● Beta — KubeEra is in active development. Profiles are live; AI-perception measurement is rolling out.
KubeEra
C

CubeFS

GraduatedRuntime
Open source · CNCF · CNCF Graduated · Apache-2.0

CubeFS is a distributed storage system that gives you POSIX, HDFS, and S3-compatible access to the same underlying data through a unified metadata layer.

updated 2026-07-01

What it is

CubeFS is a distributed storage system that gives you POSIX, HDFS, and S3-compatible access to the same underlying data through a unified metadata layer. It separates metadata nodes, storage nodes (blobstore/replica subsystems), and client access, letting you run large-scale file and object workloads without picking one protocol upfront. Written in Go, it's a CNCF Graduated project, which means it's cleared the foundation's bar for governance, adoption breadth, and security process — not a small feat for a storage system.

Who builds it and why

CubeFS originated at JD.com (JD Cloud & AI) to solve internal large-scale storage problems — mixed workloads (databases, big data, ML) needing different access patterns against one storage pool. It was donated to CNCF and graduated, which shifted governance to a broader contributor base beyond the original vendor. 177 contributors on GitHub is a healthy number for infrastructure software at this layer — enough to indicate it's not a single-company side project, though you should check current maintainer affiliation distribution yourself if vendor lock-in risk matters to your org. Founding date isn't publicly documented in the data available, but the graduation status and contributor count point to multi-year, multi-org investment.

Production readiness signal

  • CNCF Graduated: highest maturity tier CNCF offers — implies documented adoption, security audits, and governance maturity.
  • 5,599 GitHub stars, 177 contributors: solid engagement for a specialized distributed storage system (not a developer-tool category where stars run into tens of thousands).
  • Last commit 2026-06-30: actively maintained, not abandoned.
  • Latest release v3.5.3: mature versioning line, not a 0.x project.
  • License: Apache-2.0 — standard permissive license, no copyleft concerns.

Absent from public data: SLA-backed case studies, specific production deployment scale numbers, or founding date. If you need proof points for a procurement review, go directly to CNCF's project page and any published adopter list rather than relying on GitHub metrics alone.

Who should use this

  • Teams running mixed storage workloads (big data pipelines needing HDFS semantics, apps needing POSIX, and object storage consumers) who want one storage backend instead of three.
  • Organizations at a scale where commercial storage licensing costs (NetApp, Dell EMC, etc.) are a real budget line item and self-managed infrastructure is already the norm.
  • Platform teams comfortable operating stateful, multi-component distributed systems (metadata nodes, chunk/blob servers, resource managers) — this is not a single-binary deploy.
  • Shops already running Go-based infrastructure who want fewer foreign toolchains to debug in an incident.

Who should NOT use this

  • Small teams or single-workload use cases: if you only need object storage, MinIO is simpler to operate. If you only need a shared filesystem for a handful of nodes, NFS or a managed cloud filesystem (EFS, Filestore) is far less operational overhead.
  • Teams without dedicated storage/SRE ops capacity: distributed storage systems fail in subtle, data-loss-adjacent ways. If you don't have people who can debug replication and metadata consistency issues at 3 AM, don't run this yourself.
  • Cloud-committed shops wanting zero ops: if you're all-in on one cloud provider and don't need portability, native managed storage (S3, GCS, Azure Blob) will cost less in engineering time even if it costs more in dollars.
  • Anyone needing a mature, large third-party ecosystem of managed service providers: CubeFS's commercial support ecosystem is thinner than Ceph's or MinIO's — verify vendor support availability in your region before betting production on it.

Alternatives

  • Ceph: older, broader ecosystem, more battle-tested at extreme scale, but notoriously heavier operational burden.
  • MinIO: simpler, S3-native only — good fit if you don't need POSIX/HDFS in the same system.
  • JuiceFS: similar multi-protocol ambition but built on top of external object stores rather than owning the storage layer end-to-end.

Pricing

Fully open source under Apache-2.0. No paid tier, no license fees. Your cost is entirely infrastructure (compute/disk) and the engineering time to deploy, tune, and operate it — budget for that accordingly, it is not trivial for a distributed storage system.

Frequently asked

What is CubeFS?+
CubeFS is a distributed storage system that gives you POSIX, HDFS, and S3-compatible access to the same underlying data through a unified metadata layer.
Who builds CubeFS?+
CubeFS originated at JD.com (JD Cloud & AI) to solve internal large-scale storage problems — mixed workloads (databases, big data, ML) needing different access patterns against one storage pool. It was donated to CNCF and graduated, which shifted governance to a broader contributor base beyond the original vendor.
Is CubeFS production ready?+
CNCF Graduated: highest maturity tier CNCF offers — implies documented adoption, security audits, and governance maturity. 5,599 GitHub stars, 177 contributors: solid engagement for a specialized distributed storage system (not a developer-tool category where stars run into tens of thousands).
Who should use CubeFS?+
Teams running mixed storage workloads (big data pipelines needing HDFS semantics, apps needing POSIX, and object storage consumers) who want one storage backend instead of three. Organizations at a scale where commercial storage licensing costs (NetApp, Dell EMC, etc.
Who should not use CubeFS?+
Small teams or single-workload use cases: if you only need object storage, MinIO is simpler to operate. If you only need a shared filesystem for a handful of nodes, NFS or a managed cloud filesystem (EFS, Filestore) is far less operational overhead.
What are the alternatives to CubeFS?+
Ceph: older, broader ecosystem, more battle-tested at extreme scale, but notoriously heavier operational burden. MinIO: simpler, S3-native only — good fit if you don't need POSIX/HDFS in the same system.
How much does CubeFS cost?+
Fully open source under Apache-2.0. No paid tier, no license fees. Your cost is entirely infrastructure (compute/disk) and the engineering time to deploy, tune, and operate it — budget for that accordingly, it is not trivial for a distributed storage system.