Hi
I have also been talking with diskimage-builder cores and they suggested we better start with the element on magnum. So I moved the change here:
https://review.openstack.org/#/c/296719/

About diskimage-size, I'll dig a bit more on it. It can be for two causes:
1. To generate the fedora-atomic image, I first install a fedora-minimal image, and install ostree packages and the dependencies there. Based on that, i execute the ostree commands, and then I update the grub config to boot with the new image. But the old contents are still there, so this is potentially increasing the size. I will take a look at possible cleanup solutions. 2. The way diskimage-builder compresses the image. It can be different from the one that Fedora is using, and less effective. I saw discussions in diskimage-builder about that, and there have been recent improvements as https://review.openstack.org/290944 . Is worth investigating a bit more as well.

Best
Yolanda

El 23/03/16 a las 18:10, Ton Ngo escribió:

Hi Yolanda,
Thank you for making a huge improvement from the manual process of building the Fedora Atomic image. Although Atomic does publish a public OpenStack image that is being considered in this patch:
https://review.openstack.org/#/c/276232/
in the past we have come to many situations where we need an image with specific version of certain software for features or bug fixes (Kubernetes, Docker, Flannel, ...). So the automated and customizable build process
will be very helpful.

With respect to where to land the patch, I think diskimage-builder is a reasonable target. If it does not land there, Magnum does currently have 2 sets of diskimage-builder elements for Mesos image and Ironic image, so it is also reasonable to submit the patch to Magnum. With the new push to reorganize into drivers for COE and distro, the elements would be a natural fit for Fedora Atomic.

As for periodic image build, it's a good idea to stay current with the distro, but we should avoid the situation where something new in the image breaks a COE and we are stuck for awhile until a fix is made. So instead of an automated periodic build, we might want to stage the new image to make sure it's good before switching.

One question: I notice the image built by DIB is 871MB, similar to the manually built image, while the public image from Atomic is 486MB. It might be worthwhile to understand the difference.

Ton Ngo,

Inactive hide details for Yolanda Robla Mota ---03/23/2016 04:12:54 AM---Hi I wanted to start a discussion on how Fedora AtomicYolanda Robla Mota ---03/23/2016 04:12:54 AM---Hi I wanted to start a discussion on how Fedora Atomic images are being

From: Yolanda Robla Mota <yolanda.robla-m...@hpe.com>
To: <openstack-dev@lists.openstack.org>
Date: 03/23/2016 04:12 AM
Subject: [openstack-dev] [magnum] Generate atomic images using diskimage-builder

------------------------------------------------------------------------



Hi
I wanted to start a discussion on how Fedora Atomic images are being
built. Currently the process for generating the atomic images used on
Magnum is described here:
http://docs.openstack.org/developer/magnum/dev/build-atomic-image.html.
The image needs to be built manually, uploaded to fedorapeople, and then
consumed from there in the magnum tests.
I have been working on a feature to allow diskimage-builder to generate
these images. The code that makes it possible is here:
https://review.openstack.org/287167
This will allow that magnum images are generated on infra, using
diskimage-builder element. This element also has the ability to consume
any tree we need, so images can be customized on demand. I generated one
image using this element, and uploaded to fedora people. The image has
passed tests, and has been validated by several people.

So i'm raising that topic to decide what should be the next steps. This
change to generate fedora-atomic images has not already landed into
diskimage-builder. But we have two options here:
- add this element to generic diskimage-builder elements, as i'm doing now
- generate this element internally on magnum. So we can have a directory
in magnum project, called "elements", and have the fedora-atomic element
here. This will give us more control on the element behaviour, and will
allow to update the element without waiting for external reviews.

Once the code for diskimage-builder has landed, another step can be to
periodically generate images using a magnum job, and upload these images
to OpenStack Infra mirrors. Currently the image is based on Fedora F23,
docker-host tree. But different images can be generated if we need a
better option.

As soon as the images are available on internal infra mirrors, the tests
can be changed, to consume these internals images. By this way the tests
can be a bit faster (i know that the bottleneck is on the functional
testing, but if we reduce the download time it can help), and tests can
be more reilable, because we will be removing an external dependency.

So i'd like to get more feedback on this topic, options and next steps
to achieve the goals. Best

--
Yolanda Robla Mota
Cloud Automation and Distribution Engineer
+34 605641639
yolanda.robla-m...@hpe.com


__________________________________________________________________________
OpenStack Development Mailing List (not for usage questions)
Unsubscribe: openstack-dev-requ...@lists.openstack.org?subject:unsubscribe
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev






__________________________________________________________________________
OpenStack Development Mailing List (not for usage questions)
Unsubscribe: openstack-dev-requ...@lists.openstack.org?subject:unsubscribe
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev

--
Yolanda Robla Mota
Cloud Automation and Distribution Engineer
+34 605641639
yolanda.robla-m...@hpe.com

__________________________________________________________________________
OpenStack Development Mailing List (not for usage questions)
Unsubscribe: openstack-dev-requ...@lists.openstack.org?subject:unsubscribe
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev

Reply via email to