The following is a transcript of an interview by Alan Shimel of DevOps.com with three leaders in the DevOps community — John Willis, Cornelia Davis, and Damon Edwards. This interview was conducted during the DevOps Enterprise Summit in London 2018. The topic at hand is the future of where Ops is heading and next gen ops.
Some of the responses have been edited for brevity and clarity. Watch the full interview here.
|Alan Shimel:||So, guys, little different than maybe some of the interviews we’ve done in the past. We’d like to do a deeper dive today and we’re going to to explore a topic. Our topic for today is next gen ops.
Damon, I’ve seen you champion this, so I’m going to give you the chance to kind of define it. What do we mean by next gen ops?
|Damon Edwards:||I don’t think of it like ‘a thing.’
I think it’s just saying how do we, in this modern world, go fast and the lock it down? We have this next generation of infrastructure platform technologies, so how do we operate this new stuff so that we can do the new business, and meet the new business demands?
Sort of the problem we’re solving is the development side of the house has had almost 20 years of agile sort of seeping into their brains. Ops hasn’t had that.
Ops has had since 1990 Itel seeping into their brains, which is a different way of viewing the world. So this modern operations idea is about how do we take these DevOps and agile ideas and break them down into the first principles and apply them to all this new technology, and these new business demands, so that we can transform operations the way we’ve done a good job of transforming dev in the DevOps world.
|Alan Shimel:||Cornelia, your world is more the developer’s side of things. What do you think of that?|
|Cornelia Davis:||Yeah, so it’s interesting. Because, as you know, I work for Pivotal, and when I started my work at Pivotal, I very much self-identified as a developer because that’s what I had been my entire career, you know, 30 years or something like that.
And then I got there, I was working on a platform as a service and realized that it was just as much an ops story as it was a developer story. So I’ve had that benefit of having come from the development side and then seeing some of the challenges we have on the ops side.
And to bring those two things together, so it’s kind of from the above and coming up from the below, the below being all the primitives that you have, the platforms that Damon mentioned, and from above being the requirements. And so it’s the requirements that are changing. We’re moving a lot faster and the developers are actually bringing things to us more quickly, us being operations folks, and I have new platforms. And that’s very intriguing.
|John Willis:||Can I chime in? Thank you. Now, you know what, one of the things I look for these signposts, right?
I think being part of IT Revolution has created an incredible network for all of us, being part of DevOps Days has created a network, being at Docker. So a lot of people, all my best friends are all taking jobs as VP of Next Generation Platform Development in large corporations. So I see this macro trend within the last two years. And we’ve had the pass, and some of that is still in play … But large corporations are starting to ask this question about, “Okay, in 2018, what do we need to look like in the next three years?”
This is more of a technology. I totally agree with both of you on the meta side, but I see Nick Weaver going over to Nike, and he’s a platform genius, right? And different people like that are taking these new kind of roles back in large corporations. So that tells me there’s this yearning in the enterprise to whatever past is prologue. And so then now I’m interested in what does that look like. I have my opinions, of course, but I think that’s to me the fabric of the future and how are they going to evolve?
|Cornelia Davis:||You know what I think is so interesting about that is that in the past, and still existing in enterprises today is a lot of the ‘plan, build, run.’ And you’re pointing out that what we’re seeing now is a shift from plan, build, run to platform teams. Platform engineering teams. And I was noticing like Jason Cox has the title of Systems Engineer. And so we’re starting to see titles like VP of Platform, VP of Systems Engineering. And that’s something that didn’t exist in IT before.|
|Alan Shimel:||So, let me play Devil’s Advocate if I can. You know for too long I heard from friends of mine in the IT space in general, security specifically, “DevOps is just some evil scheme to let Devs take control of Ops, and it’s a Dev-centric world.” But yet I’ve heard from devs, “Is DevOps giving all the power to Ops? The Ops control everything.” Maybe that’s always the sign of a real compromise, when both sides are a little unhappy. But, did we fall out of equilibrium here, where we need to have a new thing?|
|John Willis:||So there was this phenomenon that happened what, four years ago. It was called Docker. Right? And it actually gave developers for the first time, this ability to go so far into the stack without any oper — or minimal operational help. Right? At least that the promise and it was happening and it’s why Docker went incredibly viral the first two years, right? Developers, I believe, had the first time in their life, felt like they had more control in the delivery of what they were doing.
Life goes on and the interesting question about how do you manage that at scale comes along, then a technology called Kubernetes comes in, and now basically it’s a mess because nobody knows who owns that. So like with Docker the developers now have this freedom to drive infrastructure pretty deep into it and some places have a lot of control of the delivery for an infrastructure standpoint.
