good job I think what you have said is very comprehensive, you can start with this proposal
Best Regards --------------- Apache DolphinScheduler PMC Chair David Dai lidong...@apache.org Linkedin: https://www.linkedin.com/in/dailidong Twitter: @WorkflowEasy <https://twitter.com/WorkflowEasy> --------------- On Thu, Sep 2, 2021 at 3:48 PM Xingjie Wang(联通集团联通数字科技有限公司) < wangxj...@chinaunicom.cn> wrote: > Hi community: > > > The current workflow instances are scheduled to run in parallel, plan to > submit workflow serial functionality. > > The functions include: > 1. ProcessInstance of processDefinition that defines a serial plan will > execute sequentially, that is, there will be at most one processInstance > running. > 2. Serial execution is divided into three policies (the processInstance > described below belongs to the same processDefinition and defines the > corresponding serial rules) > > (1) Serial waiting: When a new processInstance is generated, it is > executed in sequence according to the order of the progress queue. > (2) Serial discard: When a new processInstance is generated, if a > processInstance (under the same processDefinition) is found , abandon > thisprocessInstance , the state is stopped. > (3) Serial priority: When a new processInstance is generated, if a running > instance is found, stop this running processInstance. > > Implementation: > > Serial wait: > 1. After the master reads the command, it generates a processInstance > according to the existing logic, > 2. Before saving, determine whether the serial plan is defined in the > processDefinition, if not defined, continue to execute according to the > original logic. If defined: Determine whether the processInstance under the > processDefinition has a processInstance with the status of [Running: 1] or > [Waiting Queue: 14] (newly added), and if there is, update the status to > [Waiting Queue]. > 3. Save the processInstanceId to the processInstance where the > nextProcessInstanceId field (newly added) is empty (here, if the > processInstance just ends running, there will be problems, so before the > end of the processInstance, judge if the serial plan is defined and the > nextProcessInstanceId is empty, then Find the processInstance with the > status of [Waiting Queue: 14] under the processDefinition and start it). > 4. After the processInstance is executed, it is judged that if > nextProcessInstanceId is not empty, a command is generated to resume the > operation of the processInstance. > Note: The operation of Part 3 will have concurrency after 2.0, so the > database optimistic lock is used to solve the concurrency problem when > saving the nextProcessInstanceId field. > > Serial discard: > 1. After the master reads the command, it generates a processInstance > according to the existing logic, > 2. Before saving, determine whether the serial plan is defined in the > processDefinition, if not defined, continue execution according to the > original logic. If defined: Check whether there is a running process, and > if there is a current process state, set it to stop. > > Serial priority: > 1. After the master reads the command, it generates a processInstance > according to the existing logic, > 2. Before saving, determine whether the serial plan is defined in the > processDefinition, if not defined, continue execution according to the > original logic. If defined: Check whether there is a running process, and > if there is, set the running process status to stop. > > Best Wishes！ > > > -------------------------------------------------------------------------------------------------------------------------------- > > 目前工作流实例的运行计划均为并行，计划提交工作流串行功能 > > 功能主要包括 > > 1.定义了串行计划的processDefinition的processInstance会顺序执行，即最多只会有一个正在运行的processInstance。 > > 2.串行执行分为三种策略（以下所描述的processInstance在属于同一processDefinition，并且该processDefinition定义了对应串行规则） > （1）串行等待：生成新的processInstance时，按照进度队列的先后顺序依次执行。 > > （2）串行抛弃：生成新的processInstance时，如果发现有正在运行的工作流实例（同一个processDefinition下）放弃运行，状态为停止。 > （3）串行优先：生成新的processInstance时，如果发现有正在运行的实例，停止该processInstance的运行。 > > 实现方案： > 串行等待： > 1.master在读取command后，按照现有逻辑生成processInstance， > > 2.在保存之前判断该processDefinition中是否定义了串行计划，如果没有定义，按照原有逻辑继续执行。如果定义：判断该processDefinition下的processInstance是否有状态为【正在运行：1】或【等待队列：14】（新增）的processInstance，如果有将状态更新为【等待队列】。 > > 3.将该processInstanceId保存到nextProcessInstanceId字段（新增）为空的processInstance中（此处如果该processInstance刚好结束运行会出现问题，所以在processInstance结束之前判断如果定义了串行计划，并且nextProcessInstanceId为空，则查找该processDefinition下状态为【等待队列：14】的processInstance并且启动）。 > > 4.processInstance执行结束后判断nextProcessInstanceId如果不为空则生成command恢复该processInstance的运行。 > 注：第3部的操作在2.0后会出现并发情况，所以在保存nextProcessInstanceId字段时利用数据库乐观锁解决并发问题。 > > 串行抛弃： > 1.master在读取command后，按照现有逻辑生成processInstance， > > 2.在保存之前判断该processDefinition中是否定义了串行计划，如果没有定义，按照原有逻辑继续执行。如果定义：检查是否有正在运行的流程，如果有当前流程状态置为停止。 > > 串行优先： > 1.master在读取command后，按照现有逻辑生成processInstance， > > 2.在保存之前判断该processDefinition中是否定义了串行计划，如果没有定义，按照原有逻辑继续执行。如果定义：检查是否有正在运行的流程，如果有将正在运行的流程状态置为停止。 > > 附件为处理串行部分的流程，如果您有更好的解决方案，或感觉本次方案存在问题请随时联系我。 > > 祝好！ > 如果您错误接收了该邮件，请通过电子邮件立即通知我们。请回复邮件到 > hqs-s...@chinaunicom.cn，即可以退订此邮件。我们将立即将您的信息从我们的发送目录中删除。 > If you have received this email in error please notify us immediately by > e-mail. Please reply to hqs-s...@chinaunicom.cn ,you can unsubscribe from > this mail. We will immediately remove your information from send catalogue > of our. >
