Document DB sync prototype scope

This commit is contained in:
Eric Wendland 2026-05-22 14:55:31 +02:00
commit 59c86aba6a
4 changed files with 40 additions and 14 deletions

View file

@ -6,11 +6,25 @@ Accepted.
## Decision
Database sync will use local SQLite databases with cr-sqlite. Changes from
`crsql_changes` will be exchanged over Iroh and applied through cr-sqlite, with
schema hash gating and resource authorization.
Database sync uses local SQLite databases with cr-sqlite-compatible
`crsql_changes` tables or views. Geth extracts typed change batches, exchanges
them over the protected Iroh control path, checks schema metadata before
application, and applies compatible batches through the local `crsql_changes`
interface.
The prototype does not add a second snapshot transport for DBs. CAS-backed DB
snapshots or change-batch blobs are deferred until the change path proves too
large for the control path or until initial peer catch-up needs snapshots.
## Consequences
The MVP stores DB resource metadata and leaves synchronization as explicit future
work.
The prototype has deterministic fixture coverage for extraction, application,
two-node Iroh exchange, authorization, schema mismatch detection, and live-sync
watermarks.
A real cr-sqlite extension integration test is intentionally not marked done in
this environment because there is no `sqlite3` CLI or cr-sqlite extension
artifact available to load. The blocker is environmental, not a protocol
decision. When the extension is available, add a two-node test that creates a
real cr-sqlite-enabled table, mutates one node, runs `geth db sync`, and proves
the second node applies the change through cr-sqlite.