[
https://issues.apache.org/jira/browse/HDFS-7800?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14345343#comment-14345343
]
Steve Loughran commented on HDFS-7800:
--------------------------------------
bq. Wow, that's a fantastic piece of documentation.
Thank you
bq. Why is this seperate from the javadoc I mentioned
it's trying to be independent of the current HDFS impl, even though that
defines the semantics.
-now that is something worth doing
bq. can the javadoc atleast point to the more detailed url?
that would be good.
incidentally, HDFS's new truncate operation has apparently changed the rule on
append; you will be able to append incomplete blocks in future. So the FS spec
are obsolete too
> Improve documentation for FileSystem.concat()
> ---------------------------------------------
>
> Key: HDFS-7800
> URL: https://issues.apache.org/jira/browse/HDFS-7800
> Project: Hadoop HDFS
> Issue Type: Task
> Affects Versions: 2.2.0, 2.6.0
> Reporter: Steve Armstrong
> Attachments: HDFS-7800-1.patch
>
>
> This is a documentation request.
> [FileSystem.concat()|https://hadoop.apache.org/docs/current/api/org/apache/hadoop/fs/FileSystem.html]
> says it will "Concat existing files together". It seems to be a
> Namenode-only operation though, mapping the data blocks into a single file.
> This means:
> # The destination must exist
> # The destination must be non-empty
> # The destination must have it's last block exactly full
> # All but the last of the source files must have their last block full
> # All the source file will be deleted by this operation
> HDFS-6641 brought up some of these limitations, but was closed as not a
> problem. I think the javadoc should be improved so it's clear this function
> was never intended to work the same as a general purpose file concatenation.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