Then you get into how do you scale the architecture? Kubernetes starts getting incredibly popular, and now really from what I see, in most cases, the developers are responsible for maintaining Kubernetes and now they’re like kind of worse off than they were. And I’m pro-Kubernetes all the way. I mean, I think that is the future, I’ll say that right now.
But it is kind of messy right now, because what looked like a clean path just got awfully ugly all over again. And developers are like, “Well, this is a lot harder than the Docker’s stuff.” You know, especially when you get into stateful stuff, right? Anyway, so there is this weird, developers looked like they had this clearer path, and now we’re back kind of being messy all over again.
|Damon Edwards:||I mean, that battle’s always been fought, which is what’s application and what’s the platform? What’s the infrastructure? It’s like what we’re seeing is because the same operating model is still baked in everyone’s heads, so now we have this new technology, we’re still suffering from the old operating model, right?
And I think that now what we’re seeing with the platform, the smarter companies going sort of to platform organization, right? It’s no longer about operations. It’s about how do you bake the best practices of operations into as much of that platform as possible? And then give control to the rest of the organization through kind of pull-based self-service, but it’s that ability to build just the right amount of platform. But they have to re-think it.
On Operations we’ve got all these different silos, and we have ticket queues. And that’s how the Operations in a lot of organizations still works. So now, the challenge of the platform organization is to try to bake all of that, send the ticket queues into a self-service platform. And really offer that operations, but for the rest of the organization.
And that’s a new operating model. It totally, fundamentally breaks how people in operations think about their job and their specialties and their silos.
So I just think it hasn’t caught up yet. That’s kind of my point about dev, the agile side, has been seeping in for a long time, and ops, it’s still the Itel mindset that’s been baked in for the last 20, 30 years.
|Cornelia Davis:||What’s interesting about that is the contracts. So it’s what are the contracts that the platform is going to project up that the developers and the application teams are going to consume? And getting that contract right is really, really important. And when we did it in a smaller setting, when we were just doing Application Dial Tone, platform as a service, we came up with some good contracts that worked well.
Back to your [John’s] point of Kubernetes, that has completely made fuzzy that, what that contract is, and so I actually think of it not so much as a flattening, but what we’ve done is added an extra layer in there and we haven’t figured out what the right contract is for the application teams and for the developers.
|Alan Shimel:||Could be. I mean, certainly I think we’re in a state of flux. And there is that push-pull, right, between these different constituencies, if you will.|
|Damon Edwards:||Just wondering about that. I think the DevOps idea was great around the edges. And it was really fine until suddenly you have to change. Then the pushback comes. So like, “It sounds so awesome! We’re going to actually, dev, we’re going to take over, we’re going to run our own things.” And then they actually have to do it, and it’s like, “I didn’t really want to do this. Can I give this back?” And then operations side is like, “Aw, that’s great, like we’ll do those things” They actually have to change how you work? And then it’s like, “Oh, man.”
So I think a lot of the counter-reformation or pushback you’re getting from these different sides, ’cause it doesn’t make sense that they’re each saying different things. It’s just a reaction to that it’s actually cutting deep. Actually getting deeper into these organizations and affecting broader change, and that’s just hard.
|John Willis:||I want to ask Cornelia something, so I’m going to play the “I actually know this, but it’s for our listeners.” *Wink, wink.*
But can you go a little deeper into contracts? Because I think that’s incredibly interesting. Like the transition of where we kind of felt safe, and now we’re kind of unsafe?
|Cornelia Davis:||Yeah. So, for example, the thing I know best is Cloud Foundry, which is the product that I’ve been working on. And there, the fundamental contract between the app team and the platform team — the platform team provided a contract that said, “We’re going to give you Application Dial Tone.” What you as the developer do, or as the DevOps, the app team —|
|John Willis:||— kind of like the Heroku, like you put your date code here?|
|Cornelia Davis:||Exactly, you say, “Here’s my code, run it in the cloud for me, I do not care how.” And that was the contract. And then there’s a few other things that you can also control, the services you bind to the app and so on.
And now enter Kubernetes, and there’s this idea that the developers are going to interact with the Kuber —and actually in practice they’re interacting with the Kubernetes APIs. So they’re generating these YAML files, where they’re creating their pods, and their creating their policies, and they’re creating their storage volume claims, and all of these things. And what we don’t, what the industry hasn’t really grappled with is this idea that Kubernetes projects much more of an infrastructure Dial Tone.
