[
https://issues.apache.org/jira/browse/HBASE-10355?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13901221#comment-13901221
]
Nicolas Liochon commented on HBASE-10355:
-----------------------------------------
bq. the information that he sees would be partial
Hum. Yes, it's not well managed in the patch as it is. It's a little bit
complex, because the contract of the getLocation is to retry, so the code
expects the locations to be good. If we don't do that, we will have two level
of retries (one in getLocation, one in the retrying callable). I can either
manage the case explicitly, either getLocation should return a full & good
location set.
On the same range, if we change dynamically (without disable / enable) the
number of replica, the client we would be surprised.
Lastly, as it is today, we stop on the first 'definitive' error: we don't wait
for all replica to fails, we stop when one of the replica says 'it does not
work'. It's a trade off, doing differently could make sense as well.
bq. Maybe if the location is null, the client should hit the meta always?
I can do that (force reload).
About the jdk 1.7, it's the completion service that is not available on 1.6.
It's not a big dependency, but makes the code much simpler and safer.
We actually didn't conclude when we discussed 1.7 in this thread, 6 months ago.
http://search-hadoop.com/m/X9ivp1d7Xoc2&subj=Re+Default+JDK+settings+for+Jenkins+projects
> Failover RPC's from client using region replicas
> ------------------------------------------------
>
> Key: HBASE-10355
> URL: https://issues.apache.org/jira/browse/HBASE-10355
> Project: HBase
> Issue Type: Sub-task
> Components: Client
> Reporter: Enis Soztutar
> Assignee: Nicolas Liochon
> Fix For: 0.99.0
>
> Attachments: 10355.v1.patch
>
>
--
This message was sent by Atlassian JIRA
(v6.1.5#6160)
