[ 
https://issues.apache.org/jira/browse/HBASE-16608?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15484054#comment-15484054
 ] 

Anastasia Braginsky commented on HBASE-16608:
---------------------------------------------

The second patch published (V04) is the minimal in-flush memory policy, which 
can be enhanced in HBASE-16417.

The simplest CompactingMemStore need to be defined by the user as having a type 
of the "index-compaction" or the "data-compaction". Any immutable segment that 
remains in the compaction pipeline is represented with CellArrayMap index.
In both cases, CompactingMemStore behaves the same until a new immutable 
segment is pushed into the compaction pipeline. After that:

*index-compaction:* merges the new segment with the old segment in the pipeline 
and that is all. The merge is the literal merge of the two indexes without 
data-copy and without SQM.
*data-compaction:* compacts the new segment with the old segment in the 
pipeline and that is all. The compaction implies the data-copy, but no 
speculative SQM scan is used.

Pay attention that this is what we suggest as a *minimal* basis to be enhanced 
in HBASE-16417, where we are going to check the performance of the other 
optimization as it is the compaction prediction, length of the pipeline, etc.

[~anoop.hbase], [~ram_krish], [~stack], what do you think about all this? Both 
patches can be seen in the RB.

> Introducing the ability to merge ImmutableSegments without copy-compaction or 
> SQM usage
> ---------------------------------------------------------------------------------------
>
>                 Key: HBASE-16608
>                 URL: https://issues.apache.org/jira/browse/HBASE-16608
>             Project: HBase
>          Issue Type: Sub-task
>            Reporter: Anastasia Braginsky
>            Assignee: Anastasia Braginsky
>         Attachments: HBASE-16417-V02.patch, HBASE-16417-V04.patch
>
>




--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Reply via email to