19 Releases 150 Tags

RSS Feed

• v0.6.18 c4c0b9abef

  Compare

  fix(raft): snapshots dumped the wrong user tables — fatal on Postgres, silent data gap on SQLite

  LastSkywalker released this 2026-06-12 13:29:09 +03:00 | 216 commits to main since this release

  managedTables listed 'users'/'refresh_tokens', but the entities map to
  'user_accounts'/'user_refresh_tokens'. SQLite swallowed the missing
  tables (isNoSuchTable → 'empty'), so every snapshot ever taken silently
  EXCLUDED accounts and sessions. Postgres aborts the transaction at the
  first failed SELECT instead — no snapshot could ever complete, the raft
  log grew without bound and the node burned CPU/disk retrying every ~80s
  (plus checkpoint storms from the orphaned aborted transactions).

  • correct table names; probe HasTable before each dump/DELETE/restore
    insert so a missing table can never poison the transaction on
    Postgres (and legacy snapshots restore cleanly);
  • bridge: the receiver now applies batches on a context detached from
    the HTTP request — a slow hub outliving the sender's timeout used to
    do all the raft-apply work and then fail MarkApplied with 'context
    canceled', so the same entries were re-shipped and re-applied
    forever; the sender's ship timeout grows 10s → 90s to match reality
    on fsync-bound disks.

  Downloads

  • Source Code (ZIP)
  • Source Code (TAR.GZ)