Adoption & Ecosystem
Who uses it
The repository's ADOPTERS.md lists the organizations below. Each entry comes from that file (source 1).
| Organisation | Use case | Source |
|---|---|---|
| Alibaba Group | Large-scale image and file distribution | ADOPTERS.md |
| Ant Group | Large-scale image and file distribution | ADOPTERS.md |
| DiDi | Large-scale image and file distribution | ADOPTERS.md |
| Kuaishou | Large-scale image and file distribution | ADOPTERS.md |
| Bilibili | Large-scale image and file distribution | ADOPTERS.md |
| JFrog | Image and file distribution | ADOPTERS.md |
| Datadog | Image distribution with lazy loading | ADOPTERS.md |
| Google Cloud | Click-to-deploy / GKE Marketplace integration | ADOPTERS.md |
| Volcano Engine | VKE / Container Registry integration | ADOPTERS.md |
| Baidu AI Cloud | CCE integration | ADOPTERS.md |
| Alibaba Cloud | ACK P2P acceleration | ADOPTERS.md |
ADOPTERS.md lists more organizations, including miHoYo, Xiaomi, Qunar, Yahoo, Meituan, JD, NetEase, Huawei, Shopee, China Unicom, ZTE, iQIYI, and Lazada.
Adoption signals
At graduation on 2025-10-28 the project reported contributions from 130 companies and 271 people across roughly 26,000 commits (source 5). As observed on 2026-06-22 via the GitHub API, dragonflyoss/dragonfly had 3,212 stars, 406 forks, and 26 open issues, with about 105 non-anonymous contributors (source 1). CNCF Graduated status was confirmed on the CNCF project page (source 3).
Ecosystem
Dragonfly integrates as a containerd or Docker registry mirror and as a Harbor-fronted distribution path. It works with Nydus for lazy image loading, and it natively handles Hugging Face and ModelScope model sources (source 8). The project ships a Helm chart (dragonflyoss/helm-charts) and a console (dragonflyoss/console), and the data-plane client lives in dragonflyoss/client.
Alternatives
Dragonfly distributes arbitrary artifacts, files, images, and AI models, and optimizes parent selection with a central scheduler that builds a DAG per task. The cost is operating several components. The alternatives below trade some of that range for simpler operation.
| Alternative | Differs by |
|---|---|
| Uber Kraken | BitTorrent-based; the tracker only mediates the connection graph while peers transfer data. Scales for large blobs, but development has been quiet since 2020. |
| Spegel | Stateless; reuses containerd's existing cache and discovers peers with a Kademlia DHT over libp2p. Built into K3s and RKE2, but scoped to a cluster-local containerd mirror. |