[
https://issues.apache.org/jira/browse/HBASE-14014?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14632330#comment-14632330
]
Lars Hofhansl commented on HBASE-14014:
---------------------------------------
Some more thoughts.
Would it be correct to batch edit for multiple WALKeys and mark the result with
both the latest write time and latest seqnum seen?
If so, I can freely recombine edits for the table table and source cluster ids
and hence be able to group Cells by row across WALEdits.
I.e. if we had two edits: write time T1, seqnum N1, and write time T2, seqnum
N2, with cells for the same table, cluster ids, and rows. I would then
recombine these into a single WALEdit with T2 and seqnum N2. As a side-effect
it would also reduce the amount of data to be sent to the sink.
> Explore row-by-row grouping options
> -----------------------------------
>
> Key: HBASE-14014
> URL: https://issues.apache.org/jira/browse/HBASE-14014
> Project: HBase
> Issue Type: Sub-task
> Components: Replication
> Reporter: Lars Hofhansl
>
> See discussion in parent.
> We need to considering the following attributes of WALKey:
> * The cluster ids
> * Table Name
> * write time (here we could use the latest of any batch)
> * seqNum
> As long as we preserve these we can rearrange the cells between WALEdits.
> Since seqNum is unique this will be a challenge. Currently it is not used,
> but we shouldn't design anything that prevents us guaranteeing better
> ordering guarantees using seqNum.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
