[
https://issues.apache.org/jira/browse/HBASE-16166?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15370277#comment-15370277
]
ramkrishna.s.vasudevan commented on HBASE-16166:
------------------------------------------------
bq.But the issue is, when we dont do this compaction, it is incorrect to just
flush the tail of the pipeline alone while flush to disk. We must flush all
segments + active then. If some way we can make this also configable this new
memstore can be done in CompactingMemstore itself?
It can be done in the CompactingMemstore but not sure if the changes will be
clean. For now added as a new MS impl.
bq.You fellows finding Segments are in the way when default case? What you
seeing in terms of friction when the segments pipeline is running (I've not
tried it).
Segments are not a friction but since we don't have any case of compaction
reading the segments and compacting into one and always flushing the tail is
not needed. Instead we could just read and flush all segments. Once flattening
is in place then it will be more useful in terms of heap usage specifically
with offheap cases.
> Create a memstore impl that does not compact but creates pipelines and
> flushes all segments from pipeline
> ---------------------------------------------------------------------------------------------------------
>
> Key: HBASE-16166
> URL: https://issues.apache.org/jira/browse/HBASE-16166
> Project: HBase
> Issue Type: Sub-task
> Affects Versions: 2.0.0
> Reporter: ramkrishna.s.vasudevan
> Assignee: ramkrishna.s.vasudevan
> Fix For: 2.0.0
>
> Attachments: HBASE-16166.patch
>
>
> A new memstore impl for the default cases, where we could create pipelines
> and then directly flush all the segments in pipelines.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
