crunch-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Chao Shi (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (CRUNCH-355) Rename jobs to show how many stages have done before job submission
Date Tue, 25 Feb 2014 02:18:20 GMT

    [ https://issues.apache.org/jira/browse/CRUNCH-355?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13911134#comment-13911134
] 

Chao Shi commented on CRUNCH-355:
---------------------------------

Hi Josh, It is not the only case with such  problem that a pipeline has completely disjoint
components. In a case I'm running into, there is a final last stage which depends on several
disjoint sub-pipelines. So I think we have to assign the label at runtime: in (m/n), let m
be the number of jobs we have submitted.

> Rename jobs to show how many stages have done before job submission
> -------------------------------------------------------------------
>
>                 Key: CRUNCH-355
>                 URL: https://issues.apache.org/jira/browse/CRUNCH-355
>             Project: Crunch
>          Issue Type: Bug
>          Components: Core
>            Reporter: Chao Shi
>            Assignee: Josh Wills
>
> The naming mechanism introduced in CRUNCH-262 has a flaw. It adds (m/n) to the end of
job name, where m is the current stage number at planning time and n is the total number of
stages.
> Suppose in the following case, where A takes a long time to run, when B is submitted,
C and D have completed. As there are 3 jobs done, we should expect B to be (4/5) rather than
(2/5) or (3/5).
> {code}
> A   C
> |   |
> B   D
>  \ /
>   E
> {/code}
> In planning time, we don't have a clue which stage will complete earlier. So I think
it better to assign it at runtime.
> One problem here is that the user may access to Job instance before it submits. So he
may observe a sudden change of job name.



--
This message was sent by Atlassian JIRA
(v6.1.5#6160)

Mime
View raw message