Re: [DISCUSS] Dependent task check rules

[Hemin Wen <we...@apache.org>]

Yes, There are two scenarios.
1.Add a switch, Whether to check task retry.
2.Increase cycle strategy, recent success or all success
Do you think it is ok.

--------------------
DolphinScheduler(Incubator) Commtter
Hemin Wen  温合民
wenhemin@apache.org
--------------------


Jave-Chen <ke...@foxmail.com> 于2020年6月7日周日 上午11:01写道：

> I think you are right.
>
>
> BTW, would you also consider the following two scenarios：
> 1.&nbsp;The relied node has a retry attribute.
> 2.&nbsp;The dependent node needs all task instances of relied node to
> be&nbsp;&nbsp;successful&nbsp;
> in dependent cycle. For example:&nbsp;
> &nbsp; - node A run hourly
> &nbsp; - node B run daily. And when all tasks of node A are executed
> successfully,&nbsp;
> &nbsp; node B executes
> &nbsp; &nbsp;
> For second scenario, add some tags to successfully
> executed&nbsp;&nbsp;task may be a
> possible solution.&nbsp;
>
>
>
>
>
> ------------------&nbsp;原始邮件&nbsp;------------------
> 发件人:&nbsp;"wenhemin"<whm_777@163.com&gt;;
> 发送时间:&nbsp;2020年5月12日(星期二) 下午5:26
> 收件人:&nbsp;"dev"<dev@dolphinscheduler.apache.org&gt;;
>
> 主题:&nbsp;[DISCUSS] Dependent task check rules
>
>
>
> Hello!
>
> Abort dependent task check rules, I think the current implementation is
> problematic.
>
> E.g. check logic: this week
> The current implementation:
> 1.get this Monday
> 2.the date list of this Monday to current date（asc order）
> 3.Iterate through the list of dates in sequence to check the status of all
> process instances. There is a failure, the ultimate is failure state.
>
> I think, the check logic should be:
> 1.get date range, start date and end date
> 2.if type is DEPENDENT_ALL
>         query the first process instance in reverse order, check this
> instance state.
> 3.if type is non DEPENDENT_ALL
>         query the first task instance in reverse order, check this
> instance state.