Dragonfly v2.5.0 is released

2026-06-30 · Source: Cloud Native Computing Foundation · Field: Technology & Digital — Cloud Computing & IT Infrastructure, Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Advanced, short

Summary

Dragonfly v2.5.0, released on June 30, 2026, introduces significant enhancements for P2P content distribution. Key updates include direct repository downloads from Hugging Face and ModelScope, enabling users to fetch models like "deepseek-ai/DeepSeek-OCR" with `dfget`. The release also features the Dragonfly Injector, a Kubernetes Mutating Admission Webhook for automatic P2P capability injection into Pods without image rebuilding. New control mechanisms include a Manager console blocklist for disabling specific downloads and comprehensive rate limiting across the control plane and client for improved stability. The `dfctl` command-line tool now manages local client tasks, while container registry proxy configuration is simplified. Client download efficiency, file transfer reliability, and HTTP handling security are also improved. Additionally, the Nydus component gains features like prefetch-optimized layer blobs and zero-disk transfer, alongside numerous bug fixes across both Dragonfly and Nydus.

Key takeaway

For MLOps Engineers managing large model deployments, Dragonfly v2.5.0 significantly streamlines content delivery. You can now directly download Hugging Face and ModelScope repositories with P2P acceleration, reducing bandwidth strain and speeding up model distribution. Consider deploying the Kubernetes Injector via Helm Charts to automatically enable P2P downloads for your application Pods, avoiding manual image modifications. This update enhances operational efficiency and system stability for your AI infrastructure.

Key insights

Dragonfly v2.5.0 enhances P2P content delivery with direct model repository downloads and Kubernetes integration.

Principles

• P2P acceleration improves large file distribution.
• Webhook injection simplifies client deployment.
• Granular control enhances system stability.

Method

Dragonfly Injector uses Kubernetes webhooks to inject client binaries and configurations into Pods via annotations, enabling P2P downloads.

In practice

• Use `dfget hf://...` for direct Hugging Face model downloads.
• Deploy Dragonfly with Helm Charts for webhook injection.
• Configure blocklists in Manager for download control.

Topics

• P2P Content Delivery
• Kubernetes Webhooks
• Hugging Face Models
• ModelScope
• Rate Limiting
• Nydus

Code references

• dragonflyoss/dragonfly
• dragonflyoss/dragonfly-injector
• dragonflyoss/client
• dragonflyoss/d7y.io
• dragonflyoss/console

Best for: AI Architect, NLP Engineer, Computer Vision Engineer, MLOps Engineer, AI Engineer, Software Engineer

Related on AIssential

• Build an AI app with FastAPI and Docker - Coding Tutorial with Tips — Integrate Hugging Face models with FastAPI and Docker for efficient, production-ready ML applications.
• Kubernetes v1.36: Security Defaults Tighten as AI Workload Support Matures — Kubernetes v1.36 enhances security, AI/ML workload management, and API scalability through new stable and beta features.
• Deploying a Multistage Multimodal Recommender System on Amazon Elastic Kubernetes Service — A robust multistage, multimodal recommender system can be deployed on EKS, integrating diverse models and MLOps for scalability and real-time adaptation.
• Docker Kanvas Challenges Helm and Kustomize for Kubernetes Dominance — Docker Kanvas simplifies cloud-native deployments by bridging Docker Compose to Kubernetes and IaC.
• Why PyTorch’s Ecosystem is Driving the Next Wave of Enterprise Machine Learning Solutions — PyTorch's dynamic graph, rich ecosystem, and robust deployment tools drive enterprise ML adoption and scalability.
• NVIDIA Dynamo Snapshot: Fast Startup for Inference Workloads on Kubernetes — NVIDIA Dynamo Snapshot significantly reduces AI inference cold-start times on Kubernetes through optimized checkpoint/restore.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Cloud Native Computing Foundation.