Nodepool

Nodepool is a service used by the OpenStack CI team to deploy and manage a pool of devstack images on a cloud server for use in OpenStack project testing.

Overview

Once per day, for every image type (and provider) configured by nodepool, a new image with cached data is built for use by devstack. Nodepool spins up new instances and tears down old as tests are queued up and completed, always maintaining a consistent number of available instances for tests up to the set limits of the CI infrastructure.

Zookeeper

Nodepool stores image metadata in ZooKeeper. We have a three-node ZooKeeper cluster running on zk01.openstack.org - zk03.openstack.org.

The Nodepool CLI should be sufficient to examine and alter any of the information stored in ZooKeeper. However, in case advanced debugging is needed, use of zk-shell (“pip install zk_shell” into a virtualenv and run “zk-shell”) is recommended as an easy way to inspect and/or change data in ZooKeeper.

Bad Images

Since nodepool takes a while to build images, and generally only does it once per day, occasionally the images it produces may have significant behavior changes from the previous versions. For instance, a provider’s base image or operating system package may update, or some of the scripts or system configuration that we apply to the images may change. If this occurs, it is easy to revert to the last good image.

Nodepool periodically deletes old images, however, it never deletes the current or next most recent image in the ready state for any image-provider combination. So if you find that the ubuntu-precise image is problematic, you can run:

$ sudo nodepool dib-image-list

+---------------------------+----------------+---------+-----------+----------+-------------+
| ID                        | Image          | Builder | Formats   | State    | Age         |
+---------------------------+----------------+---------+-----------+----------+-------------+
| ubuntu-precise-0000000001 | ubuntu-precise | nb01    | qcow2,vhd | ready    | 02:00:57:33 |
| ubuntu-precise-0000000002 | ubuntu-precise | nb01    | qcow2,vhd | ready    | 01:00:57:33 |
+---------------------------+----------------+---------+-----------+----------+-------------+

Image ubuntu-precise-0000000001 is the previous image and ubuntu-precise-0000000002 is the current image (they are both marked as ready and the current image is simply the image with the shortest age.

Nodepool aggressively attempts to build and upload missing images, so if the problem with the image will not be solved with an immediate rebuild, image builds must first be disabled for that image. To do so, add pause: True to the diskimage section for ubuntu-precise in nodepool.yaml.

Then delete the problematic image with:

$ sudo nodepool dib-image-delete ubuntu-precise-0000000002

All uploads corresponding to that image build will be deleted from providers before the image DIB files are deleted. The previous image will become the current image and nodepool will use it when creating new nodes. When nodepool next creates an image, it will still retain build #1 since it will still be considered the next-most-recent image.

vhd-util

Creating images for Rackspace requires a patched version of vhd-util to convert the images into the appropriate VHD format. A package is manually managed at ppa:openstack-ci-core/vhd-util and is based on a git repo at https://github.com/emonty/vhd-util

Updating vhd-util

Should it become required to update vhd-util before Infra has a proper packaging repo or solution in place, one should clone from the git repo:

$ git clone git://github.com/emonty/vhd-util
$ cd vhd-util

Then perform whatever updates and packaging work are needed. The repo is formatted as a git-buildpackage repo with –pristine-tar. When you’re ready to upload a new verion, commit, create a source package and a tag:

$ git-buildpackage --git-tag --git-sign-tags -S

This will make a source package in the parent directory. Upload it to launchpad:

$ cd ..
$ dput ppa:openstack-ci-core/vhd-util vhd-util_$version_source.changes

Then probably pushing the repo to github and submitting a pull request so that we can keep up with the change is not a terrible idea.