[
https://issues.apache.org/jira/browse/HBASE-12988?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14541365#comment-14541365
]
Lars Hofhansl commented on HBASE-12988:
---------------------------------------
Looking into this. This is a bit more intricate than expected. An Entry in the
WAL is a WALKey/WALEdit pair. The WALKey (among other stuff has table and
region name), the WALEdit has the cells, which in turn have the row. A WALEdit
can contain Cells for many rows.
So to group by row we would need to pull WALEdits apart. I'd have to think
through the implication to convince myself that that's OK. On top of that after
the row-grouping we'd now have to build _new_ WALEdits, since they'd have a
different set of Cells. Ugh...
Grouping entire WALEdits on the other hand is easier, but those I can only
group by table, and - as discussed above - that might not be effective in many
scenarios.
I can easily do a size based breakup and then rely on the fact that we keep
deletes around for a little bit, but that is brittle... I do have a test patch
for this.
> [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)
