Create a listener connection supervisor so we can recover in case of database connection failure

[diogob]

For more information on what this feature was designed to address check the issue #71

Goal

The user should be able to configure a supervisor thread for the listener connecion. When a listener connection is not detected by the supervisor it will open a new listener connection and reconnect the multiplexer.

Configuration

We can use a variable such as PGWS_CHECK_LISTENER_INTERVAL where the user can configure a number of milliseconds between checks, and keep the current system behaviour by setting it to 0. The default configuration would be 0 so the feature is backwards compatible with the current behaviour.

The time configured is roughly the maximum amount of time where messages could be lost due to the lack of listener.

Implementation details

We will have to ensure that the producer thread for the Multiplexer is dead to avoid leaving any dangling resource. It seems that the supervisor and the producer termination behaviour are orthogonal but to avoid additional complexity the new setting will bypass the existing shutdown the whole server and just re-spawn the producer.

In order to cancel the old thread we need to call the supervisor where the thread id of the open producer is available. Also, using the same openProducer function might simplify the code. One way to achieve both things is to keep both the openProducer function and the producer threadId in the Multiplexer and have another function such as superviseMultiplexer :: Multiplexer -> IO () which would take a multiplexer and launch the supervisor. We could also bake the supervisor inside either newHasqlBroadcasterForChannel or newMultiplexer but those functions are complex enough as they are.

TODO:

• Write a good description here of how the new feature is supposed to behave
• Use that info to adjust the README so we have a clear target

[diogob]

Instead of using the value 0 to encode the disabled parameter I just used an optional parameter. So when PGWS_CHECK_LISTENER_INTERVAL is not set the server behaves as the previous version.

[diogob]

Good news @W1M0R , I have finally finished the feature 😄
I understand that you are using an external supervisor as a workaround, but the new feature should make the database connection recovery seamless by keeping all websocket connections intact.

I'll just let the code soaking for a couple of days but I should release a new version very soon. In case you compile from the branch and test it out please let me know the results.

[W1M0R]

Thanks@diogob!