Posted on

Scaling DevOps and the Revival of Operations

Operations is going through a renaissance right now. With the move to cloud, the increasing amount of automation, and the increasing importance of automation, Ops as we know it is reinventing itself out of necessity. Infrastructure is becoming more and more sophisticated—and commoditized—and practices are just now starting to grow up around that. So while some worry about robots taking our jobs, the reality is more about how automation will help augment us to build better software and focus on higher-value things. It’s not so much about the distant future—whatever that may hold—so much as it is about the next five to ten years, what Operations looks like in that timeframe, and why I think it has to retool.

When we think about traditional Operations, we probably think about hardware and servers, managing networks and databases, application servers and runtimes, disaster recovery, Nagios checks, as well as the business side—vendor management, procurement, and so on. Finally, we have applications built on top by development teams.

We have a nice, clean separation—developers focus on building features and products, and Ops focuses on making sure the lights stay on. Of course, we know the reality is this separation also creates a lot of problems, so DevOps was borne out of this as a way to bring these two groups into alignment by improving communication and feedback loops.

Now, with the move to cloud, many of these traditional Ops functions are effectively being outsourced to cloud providers, i.e. the idea of NoOps. We get unprecedented elasticity and on-demand compute with far less overhead than we ever had before—shrinking procurement time from days or weeks to seconds or minutes.

What this leaves is a thin but important slice between Google or Amazon and those products built by developers—the glue, essentially, between cloud and product. I call this NewOps (which I use facetiously in reference to NoSQL/NewSQL), and it’s the future of Ops. This encompasses infrastructure automation, deployment automation, configuration management, logging, monitoring, and many other things. When Marc Andreessen said software is eating the world, he really meant it. The future of Ops—and many other things—is software. It’s killing the boring, repetitive things we really don’t want to be doing anyway and letting us shift our focus elsewhere.

Certainly, automation is nothing new and is, I think, an important part of DevOps, so I’m going to explain what I mean by NewOps and why I’m distinguishing it. I also don’t want to mischaracterize by having these neatly delineated Ops models. The truth is, your company doesn’t just one day graduate and gets its DevOps diploma. Instead, it might evolve through various manifestations of these different models. DevOps is a journey, not a destination in and of itself.

I like to think of a DevOps scale of automation, from manual provisioning all the way to fully self-service. Next, I add a second dimension, org size, from the smallest startups to the biggest enterprises.

Scaling DevOps

Scaling a business is probably one of the hardest things a company has to go through. In particular, dealing with the problem of silos. They happen at every company as it grows, but why is it that silos form in the first place?

Many companies start with a “DevOps” approach, often out of necessity more than anything. As a small startup, we can’t afford to have dedicated developers, QA, Ops, and security people. We just have people, and those people wear many different hats. Developers might be pushing their own code to production. They might even be managing the infrastructure that code runs on. There’s probably not a lot of stability, probably a lot of risk, and probably not a whole lot of thought towards controlling costs.

But as the product scales, we specialize. And as the business scales, we add various safety checks, controls, and processes. Developers write code, Ops people run it, QA gets blamed for defects, security blocks everything, and management wonders why nothing gets shipped.

And so we end up in the top left-hand quadrant with Ops as gatekeepers. Ops is fighting for stability and, at the same time, devs are basically fighting for change. More or less, we have a stable, cost-controlled, risk-averse environment—hopefully. But we also have a significant delivery and innovation bottleneck.

Specialization is good! But misalignment is not good. The question is, then, how do we scale specialization? Cross-functional teams come to mind. After all, DevOps encourages cooperation! We add an Ops engineer to each team, and maybe a reliability engineer, and perhaps a few extra for on-call backup, and of course a QA engineer too. Problem solved, right?

But hold on. What if we have 40 development teams? And all those teams are doing microservices. And, of course, all of those microservices are special snowflakes each with their own stacks, infrastructure, databases, and so on. This quickly gets out of control, but moreover, that’s a lot of teams and specialized roles on those teams. That’s a lot of headcount which equates to a lot of hiring and a lot of time and money. If you’re Google and you can just throw money at the problem, this might work out okay. For the rest of us, it might not be such a realistic option.

We go back to the drawing board and again ask ourselves how do we scale specialization? My thought to how we do this is with vision and product.

