falcon-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Nikhil Mulley (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (FALCON-47) Falcon Replication should support configurable delays in feed, parallel, timeout and bulk transfer with variable frequency
Date Sun, 21 Jul 2013 11:56:48 GMT

    [ https://issues.apache.org/jira/browse/FALCON-47?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13714687#comment-13714687

Nikhil Mulley commented on FALCON-47:


In the case of bulk transfers, please also consider to induce a random delay in the range
of 0-x, specifiable in seconds/minutes. 

I see the delay configured here is going to be static, which is as good as changing the base
frequency altogether.
In case of not needing to change the base frequency, I think of randomized delays would be
something to look at.

Please feel free to correct if my whole understanding is wrong though.


> Falcon Replication should support configurable delays in feed, parallel, timeout and
bulk transfer with variable frequency
> --------------------------------------------------------------------------------------------------------------------------
>                 Key: FALCON-47
>                 URL: https://issues.apache.org/jira/browse/FALCON-47
>             Project: Falcon
>          Issue Type: New Feature
>    Affects Versions: 0.3
>         Environment: oozie 3.x distcp 2.x (custom)
>            Reporter: Shaik Idris Ali
>              Labels: Replication
>             Fix For: 0.3
>         Attachments: Falcon-47.patch, Falcon-47-v2.patch
> Falcon Replication should support configurable delays in feed.
> By default the replication/distcp works without any delay, i.e. as soon as a feed is
scheduled, it will start replicating with the current nominal time.
> 1. We need to support usecase of delayed replication, for example, a feed can be produced
with a delay of n hours based on the process generating it, however the replication kicks
of immediately and oozie goes into waiting state and might timeout.
> 2. We need to support parallel/concurrency for feed replication by capturing it from
properties, user may want to run distcp parallelly to backfill data on another cluster.
> 3. timeout can also be set as special property.
> All these can be back-ported from Ivory release.
> 4. Need to support replication with a frequency different from the feed frequency, ex:
a feed can have hours(1) as frequency, by default a scheduled feed will distcp every hour,
however users should be able to set larger frequency like hours(6) which bulk replicates 6
hours of data in one distcp operation.

This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira

View raw message