[
https://issues.apache.org/jira/browse/HADOOP-13946?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15796303#comment-15796303
]
Chris Nauroth commented on HADOOP-13946:
----------------------------------------
Sorry to come late to the review, but I would have liked to see a mention of
how HDFS rename updates the modification time of both the source and the
destination folder (though not the modification time of the renamed file
itself).
Also, regarding this:
{quote}
Object stores have a significantly simpler view of time:
...
+ A file's modification time is always the same as its creation time.
{quote}
This makes it sound like this section covers all object stores, but the
statement about modification time is not necessarily true universally. For
example, on WASB, the {{FileStatus}} on read is always populated with the last
modified time field of the blob as reported by the Azure Storage service. I
think any kind of modification of the blob will result in a change in that
value. I specifically tested {{hadoop fs -chmod}} against WASB, and it updated
the blob's modification time, which is different from HDFS. Out-of-band blob
modifications directly through the Azure Storage service, bypassing the
{{FileSystem}} API, could be another source of perceived changes in the last
modification time.
I expect this is not consistent across services, and therefore it's unlikely we
can make accurate statements in the file system spec beyond just saying "it's
different." :-)
Please feel free to address this either by reverting and revising or filing a
new JIRA to track an addendum.
Thanks!
> Document how HDFS updates timestamps in the FS spec; compare with object
> stores
> -------------------------------------------------------------------------------
>
> Key: HADOOP-13946
> URL: https://issues.apache.org/jira/browse/HADOOP-13946
> Project: Hadoop Common
> Issue Type: Improvement
> Components: documentation, fs
> Affects Versions: 2.7.3
> Reporter: Steve Loughran
> Assignee: Steve Loughran
> Priority: Minor
> Fix For: 2.8.0, 3.0.0-alpha2
>
> Attachments: HADOOP-13946-001.patch
>
>
> SPARK-17159 shows that the behavior of when HDFS updates timestamps isn't
> well documented. Document these in the FS spec.
> I'm not going to add tests for this, as it is so very dependent on FS
> implementations, as in "POSIX filesystems may behave differently from HDFS".
> If someone knows what happens there, their contribution is welcome.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
