helix-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "ASF GitHub Bot (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HELIX-601) Allow work flow to schedule dependency jobs in parallel
Date Fri, 03 Jul 2015 21:05:04 GMT

    [ https://issues.apache.org/jira/browse/HELIX-601?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14613456#comment-14613456
] 

ASF GitHub Bot commented on HELIX-601:
--------------------------------------

Github user kanakb commented on the pull request:

    https://github.com/apache/helix/pull/30#issuecomment-118425221
  
    1. Let's say we have a target resource with 2 partitions and 0 replicas, with one partition
assigned to node A, and one partition assigned to node B. Job 0 runs on nodes A and B, it
finishes on node A, and then Job 1 starts on node A. Then imagine node B fails, and both partitions
are now on node A. Job 1 is running on node A, but Job 0 did not finish for the partition
that was reassigned to node A. We have a dependency inversion, and that's why this is unsafe.
    
    2. If the job does not have a target resource, this change doesn't make sense. An exception
should be thrown if you attempt to submit an untargeted workflow that has parallelism set.


> Allow work flow to schedule dependency jobs in parallel
> -------------------------------------------------------
>
>                 Key: HELIX-601
>                 URL: https://issues.apache.org/jira/browse/HELIX-601
>             Project: Apache Helix
>          Issue Type: New Feature
>            Reporter: Congrui Ji
>
> Currently, Helix won't schedule dependency jobs in a same work flow. For example, if
Job2 depends on Job1, Job2 won't be scheduled until every partition of Job1 is completed.
> However, if some participant is very slow, then all dependency jobs is waiting for that
single participant.
> Helix should be able to schedule multiple jobs according to a parameter.
> A.C.
> 1. Introduce parallel count parameter in work flow and job queue.
> 2. Dependency jobs can be scheduled according to the parameter (Now the parameter is
always 1, so no parallel)
> 3. If Job2 depends on Job1, Job1 is scheduled before Job2.
> 4. No parallel jobs on the same instance. If a instance is running Job1, it won't run
Job2 until Job1 is finished. 



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Mime
View raw message