[DO NOT MERGE] Use RLNC for block propagation #14813
Conversation
- Add hooks for the new topic "beacon_block_chunk" - Add protos for the new messages - Handle the sync package logic of signature validation - Create native types for the chunks to be handled in the sync package Still TODO: - Handle chunks for pending blocks. - Handle nodes with incoming chunks, avoid verifying signature twice - Decode the block and send it to the blockchain package
- Finish validation of each incoming chunk including signature verification when needed. Still TODO: - Handle chunks for pending blocks - Convert the decoded block to an actual block and send it to the blockchain package.
Still TODO - Handle chunks for pending blocks - Handle state transition logic - Handle rebroadcasting - Handle block production
Also make `ReadOnlyBeaconBlockChunk` actually read only and copy the return values.
Load the setup from a file and use it across all clients.
The implementation of Broadcast is blocked as it needs to be changed
| if err := s.cfg.p2p.Broadcast(ctx, msg); err != nil { | ||
| logrus.WithError(err).Error("could not broadcast chunk") | ||
| } |
There was a problem hiding this comment.
This is wrong, instead of broadcasting the node should send a different chunk to each one of its peers.
| @@ -362,6 +416,11 @@ func (vs *Server) blobSidecarsFromUnblindedBlock(block interfaces.SignedBeaconBl | |||
| return BuildBlobSidecars(block, rawBlobs, proofs) | |||
| } | |||
|
|
|||
| // broadcastReceiveChunkedBlock broadcasts a chunked block and handles its reception. | |||
| func (vs *Server) broadcastReceiveChunkedBlock(ctx context.Context, req *ethpb.ChunkedBeaconBlock, root [32]byte) error { | |||
| return nil | |||
There was a problem hiding this comment.
This needs to be implemented like mentioned on gossip: instead of calling broadcast for a single message, each peer needs to get a different chunk.
…g the state transition
| // We overwrite the signature in the block to keep it to be the original one in the database | ||
| // Unfortunately to do this without reflection we make extra copies of the full block. We could switch over the type instead. |
There was a problem hiding this comment.
This is unfortunate, depending on the cost here we may want to just overwrite directly on msg but not clear what's the best way
| @@ -92,8 +92,11 @@ func (s *Service) validateBlob(ctx context.Context, pid peer.ID, msg *pubsub.Mes | |||
| return pubsub.ValidationIgnore, err | |||
| } | |||
|
|
|||
| if err := vf.ValidProposerSignature(ctx); err != nil { | |||
| return pubsub.ValidationReject, err | |||
| // TODO: verify the proposer signature here by fetching it from the chunks cache, or putting the blob in a queue | |||
There was a problem hiding this comment.
We need to update the logic to put in a pending queue when no chunk has been seen, and if a chunk has been seen, we just need to check that the signature we keep in the node is the same as in the sidecar.
There was a problem hiding this comment.
Ultimately the right solution is to change the BeaconBlockHeader structure to be the one in the chunk, and also change the layout of the BeaconBlockChunkHeader to keep only the commitments roots and not the full commitments.
Adds a feature flag to allow for block propagation using random linear network coding.