A vision is simply a mental image of what the future could be like. It enables independent decision making and alignment. Vision allows all of those teams, and the people on those teams, to make decisions without having to constantly coordinate with each other. Without vision, you’re just iterating to nowhere fast.

But vision without execution is just hallucination. Products are how we scale execution. Specifically, this idea of Operations through the lens of product, which I’ll describe after showing the parallel with what’s happening in QA.

In a lot of engineering organizations, many QA roles have been quietly disappearing. I think what’s happening is this evolution of QA, particularly, this shift from being test-focused to tools-focused.

We can look at companies like Amazon and Microsoft who popularized the SDET (Software Development Engineer in Test) model. These companies recognized that having a separate QA and development group causes a lot of problems, just like how having a separate Ops group does. We end up with SDEs (Software Development Engineers) who still focus on the development aspects of building software and SDETs who focus on the quality aspects, but rather than having two wholly separate groups, we just have development teams with SDETs embedded in them.

More recently, Microsoft moved to what they call a “Combined Engineering” model—effectively combining the SDE and SDET roles into a single role called a Software Engineer. Software Engineers write the product code, test code, and tools code needed to deliver their service. They are responsible for everything. Quality is a core concern of software development anyway.

Software Engineers write the code, unit tests, and integration tests. Those tests run in CI. The code moves through a CD pipeline before finally going out to production in some fashion. QA teams are shrinking, but what’s growing are the teams building the tools—the CI environments, the CD pipelines, the automated testing frameworks, the production tooling and automation, etc. The same is becoming true of Ops.

This is what I mean by “Operations through the lens of product.” The build, release, deploy automation, configuration management, infrastructure automation, logging, monitoring—these are all products.

Constraints often make problems easier. At Workiva, as we were struggling through that scaling phase, we placed a constraint on ourselves. We capped our infrastructure engineering headcount at 15% of R&D. This forced us to solve the problem using technology, and technical problems tend to be easier than people problems. In effect, this required us to productize our infrastructure. In doing so, we scaled. We controlled costs. We kept our headcount in check. We reduced risk. We accelerated development. Ultimately, we delivered value to customers faster, going from about three to four releases per year to multiple releases per day. In the end, this is really the goal of DevOps—to deliver value to customers continuously and to do it rapidly and reliably.

Rethinking Ops

It’s time we start to rethink Operations because clearly this model of Ops as cluster or infrastructure admins does not scale. Developers will always out-demand their capacity to supply. Either your headcount is out of control or your ability to innovate and deliver is severely hamstrung. Operations becomes this interrupt-driven thing where we’re just fighting fires as they happen. Ops as masters of production usually devolves to Ops becoming human incident routers, trying to figure out what team or person can help resolve problems because, being responsible for everything, they don’t have the insight to fix it themselves.

Another path that many companies take is Platform as a Service. Workiva is an example of this. For a very long time, Workiva didn’t have a traditional Ops team because the Ops team was Google. The first product was built on Google App Engine. This helped immensely to deliver value to customers quickly. We could just focus on the product and not the surrounding operational aspects, but there is a very real innovation bottleneck that comes with this.

The idea of “Ops lock-in” can be a major problem, whether it’s a PaaS like App Engine locking you in or your own Ops team who just isn’t able to support the kind of innovation that you’re trying to do.

My vision for the future of Operations is taking Combined Engineering to its logical conclusion. Just like with QA, Ops capabilities should be embedded within development teams. The reality is you can’t be an effective software engineer today without some Ops skills, and I think every role should be working towards automating itself out of a job. Specifically, my vision is enabling developers to self-service through tooling and automation and empowering them to deploy and operate their services.

The knee-jerk reaction to this idea is usually fully embracing Infrastructure as a Service, infrastructure as code, and giving developers freedom—and usually the consequences are dire. The point here is that the pendulum can swing too far in the other direction. This was a problem for a brief period of time at Workiva. As we were building new products off of App Engine, developers had this newfound freedom, so teams all went different directions introducing new tech, new infrastructure, new services, and so forth. It was a free-for-all, an explosion of stuff, and the cost explosion that comes with it.

There has to be some control around that, so we tweak the vision statement a bit: enabling developers to self-service through tooling and automation and empowering them to deploy and operate their services…with minimal Ops intervention. We have to have some checks and balances in place.

