GraphSync Meeting Action Items

[momack2]

There was a meeting on 1/16/19 about the various GraphSync proposals including sharing state and motivation, and how to unblock work on this. Full notes here.

Quick meeting summary

Motivations of GraphSync

• Blockchains and large trees with similar characteristics want to quickly be able to specify graphs to sync between peers
• Need room in the protocol for overlapping requests which dedupe into a single response
  • Ability to ask for two graphs that are highly overlapping and have people start sending responses

Constraints of GraphSync proposal

• Ability to change selectors over time (be extensible) by adding new types
• Simple, stateless wire protocol
  • A request shouldn’t require knowledge of prior requests - possible to use outside info, but no requirement
• Need ways to deal with rebooting a connection that wasn't closed down properly
• Want a trustless protocol - hard to send the structure of the graph along with the data because very large
• Unions and multiple selectors at once are both not rooted at the same node - so we'll need to augment Proposal B with a way to send a list of selectors (not based on the same root)

Timeline / Next steps

• Goal: MVP of Proposal B implemented and solidified by April
• Plan: break apart selectors work from wire protocol and prototype/implement in parallel
  • IPLD will handle MVP Selectors implementation (starting with cid, path, glob, + a recursive one that supports blockchain use case)
  • IPFS will handle wire protocol implementation

Action items (AIs) from this meeting include:

• AI(@vmx): flesh out js selectors implementation to include full MVP list
• AI(@vmx, @whyrusleeping): define set of selector test vectors for both js and go to test against
• AI(tentatively @warpfork): pick up go-ipld selectors work matching same tests/MVP constraints
• AI(tentatively @hannahhoward): take on wire protocol implementation in go-ipfs
  • AI(@Stebalien, @whyrusleeping, @hannahhoward): Sync on the requirements of the go-ipfs wire protocol work and whether there is bandwidth/sufficient definition to take this on
• ?? AI(TBD): wire protocol implementation in js-ipfs

[mikeal]

Follow up: I've added/adjusted the quarterly OKR's for IPLD to reflect this work and its priority https://github.com/ipld/roadmap/pull/4