[
https://issues.apache.org/jira/browse/HDFS-7889?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14383282#comment-14383282
]
Li Bo commented on HDFS-7889:
-----------------------------
Thanks Zhe’s careful review.
bq. Isn’t currentPacket always null after super.writeChunk?
{{super.writeChunk}} just writes a chunk(typically 512 bytes) to
{{currentPacket}}, a packet is typically about 64K bytes, so {{currentPacket}}
may not be null if it is not full.
bq. Shouldn't we call super.writeChunkImpl instead? We definitely don't want to
handle the tracing scope twice
You’re right. I will replace it with {{super.writeChunkImpl()}} in next patch.
bq. Maybe streamer() should be getStreamer(), and leadingStreamer() should be
getLeadingStreamer()?
Agree. {{getStreamer()}} and {{getLeadingStreamer()}} are better. I will fix it
in next patch.
bq.Can we get rid of sizeOfCellInBuffer, by using cellBuffers[i].array().length?
You mean {{ cellBuffers[i].position()}}? I will add a new function
{{getSizeOfCellInBuffer()}} to get rid of {{sizeOfCellInBuffer}}
bq.hookAfterRetrievingBlock looks complex and I'm not sure if it's the best way
to allocate new blocks. Can we move it to DataStreamer?
I think DataStreamer should not be aware of the existence of other streamers,
so I add a hook to DataStreamer. We can further discuss about a better
implementation of {{hookAfterRetrievingBlock}}
> Subclass DFSOutputStream to support writing striping layout files
> -----------------------------------------------------------------
>
> Key: HDFS-7889
> URL: https://issues.apache.org/jira/browse/HDFS-7889
> Project: Hadoop HDFS
> Issue Type: Sub-task
> Reporter: Li Bo
> Assignee: Li Bo
> Attachments: HDFS-7889-001.patch
>
>
> After HDFS-7888, we can subclass {{DFSOutputStream}} to support writing
> striping layout files.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
