Gravity Sync

Gravity Sync keeps a set of machines holding the same project tree without putting anything in charge. Each node runs its own Postgres and its own HTTP and WebSocket endpoint, watches a directory, and replicates file metadata and change logs peer to peer with whichever other nodes it is told about. There is no coordinator to lose.

The interesting parts are the failure cases. A peer that disconnects accumulates a pending-change queue and drains it on reconnect. A peer running an older schema — missing a table or a column — degrades gracefully rather than refusing to speak, and migrations propagate between peers rather than needing a synchronised deploy.

Its file tracking is borrowed, and openly so — the schema describes it as hash-based sync like Conduit. That is a design lineage rather than a dependency: the two share an idea about identifying files by content, and no code at all.

It is a prototype, and we would rather call it that than dress it up — complete enough to run, documented and containerised, but a working model of the idea rather than a product. The tests are not written, and nothing in production depends on it.