Coordinates the global Roger partition state

Each Roger partition has a single place where global state is kept. Global state (and global coordination) is needed for the following things:

• Job cancellation; when cancelling a job, we store the job ID globally; when the cancelled job is started, we check the job id against this list of cancelled ids.

• Queue keys; some jobs dictate that they cannot be queued when there is already a job queued with an identical queue key; if so, the job fails to enqueue.

• Execution keys; jobs which have the same execution key cannot be executed concurrently and need to wait on one another.

• Pause states; it is globally stored which queues are currently paused.

The per-partition Global process stores all this information. It provides hooks to persist the information between partition / node restarts. By default, the global state is loaded from and written to the filesystem, but it is possible to override the persister, like this:

config :roger,
  persister: Your.PersisterModule

The persister module must implement the Roger.Partition.Global.StatePersister behaviour, which provides simple load and save functions.