[
https://issues.apache.org/jira/browse/HDFS-14844?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16928745#comment-16928745
]
Íñigo Goiri commented on HDFS-14844:
------------------------------------
The description is a little generic when saying "The buffer size of DFSClient
remote read.".
We should be a little more specific and mention what this actually buffers.
This is kind of a readahead buffer that reads at least this much data from the
DN, right?
Then the value is 8KB by then mentions that 512 is reasonable.
It should say that for regular sequential reads a value like 8KB should be used.
However, if the access pattern is more random (HBase style, I guess) we should
use smaller values.
As we are changing the path, a minor style request:
{code}
int bufferSize = configuration.getInt(
DFS_CLIENT_BLOCK_READER_REMOTE_BUFFER_SIZE_KEY,
DFS_CLIENT_BLOCK_READER_REMOTE_BUFFER_SIZE_DEFAULT);
{code}
I feel this is usually more readable.
> Make buffer of BlockReaderRemote#newBlockReader#BufferedOutputStream
> configurable
> ----------------------------------------------------------------------------------
>
> Key: HDFS-14844
> URL: https://issues.apache.org/jira/browse/HDFS-14844
> Project: Hadoop HDFS
> Issue Type: Improvement
> Reporter: Lisheng Sun
> Assignee: Lisheng Sun
> Priority: Minor
> Attachments: HDFS-14844.001.patch, HDFS-14844.002.patch
>
>
> details for HDFS-14820
--
This message was sent by Atlassian Jira
(v8.3.2#803003)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
