hbase-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Anoop Sam John (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HBASE-13916) Create MultiByteBuffer
Date Tue, 16 Jun 2015 17:46:01 GMT

    [ https://issues.apache.org/jira/browse/HBASE-13916?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14588446#comment-14588446
] 

Anoop Sam John commented on HBASE-13916:
----------------------------------------

bq,Is the fact that the nature of MBB changes based on whether there is one BB or many BBs
going to confuse? Why not just throw unsupported whatever the BB count in methods like array,
arrayOffset, etc.
HFileBlock is going to be backed by MBB.  I wanted like when the block comes NOT from L2 cache,
we will go like in existing way and make KeyValue. when we make Cell objects in read path,
based on this hasArray() we will have to make either byte[] backed KVs or BB backed new objects.
Just because of this I thought of keeping this hasArray() true when MBB is backed by a single
on heap BB.
bq.I think the class comment should describe when to use this rather than ByteBufferArray
from util.
I will clean it. 
bq.Why a getVLong in here? That belongs outside this class?
Just did not want a new util class for it. Let me see. 
bq.Is the unsafe work going on in BBUtils dupe of Bytes unsafeing? Should the BBUtils go to
Bytes for Unsafe ops?
Not really. When it is BB we will have to deal with isDirect case also. There is an UnsafeAccess
class now.. May be all these byte[] and BB related unsafe access has to go in there and Bytes
and BBUtils etc has to use that. (?)

> Create MultiByteBuffer
> ----------------------
>
>                 Key: HBASE-13916
>                 URL: https://issues.apache.org/jira/browse/HBASE-13916
>             Project: HBase
>          Issue Type: Sub-task
>          Components: regionserver, Scanners
>            Reporter: Anoop Sam John
>            Assignee: Anoop Sam John
>             Fix For: 2.0.0
>
>         Attachments: HBASE-13916.patch, HBASE-13916_V2.patch
>
>
> This is an aggregation of N ByteBuffers. The block when served directly by block cache
buckets memory, we have the block data split across multiple byte buffers. This aggregate
type (like ByteBuffer) will serve the HFileBlock data.
> This jira wil just provide the new data structure



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

Mime
View raw message