[
https://issues.apache.org/jira/browse/HADOOP-11687?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15179718#comment-15179718
]
Thomas Demoor commented on HADOOP-11687:
----------------------------------------
This should also fix another bug. The current clone implementation also copies
reponse headers (f.i. ETag, Accept-Ranges) from the source object to the new
copy request. AWS can handle these superfluous headers but other
implementations might not handle this well.
Looks OK at fist glance, slight concerns:
* have we got everything? Will review in more detail later.
* we'll need to keep this up to date with every AWS version bump (check if they
have added / changed something).
> Ignore x-emc-* headers when copying an Amazon S3 object
> -------------------------------------------------------
>
> Key: HADOOP-11687
> URL: https://issues.apache.org/jira/browse/HADOOP-11687
> Project: Hadoop Common
> Issue Type: Improvement
> Components: fs/s3
> Affects Versions: 2.7.0
> Reporter: Denis Jannot
> Attachments: HADOOP-11687.001.patch
>
>
> The EMC ViPR/ECS object storage platform uses proprietary headers starting by
> x-emc-* (like Amazon does with x-amz-*).
> Headers starting by x-emc-* should be included in the signature computation,
> but it's not done by the Amazon S3 Java SDK (it's done by the EMC S3 SDK).
> When s3a copy an object it copies all the headers, but when the object
> includes x-emc-* headers, it generates a signature mismatch.
> Removing the x-emc-* headers from the copy would allow s3a to be compatible
> with the EMC ViPR/ECS object storage platform.
> Removing the x-* which aren't x-amz-* headers from the copy would allow s3a
> to be compatible with any object storage platform which is using proprietary
> headers
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
