DolphinSchedulerProposal

 FINAL 

This proposal is now complete and has been submitted for a VOTE.

Abstract

DolphinScheduler is a distributed ETL scheduling engine with powerful DAG visualization interface. DolphinScheduler focuses on solving the problem of 'complex task dependencies & triggers' in data processing. Just like its name, we dedicated to making the scheduling system out of the box.

Current project name of DolphinScheduler is EasyScheduler, will change it after it is accepted by Incubator.

Proposal

DolphinScheduler provides many easy-to-use features to accelerate the engineering efficiency on data ETL workflow job. We propose a new concept of 'instance of process' and 'instance of task' to let developers to tuning their jobs on the running state of workflow instead of changing the task's template. Its main objectives are as follows:

• Define the complex tasks' dependencies & triggers in a DAG graph by dragging and dropping.
• Support cluster HA.
• Support multi-tenant and parallel or serial backfilling data.
• Support automatical failure job retry and recovery.
• Support many data task types and process priority, task priority and relative task timeout alarm.

For now, DolphinScheduler has a fairly huge community in China. It is also widely adopted by many companies and organizations as its ETL scheduling tool.

We believe that bringing DolphinScheduler into ASF could advance development of a much more stronger and more diverse open source community.

Analysys submits this proposal to donate DolphinScheduler's source codes and all related documentations to Apache Software Foundation. The codes are already under Apache License Version 2.0.

• Code base: https://www.github.com/analysys/easyscheduler
• English Documentations: https://analysys.github.io/easyscheduler_docs
• Chinese Documentations: https://analysys.github.io/easyscheduler_docs_cn

Background

We want to find a data processing tool with the following features:

• Easy to use，developers can build a ETL process with a very simple drag and drop operation. not only for ETL developers，people who can't write code also can use this tool for ETL operation such as system adminitrator.
• Solving the problem of "complex task dependencies" , and it can monitor the ETL running status.
• Support multi-tenant.
• Support many task types: Shell, MR, Spark, SQL (mysql, postgresql, hive, sparksql), Python, Sub_Process, Procedure, etc.
• Support HA and linear scalability.

For the above reasons, we realized that no existing product met our requirements, so we decided to develop this tool ourselves. We designed DolphinScheduler at the end of 2017. The first internal use version was completed in May 2018. We then iterated several internal versions and the system gradually became stabilized.

Then we open the source code of DolphinScheduler on March 2019. It soon gained lot's of ETL developers interest and stars on github.

Rationale

Many organizations (>30) (refer to Who is using DolphinScheduler ) already benefit from running DolphinScheduler to make data process pipelines more easier. More than 100 feature ideas come from DolphinScheduler community. Some 3rd-party projects also plan to integrate with DolphinScheduler through task plugin, such as Scriptis, waterdrop. These will strengthen the features of DolphinScheduler.

Current Status

Meritocracy

DolphinScheduler was incubated at Analysys in 2017 and open sourced on GitHub in March 2019. Once open sourced, we have been quickly adopted by multiple organizations，DolphinScheduler has contributors and users from many companies; we have set up the Committer Team. New contributors are guided and reviewed by existed committer members. Contributions are always welcomed and highly valued.

Community

Now we have set development teams for DolphinScheduler in Analysys, and we already have external developers who contributed the code. We already have a user group of more than 1,000 people. We hope to grow the base of contributors by inviting all those who offer contributions through The Apache Way. Right now, we make use of github as code hosting as well as gitter for community communication.

Core Developers

The core developers, including experienced senior developers, are often guided by mentors.

Known Risks

Orphaned products

DolphinScheduler is widely adopted in China by many companies and organizations. The core developers of DolphinScheduler team plan to work full time on this project. Currently there are 10 use cases with more that 1000 activity tasks per day using DolphinScheduler in the user's production environment. There is very little risk of DolphinScheduler getting orphaned as at least two large companies (xueqiu、fengjr) are widely using it in their production, and developers from these companies have also joined Easy Scheduler's team of contributors, DolphinScheduler has eight major releases so far, and and received 373 pull requests from contributors, which further demonstrates DolphinScheduler as a very active project. We also plan to extend and diversify this community further through Apache.

Thus, it is very unlikely that DolphinScheduler becomes orphaned.

