helix-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Kanak Biscuitwala (JIRA)" <j...@apache.org>
Subject [jira] [Updated] (HELIX-148) Current preferred placement for auto rebalace is suboptimal for n > p
Date Fri, 26 Jul 2013 23:35:48 GMT

     [ https://issues.apache.org/jira/browse/HELIX-148?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]

Kanak Biscuitwala updated HELIX-148:
------------------------------------

    Attachment: preferredchk.py

Attached a python script I was using to try some assignment strategies, in case anyone's interested.

At a high level, the proposed strategy is:

if n > p: assign in partition order, so n0 gets p0r0, n1 gets p1r0 until replicas are exhausted,
then the next node gets p0r1 and so on
else if n == p: similar to above, but add an offset of replica id so that a node won't prefer
all the replicas of the same partition
else: keep the original strategy

Fix is out for review: https://reviews.apache.org/r/12984
                
> Current preferred placement for auto rebalace is suboptimal for n > p
> ---------------------------------------------------------------------
>
>                 Key: HELIX-148
>                 URL: https://issues.apache.org/jira/browse/HELIX-148
>             Project: Apache Helix
>          Issue Type: Bug
>          Components: helix-core
>            Reporter: Kanak Biscuitwala
>            Assignee: Kanak Biscuitwala
>            Priority: Minor
>         Attachments: preferredchk.py
>
>
> The new auto-rebalance algorithm computes suboptimal preferred placement if there are
more nodes than partitions because it fills up nodes before moving onto more nodes.

--
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

Mime
View raw message