With this, Ops become force multipliers. We move away from the reactive, interrupt-driven model where Ops are masters of production responsible for everything. Instead, we make dev teams responsible for their services but provide the tools they need to actually own their systems end-to-end—from the code on their laptops to operating it in production.

Enabling developers to self-service through tooling and automation means treating Ops as a product team. The infrastructure automation, deployment automation, configuration management, logging, monitoring, and production tools—these are all products. It’s these products that allow teams to fully own their services. This leads to empowerment.

I have this theory that all engineering organizations operate in this fashion which I call pain-driven development. As a company grows, it starts to develop limbs—teams or silos. Each of these limbs has its own pain receptors. Teams operate in a way that minimizes the amount of pain that they feel, it’s human instinct. We make locally optimal decisions to minimize pain and end up following a path of least resistance.

Silos promote pain displacement, which results in a “bulkhead” effect. Product development feels the pain of building software, QA feels the pain of testing software, and Ops feels the pain of running software. This creates broken feedback loops. For instance, developers aren’t feeling the pain Ops is feeling trying to run their software. We just throw things over the wall and it becomes an empathy problem.

This leads to misaligned incentives because each team will optimize for the pain that they feel. How do you expect developers to care about quality if they’re not on the hook? Similarly, how do you expect them to care about operability if they’re not on the hook? Developers won’t build truly reliable software until they are on-call for it and directly responsible. However, responsibility requires empowerment. You can’t have one without the other. You can’t ask someone to care about something and fix it without also giving them the power to do so. Most Ops teams simply haven’t done enough to empower and offload responsibility onto development teams.

Products enable ownership. We move away from Ops as masters of production responsible for everything and push that responsibility onto dev teams. They are the experts for their services. They are best equipped to deal with problems that arise. But we provide the tools they need to diagnose and resolve those problems on their own.

Products maintain control through enablement—enabling teams to follow best practices for builds, testing, deploys, support, and compliance. Compliance and other SDLC requirements have to be encoded into the tools and processes. These are things developers won’t empathize with or simply won’t understand. Rather than giving them a long list of things they have to do, we take as many of those things as we can and bake them into our products. If you use these tools or follow these processes, you’ll get a lot of this stuff for free. This reduces risk and accelerates development.

Similarly, we can’t allow all of the special snowflakes to happen. We have to control that explosion of stuff. To do this, we use pain-driven development to our advantage by creating paths of least resistance. Using standardized patterns, application shapes, and infrastructure services, we can setup “paths” to both make it easier to reach production and meet the goals of the business. As a developer, if you follow this path, your life will be a lot easier and you’ll feel less pain. If you deviate from that path, things get much harder—and painful.

We end up with a set “menu” of standard application shapes and infrastructure. If teams want to deviate and go off-menu, it’s on them to make a case for it. For example, if I want to introduce Erlang into our stack, it’s on my team and me to present the case for that. Part of this might mean we help build and maintain the tools needed to support that. If there is a compelling enough case or enough teams are making similar asks, we can start to standardize new shapes.

Note that we aren’t necessarily mandating technologies, but we’re leveraging pain-driven development to work in our favor.

Products in Practice

Next, I’m going to look at this idea of Operations through the lens of product in a bit more detail. We’ll see what this might actually look like in practice, again using Workiva as a bit of a case study.

Below is the high-level flow that I think about, from code on laptop to code in production.

Starting with the Build and continuous integration stage, this workflow tends to look something like the following. A developer pushes a change to a branch in a code repository, e.g. GitHub. This triggers a few things to happen. First, the build process, which runs unit/integration tests and builds artifacts. This, in turn, might trigger a QA and/or compliance process. At the same time, we have code reviews happening. All of these processes provide feedback to the developer to quickly iterate.

Workiva has a lot of automated processes built into the developer workflow, some off-the-shelf and some built in-house. For example, when a PR is opened, a security scanner runs which does static analysis and looks for various security vulnerabilities. This can flag a security review when a closer look is needed. Likewise, there is code coverage, automated builds, unit tests, and integration tests, Docker image builds, and compliance checks. The screenshots below come from an open-source repo showing some of these products in practice.

