuff to other projects like any linux-ha I want.
3.) LVS + Keepalived/VRRP (variant on Cisco's redundancy protocol). I don't
know much about it, but check the archives and look for Alexandre Cassen.
Alexandre seems to be pretty close to the LVS team, so I think it is worth
looking into. Note of caution : someone said the code was a little beta,
but I'm not certain...
No matter what you do, I recommend using mon. Mon allows you to setup
custom pages, failover scripts when xyz occurs, etc. It is very
customizable - exactly what you would expect from unix.
|