Inexperience with Open Source

DolphinScheduler's core developers have been running it as a community-oriented open source project for some time, several of them already have experience working with open source communities, they are also active in presto, alluxio and other projects. At the same time, we will learn more open source experiences by following the Apache way in our incubator journey.

Homogenous Developers

The current developers work across a variety of organizations including Analysys, guandata and hydee; some individual developers are accepted as developers of DolphinScheduler as well. Considering that fengjr and sefonsoft have shown great interests in DolphinScheduler, we plan to encourage them to contribute and invite them as contributors to work together.

Reliance on Salaried Developers

At present, eight of the core developers are paid by their employer to contribute to DolphinScheduler project. we also have some other developers and researchers taking part in the project, and we will make efforts to increase the diversity of the contributors and actively lobby for Domain experts in the workflow space to contribute.

Relationships with Other Apache Products

DolphinScheduler integrates Apache Zookeeper as one of the service registration/discovery mechanisms. DolphinScheduler is deeply integrated with Apache products. It currently support many task types like Apache Hive, Apache Spark, Apache Hadoop, and so on

A Excessive Fascination with the Apache Brand

We recognize the value and reputation that the Apache brand will bring to DolphinScheduler. However, we prefer that the community provided by the Apache Software Foundation will enable the project to achieve long-term stable development. so DolphinScheduler is proposing to enter incubation at Apache in order to help efforts to diversify the community, not so much to capitalize on the Apache brand.

Documentation

A complete set of DolphinScheduler documentations is provided on github in both English and Simplified Chinese.

• English
• Chinese

Initial Source

The project consists of three distinct codebases: core and document. The address of two existed git repositories are as follows:

• https://github.com/analysys/easyscheduler
• https://github.com/analysys/easyscheduler_docs
• https://github.com/analysys/easyscheduler_docs_cn

Source and Intellectual Property Submission Plan

As soon as DolphinScheduler is approved to join Apache Incubator, Analysys will provide the Software Grant Agreement(SGA) and intial committers will submit ICLA(s). The code is already licensed under the Apache Software License, version 2.0.

External Dependencies

As all backend code dependencies are managed using Apache Maven, none of the external libraries need to be packaged in a source distribution.

Most of dependencies have Apache compatible licenses，and the detail as follows:

Most of dependencies have Apache compatible licenses，and the core dependencies are as follows:

Backend Dependency

The front-end UI currently relies on many components, and the core dependencies are as follows:

UI Dependency

Required Resources

Git Repositories

• https://github.com/analysys/EasyScheduler.git
• https://github.com/analysys/easyscheduler_docs.git
• https://github.com/analysys/easyscheduler_docs_cn.git

Issue Tracking

The community would like to continue using GitHub Issues.

Continuous Integration tool

Apache Jenkins

Mailing Lists

• DolphinScheduler-dev: for development discussions
• DolphinScheduler-private: for PPMC discussions
• DolphinScheduler-notifications: for users notifications

Initial Committers

• William-GuoWei(guowei20m@outlook.com)
• Lidong Dai(lidong.dai@outlook.com)
• Zhanwei Qiao(qiaozhanwei@outlook.com)
• Liang Bao(baoliang.leon@gmail.com)
• Gang Li(lgcareer2019@outlook.com)
• Zijian Gong(quanquansy@gmail.com)
• Jun Gao(gaojun2048@gmail.com)
• Baoqi Wu(wubaoqi@gmail.com)

Affiliations

• Analysys Inc: William-GuoWei，Zhanwei Qiao，Liang Bao，Gang Li，Jun Gao，Lidong Dai

• Hydee Inc: Zijian Gong

• Guandata Inc: Baoqi Wu

Sponsors

Champion

• Sheng Wu ( Apache Incubator PMC, Apache officer, wusheng@apache.org)

Mentors

• Sheng Wu ( Apache Incubator PMC, Apache officer, wusheng@apache.org)

• ShaoFeng Shi ( Apache Incubator PMC, shaofengshi@apache.org)

• Liang Chen ( Apache Incubator PMC, Apache member, chenliang613@apache.org)

• Furkan KAMACI ( Apache Incubator PMC, Apache member, kamaci@apache.org)
• Kevin Ratnasekera ( Apache Incubator PMC, Apache member, djkevincr@apache.org)

Sponsoring Entity

We are expecting the Apache Incubator could sponsor this project.