[
https://issues.apache.org/jira/browse/FALCON-1525?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14951819#comment-14951819
]
Srikanth Sundarrajan commented on FALCON-1525:
----------------------------------------------
[~jakub_havlik], We felt the need for having an invariant identifier that is
easy to arrive at for a particular instance of execution and instance time
seemed like a good candidate. If we were to use any other identifier that may
be generated by the system, you will need a mechanism to resolve/figure that
identifier and that seemed cumbersome. Some of us have used vanilla Oozie in
the past and have had trouble working with action-id for a coordinator, as the
relationship between action-id and time was not intuitive. Having said that
FALCON-1168 will soon support a-periodic instances and time doesn't seem like a
natural way to identify instances in those cases. In near future we might be
seeing non-time based instance-id (such as revision/version) for such
executions, but the approach of using an time for periodic instances may not
change any time soon.
A questions relating to your use case/application: Since you were concerned
about more instances qualifying with time being slightly off, forces me to
assume that the frequency of your process/feed may be too high (such as
minutes(1)). Can you provide a bit more information about the application and
the issue you were running into to help us appreciate the problem better?
> Add possibility to rerun process instances by an id
> ---------------------------------------------------
>
> Key: FALCON-1525
> URL: https://issues.apache.org/jira/browse/FALCON-1525
> Project: Falcon
> Issue Type: Improvement
> Components: client
> Affects Versions: 0.7
> Reporter: Jakub Havlík
> Priority: Critical
>
> As a user I need to specifically rerun individual instances by their ids.
> Doing rerun by name and time slot (where minimum resolution means minutes) is
> not usable in a corporate environment. FalconClient class needs to have a
> rerunInstance method which accepts some unique identifier for each instance.
> As it stands now there is possibility to inadvertently restart instance if it
> somehow fits into a time slot (parallel instances or instances long only a
> few seconds for example).
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
