GenServer that handles network split recovery for Pow.Store.Backend.MnesiaCache.

This should be run on node(s) that has the Pow.Store.Backend.MnesiaCache GenServer running. It'll subscribe to the Mnesia system messages, and listen for :inconsistent_database system events. The first node to set the global lock will find the island with the oldest node and restore that nodes table into all the partitioned nodes.

If a table unrelated to Pow is also affected, an error will be logged and the network will stay partitioned. If you don't mind potential data loss for any of your tables in Mnesia, you can set flush_tables: :all to restore all the affected tables from the oldest node in the cluster.

For better control, you can use unsplit instead of this module.

Usage

To start the GenServer, add it to your application start/2 function:

defmodule MyApp.Application do
  use Application

  def start(_type, _args) do
    children = [
      MyApp.Repo,
      MyAppWeb.Endpoint,
      {Pow.Store.Backend.MnesiaCache, extra_db_nodes: {Node, :list, []}},
      Pow.Store.Backend.MnesiaCache.Unsplit
    ]

    opts = [strategy: :one_for_one, name: MyAppWeb.Supervisor]
    Supervisor.start_link(children, opts)
  end

  # ...
end

Auto initialize cluster

If nodes are lazily connected a race condition can occur in which the Pow.Store.Backend.MnesiaCache is running on each node without being connected in a Mnesia cluster.

To ensure that cluster will automatically initialize, Pow.Store.Backend.MnesiaCache.Unsplit will reset the most recent node's Mnesia schema when connecting to another node or a cluster. This will only occur if the Mnesia node has never been connected to the other node(s) and the other node currently runs the Mnesia cache GenServer.

The Pow.Store.Backend.MnesiaCache GenServer will be restarted, using the same :extra_db_nodes configuration as when it was first initialized. Therefor it's important that a MFA is used like {Node, :list, []} for the auto initialization to succeed.

Please be aware the reset of the Mnesia node will result in data loss for the node.

Strategy for multiple libraries using the Mnesia instance

It's strongly recommended to take into account any libraries that will be using Mnesia for storage before using this module.

A common example would be a job queue, where a potential solution to prevent data loss is to simply keep the job queue table on only one server instead of replicating it among all nodes. When a network partition occurs, it won't be part of the affected tables so this module can self-heal without the job queue table set in :flush_tables.

There may still be data loss if nodes are lazily connected. Please read the "Auto initialize cluster" section above.

Initialization options

• :flush_tables - list of tables that may be flushed and restored from the oldest node in the cluster. Defaults to false when only the MnesiaCache table will be flushed. Use :all if you want to flush all affected tables. Be aware that this may cause data loss.

• :auto_initialize_cluster - a boolean value to automatically initialize the Mnesia cluster by resetting the node Mnesia schema when new nodes are connected, defaults to true.