Skip to content

v2.1.0-alpha.1: Distributed Correctness Hardening

Pre-release
Pre-release

Choose a tag to compare

View all tags
@MartinKalema MartinKalema released this 10 Jun 11:32
v2.1.0-alpha.1
a52a478
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

This release is a distributed-correctness hardening milestone for the horizontal-scaling fork.

It is intentionally marked as an alpha prerelease. The project has expanded from early read/write scaling prototypes into a broader clustered architecture with Raft-backed partition durability, 2PC recovery, replica replay safeguards, route authority, and explicit tracking of remaining distributed-database work. This release promotes that current state to main while keeping the production-readiness boundary clear.

What changed since v2.0.0

  • Added Raft-gated write acknowledgement work and fatal handling for Raft persistence failures.
  • Added durable 2PC recovery logs, prepared-intent queue separation, and placement-version guarding.
  • Added global table-number allocation in clustered mode so document IDs can remain portable across nodes.
  • Hardened NATS/JetStream replica delivery with supervised consumers, safer replay, idempotent apply, unmapped-table retry behavior, and retention-gap detection.
  • Added bounded remote-read frontier waits and remote-read frontier heartbeat cleanup.
  • Added route-authority classification, fail-closed clustered request surfaces, cluster-aware deploy/admin/action/import/export/log route migration, and sync-route authority rejection.
  • Added write-owner mutation routing so clients are abstracted from physical partition topology.
  • Added selective-delivery groundwork and documentation of its correctness boundaries.
  • Updated the issue journal, project goals, and placement/routing docs to reflect the current architecture and active roadmap.

Validation

  • cargo check -p local_backend

Active roadmap

Follow-up work remains open, including internal gRPC authentication, commit hot-path optimization, pooled forwarding/2PC clients, stronger Raft visible-apply semantics, full intra-partition state replication, background-worker authority/lease gating, stronger 2PC recovery, distributed index correctness, timestamp-domain unification, and production-grade benchmarking.

Treat this as: a meaningful clustered-database hardening milestone with an explicit alpha boundary and active correctness roadmap.

Assets 2
Loading