[
https://issues.apache.org/jira/browse/HBASE-11908?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14124717#comment-14124717
]
Enis Soztutar commented on HBASE-11908:
---------------------------------------
The solution is easy enough. We can just add a server name columns for region
replicas, but leave the server name as empty. This will allow the clients to
know about the region replica, but the location server name will still be empty
until it is assigned. Any RPC attempted for these replica locations will
eventually succeed using the regular RPC retry mechanism.
> Region replicas should be added to the meta table at the time of table
> creation
> -------------------------------------------------------------------------------
>
> Key: HBASE-11908
> URL: https://issues.apache.org/jira/browse/HBASE-11908
> Project: HBase
> Issue Type: Sub-task
> Reporter: Enis Soztutar
> Assignee: Enis Soztutar
> Fix For: 2.0.0
>
>
> While testing async replication handling and failover handling for region
> replicas, we've found that sometimes the secondary region replicas do not
> open and update meta quickly enough, and hence meta would not contain any
> information about the region replica id.
> If a reader caches the meta row before all region replicas are open the first
> time (such as the async wal replication endpoint), then it may miss the
> region replica and won't know about it until it refreshes it's cache again.
> Instead, we should add entries to the meta for all of the region replicas at
> the time of create table (just like primary regions).
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
