JuiceFS vs Kubernetes

The v1.4 cycle is running in the open across three pre-releases — beta1 in May with 373 commits since v1.3, beta2 two weeks later, and rc1 in June. The recurring subject is the metadata layer: quota keys no longer create mass tombstones, quota lookups are batched to save a round trip, transactional key-value lookups collapse into a single transaction, and chunks commit in write order. Feature additions are narrow — custom tags in tier configuration, an upload-part stream API, and checkpoint support for multipart uploads in sync.

This is a release cycle about cost at scale rather than new capability. Every metadata change removes a round trip, a tombstone, or a transaction from paths that run constantly, which is where a filesystem backed by object storage and an external metadata engine actually gets expensive. The sync and upload work points the same direction: multipart and streaming paths make large-object transfers resumable instead of restarting them. Contributor counts stay high across releases, so the pace is sustained rather than a push by one maintainer.

With rc1 cut and the changes since beta2 already down to 50 commits, a v1.4.0 final is the likely next step rather than further feature work.