Skip to content

Adoption & Ecosystem

Who uses it

The repository keeps an ADOPTERS.md file with a production and testing split. The named adopters below all come from that file (S1) or from CNCF's graduation announcement (S3).

OrganisationUse caseSource
JD.comIn production since 2018: 3000+ workloads, 50+ PB, 5000+ servers; ads, search, AI trainingADOPTERS.md
OPPOBackend storage for a Kubernetes-based AI platformADOPTERS.md
NetEase (网易)Backend for Elasticsearch, 2+ PBADOPTERS.md
Meizu, BEIKE, LinkSure, Reconova, BIGO, VipshopListed under productionADOPTERS.md
Xiaomi, Shopee, CreditEase, TD TechListed in production or testing tablesADOPTERS.md

At graduation CNCF reported over 200 organizations using CubeFS and about 350 PB under management as of January 2025 (S3, S4).

Adoption signals

  • GitHub: 5,593 stars and 703 forks (gh API, 2026-06-22).
  • Contributors: the GitHub contributors API paginated to 127 (about 127 accounts, 2026-06-22). CNCF reports a separate count growing from 27 to 379 contributors across 42 companies (S3, S4).
  • Organizations: CNCF states adoption grew from roughly 10 to over 200 across the Sandbox-to-graduation period (S3).
  • Releases: latest is v3.5.3, dated 2025-12-23 (S10).

Ecosystem

  • CSI driver: cubefs/cubefs-csi, a separate sub-project repo (S1).
  • Helm chart: cubefs/cubefs-helm, a separate sub-project repo (S1).
  • Client surfaces: S3 SDKs through ObjectNode, Hadoop FileSystem, and POSIX FUSE.
  • Observability: Prometheus metrics through util/exporter.

Alternatives

CubeFS is compared directly with Ceph in its origin paper, which benchmarks metadata operations against CephFS and RADOS (S7). Other systems in the same space:

AlternativeDiffers by
Ceph (CephFS/RADOS)Unified object, block, and file with CRUSH placement; the SIGMOD baseline CubeFS measured against (S7)
MinIOS3 object storage only; no POSIX or HDFS file semantics
JuiceFSPOSIX file system that keeps metadata in an external database rather than in-memory B-Trees
AlluxioA caching and data-orchestration layer over existing stores, not a primary store
HDFSThe HDFS-compatible API CubeFS offers is meant to let it stand in for HDFS

The distinctions that matter: CubeFS serves file and object in one system and lets a volume choose multi-replica or erasure coding; it keeps metadata in memory and places it by memory usage, which it argues avoids rebalancing on capacity growth (S7); and the Master stays off the data path.