Distributed supervisors test
Background
We need a worker which performs a task periodically.
What we want
If we distribute the application in multiple nodes, we want the worker process to be running just once. If the process dies, we want the current node to start it again. If the node running the process dies, we want a different node to take care of restarting the worker again.
Alternative solutions
- https://github.com/bigardone/elixir-distributed-supervisors-test/tree/princemaple-version by princemaple
Original solution
We have 3 different components:
- The
DistributedTest.Serverworker, which is registered globally. - The
DistributedTest.ServerSupervisorwhich tries to start the previous worker. - The
DistributedTest.NodeMonitorwhich checks for nodes dying.
The application starts the three of them.
When the DistributedTest.Server gets started, it will return the new pid or the existing one, if it was started on a different node, and it will begin doing its periodical task.
The DistributedTest.NodeMonitor, monitors the existing nodes, and in case one dies, it tries to start a new DistributedTest.Server process using the DistributedTest.ServerSupervisor.
Test
Open three different terminal windows and start one different node in each of them:
$ iex --name [email protected] --erl "-config sys.config" -S mix
$ iex --name [email protected] --erl "-config sys.config" -S mix
$ iex --name [email protected] --erl "-config sys.config" -S mix
Try to kill the DistributedTest.Server process and watch it restart again. Finish the node session where it is currently running, and watch how any of the other nodes starts the process again.
Feedback and suggestions are welcomed :)
