Spread the store across nodes

After the OTP chapter, KV is a solid single-node service: supervised buckets, fast reads from ETS, expiring keys. But every key still lives on one machine. When that machine restarts, crashes or is replaced during a deploy, the data is gone and nobody can be served.

This chapter spreads the store across several nodes, using only what ships with Erlang/OTP:

1. Replicate writes. Every node runs a replica; replicas find each other through a :pg group; a write goes to all of them and reads are served locally. Lose a node and the others still have the data.

Coming next in this chapter, once the first part has had feedback: bringing a new node up to date, Mnesia's replicated tables, network partitions, and forming clusters automatically with libcluster.

Running these examples

Distributed examples need several connected nodes, which browsers can't provide (they can't open the network sockets nodes use), so they have no Run button. Every page still shows the real output from running it on multiple nodes when the site was built.

To run one yourself, download it and start it with a node name, as shown at the top of each page, for example:

elixir --name kv08@127.0.0.1 08-replicated-buckets.exs

The script starts the other nodes it needs with OTP's :peer module, so one command is enough.