[
https://issues.apache.org/jira/browse/HAWQ-1624?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16514003#comment-16514003
]
Sergei Lebedev commented on HAWQ-1624:
--------------------------------------
> hdfsPread cannot be implemented efficiently, that is why it is not in
> libhdfs3. For the ABI compatible reason, it is OK to me to implement it with
> seek and read.
Could you elaborate a bit on this, please? Do you mean that the client would
need to read (and discard) everything from current offset to the requested one?
Do you know if the JVM library does the same?
> Make libhdfs3 binary compatible with libhdfs
> --------------------------------------------
>
> Key: HAWQ-1624
> URL: https://issues.apache.org/jira/browse/HAWQ-1624
> Project: Apache HAWQ
> Issue Type: Improvement
> Components: libhdfs
> Reporter: Sergei Lebedev
> Assignee: Radar Lei
> Priority: Minor
>
> libhdfs3 is almost binary compatible with libhdfs. The missing bits are:
> * libhdfs3 does not define hdfsPread which nonetheless can be implemented in
> terms of hdfsTell, hdfsSeek and hdfsRead;
> * hdfsSync is called hdfsHSync in libhdfs3;
> * In libhdfs hdfsFlush flushes the internal buffer of the
> BufferedInputStream; in libhdfs3 it always flushes to HDFS, i.e. it does what
> hdfsHFlush should do.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
