For each peer, there should be an error channel that reports all peer errors. The channel should be consumed by an error handler that for each error assigns a weighted score (naughty points) to the peer. Points should decay over time exponentially and be added linearly (leaky bucket). Once a peer's error score surpasses a specific threshold, they are disconnected. The scores need to be preserved after diconnection and decay continue in case the peers attempts to reconnect. (e.g reconnect should not be a reset) These points accrue for both sync and gossip services.
sgerbino
commented
2 years ago
For each peer, there should be an error channel that reports all peer errors. The channel should be consumed by an error handler that for each error assigns a weighted score (naughty points) to the peer. Points should decay over time exponentially and be added linearly (leaky bucket). Once a peer's error score surpasses a specific threshold, they are disconnected. The scores need to be preserved after diconnection and decay continue in case the peers attempts to reconnect. (e.g reconnect should not be a reset) These points accrue for both sync and gossip services.