Obviously if a single node has a disk failure or OS crashes, and you are connected to that node, the blockchain is down for you and everybody else connected to that node .... but most of the other nodes are still up and the system as a whole is up.
"Availability" means being able to access the blockchain/system/network and get the data you want, when you want it.
The way most blockchain systems find out which nodes are on the network is usually through a gossip protocol, so typically your app will start off with some "seed nodes", which it contacts, then gets information about other nodes on the network.
Once you have the IPs of say 100 nodes scattered around the world, on different networks, running different OSes, etc. the chances of losing access to all those nodes is virtually zero.
Edit: You seem to be thinking about blockchains in terms of how traditional cloud/web based systems work. That's the wrong way to think about it. Blockchains have fundamentally different architecture. They do not use load balancers.