[
https://issues.apache.org/jira/browse/PHOENIX-4601?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16445032#comment-16445032
]
Vincent Poon commented on PHOENIX-4601:
---------------------------------------
[~jamestaylor] Attached a patch where I set the
DEFAULT_INDEX_WRITER_RPC_RETRIES_NUMBER back to it's old value of 11. This is
in case existing installations were relying on the default. I deprecated it
along with the index_writer_rpc configs, as eventually those configs won't do
anything. We won't have a way of increasing the number of server side retries
anymore, though, unless we want to invent a new config to control that. But I
figure we can do a new Jira for that if we ever need it.
> Perform server-side retries if client version < 4.14
> ----------------------------------------------------
>
> Key: PHOENIX-4601
> URL: https://issues.apache.org/jira/browse/PHOENIX-4601
> Project: Phoenix
> Issue Type: Bug
> Reporter: James Taylor
> Assignee: Vincent Poon
> Priority: Major
> Attachments: PHOENIX-4601.v1.master.patch
>
>
> The client version is now available on the server side when index maintenance
> is being performed. Given that this information is available, we should
> conditionally retry on the server depending on the client version (instead of
> relying on the operator to manually update the config after clients have been
> upgraded).
> With PHOENIX-4613, the client version has been threaded through to the
> IndexCommitter.write() method. All that's left to do is:
> - Always set the config on the server side to have no HBase retries.
> - Add catch of IOException and conditionally call the retrying exception
> handler code based on clientVersion < 4.14.0 in
> TrackingParallelWriterIndexCommitter and ParallelWriterIndexCommitter.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
