JuiceFS
An open-source (Apache 2.0), high-performance distributed POSIX filesystem for the cloud: data in object storage, metadata in a separate database — mountable on thousands of nodes at once. Source: juicefs.com, 2026-06-15
What it is
JuiceFS (Juicedata, ~14K★, Go) decouples data and metadata: Source: github.com/juicedata/juicefs, 2026-06-15
- Data → object storage (S3, GCS, Azure Blob, MinIO, Ceph, …), split into chunks/blocks.
- Metadata → a pluggable engine: Redis, PostgreSQL, MySQL, TiKV, SQLite, or a proprietary distributed service (millions of req/s, ~hundreds-of-µs latency).
- Access → FUSE POSIX mount, S3 gateway, HDFS API, and a Kubernetes CSI driver. Supports flock/fcntl locks and distributed caching.
Designed for massive data: ML/AI training datasets, shared workspaces mountable across hosts/regions, and big-data tiering, with cross-cloud replication for multi-cloud resilience. Source: juicefs.com/docs/community/introduction, 2026-06-15
Relevance to Dedalus
Another comparison point for Dedalus FS: JuiceFS optimizes for multi-node concurrent mounts (distributed POSIX), the opposite end from Dedalus FS's single-writer, clone-first agent-workspace model. The shared object-store-plus-metadata-DB pattern mirrors Dedalus FS (chunks in a blob store + SQLite metadata) — JuiceFS just makes it multi-writer and distributed. See also ZeroFS (S3-as-primary, single-host). Source: compiled, 2026-06-15
Timeline
- 2026-06-15 | Added (Kevin) as a distributed-POSIX-on-object-storage reference; cross-linked to the Dedalus storage stack and ZeroFS. Source: User, 2026-06-15; juicefs.com