[ 
https://issues.apache.org/jira/browse/HBASE-7413?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13642134#comment-13642134
 ] 

stack commented on HBASE-7413:
------------------------------

bq. If CellBlock is only a mechanism to write this stuff to RPC, then why do we 
need to construct it in memory. We can just dump it into some stream.

An implementation could do that yeah.  Other implementations will want to do 
encoding or compression and will need to run with buffers to store up 
intermediary forms in between flushes.

Over in RPC, our RPC as is prefixes all messages with a total length.  While 
this constraint is in place, we can only know the cellblock size after its 
creation.

bq. We can put these on the wire directly too, though.

Writing WAL, yeah (unless you need to know size in advance), on the wire, not 
currently (After this refactor of rpc is done, the next refactor should be to 
do chunking/stream.. undo the need of the prefatory length).

bq. The count, yes. For the certain types, perhaps cellblock should be 
cellsink, and it can do whatever internally - construct in memory, write to 
stream, etc.?

Is it not now?  Or, is there something we need to do to make it a "CellSink"?

bq. That will depend on what operations client performs. WAL currently is 
already compressed using dictionary compression (if enabled).

Not the values though?

bq. newCodedOutput

We could, yeah, or not put data into pb at all (which is what rpc is trying to 
do)
                
> Convert WAL to pb
> -----------------
>
>                 Key: HBASE-7413
>                 URL: https://issues.apache.org/jira/browse/HBASE-7413
>             Project: HBase
>          Issue Type: Sub-task
>          Components: wal
>            Reporter: stack
>            Assignee: Sergey Shelukhin
>            Priority: Critical
>             Fix For: 0.95.1
>
>         Attachments: HBASE-7413-v0.patch, HBASE-7413-v1.patch, 
> HBASE-7413-v2.patch, HBASE-7413-v3.patch
>
>
> From HBASE-7201

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira

Reply via email to