And that’s probably not what application teams, DevOp teams, really need, because that’s actually going to slow them down. And so the platform—
|John Willis:||That was my meta point. You explained it, I love, you’re better, more technical, but that’s exactly my point.|
|Cornelia Davis:||And so now, the question is, I’m 100% convinced that those low-level primitives and Kubernetes, while I adore them, I think they’re fantastic and all the engines underneath and the control loops that do great things with those abstractions are fantastic. What we need to do as an industry is start to figure out how do we up-level those things? How do we layer on top of those, take advantage of those low-level primitives, but present an API, present a contract to the app/DevOps teams that allows them to stay efficient?|
|John Willis:||Do you think we’re on the path at least?|
|Cornelia Davis:||I think, very, very early days. For the most part, we’re just grappling with this idea for the moment.|
|Alan Shimel:||When you’re talking about something that so fundamentally changes the way we do things, is it realistic to expect it would be mature while it’s still immature?|
|Cornelia Davis/ John Willis:||Not at all.|
|John Willis:||But I think, Cornelia makes a great point. I was just saying vaguely that the developers are now confused, but you’re right. We had that Heroku-like contract. The purity of a pass was this Heroku-like contract where literally you put the code here, you hit a button, and it happens. And if you wanted some type of service with Cloud Foundry, I mean, I was a big fan of Cloud Foundry early on. I do think it has to evolve in the Kubernetes world —|
|John Willis:||And I know you agree. But you’re right. We’ve lost all of that, ’cause again, I’ll show in a second why they love Docker so much, because from the pre-How do you scale/ how do you orchestrate, it was create a mutable object, I’d push it and I’m good to go.|
|Damon Edwards:||So put it this way. Kubernetes might be 1.7. Figuring out how to actually work with this stuff is like 0.1.
We’re changing how you work, and that’s totally different, and that parts has to be figured out. Otherwise you can’t escape this sort of infrastructure dial-tone. You can’t escape those primitives, because then anything you bake in is just going to be … if everyone is doing ten different things, you’re going to satisfy one person, and nine people you’re going to upset.
So everything stays at that low level until people do the hard work and the false starts to figure it out, and then you can kind of jump up a level and say, “Okay, now we’ve kind of got a common idea on that. That’s gotta-“.
|Alan Shimel:||But nature abhors a vacuum, right? So, let me bring you full circle first — is that the vacuum that creates this New Ops, or Modern Ops, or whatever you want to call it? Is that Ops new mission? Or is it just a maturation of Kubernetes?|
|Damon Edwards:||No, I think the Modern Ops in the context at least we’re talking about in this conference is, yeah, we’ve got to figure that out. We have to figure out what to do so we can start to build those primitives and standardize things. So, the stuff that’s now application, the stuff that’s now like the things we do individually in a company, can kind of melt into the background. And you kind of saw that sort of the cloud primitives, like with the rise of especially AWS, people just kind of, “Alright, this is sort of how things are supposed to work, how we work with it.” We kind of had a sense for how that feels, and now it’s no longer really contentious. And it’s got to happen with the container, you know, the kind of Cloud Native we always wanted.|
|John Willis:||I think you described really well. We have to have two brains here. We have to have the operations meta, of how we think about operationalization in the new world, not Itel, but then there’s the technology at the same time.|
|Damon Edwards:||Yeah, it’s both.|
|John Willis:||It’s two pieces that have to be matured in balance|
|Damon Edwards:|| Yeah, it’s like a complex, complicated system of technologies and users and platforms, interacting with a complex system called human beings at the same time. But it’s a solvable problem. I think people love to fall into one or the other and kind of forget you’ve got to do both. Just giving everybody Kubernetes is not going to solve our problems, in fact, you’re encasing it in this concrete of these old ways.
But there’s people out there that think that. Just get my hands on these new tools, all this stuff’s going to go away. Then you see the chaos and the blow back when people do, try to do that.
