Goal
Today the relay is kept off the data path for connectivity, but a write still needs the origin to have a path to each peer. Epidemic/anti-entropy gossip lets a peer that received a changeset re-disseminate it to other group members — so data can reach a peer the origin can't directly reach, by hopping through a closer peer (LAN/direct) instead of the paid relay.
This is the lever that makes "closer-peer-first" true for data, not just connectivity, and is the main way to cut relay infrastructure cost across partial meshes.
Proposed approach (Phase 3 — after durable outbox + RBSR are proven)
A peer re-broadcasts received changesets to other group members, preferring peers reachable over LAN/direct. Bound fan-out amplification to avoid storms (reuse the Rule 2.8 anti-storm thinking). Idempotent CRDT apply makes duplicate receipt safe.
Files
wavesyncdb/src/engine/sync_handler.rs, wavesyncdb/src/engine/push_protocol.rs, engine fan-out.
Validation
On-device validation of fan-out amplification limits.
Ref: docs/research/sync-reliability.md §5 (gossip) + §6 P2.
Goal
Today the relay is kept off the data path for connectivity, but a write still needs the origin to have a path to each peer. Epidemic/anti-entropy gossip lets a peer that received a changeset re-disseminate it to other group members — so data can reach a peer the origin can't directly reach, by hopping through a closer peer (LAN/direct) instead of the paid relay.
This is the lever that makes "closer-peer-first" true for data, not just connectivity, and is the main way to cut relay infrastructure cost across partial meshes.
Proposed approach (Phase 3 — after durable outbox + RBSR are proven)
A peer re-broadcasts received changesets to other group members, preferring peers reachable over LAN/direct. Bound fan-out amplification to avoid storms (reuse the Rule 2.8 anti-storm thinking). Idempotent CRDT apply makes duplicate receipt safe.
Files
wavesyncdb/src/engine/sync_handler.rs,wavesyncdb/src/engine/push_protocol.rs, engine fan-out.Validation
On-device validation of fan-out amplification limits.
Ref:
docs/research/sync-reliability.md§5 (gossip) + §6 P2.