DolphinScheduler JSON拆解详解

        本次活动邀请DolphinScheduler社区活跃贡献者，开源积极分子，现就职于政采云大数据部门，从事大数据平台架构工作的李进勇同学给大家分享相关内容。

        同时也特别感谢示说网对本次直播活动的大力支持。

PS：本章内容篇幅较长，请大家耐心阅读。

为什么拆解json

        在DolphinScheduler 1.3.x及以前的工作流中的任务及任务关系保存时是以大json的方式保存到数据库中process_definiton表的process_definition_json字段，如果某个工作流很大比如有100或者1000个任务，这个json字段也就非常大，在使用时需要解析json，非常耗费性能，且任务没法重用；基于大json，在工作流版本及任务版本上也没有好的实现方案，否则会导致数据大量冗余。

故社区计划启动json拆解项目，实现的需求目标：

• 大json完全拆分

• 新增工作流及任务版本

• 引入全局唯一键(code)

如何设计拆解后的表

3.6版本工作流

        比如在当前1.3.6版本创建个a-->b的工作流以下是processDefiniton save 接口在controller入口打印的入参日志

create  process definition, project name: hadoop, process definition name: ab, process_definition_json: {"globalParams":[],"tasks":[{"type":"SHELL","id":"tasks-77643","name":"a","params":{"resourceList":[],"localParams":[{"prop":"yesterday","direct":"IN","type":"VARCHAR","value":"${system.biz.date}"}],"rawScript":"echo ${yesterday}"},"description":"","timeout":{"strategy":"","interval":null,"enable":false},"runFlag":"NORMAL","conditionResult":{"successNode":[""],"failedNode":[""]},"dependence":{},"maxRetryTimes":"0","retryInterval":"1","taskInstancePriority":"MEDIUM","workerGroup":"default","preTasks":[]},{"type":"SHELL","id":"tasks-99814","name":"b","params":{"resourceList":[],"localParams":[{"prop":"today","direct":"IN","type":"VARCHAR","value":"${system.biz.curdate}"}],"rawScript":"echo ${today}"},"description":"","timeout":{"strategy":"","interval":null,"enable":false},"runFlag":"NORMAL","conditionResult":{"successNode":[""],"failedNode":[""]},"dependence":{},"maxRetryTimes":"0","retryInterval":"1","taskInstancePriority":"MEDIUM","workerGroup":"default","preTasks":["a"]}],"tenantId":1,"timeout":0}, desc:  locations:{"tasks-77643":{"name":"a","targetarr":"","nodenumber":"1","x":251,"y":166},"tasks-99814":{"name":"b","targetarr":"tasks-77643","nodenumber":"0","x":533,"y":161}}, connects:[{"endPointSourceId":"tasks-77643","endPointTargetId":"tasks-99814"}]

依赖节点的工作流，dep是依赖节点

以下是processDefiniton save 接口在controller入口打印的入参日志

 create  process definition, project name: hadoop, process definition name: dep_c, process_definition_json: {"globalParams":[],"tasks":[{"type":"SHELL","id":"tasks-69503","name":"c","params":{"resourceList":[],"localParams":[],"rawScript":"echo 11"},"description":"","timeout":{"strategy":"","interval":null,"enable":false},"runFlag":"NORMAL","conditionResult":{"successNode":[""],"failedNode":[""]},"dependence":{},"maxRetryTimes":"0","retryInterval":"1","taskInstancePriority":"MEDIUM","workerGroup":"default","preTasks":["dep"]},{"type":"DEPENDENT","id":"tasks-22756","name":"dep","params":{},"description":"","timeout":{"strategy":"","interval":null,"enable":false},"runFlag":"NORMAL","conditionResult":{"successNode":[""],"failedNode":[""]},"dependence":{"relation":"AND","dependTaskList":[{"relation":"AND","dependItemList":[{"projectId":1,"definitionId":1,"depTasks":"b","cycle":"day","dateValue":"today"}]}]},"maxRetryTimes":"0","retryInterval":"1","taskInstancePriority":"MEDIUM","workerGroup":"default","preTasks":[]}],"tenantId":1,"timeout":0}, desc:  locations:{"tasks-69503":{"name":"c","targetarr":"tasks-22756","nodenumber":"0","x":597,"y":166},"tasks-22756":{"name":"dep","targetarr":"","nodenumber":"1","x":308,"y":164}}, connects:[{"endPointSourceId":"tasks-22756","endPointTargetId":"tasks-69503"}]