For compliance reasons, Workiva requires at least one other person sign-off on code changes. GitHub provides pretty good support for this. Code reviewers provide their feedback, developers work through that feedback, and, once satisfied, reviewers give their “plus one.”

The screenshot below shows some of the automated processes Workiva relies on in the developer workflow: Travis CI, Codecov, Smithy (which is Workiva’s internal build system), Skynet (automated testing), Rosie (automated compliance controls, e.g. do you have code reviews, security reviews, other SDLC compliance requirements?), and Aviary (the security scanner). Once all of these have passed, the PR is automatically labeled with “Merge Requirements Met” and the change can be merged into master.

There are a couple things worth pointing out with this workflow. First, the build plan is part of the code and not baked into some build tool. This allows dev teams to fully control their builds. Second, you noticed that Workiva has very deep integration with GitHub. This has allowed them to build automated controls into the development process, which speeds up the developer’s workflow while reducing risk.

Next, we move on to the Release stage. This flow looks something like the following:

The developer tags a branch for release, which triggers a build process for creating the artifact. This may have a QA process which then promotes the artifact to a development artifact repository. As you may have noticed, Workiva has a lot of compliance requirements since they deal with companies’ pre-financial data, so there is typically a sign-off process at various stages involving different parties like Release Management, QA, Security, etc. Depending on your compliance controls, this might just be clicking a button to promote an artifact to a production repository. From there, it can actually be deployed to a production environment.

With this workflow, artifact tagging, building, and promotion is all automated. It’s also important we have processes around security. Container and machine image auditing is automated as well as security patching for OS updates, etc. For example, this workflow might use something like Packer to automate AMI building. Finally, the artifact sign-off is streamlined for the various parties involved, if not fully automated.

Now we’re ready to actually deploy our application. This is a key part of self-service and “owning” a product. This allows a team to configure their application and, ideally, deploy it themselves to production. Initially, this might be handled by a Release Management team who actually clicks the deploy button, but as you become more confident in your processes and your tools become more mature, more of this responsibility can be pushed onto the development teams.

This is also where control comes into play. For instance, I may be allowed to configure my application to use 1GB of RAM, but if I need 1TB, I may need to get additional sign-off.

Self-service deploys and self-service configuration—with guard rails—are an important part of continuous deployment. Additionally, infrastructure provisioning should be automated. No more submitting tickets for a nameless Ops person to provision and configure servers, VMs, or other resources—no ticket-driven development.

I’ve been deliberate about not prescribing particular solutions for some of these problems. You might be using Kubernetes or ECS to orchestrate containers, it doesn’t really matter. These should mostly be implementation details. What does matter, though, is having good abstractions around certain implementation details. For example, Workiva was meticulous about building some layers around workload scheduling. This allowed them at one point to switch from using Fleet to ECS to manage containers with virtually no impact to developers. With the amount of churn that happens in tech, it’s important not to tie yourself too heavily to any one implementation. Instead, think about the APIs you expose for your infrastructure and consider those the deliverable.

Finally, we need to operate our service in production, another important part of ownership. There are a lot of products here, so we’ll just look at a cross section.

Logging is arguably the most important part of how we figure out what is happening in our systems. For this reason, Workiva built structured logging and metrics specs and language libraries implementing these specs. As a developer, this made it easy to simply pull in the library for your language and get structured, contextual logging for free. The other half to this was building out a data pipeline. Basically all metadata at Workiva went into Amazon Kinesis, including logs, metrics, and traces. First, this allowed us to reuse the same infrastructure for all of this data, from the agents running on the machines to the pipeline itself. Second, it allowed us to fan this data out to different backend systems—Splunk, SumoLogic, Datadog, Stackdriver, BigQuery, as well as various internal tools. This is probably one of the most important things you can do with your infrastructure.

Other continuous operations tools include telemetry, tracing, health checks, alerting, and more sophisticated production tools like canary deploys, A/B testing, and traffic shadowing. Some might refer to these as tools for testing in production. Realistically, once you reach a certain scale, testing in production is the only real alternative to the proliferation of deployment environments.

It’s worth mentioning that you do not need to build all of these products yourself. In fact, you shouldn’t. Many off-the-shelf solutions just need glued together. However, I’ve also come to realize that it’s often the “glue” that is important. That is to say, taking some large, commercial off-the-shelf solution and introducing it into a company is frequently rife with headaches. It’s like Jira, a big Frankenstein product that attempts to solve everyone’s problems and, in doing so, solves none of them particularly well. This is why I tend to favor small, modular solutions that can be composed. But it also highlights why there is a cultural aspect to this.

