[ 
https://issues.apache.org/jira/browse/HBASE-11874?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14125104#comment-14125104
 ] 

stack commented on HBASE-11874:
-------------------------------

Is estimatedLengthOf used for allocating buffers copying Cells or KVs or just 
an 'estimate'?  If former, we should call it out as so because then 
implementations will be more careful than if just doing a rough length estimate?

So 'writeFlatKey' means write it out as we used serialize a KV? Thats a decent 
definition.  Could probably do with a more descriptive method name but can do 
that in another patch since it is in a few places, right?

Hmm the method names in here are 'interesting'.  This one is a bit hard to 
grok: writeFlatKeyWithoutRowKey  It needs to be public?  Fix ' par,t' in 
javadoc.  I don't get this one: writeRowKeyPart  What is the common part.  Are 
we writing out the Cell as we would a KV?

The rest of the patch looks great after getting over the above awkwardness (I 
did not look too close, just at the nice API changes).  The above awkwardness 
is almost ok going by the nice changes that follow.  What you thinking 
[~anoop.hbase]?  You are close in.  We should commit this and then work on 
undoing the backing array dependency?  Are we doing more copies in this patch 
than previous?  I can look closer and give a better review no problem.


> Support Cell to be passed to StoreFile.Writer rather than KeyValue
> ------------------------------------------------------------------
>
>                 Key: HBASE-11874
>                 URL: https://issues.apache.org/jira/browse/HBASE-11874
>             Project: HBase
>          Issue Type: Sub-task
>            Reporter: Anoop Sam John
>            Assignee: Anoop Sam John
>             Fix For: 0.99.0, 2.0.0
>
>         Attachments: HBASE-11874.patch, HBASE-11874_V2.patch, 
> HBASE-11874_V3.patch, HBASE-11874_V3.patch, HBASE-11874_V4.patch
>
>
> This is the in write path and touches StoreFile.Writer,  HFileWriter , 
> HFileDataBlockEncoder and different DataBlockEncoder impl.
> We will have to avoid KV#getBuffer() KV#getKeyOffset/Length() calls.



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

Reply via email to