Skip to content
This repository has been archived by the owner on Feb 27, 2023. It is now read-only.

Latest commit

 

History

History
104 lines (66 loc) · 4.51 KB

ROADMAP.md

File metadata and controls

104 lines (66 loc) · 4.51 KB
Raw

Dragonfly Roadmap

This document is used for contributors to have a better understanding of project's progress.

For Dragonfly, the following versions are tentatively defined as project milestones. The following three versions are being planned currently:

  1. v0.4.x, improved performance and stability.
  2. v0.5.x, more adoptive with cloud env.
  3. v1.0.x, GA version.

v0.4.x

This version will focus on to improved the performance and stability of Dragonfly. Including:

  • Golang SDK
  • Supernode HA
  • Merge the dfdaemon and dfget components
  • Decentralized distribution

We will try to finish it before October 31, 2019.

Golang SDK

Dragonfly needs a Golang SDK which is stored in https://github.com/dragonflyoss/Dragonfly/tree/master/client. And then we can achieve the following goals:

  • every caller of Dragonfly can take advantages of this SDK by integrating it
  • golang sdk encapsulates of the details of HTTP communication
  • it could provide scalability and portability for other system to integrate Dragonfly.

Related issue: #348

Supernode HA

As for now, we can specify multiple supenodes for dfclient at the same time. However, supernode is not aware of each other. When one of the supernode server is unavailable, dfclient will try to register to the one of the remained supernodes. The supernode registered will download the file from source server, reschedule pieces and re-build a P2P network because it lacks the information about the task and the peer node which will cost a lot of time. So we should make multiple supernodes be aware of each other.

Related issue: #232

Merge the dfdaemon and dfget components

Both dfdaemon and dfget are meant for the client and packaged together into dfclient. So we can merge them into one single command to make the Dragonfly look simpler and easier to be deployed.

Related issue: #806

Decentralized distribution

Now, dfget talks to supernode to discover peers information and publish its own status. It works well, however, the user has to maintain a supernode server and supernode will become the bottleneck of the cluster. So we can use a new approach where dfget can discover the same information without talking to supernode and all the clients can form a gossip cluster, where they can fire and listen for events.

Related PR: #594

v0.5.x

This version will focus on enhancing the Dragonfly's features to make it more adoptive with cloud env. Including:

  • Support for more deployment options
  • Support multi-server computing framework.
  • Development of operation and maintenance tools.

We will try to finish it before December 30, 2019.

Support for more deployment options

As a cloud native project, we should do more work to support deploy the Dragonfly on the kubernets platform. Including but not limited to the following list:

  • Deploy supernode using Helm in Kubernetes to simplify the complexity of scaling SuperNodes in Kubernetes.
  • Deploy supernode cluster using Operator.
  • Deploy dfget & dfdaemon using DaemonSet in Kubernetes.

Related issue: #346

Support multi-server computing framework

With the rise of ARM servers, x86 servers will not be the only choice. However, the images of the two computing frameworks are incompatible for one application. So the user will have to maintain the different versions for the same image which will bring challenges to management.

Related issue: #775

Development of operation and maintenance tools

For better to use Dragonfly, we plan to provide more operation and maintenance tools to make it easy, including:

  • A tool to validate the config file.
  • A tool for performance testing of Dragonfly clusters.
  • A dashboard tool for Dragonfly.

v1.0.x

This is a stable version in GA stage. And we will focus on making Dragonfly support more scenarios. Including:

  • Support different encryption algorithm in data transmission.
  • Support multiple file transfer protocols, such as ftp, etc.
  • Distribution images across cloud vendors.
  • Support publish and subscribe mechanism.
  • ......

This document give us a trunk direction, however it's not perfect, we may adjust the plan appropriately according to the urgency of demand.