[
https://issues.apache.org/jira/browse/HBASE-10347?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13883706#comment-13883706
]
Enis Soztutar commented on HBASE-10347:
---------------------------------------
After some discussions in the parent jira, and in the design proposal for this
jira at
https://docs.google.com/document/d/1ChXchhv1bWpcA0kU4osJotZ_QUcRGSCONXgtsY3M1D4/edit#
, I would like to propose the following review as possible route to continue.
It contains a patch with all of the structural changes as in proposal #3. In
the document, I also tried to write up a summary of the changes, and some
justification for those.
I think we can use this jira to decide on one particular route. The patch
attached at RB contains the structural changes for #3, which we think are a
good balance between abstraction / cleanness and amount of refactoring needed
but as always all feedback is welcome.
The review is at:
https://reviews.apache.org/r/17444/
> HRegionInfo changes for adding replicaId and MetaEditor/MetaReader changes
> for region replicas
> ----------------------------------------------------------------------------------------------
>
> Key: HBASE-10347
> URL: https://issues.apache.org/jira/browse/HBASE-10347
> Project: HBase
> Issue Type: Sub-task
> Components: Region Assignment
> Reporter: Enis Soztutar
> Assignee: Enis Soztutar
> Fix For: 0.99.0
>
>
> As per parent jira, the cleanest way to add region replicas we think is to
> actually create one more region per replica per primary region. So for
> example, if a table has 10 regions with replication = 3, the table would
> indeed be created with 30 regions. These regions will be handled and assigned
> individually for AM purposes.
> We can add replicaId to HRegionInfo to indicate the replicaId, and use this
> to differentiate different replicas of the same region. So, primary replica
> would have replicaId = 0, and the others will have replicaId > 0.
> These replicas will share the same regionId prefix, but differ in an appended
> replicaId. The primary will not contain the replicaId so that no changes
> would be needed for existing tables.
> In meta, the replica regions are kept in the same row as the primary ( so for
> above example, there will be 10 rows in meta). The servers for the replicas
> are kept in columns like "server+replicaId".
--
This message was sent by Atlassian JIRA
(v6.1.5#6160)
