[
https://issues.apache.org/jira/browse/HADOOP-1707?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#action_12533875
]
dhruba borthakur commented on HADOOP-1707:
------------------------------------------
Thanks Doug for your comments.
1. My thinking is as follows: the client has a bunch of small buffers. Say 2
buffers each of size 16K. When the first buffer is full, it writes that buffer
to the first datanode in the pipeline. The client meanwhile can continue to
fill up the remaining buffer(s). The first datanode, on receipt of this buffer,
sends it to the next datanode in the pipeline and also writes it to its local
disk.
2. If a datanode fails to write a buffer to its disk, it is reported back to
the client. The client removes this datanode from the pipeline and continues to
write to the remaining two datanodes. The file in the bad datanode remains in
the "tmp" directory.
3. When the file is closed, the under-replicated blocks will be replicated by
the namenode.
> DFS client can allow user to write data to the next block while uploading
> previous block to HDFS
> ------------------------------------------------------------------------------------------------
>
> Key: HADOOP-1707
> URL: https://issues.apache.org/jira/browse/HADOOP-1707
> Project: Hadoop
> Issue Type: Bug
> Components: dfs
> Reporter: dhruba borthakur
> Assignee: dhruba borthakur
>
> The DFS client currently uses a staging file on local disk to cache all
> user-writes to a file. When the staging file accumulates 1 block worth of
> data, its contents are flushed to a HDFS datanode. These operations occur
> sequentially.
> A simple optimization of allowing the user to write to another staging file
> while simultaneously uploading the contents of the first staging file to HDFS
> will improve file-upload performance.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
