v3.0.0

Summary of changes:

• In this release, we have made several significant updates to improve write performance and user experience.
  • We have revised our implementation of key-value store that translates key-value pairs in the request to block layer data written to device.
  • Specifically, we have replaced RocksDB with an internal implementation of a hash-based key-value translator and a hint-based logical block manager sensitive to optimal NVMe drive IO boundary to improve write performance.
• For impacted users, these updates should provide a smoother, more intuitive experience with a noticeable improvement to write operation.
  • We have deprecated compaction and the requirement for written data to be compacted before reading to achieve maximum read performance is no longer needed.
  • We have introduced a new format tool called dss-formatter to format devices; similar to existing mkfs-blobfs tool which has now been deprecated.

Limitations

• Persistent key listing is not yet implemented. This will result in MinIO failing to return object listing after a reboot.
• Target startup time is presently longer than expected - and will be addressed in a future release.

What's Changed

• Cluster UUID and disk capacity by @aaronlee121 in #17
• Delete was broken for large object >= 70MB by @somnathr in #14
• Memory pool support for Replication IO by @somnathr in #18
• Removing s3 benchmark from nkv-minio artifact by @nsarras in #16
• Instantaneous BW and IOPS reporting to Prometheus by @aaronlee121 in #19

New Contributors

• @aaronlee121 made their first contribution in #17
• @nsarras made their first contribution in #16

Full Changelog: v2.0.0...v3.0.0