[
https://issues.apache.org/jira/browse/HBASE-15631?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16211443#comment-16211443
]
Andrew Purtell edited comment on HBASE-15631 at 10/19/17 6:37 PM:
------------------------------------------------------------------
bq. I see that randomAssignment is overridden in RSGroupBasedLoadBalancer, so
it might be relevant, ... hard to say without studying the complete patch.
Thanks. It will be good to get a second pair of eyes on it.
For what it is worth, I would be ok with leaving that hunk out of the backport
for now. Pass criteria is all unit tests passing. We are going to mark this
feature as experimental. The change may need to be applied later, should we
find in cluster chaos testing a problem that is resolved by it, but it is
reasonable to say any change to core for RSgroups should have a test that
covers whatever it is required for. There isn't one for this hunk. We can put
this on a list of post backport follow ups. It can go back in with a test.
Alternatively, I don't mind including it either, if someone feels strongly
about it. That change to HMaster didn't cause any HBase test failures nor any
trouble when RSGroups were (lightly) tested on a cluster with chaos for an
earlier backport to an internal branch based on 1.3.
was (Author: apurtell):
bq. I see that randomAssignment is overridden in RSGroupBasedLoadBalancer, so
it might be relevant, ... hard to say without studying the complete patch.
Thanks. It will be good to get a second pair of eyes on it.
For what it is worth, I would be ok with leaving that hunk out of the backport
for now. Pass criteria is all unit tests passing. We are going to mark this
feature as experimental. The change may need to be applied later, should we
find in cluster chaos testing a problem that is resolved by it, but it is
reasonable to say any change to core for RSgroups should have a test that
covers whatever it is required for. There isn't one for this hunk. We can put
this on a list of post backport follow ups.
Alternatively, I don't mind including it either, if someone feels strongly
about it. That change to HMaster didn't cause any HBase test failures nor any
trouble when RSGroups were (lightly) tested on a cluster with chaos for an
earlier backport to an internal branch based on 1.3.
> Backport Regionserver Groups (HBASE-6721) to branch-1
> ------------------------------------------------------
>
> Key: HBASE-15631
> URL: https://issues.apache.org/jira/browse/HBASE-15631
> Project: HBase
> Issue Type: New Feature
> Affects Versions: 1.4.0
> Reporter: Francis Liu
> Assignee: Andrew Purtell
> Fix For: 1.4.0, 1.5.0
>
> Attachments: HBASE-15631-branch-1.patch, HBASE-15631-branch-1.patch,
> HBASE-15631.branch-1.patch, HBASE-15631.patch
>
>
> Based on dev list discussion backporting region server group should not be an
> issue as it does not: 1. destabilize the code. 2. cause backward
> incompatibility.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
