ChainSafe
Apply peer action when there are gossip validation errors
Describe the bug
Hetzner-s2 shows a lot of rejected ContributionAndProof gossip messages but other nodes show nothing
The lodestar_gossip_validation_accept_total metric is the same for all nodes.
ls 2022-02-16 | grep "contribution" | wc -l
174
ls 2022-02-15 | grep "contribution" | wc -l
125
Expected behavior
- It should be consistent across nodes
all errors are the same: SYNC_COMMITTEE_ERROR_VALIDATOR_NOT_IN_SYNC_COMMITTEE
Feb-16 00:14:20.583 [NETWORK] ^[[34mdebug^[[39m: The invalid gossip contribution and proof was written to .../contributionAndProof_0x6bb15ca4bbe8160f97304dfe42521a0f3df6cd518ccefe091f3a6f9f929d784a_1644966860583.ssz
all of the errors are correct as the validator is really not in the sync committee at the specified slot.
This works as designed since some nodes could catch invalid gossip messages and not spread them to the network, other nodes may not see them.
This brings another concern: if a node keeps receiving invalid messages from another node, right now it only decreases gossip score of peer, should we also decrease score at lodestar side, and in the end disconnect the peer?
if a node keeps receiving invalid messages from another node, right now it only decreases gossip score of peer, should we also decrease score at lodestar side, and in the end disconnect the peer?
Maybe the logic to decide what peers to disconnect should consider gossip score too
if a node keeps receiving invalid messages from another node, right now it only decreases gossip score of peer, should we also decrease score at lodestar side, and in the end disconnect the peer?
Maybe the logic to decide what peers to disconnect should consider gossip score too
What lighthouse does is to also penalize peer at node side depending on the gossip error https://github.com/sigp/lighthouse/blob/c3a793fd73a3b11b130b82032904d39c952869e4/beacon_node/network/src/beacon_processor/worker/gossip_methods.rs#L1339
for example if a peer sends invalid signature attestations for 5 times, it'll get banned at node (lodestar) side due to LowToleranceError PeerAction
