[
https://issues.apache.org/jira/browse/HBASE-16417?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15443151#comment-15443151
]
Anastasia Braginsky commented on HBASE-16417:
---------------------------------------------
bq. Though on the general use case (where not many updates/deletes) why cannot
we flush all the segments in pipeline together when a flush to disk arise? In
that case also, doing an in memory compaction for segments in pipeline (eg: You
say when segments# >3) is to reduce #files flushed to disk. So another way for
that is flush whole pipeline together.
As [~ebortnik] has said, having long list of small segments in the pipeline
affects the read path. We indeed see that this list might me very long, i.e.
tens of segments. This list should be managed before flushing to disk. Please
note, that the merge that we are suggesting doesn't includes data copying or
SQM.
Of course, there is no problem to flush everything in the pipeline when flush
to disk is requested. Actually, when flush is requested first thing is to push
the active segment to pipeline. So the active is going to be flushed all
together with the pipeline upon the flush to disk.
> In-Memory MemStore Policy for Flattening and Compactions
> --------------------------------------------------------
>
> Key: HBASE-16417
> URL: https://issues.apache.org/jira/browse/HBASE-16417
> Project: HBase
> Issue Type: Sub-task
> Reporter: Anastasia Braginsky
>
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
