The OpenStack Developer Summit has just concluded in Portland, OR, and by many measures was an enormous success. Enormous because there were a reported >2800 people in attendance. I never heard a breakdown of how many of those were developers but I'd guess that that number was up too. The rooms were generally full but usually only a handful of people actively participated in most sessions. And speaking of full rooms, most of the 'project-formerly-known-as-Quantum' sessions were SRO. I skipped those due to other interests but the word is none of the network vendors did. Most of my time was spent looking for things that affected my three primary projects: DevStack, Grenade and OpenStackClient.
There was no dedicated session to DevStack but it did not escape mention. Most of the talk I heard about it was in conjunction with its use in the CI gating tests and even then there wasn't much. The hallway talk seemed to always revolve around adding functionality that really doesn't belong in DevStack. It was nice to be able to explain in person that the requested feature wasn't core to DevStack's purpose and how they can easily hook in to stack.sh to add the feature.
Grenade had a session on Tuesday led my moi that concentrated on the project upgrade strategy and gate issues. Fortunately the discussion around run time was short, Grenade's <5 minutes is dwarfed by Tempest so nobody really cares.
The upgrade strategy noted the change from rewriting new config files based on the samples included in the project repo to only making upgrade-necessary changes to the existing config files. The current Grenade implementation does this for all but Keystone and Swift. The consensus in the room was that both of those should be fixable with some knowledgable input.
There are some improvements that can be made to devstack-vm-gate*.sh to simplify the configuration that is actually run in the gate test. The goal here is to make the gate closer to the configuration that is typically used by devs.
The first (and last) two sessions on OpenStackClient at previous summits were dominated by talk about what it should be and how people wanted to use it. The project is at a stage where it needs to get some things completed so we can make a 0.1 pre-alpha release. We grabbed an Unconference slot on Wednesday to get the primary developers together and decide on a few things, like the tasks to be completed prior to the 0.1 release. There were others present who I'm afraid we left wondering where the presentation was.
Command parity with the project clients, save for that name-that-used-to-start-with-a-q project's client, is the major task for release. Compute is not complete (dtroyer), Object (Swift) is non-existant, and the remainder are close but need an audit to see what may be missing since they were first implemented.
The major news in the command front is the change of my beloved "verb object" form to "object verb".  There is finally a good reason to change: to facilitate bash tab-completion. I'm taking dhellmann's word on this as tab-completion has not been done yet. And I'll be dipped if I can't come up with anything to counter it. So to ease my pain, Doug volunteered to make the change.
This session kept getting sidetracked by people wanting to make the common image format include the contents of the system in addition to the format of the disk image file. smoser did his best to remind everyone that Glance's concern with image formats stopped at the file blocks that contain filesystem bits. The contents of those blocks is off-limits.
It should be remembered that in previous summits the opening of user-submitted disk images to inject files and user data was specifically removed from Nova due to the potential security issues involved. Also there is the matter of knowing just what to do to a random filesystem. This led to the adoption of cloud drive as an alternative to AWS-style metadata service.
The intention of the session was to work out details of exchanging images between clouds, including the disk format used. Apparently there are OpenStack implementations that do not support the qcow2 format because otherwise that is what should be used as it has both compression and sparse capabilities. Of course raw would also work but let's be realistic here.
This is one session that may have some positive effects on DevStack as a side effect of documenting OS-level package dependencies for each project. The session primarily concerned opestack/requirements as a gate for Python dependencies and the use of an OpenStack-specific PyPi mirror.
The interesting bit to me is the need to handle non-pip-installable dependencies. DevStack already maintains some of this information for distros and releases. Moving this into the project repos would allow it to be used by other processes and get DevStack completely out of tracking project dependencies. It remains TBD as to who actually does the install.
A recent issue with one of the OpenStack projects that once had a code name with a 'Q' and two 'u's is no longer calling itself that. I'm not touching the reasons why, that is for others to explain. What I am mostly interested in is the amount of work required to change a projects name, even when it is done at phase boundaries such as moving into and out of incubation.
This means that choosing a project name has more pressure to get it right the first time than before. I am assuming here that the common sentiment expressed in the session to not stop using project names is the path chosen even if the public use of project names is ramped down.
Two of the three projects that I am deeply involved in (read as 'would do more than a little of the name changing work') have names that are invented terms so the chance that there might be confusion in the industry is low. And a different set of two of those projects are primarily used internally and should get little public attention.
That does it for the highlights of my week in Portland. I'll leave it to others to summarize the rest, save for pimping Terrence and Bertram's latest video that provided an energetic open to the keynotes. Stay tuned for some deeper thoughts on a few of the topics mentioned here.
|||Sorry VMS DCL fans, I tried.|