Prevalence is the fastest possible and third simplest ACID persistence technique, combining the two simplest ones: http://en.wikipedia.org/wiki/System_Prevalence
"Simplicity is prerequisite for reliability." Dijkstra (1970)
- Get enough RAM to hold all your data.
- Model your business system as a pure (no I/O) event handling function.
- Guarantee persistence by applying all events to you system through Prevayler, like this:
(let [my-file (File. "my-file")
my-handler (fn [state event] [(str state "Event:" event " ") nil]) ; Any function returning a pair [new-state result]
initial-state ""] ; Any value
(with-open [p1 (prevayler! my-handler initial-state my-file)]
(assert (= @p1 initial-state))
(handle! p1 "A") ; Your events
(handle! p1 "B")
(assert (= @p1 "Event:A Event:B "))) ; Your system state
(with-open [p2 (prevayler! my-handler initial-state my-file)] ; Next time you run,
(assert (= @p2 "Event:A Event:B ")))) ; the state is recovered.Calling the prevayler! function without the file argument returns a transient prevayler the you can use for fast testing.
Prevayler-clj implements the system prevalence pattern: it keeps a snapshot of your business system state followed by a journal of events. On startup or crash recovery it reads the last state and reapplies all events since.
- RAM: Requires enough RAM to hold all the data in your system.
- Start-up time: Entire system is read into RAM.
If your persistence file is called "my-file", for example, prevayler-clj will write files like this for you:
A Java serialization object stream with the system state at the moment the file was created followed by events.
On startup, my-file is renamed to my-file.backup and a new my-file is created. This new my-file will only be consistent after the system state has been written to it so when my-file.backup exists, it takes precedence over my-file.
After a new consistent my-file is written, my-file.backup is renamed to this. You can keep these old versions or delete them.