[
https://issues.apache.org/jira/browse/JCLOUDS-923?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14579610#comment-14579610
]
Jonathan Cobb edited comment on JCLOUDS-923 at 6/9/15 9:48 PM:
---------------------------------------------------------------
This is a decent workaround for now, but it means I now have to manage config
(which PROPERTY_ZONES to declare) based on inspecting Exceptions thrown when
launching instances and looking for telltale signs that a region is behaving
badly. Then I would have to re-check periodically to see if it's back to normal
and re-add it. It seems really hacky and more like something that should be
abstracted away by the library. A failure to fetch images from one region API
should not be a fatal error that prevents users from launching an instance in
any other region that is behaving normally. At worst, you cannot launch
instances in the region whose API is misbehaving, right?
>> Upon reflection, it's really not that bad. I can just pass in a single
>> region as a parameter every time. I end up not being able to reuse some
>> objects, but that's a much cleaner solution. Sorry about the above :)
was (Author: cobbzilla):
This is a decent workaround for now, but it means I now have to manage config
(which PROPERTY_ZONES to declare) based on inspecting Exceptions thrown when
launching instances and looking for telltale signs that a region is behaving
badly. Then I would have to re-check periodically to see if it's back to normal
and re-add it. It seems really hacky and more like something that should be
abstracted away by the library. A failure to fetch images from one region API
should not be a fatal error that prevents users from launching an instance in
any other region that is behaving normally. At worst, you cannot launch
instances in the region whose API is misbehaving, right?
> NovaComputeServiceAdapter.listImages fails if any zone API fails
> ----------------------------------------------------------------
>
> Key: JCLOUDS-923
> URL: https://issues.apache.org/jira/browse/JCLOUDS-923
> Project: jclouds
> Issue Type: Bug
> Components: jclouds-labs-openstack
> Affects Versions: 1.8.0
> Environment: Ubuntu 14.04 x64, OpenJDK 7, Rackspace cloud
> Reporter: Jonathan Cobb
>
> I have an integration test that launches an instance, runs a simple echo
> command on it to ensure SSH is working, then destroys the instance.
> This test passes for EC2 and DigitalOcean, and until yesterday it also passed
> for Rackspace. Today it is failing. I have tracked down the error to
> NovaComputeServiceAdapter.listImages
> In my test, I call TemplateBuilder.build to create a Template, then use that
> Template in a call to ComputeService.createNodesInGroup to create a node
> based on the template.
> Within TemplateBuider.build, it calls NovaComputeServiceAdapter.listImages
> and iterates over each zone to get a list of images for each zone.
> [Side question-- why does the TemplateBuilder look up all images for all
> zones, even though I have already configured it to use a particular zone (in
> this case, "IAD")?]
> In any case, even though I am trying to launch in IAD, the call to the SYD
> zone is failing (today, was working yesterday), and thus no instances can be
> launched at all, since the resulting HttpResponseException is uncaught and
> bubbles all the way up and out of TemplateBuilder.build. The full stack trace
> is here: http://cloudstead.io/downloads/jclouds-exception.txt
> I have made a simple patch that catches the HttpResponseException, logs an
> error, and continues loading images from other zones. The patch is here:
> http://cloudstead.io/downloads/NovaComputeServiceAdapter.java.patch
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