条件判断的工作流

以下是processDefiniton save 接口在controller入口打印的入参日志

create  process definition, project name: hadoop, process definition name: condition_test, process_definition_json: {"globalParams":[],"tasks":[{"type":"SHELL","id":"tasks-68456","name":"d","params":{"resourceList":[],"localParams":[],"rawScript":"echo 11"},"description":"","timeout":{"strategy":"","interval":null,"enable":false},"runFlag":"NORMAL","conditionResult":{"successNode":[""],"failedNode":[""]},"dependence":{},"maxRetryTimes":"0","retryInterval":"1","taskInstancePriority":"MEDIUM","workerGroup":"default","preTasks":[]},{"type":"SHELL","id":"tasks-58183","name":"e","params":{"resourceList":[],"localParams":[],"rawScript":"echo 22"},"description":"","timeout":{"strategy":"","interval":null,"enable":false},"runFlag":"NORMAL","conditionResult":{"successNode":[""],"failedNode":[""]},"dependence":{},"maxRetryTimes":"0","retryInterval":"1","taskInstancePriority":"MEDIUM","workerGroup":"default","preTasks":["cond"]},{"type":"SHELL","id":"tasks-43996","name":"f","params":{"resourceList":[],"localParams":[],"rawScript":"echo 33"},"description":"","timeout":{"strategy":"","interval":null,"enable":false},"runFlag":"NORMAL","conditionResult":{"successNode":[""],"failedNode":[""]},"dependence":{},"maxRetryTimes":"0","retryInterval":"1","taskInstancePriority":"MEDIUM","workerGroup":"default","preTasks":["cond"]},{"type":"CONDITIONS","id":"tasks-38972","name":"cond","params":{},"description":"","timeout":{"strategy":"","interval":null,"enable":false},"runFlag":"NORMAL","conditionResult":{"successNode":["e"],"failedNode":["f"]},"dependence":{"relation":"AND","dependTaskList":[{"relation":"AND","dependItemList":[{"depTasks":"d","status":"SUCCESS"}]}]},"maxRetryTimes":"0","retryInterval":"1","taskInstancePriority":"MEDIUM","workerGroup":"default","preTasks":["d"]}],"tenantId":1,"timeout":0}, desc:  locations:{"tasks-68456":{"name":"d","targetarr":"","nodenumber":"1","x":168,"y":158},"tasks-58183":{"name":"e","targetarr":"tasks-38972","nodenumber":"0","x":573,"y":82},"tasks-43996":{"name":"f","targetarr":"tasks-38972","nodenumber":"0","x":591,"y":288},"tasks-38972":{"name":"cond","targetarr":"tasks-68456","nodenumber":"2","x":382,"y":175}}, connects:[{"endPointSourceId":"tasks-68456","endPointTargetId":"tasks-38972"},{"endPointSourceId":"tasks-38972","endPointTargetId":"tasks-58183"},{"endPointSourceId":"tasks-38972","endPointTargetId":"tasks-43996"}]

从以上三个案例中，我们知道controller的入口参数的每个参数都可以在t_ds_process_definition表中找到对应，故表中数据如下图

拆解后的表设计思路

    工作流只是dag的展现形式，任务通过工作流进行组织，组织的同时存在了任务之间的关系，也就是依赖。就好比一个画板，画板上有些图案，工作流就是画板，图案就是任务，图案之间的关系就是依赖。而调度的核心是调度任务，依赖只是表述调度的先后顺序。当前定时还是对整个工作流进行的定时，拆解后就方便对单独任务进行调度。正是基于这个思想设计了拆解的思路，所以这就需要三张表，工作流定义表、任务定义表、任务关系表。

• 工作流定义表：描述工作流的基本信息，比如全局参数、dag中节点的位置信息

• 任务定义表：描述任务的详情信息，比如任务类别、任务容错信息、优先级等