You are viewing a plain text version of this content. The canonical link for it is here.
Posted to hdfs-dev@hadoop.apache.org by "Hairong Kuang (JIRA)" <ji...@apache.org> on 2009/10/07 21:15:31 UTC

[jira] Created: (HDFS-681) A client should not send buffered data to datanodes on close of a file if the data have already been flushed

A client should not send buffered data to datanodes on close of a file if the data have already been flushed
------------------------------------------------------------------------------------------------------------

                 Key: HDFS-681
                 URL: https://issues.apache.org/jira/browse/HDFS-681
             Project: Hadoop HDFS
          Issue Type: Improvement
            Reporter: Hairong Kuang


Suppose we have the following lines of pseducode:
{noformat}
1: DFSOutputstream out = dfs.create("file");
2: out.write("abc");
3: out.hflush("abc");
4: out.close();
{noformat}

Line 3 will force "abc" to be pushed to datanodes but "abc" are still kept in the buffer for checksum purpose. Line 4 will force "abc" to be sent to datanodes again. If the client has an additional check if the buffered data have been flushed or not, we can eliminate this unnecessary packet. 

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.