If you think the solution to your ailments is some magical product—maybe a CI/CD pipeline or Kubernetes or something else—you’re misguided. If anything, most problems are cultural, not technical in nature. Technology will not fix your broken culture! The products are not the endgame, they are a means to an end. And the products need to fit the company, its culture, its architecture, and its constraints. It’s tempting to take something you see on Hacker News and introduce it into your stack, but you have to be careful.

Likewise, it’s tempting to dive straight into the deep-end, automate everything, and build out a highly sophisticated infrastructure. But it’s important to start small and evolve over time. My approach to this is get the workflow correct, start manual, then automate more and more over time.

Wrapping Up

Specialization leads to misalignment and broken feedback loops, but it’s an important part of scaling a business. The question is: how do we specialize?

We know the traditional Ops model does not scale—devs will always out-demand capacity in this reactive model. Not only this, the siloing creates an empathy problem. DevOps attempts to help with this by tightening feedback loops and building empathy. NewOps takes this further by empowering teams and providing autonomy. It’s not a replacement for DevOps, it’s an evolution of it. It’s applying a product mindset to the traditional Ops model.

The future of Ops is taking Combined Engineering to its logical conclusion. As such, Ops teams should be redefining their vision from being masters of production to enablers of production. Just like with QA, Ops capabilities need to be embedded within dev teams, but the caveat is they need to be enabled! This is the direction Operations is headed. Software is eating the world, which means both up and down the stack. NewOps treats Ops like a product team whose product, effectively, is infrastructure. It’s creating guard rails, not walls—taking SDLC and compliance controls and encoding them into products rather than giving devs a laundry list of things, having them run the gauntlet through a long, drawn-out development process, and having a gatekeeper at the end.

Offloading responsibility helps correct and scale feedback loops. In my opinion, this is how we scale specialization. Operations isn’t going away, it’s just getting a product manager.

Posted on

More Environments Will Not Make Things Easier

Microservices are hard. They require extreme discipline. They require a lot more upfront thinking. They introduce integration challenges and complexity that you otherwise wouldn’t have with a monolith, but service-oriented design is an important part of scaling organization structure. Hundreds of engineers all working on the same codebase will only lead to angst and the inability to be nimble.

This requires a pretty significant change in the way we think about things. We’re creatures of habit, so if we’re not careful, we’ll just keep on applying the same practices we used before we did services. And that will end in frustration.

How can we possibly build working software that comprises dozens of services owned by dozens of teams? Instinct tells us full-scale integration. That’s how we did things before, right? We ran integration tests. We run all of the services we depend on and develop our service against that. But it turns out, these dozen or so services I depend on also have their own dependencies! This problem is not linear.

Okay, so we can’t run everything on our laptop. Instead, let’s just have a development environment that is a facsimile of production with everything deployed. This way, teams can develop their products against real, deployed services. The trade-off is teams need to provide a high level of stability for these “development” services since other teams are relying on them for their own development. If nothing works, development is hamstrung. Personally, I think this is a pretty reasonable trade-off because if we’re disciplined enough, it shouldn’t be hard to provide stable APIs. In fact, if we’re disciplined, it should be a requirement. This is why upfront thinking is critical. Designing your APIs is the most important thing you do. Service-oriented architecture necessitates API-driven development. Literally nothing else matters but the APIs. It reminds me of the famous Jeff Bezos mandate:

  1. All teams will henceforth expose their data and functionality through service interfaces.

  2. Teams must communicate with each other through these interfaces.

  3. There will be no other form of interprocess communication allowed: no direct linking, no direct reads of another team’s data store, no shared-memory model, no back-doors whatsoever. The only communication allowed is via service interface calls over the network.

  4. It doesn’t matter what technology they use. HTTP, Corba, Pubsub, custom protocols – doesn’t matter. Bezos doesn’t care.

  5. All service interfaces, without exception, must be designed from the ground up to be externalizable. That is to say, the team must plan and design to be able to expose the interface to developers in the outside world. No exceptions.

