Skip to content

Adoption & Ecosystem

Who uses it

The repository's ADOPTERS.md lists the organizations below. Each entry comes from that file (source 1).

OrganisationUse caseSource
Alibaba GroupLarge-scale image and file distributionADOPTERS.md
Ant GroupLarge-scale image and file distributionADOPTERS.md
DiDiLarge-scale image and file distributionADOPTERS.md
KuaishouLarge-scale image and file distributionADOPTERS.md
BilibiliLarge-scale image and file distributionADOPTERS.md
JFrogImage and file distributionADOPTERS.md
DatadogImage distribution with lazy loadingADOPTERS.md
Google CloudClick-to-deploy / GKE Marketplace integrationADOPTERS.md
Volcano EngineVKE / Container Registry integrationADOPTERS.md
Baidu AI CloudCCE integrationADOPTERS.md
Alibaba CloudACK P2P accelerationADOPTERS.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.

AlternativeDiffers by
Uber KrakenBitTorrent-based; the tracker only mediates the connection graph while peers transfer data. Scales for large blobs, but development has been quiet since 2020.
SpegelStateless; 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.