[
https://issues.apache.org/jira/browse/HBASE-12988?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14526121#comment-14526121
]
Lars Hofhansl commented on HBASE-12988:
---------------------------------------
Thinking on this again, it's important we do this. Otherwise replication is
slow, especially across high latency links across data centers.
We can significantly cut this down but by having multiple smaller chunks in
flight to multiple target region server instead of only a single large chunk to
exactly one region server at a time.
The grouping by row should be easy to do to avoid any weirdness around ordering
of deletes/put for the same row.
The place to do this might be HBaseInterClusterReplicationEndpoint.replication
or its caller.
There we get one (large) chunk, and send it to exactly one peer region server.
We can split the chunk up into N parts here and farm it to N region servers in
parallel, if all sub chunks are success the entire large chunk is successful.
We can retry individual chunks a few time and then fail the larger chunk if
that failed.
> [Replication]Parallel apply edits on row-level
> ----------------------------------------------
>
> Key: HBASE-12988
> URL: https://issues.apache.org/jira/browse/HBASE-12988
> Project: HBase
> Issue Type: Improvement
> Components: Replication
> Reporter: hongyu bi
> Assignee: hongyu bi
>
> we can apply edits to slave cluster in parallel on table-level to speed up
> replication .
> update : per conversation blow , it's better to apply edits on row-level in
> parallel
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
