This is the second in what I hope is a long-running series of interviews with the various OpenStack PTLs (Project Technical Leads), in an effort to better understand what the various projects do, what’s new in the Kilo release, and what we can expect in Liberty, and beyond.

If the audio player below doesn’t work for you, you can listen to the recording –> here or see the transcript below.

Rich: Hi, this is Rich Bowen. I am the OpenStack Community Liaison at Red Hat, and continuing my series on Project Technical Leads (PTLs) at OpenStack, I’m talking with Davanum Srinivas, who I’ve known for a few years outside of the OpenStack context, and he is the PTL for the Oslo project.

Oslo is the OpenStack Commons Library.

Thanks for speaking with me Davanum.

Dims: Thanks, Rich. You can call me Dims. You know me by Dims.

R: Yeah, I know. laughs

R: Give us a little bit of background. How long has the Oslo project been around?

D: We were doing things differently - we have a really old history, though. Some of the initial effort was started back in release B.

R: Oh, that long ago.

D: Yeah. So, what we were doing … why did Oslo come about? Oslo came about because way back when Nova started, we started splitting code from Nova into separate projects. But these projects were sharing code, so we were trying to figure out the best way to synchronize code between these sibling or child projects. So we ended up with a single repository of source code, called Oslo Incubator, where you would have the master copy, and everybody would sync from there, but what was happening was, everybody had their own sync schedule. Some people were contributing patches back, and it was becoming hard to maintain those patches. We decided we had to change the method the team worked. And we started releasing libraries, for specific purposes. What you saw in Kilo was a big bang explosion of a huge number of libraries from the Oslo team. Most of it was code in Oslo Incubator. We just had to cut the modules in a proper shape, sequence, with an API, with a correct set of dependencies, and that’s what we ended up releasing, one by one. Other projects started using things like oslo.config, oslo.messaging, oslo.db, oslo.log, and all these different libraries.

So that’s where we are today.

R: What is it that you’ll be doing in coming releases? Is it just the effort of identifying duplication, or are you actively developing new libraries.

D: Yes, we are. In Liberty, we have 5 new libraries coming up. Three of them start with ‘oslo.’ - like oslo.cache, oslo.reports, oslo.service. The other two do not have ‘oslo’ in their names. One is called Automaton, the other is called Futurist.

Automaton is a library for building state machines and things like that. Futurist is picking up some of the work that is done in upstream futures and things like that and making it available to all of the projects in the OpenStack ecosystem.

So these two projects can be used outside of Oslo, and outside of OpenStack by other people. That’s why they don’t have the Oslo name in them.

R: Do you see a lot of projects outside of OpenStack using these?

D: We hope so. For example, there is a project called Debt Collector, which we think fits well with how do you deprecate code and what are the primitives that we can provide that make it easy to mark their code being deprecated. A lot of people who work on Oslo also work in the overall Python ecosystem, so the hope is that if we design the libraries in such a way that it’s reusable, other people will pick some of our stuff up. But that’s a stretch goal. The real goal is to make sure that these libraries work well with the OpenStack projects.

And the other thing about these libraries is that they don’t drag in the Oslo baggage. For example, if you take oslo.db or oslo.messaging, they pull in a lot of other Oslo libraries, and these little libraries are designed so that they don’t drag in other Oslo libraries. So that’s the other good thing about these.

The way the Oslo project has been for the last few cycles has been that we are doing a lot of experiments in Oslo, which have been rolled out to other projects in the OpenStack ecosystem. Oslo is slightly different from other projects in the sense that people don’t work on it full time. So we have people work part time on it, but they focus mainly on other bigger projects. They come here when they need a feature or a fix, or things like that, and then they stay. We have a few cores who monitor reviews and bugs across all of the Oslo projects, but we also have people who specifically focus on individual little libraries, and they get core rights there.

People in the OpenStack ecosystem are experimenting with different structures, like for Neutron, they put everything into subrepos, and they experiement that way. And I think that what we are doing might be more useful to Nova, for example, and other projects, where they would like to keep a set of cores together, and also have subsystem maintainers and things like that.

Oslo is a good place to do this experimentation because the code base is not that huge, and the community is not that big as well. And the rate of churn, in terms of bugs and reviews, is not that high, as well. We are also experimenting with release versioning and things like that, and some of the things that you’ve seen recently driven by Doug, across the OpenStack ecosystem, we tested it here first, in terms of the versioning numbers, not having the Big Bang release, how do we do it, and things like that.

We lead the way.

The other big thing is, for example, Python 3.4 support. All the Oslo libraries have to be Python 3.4 complient first before they can be used and the other projects can adopt. So we end up being in the forefront trying to use libraries like websockify, or other libraries from the OpenStack ecosystem, which are not Python 3.4 complient, and we work with them to get them complient, and then use it in Oslo libraries, and then we roll it in. So we play an important role, I think, in the OpenStack ecosystem.

R: As PTL, is this a full time thing for you, or not? What are your responsibilities as PTL?

D: One of the very time consuming work is getting the releases out on a weekly basis. We try to make it predictable. At least, this cycle, we have started to make it predictable. Earlier, we heard complaints, we don’t know when you guys are releasing, so we were not ready, and things like that. So we have a good process this time around, where, over the weekend, we run a bunch of tests, outside of the CI, as well as inside our CI system, to make sure that the master of all Oslo libraries works well with Nova, Neutron, Glance, and things like that, and come Monday morning, we decide which projects need releases, based on what has changed in them for the last week or so. We follow the release management guildelines, working with Doug and Thierry, to generate the releases during the day on Monday.

After Tuesday you don’t have to worry about Oslo releases breaking the CI or your code. That has helped a lot of projects, especially Nova, for example. If they know there is a break late on Monday evening, they know who to ping, and we can start triaging the issue, and by Tuesday they are back on their feet.

That’s the worst-case scenario. Best-case scenario, no problem happens, and we are good to go. But there’s always one test case, or one scenario here or there. We always try to test beforehand, but like somebody said, it’s a living thing - the ecosystem, the CI system, is like an emergent behavior, it’s a living thing. It’s hard.