|John Willis:||And that’s the curse of being an evangelist. For me being an evangelist, I want to push new edge technology, be very excited about it, then you’ve got to pull back a little to make sure people understand the complexities, so it’s a little push-pull of like, I’m very high on a lot of things going Kubernetes, but then you’re right, you have to kind of pull it back|
|Alan Shimel:||But that’s classic crossing-the-chasm kind of stuff, right? You want to be a 15% early adopter, well, that’s the price you pay.|
|Cornelia Davis:||And that’s the fun work, quite frankly.|
|Alan Shimel:||For some, for others, and in some organizations, depending how risk-averse they are. I was interviewing the folks from Capital One earlier, and the business person said, their job is measuring risk, and risk management. So when they look at new technologies, the risk is, is it mature enough? Is it going to create too much change or what-have-you, right? Now, some organizations like Capital One, will go out there. They’re way out there. Others are not as risk, right there, less risk-averse.|
|Cornelia Davis:||Yup, and that kind of goes back to the Kubernetes has been out for, in open source for three years. Quite frankly, it’s been out there for a long time. But back to Damon’s point, it’s 0.01 in the enterprise. And we do see some folks, we even saw Adidas this morning who said, “Yeah, we’re running Kubernetes clusters.” But that is still pretty nascent in most organizations, and even within the organization relatively small. And so, yeah, it’s very early days still.|
|Alan Shimel:||Yes, it is, but I think back to VMware adoption, and how I thought that happened overnight. It was five years, maybe six years, so really … I think Kubernetes might be half that in terms of adoption.|
|Cornelia Davis:||You know, I’m not sure. Because the thing about VMware and that adoption is that we weren’t actually asking people to change things too, too radically. We weren’t asking them to change their application structures. We were giving them the same exact primitives they were using before in a hardware setting, virtualized. And now, we’re actually shifting to Dial Tone.|
|John Willis:||We’re forcing change, to change everything.|
|Damon Edwards:||I mean how many companies had to hit their objective to move to VM, but a lot of these had one long-running VM on top of one physical machine.
Or, okay, maybe we’re going to consolidate, which means five long-running VMs on one physical machine. So you’re not really fundamentally changing how developers create, and you’re not really changing how you operate. You’re just adding another kind of layer of complexity in there, or complication, I guess you could say.
|John Willis:||But this has been a long and winding road. You know, initial cloud was kind of ‘lift and shift,’ and then we learned that was terrible.
You know one of the greatest things I love about Pivotal, I had this kind of epiphany the other day — I was at a customer and we were debating whether Kubernetes and Pass and what it should be, and one of the things they said, ‘the greatest thing that Cloud Factory did for us is it forced us to do twelve-factor app.’So I realized, you know, I’m not paying as much tribute to a technology that, like you went to it, and to go to it, you had to be twelve factor app. And that’s kind of brilliant. You leave some people behind. But my longer point is this whole Cloud Native, twelve factor, micro-server, whatever you want to call it today, we’ve been winding down this path.
When people started going to Amazon they realized you really had to refactor stuff, you had to decouple. You know, went to containers of costs. That was a beautiful collision of micro-services and compute infrastructure.
And what you guys have done with the twelve-factor app and pushing it. So I think Kubernetes is going to move a little faster, because I mean, a lot of the green-filled world in most enterprises is kind of hip to this model. So I think there’s still going to be a lot of legacy stuff.
But I do think it’s still ugly, it’s still the wild west.
|Damon Edwards:||Here’s a thought. So we talk a lot about history. Some say that every era of history thinks they’re the penultimate history. Like they’ve solved all the problems. So there’s this question of, we though VMs, this is the future, it’s going to solve all the problems. And then it was this kind of cloud thing. And then it was like, remember Open Stack anyone? Right? That was going to solve all our problems. And then it’s like ‘Docker, that’s it, it’s done. That’s going to solve our problems,’ right? And then it’s, ‘oh, no we need Kubernetes,’ right? And now there’s the server-less thing kind of lurking out there too.
So we’ve got to think about that, it might take three years for Kubernetes to kind of mature to more of a mainstream operation, but then are we going to look back and go, “Oh, this was the Open Stack of the mid-2010s?” And there’s something five years from now where like, “Oh, yeah, remember Kubernetes?” It’s just one of those where, do we foolishly think that like, this is it, when there’s a bunch of other things coming?
|John Willis:||I’ll hold the fool flag for a minute. I think, and there are some people that believe this, and I’m leaning in this camp, that Kubernetes could be the next Linux kernel. And I think Google has this idea that they can drive it down to be completely commodified, and that it becomes the base way you run any application. I mean, it’s naïve to say that giving what the world looked like four years ago. But I do have this fundamental belief that’s a possibility, that Kubernetes could be that layering, it could be like a 10-year thing, like a Linux.|
|Damon Edwards:||Right, or is there something else we need?|
|Alan Shimel:||Or think about how easy Linux was 20, 25 years ago. We probably had similar conversations then about the primitive —|
|John Willis:||But Linux isn’t a 20 year run? There’s a chance, I think, Kubernetes could be a 10 or 20 year run. If it becomes fabric based.|
|Alan Shimel:||Could very well be. Someone much smarter than me taught me that most of our changes are evolutionary, not revolutionary. And I think with Kubernetes you’re going to see a continued evolution, don’t look for the revolution.|
|John Willis:||I like that, yeah.|
|Damon Edwards:||Unless it’s the IT revolution. It’s on the wall. The writing’s on the wall.|