Problem

Currently when a fullnode fails, a light node is not aware of this. This means that it may no longer receive messages from the filter protocol and believe that it is just not receiving messages.

Possible Solutions

There are currently 2 scenarios that I see working, we may even need to make these both work together assuming that we think certain fullnodes could be byzantine.

1. A lightnode should ping a fullnode every epoch to ensure that it still stores the filter they had passed initially, this could be ensured by making the fullnode return the full set of filters from the sending subscriber.

2. Additionally we could spec peristence for fullnode filter protocols, this would ensure that fullnodes save the filters in a database meaning if they crash and restart they could read these filters from the database.

@jm-clius if you feel differently about this feel free to activate it again, otherwise seems unlikely we'll get to in Feb

Closing, seems like discussion is complete.