Major users, including Ant Group, Alibaba, Datadog, DiDi, and Kuaishou, rely on Dragonfly for large-scale distribution of containers and AI models, with leaders noting reduced bandwidth needs and faster image and model delivery in production.

Looking ahead, Dragonfly’s roadmap includes accelerated AI model weight distribution with RDMA, optimized image layouts to cut data loading times, load-aware two-phase scheduling, automatic updates, fault recovery, and improved control of back-to-source traffic during traffic bursts.

Since joining CNCF, Dragonfly has seen community growth with code contributions rising more than 3,000% and participation from over 270 contributors across more than 130 companies.

In production, Dragonfly handles tens of millions of container launches per day, saving up to 90% of storage bandwidth and cutting image launch times from minutes to seconds, enabling scalable AI workloads and model distribution.

Dragonfly has graduated from the Cloud Native Computing Foundation after proving production readiness and broad adoption for container and AI workloads.

CNCF hosts key cloud-native components within the Linux Foundation and supports a large contributor and member ecosystem.

The graduation process involved enhanced governance, standardized maintainer lifecycles and contribution policies, community guidelines for subprojects, and technical due diligence by CNCF TOC sponsors, along with a third-party security audit.

Dragonfly’s journey began at Alibaba in 2017, entered CNCF as a Sandbox project in 2018, reached production readiness with Dragonfly 1.0 in 2019, saw the Nydus subproject open-sourced in 2020, entered incubation the same year, and released Dragonfly 2.0 in 2021.

Dragonfly is a cloud-native, open-source distribution system using peer-to-peer tech to accelerate delivery of container images, OCI artifacts, AI models, caches, logs, and dependencies, and runs on Kubernetes with integration to Prometheus, OpenTelemetry, gRPC, and Harbor preheat.