[
https://issues.apache.org/jira/browse/WHIRR-220?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Tom White updated WHIRR-220:
----------------------------
Attachment: BlobCopyOperation.java
Overall, this is looking good. A few comments:
* Does this subsume WHIRR-222? If so, this patch could incorporate that code -
e.g. not checking for a null tarball, since whirr.<service>.tarball.url is
always set. Or would it be easier to commit WHIRR-222 then do this one?
* It would be better to add a blobstore provider property, rather than
inferring it from the provider. One may want to run on one one provider and use
a different provider's blobstore. It should be possible to specify different
credentials for compute and blobstore.
* ClusterSpec should really be little more than a configuration holder, with
little to no logic. The logic for uploading the files to the blobstore is
complex enough to warrant its own class (and unit tests).
* Should uploaded files be removed at cluster destroy time? This might be a
better approach for expanding a cluster, since the tarball wouldn't need
uploading again. However, there's a question of how to persist the tarball URL,
so perhaps we shouldn't worry about this yet.
> it's not trivial to push custom headers to bash install scripts
You can copy the remote resource to the local filesystem on each node using a
SaveHttpResponseTo statement. Then set the URL to a file:/// URL pointing to
the local file. The advantage to doing this would be that you can use a signed
request, which is portable across providers.
I've attached a class that I knocked up to do this which might be useful. (Note
that SaveHttpResponseTo is in jclouds beta 9, so a copy is not needed.)
> Support local tarball upload
> ----------------------------
>
> Key: WHIRR-220
> URL: https://issues.apache.org/jira/browse/WHIRR-220
> Project: Whirr
> Issue Type: New Feature
> Components: service/cassandra, service/hadoop
> Reporter: Tom White
> Assignee: Andrei Savu
> Fix For: 0.5.0
>
> Attachments: BlobCopyOperation.java, WHIRR-220.patch,
> WHIRR-220.patch, WHIRR-220.patch, WHIRR-220.patch, WHIRR-220.patch,
> WHIRR-220.patch
>
>
> If the installation tarball URL is a file:// URL then Whirr could upload it
> to a blobstore and automatically download it onto the instances before
> running the install script. This change would make it very easy to test
> development versions of Hadoop, or other services.
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira
