Linux System Administration

Load Balancing and High Availability

|

159

scheduler=rr

protocol=tcp

checktype=negotiate

If quiescent is yes, a faulty realserver gets a weight of 0 but remains in the LVS rout-

ing table; we’ve set it to

no, so dead servers will be removed from the pool. The

weight of a server reflects its capacity relative to the other servers. For a simple LB

scheme like ours, all live servers have a weight of

1 and dead ones have a weight of 0.

If

checktype is negotiate, the director will make an HTTP request to each of the

realservers for the URL request, and see if its contents contain the string value for

receive. If the value is check, only a quick TCP check will be done, and request and

receive will be ignored.

The system startup files in /etc for ldirectord should have already been created during

the installation. Ultra Monkey also installed Heartbeat, which we aren’t using yet, so

let’s disable it for now:

# update-rc.d heartbeat remove

update-rc.d: /etc/init.d/heartbeat exists during rc.d purge (use -f to force)

The load balancer monitors the health of the web servers by regularly requesting the

file we specified in ldirectord.cf (

request="director.html").

Since this server will be responding to web requests at the VIP address (70.253.158.42),

we’d better tell the server about it. Edit /etc/network/interfaces and add these lines to

create the alias device

eth0:0:

auto eth0:0

iface eth0:0 inet static

address ...