Debian Administration

If you want a simple solution for heartbeat 1, you can use mon: http://www.linux-ha.org/mon

--
http://www.virtualconcepts.nl/

ipvs (in kernel) + ldirector

pacemaker has documentation, just nothing really coherent. Clusterlabs has some good stuff and drbd.org also has a section about Pacemaker.

However, the init scripts in Debian (Ubuntu Server 9.10 at least) of corosync+pacemaker cause split brain upon shutdown/restart (https://bugs.launchpad.net/bugs/505981). This makes reliable operation impossible.

Just my $0.02, but I've been burnt multiple times setting up h-a systems using heartbeat mechanism and assuming that the system you're failing over still works. Of course you could schedule a h-a test monthly but most sysadmins fail at this. If you need h-a, skip the heartbeat and figure out how to load balance and remove as many single points of failure as possible.

I don't think you want to. Doing so removes notification if one server should fail. You always want to monitor both, so that if your backup goes south while the primary is still running, you're notified and can take appropriate action. Likewise, if the primary goes and fails over, you want to be notified so that you can get it back up and running ASAP so that you can either put it in as backup, or switch back to it.

after examination heartbeat1 (need mon to check the services and fail over),
after work with heartbeat2 (xml config file, designed to be edited with GUI, painfull and clumsy, or you can edit by hand (recommended), but heartbeat2 has to be stopped and don't forget to erase backups and md5sums of the config, otherwise it will notice, and worse, heartbeat2 will write to that file status info (like, many lines)),

i discovered keepalived. not top notch doc, but enought to do everything : easily monitor others services (not like heartbeat)), take virtual ips, simple config, simple hooks. works with ipvs, (one config for all : virtual ips, forwarding to services). in case keepalived is used on the box with services, it will check it self.

You could use CARP to have them to share a public IP in master/slave configuration, and something like Pound (apsis.ch) on both to be able to (round-robin) loadbalance between the hosts

http://3molo.blogspot.com