Repository navigation
v2.5.4
HexDB 2.5.4 fixes a way a sliced lattice could lose documents. Upgrade every lattice that uses replication.slicing.
Fixed
- Data loss after a full sync cut short (slicing). A replica copying everything from the Overseer rewrites its data one tessellation at a time, dropping the slices it no longer keeps, and only recorded what it held at the end. If a restart or crash cut that copy short, the hex still claimed every slice it held before: other replicas counted it as a copy, and once elected Overseer it served what was left as everything, and the replicas copied it. Now a full sync first records that a copy is in progress and that it holds no slice. Such a hex is never counted as a copy, is never elected Overseer (not even alone), and copies everything again on its next start.
- Leftovers in collected slices. A new Overseer clears what it has in a slice before collecting it from a replica, so leftovers of an interrupted copy can't bring back deleted documents.
- Startup while collecting slices. An Overseer collecting slices no longer exits at startup because it can't write its security defaults yet; it sets them up once it can.
Notes
- The 2.5.1 release has no macOS disk image (its build ran out of space); 2.5.2 and later have one.
Upgrading
Restart the hexes one at a time, as usual: replicas first, the Overseer last. To check a sliced lattice's data, compare each tessellation's count on the Overseer with what its replicas hold (each hex's /status, metrics.tessellations).
Full Changelog: v2.5.3...v2.5.4