Idea: out-of-process block relayer

[abourget]

An idea has been germinating here:

• A p2p block relayer, that sits at the p2p level, and interconnects many nodes.
• Acts as a node itself, so replies with a node_id, and propagates blocks and transactions like a nodeos node.
• Has a cache of blocks, could read/write the blocks.log
• Could check signatures on blocks, and relay if those sigs are from producers in the current schedule (or some loosely defined "validation" of blocks)
• Propagate transactions or not? Could be tricky if we can't pre-validate them.. Could check well-formedness of transactions before propagating.
• Could play with topology, reconnections and instrumentation without rebooting nodes.
• We could test new ways to propagate blocks faster, without affecting the code system.

MVP

• Point-to-point connections

Phase 2

• DHT-like p2p connectivity, with slightly more random propagation paths

[fanyang1988]

i am developing a light node for eosio in https://github.com/eosforce/eos-light-node, depend on https://github.com/fanyang1988/eos-p2p, although it may need some times to complate it.

[abourget]

hey @fanyang1988 thanks for pitching in! Happy to see you guys are using eos-go like that!

Does eos-p2p work well? What about eos-light-node?

I'm interested in having a node acting like a p2p node, and feeding blocks that it could grab from a cache, or some random object store. I saw your work on the BlockStorer.. that's very interesting. Are you using that anywhere?