Full Mesh

import { Aside } from ‘@astrojs/starlight/components’;

Full Mesh

Every node ships changes to all peers directly. No coordinator, no leader. Per-peer watermark ensures changes are deleted only after all peers ACK.

    A ←→ B
    ↕ ╳ ↕
    C ←→ D

When to Use

3-7 nodes, all accepting writes
No single point of failure
LAN or low-latency WAN

Setup

./hook-sync-mesh-go -id nodeA -db a.db -listen :9001 \
  -peer http://localhost:9002 \
  -peer http://localhost:9003 \
  -peer http://localhost:9004

const mgr = attach(db, {
  id: "nodeA",
  peers: [
    "http://localhost:9002",
    "http://localhost:9003",
    "http://localhost:9004",
  ],
  batchMs: 50,
}, ["items"]);

Per-Peer Watermark

Each peer has its own last_acked entry in the _peer_state table:

CREATE TABLE _peer_state (
    peer_url TEXT PRIMARY KEY,
    last_acked INTEGER DEFAULT 0
);

Ship only sends changes with change_id > peer's last_acked
Delete from _changes only when ALL peers have ACKed (min(last_acked))
Offline peers’ changes accumulate until they reconnect — no data loss, no dead-letter

Why Full Mesh Works

INSERT OR REPLACE with UUID PK is idempotent. If A ships to B and C, and B also ships to C, C receives the same change twice — safe, no duplicates. The syncing flag is not a problem because each node receives changes directly from the writer, not through intermediaries.

Scaling Limits

The bottleneck is not connection count, but messages per second each node must handle:

msgs/sec per node = (N - 1) × writes/sec per node

Each message ≈ 200 bytes JSON. Example with 1000 writes/sec per node:

Nodes	Msgs/sec per node	Bandwidth/node	Verdict
2	1,000	0.2 MB/s	Fine
5	4,000	0.8 MB/s	Fine
8	7,000	1.4 MB/s	Fine for LAN
15	14,000	2.8 MB/s	Pushing SQLite ingest
20	19,000	3.8 MB/s	Too much

Cutoff depends on write rate, not node count alone:

Write rate per node	Full mesh OK up to	Bottleneck
100 writes/sec	15-20 nodes	HTTP overhead negligible
1,000 writes/sec	8-12 nodes	SQLite ingest ~10K msg/sec
10,000 writes/sec	3-5 nodes	SQLite can’t keep up

Benchmark

4 nodes, each ships to all 3 peers concurrently. 5 runs × 50 req per node (200 total per run):

Runtime	QPS median	QPS min	QPS max	Integrity
Go	14,853	6,108	23,570	5/5 PASS
Node	14,414	3,165	18,487	5/5 PASS
Bun	13,175	1,872	18,887	5/5 PASS

Cross-runtime mesh (Go+Bun+Node+Go) also verified — all nodes converge.

Run with: bash bench-fullmesh.sh