ZeroFS now provides smoother HA failover and recovery, including better handling when a standby temporarily becomes unavailable. Clients reconnect more reliably while preserving in-progress operations, sessions, and byte-range locks, with corresponding updates across the client libraries, CSI integration, Web UI, tests, and documentation.
Because this release updates the wire format, mixed-version rolling upgrades are not supported; upgrade servers, clients, Web UI, and HA peers together.
What's Changed
- Fix/ha replication correctness by @Barre in #537
- Define 9P mutation retry and retention windows by @Barre in #540
- Preserve 9P byte-range lock fragments for replay by @Barre in #541
- Preserve server-side 9P byte-range lock fragments by @Barre in #542
- Validate HA peer topology by @Barre in #543
- Use UUIDs for 9P operation IDs by @Barre in #544
- Define 9P mutation retry signals by @Barre in #545
- Revoke deposed HA writers before returning failures by @Barre in #546
- Map pre-apply HA rejections to 9P by @Barre in #547
- Implement retained mutation result ledger by @Barre in #548
- Enforce writer authority and replay mutations across HA failover by @Barre in #549
- Avoid redundant 9P payload copies by @Barre in #550
- Enable eager Tokio driver handoff by @Barre in #551
- Cache negative lookups by @Barre in #552
Full Changelog: v2.0.10...v2.1.0
Contributors
Barre