Wrapping Up

Data durability

The code we’ve written so far will work great as long as the BEAM stays up. If we restart the BEAM for any reason, though, the ETS table we created will be gone along with all its data. To handle BEAM restarts, we need more durability.

The way to get that durability is similar to the way we recover data from ETS, just taken one step further. To recover the state when a process restarts, we store it outside the process or any processes linked to it. To recover the state after the BEAM restarts, we need to store the state outside the BEAM.

The code’s shape we currently have will work for this. All the functions are in place and working the way we want them to. We need to swap out ETS for another storage mechanism. This is where the number of available options grows enormously. There’s no way to cover them all in a single section of this chapter. We can look at some OTP options and a strategy that will apply broadly to a lot of other options.

OTP offers two options right off the bat. DETS is a disk-based version of ETS. DETS starts tables with :dets.open_file/2 instead of :ets.new/2. However, the query API is remarkably similar to ETS. This is the durable option with the least change to the existing code.

Mnesia is OTP’s distributed database management system. It’s considerably more complex to set up and use than ETS or DETS. While it is likely to be an overkill for the needs of a game like Islands, it might be a great fit for other applications we might be working on. Take a look at the documentation if you’re interested.

Stepping outside OTP, a single broad strategy will work with a large number of data stores. The idea is to convert the game state to JSON. Once the data is in JSON, the data storage world is open to us. We can store the game state as a JSON data type practically anywhere. With any of these strategies, the game state will be stored safely out of the BEAM in case the node goes down for any reason. Using the functions and callbacks we’ve already set up, we query for it and set it back in each process’s state.

Conclusion

What we’ve done in these lessons is pretty remarkable. We started with game processes that could crash and lose all their state at any time. We’ve ended up with games that will automatically recover from any crash and restore their state to the last good version.

Try that in any other language.

Along the way, we’ve explored linking processes and trapping exits. We’ve looked at ways to customize the way supervisors behave with restart strategies and child specifications. We’ve also seen how to start and stop supervised processes. As a bonus, we’ve got an introduction to ETS as well.

We’re headed into the last part of the course now. That’s where we’ll layer on a Phoenix interface and make Islands available on the web.