  6. Anyone who doesn’t do this will be fired.

  7. Thank you; have a nice day!

If we’re not disciplined, maintaining stability in a development environment becomes too difficult. So naturally, the solution becomes doubling down—we just need more environments. If every team just gets its own full-scale environment to develop against, no more stability problems. We get to develop our distributed monolith happily in our own little world. That sound you hear is every CFO collectively losing their shit, but whatever, they’re nerds and we’ve gotta get this feature to production!

Besides the obvious cost implications to this approach, perhaps the more insidious problem is it will cause teams to develop in a vacuum. In and of itself, this is not an issue, but for the undisciplined team who is not practicing rigorous API-driven development, it will create moving goalposts. A team will spend months developing its product against static dependencies only to find a massive integration headache come production time. It’s pain deferral, plain and simple. That pain isn’t being avoided or managed, you’re just neglecting to deal with instability and integration to a point where it is even more difficult. It is the opposite of the “fail-fast” mindset. It’s failing slowly and drawn out.

“We need to run everything with this particular configuration to test this, and if anyone so much as sneezes my service becomes unstable.” Good luck with that. I’ve got a dirty little secret: if you’re not disciplined, no amount of environments will make things easier. If you can’t keep your service running in an integration environment, production isn’t going to be any easier.

Similarly, massive end-to-end integration tests spanning numerous services  are an anti-pattern. Another dirty little secret: integrated tests are a scam. With a big enough system, you cannot reasonably expect to write meaningful large-scale tests in any tractable way.

What are we to do then? With respect to development, get it out of your head that you can run a facsimile of production to build features against. If you need local development, the only sane and cost-effective option is to stub. Stub everything. If you have a consistent RPC layer—discipline—this shouldn’t be too difficult. You might even be able to generate portions of stubs.

We used Google App Engine heavily at Workiva, which is a PaaS encompassing numerous services—app server, datastore, task queues, memcache, blobstore, cron, mail—all managed by Google. We were doing serverless before serverless was even a thing. App Engine provides an SDK for developing applications locally on your machine. Numerous times I overheard someone who thought the SDK was just running a facsimile of App Engine on their laptop. In reality, it was running a bunch of stubs!

If you need a full-scale deployed environment, keep in mind that stability is the cost of entry. Otherwise, you’re just delaying problems. In either case, you need stable APIs.

With respect to integration testing, the only tractable solution that doesn’t lull you into a false sense of security is consumer-driven contract testing. We run our tests against a stub, but these tests are also included in a consumer-driven contract. An API provider runs consumer-driven contract tests against its service to ensure it’s not breaking any downstream services.

All of this aside, the broader issue is ensuring a highly disciplined engineering organization. Without this, the rest becomes much more difficult as pain-driven development takes hold. Discipline is a key part of doing service-oriented design and preventing things from getting out of control as a company scales. Moving to microservices means using the right tools and processes, not just applying the old ones in a new context.

Posted on

There and Back Again: Why PaaS Is Passé (And Why It’s Not)

In 10 years nobody will be talking about Kubernetes. Not because people stopped using it or because it fell out of favor, but because it became utility. Containers, Kubernetes, service meshes—they’ll all be there, the same way VMs, hypervisors, and switches will be. Compute is a commodity, and I don’t care how my workload runs so long as it meets my business’s SLOs and other requirements. Within AWS alone, there are now innumerable ways to run a compute workload.

This was the promise of Platform as a Service (PaaS): provide a pre-built runtime where you simply plug in your application and the rest—compute, networking, storage—is handled for you. Heroku (2007), Google App Engine (2008), OpenShift (2011), and Cloud Foundry (2011) all come to mind. But PaaS has, in many ways, become a sort of taboo in recent years. As a consultant working with companies either in the cloud or looking to move to the cloud, I’ve found PaaS to almost be a trigger word; the wince from clients upon its utterance is almost palpable. It’s hard to pin down exactly why this is the case, but I think there are a number of reasons which range from entirely legit to outright FUD.

There is often a funny cognitive dissonance with these companies who recoil at the mention of PaaS. After unequivocally rejecting the idea for reasons like vendor lock-in and runtime restrictions (again, some of these are legitimate concerns), they will describe, in piecemeal fashion, their own half-baked idea of a PaaS. “Well, we’ll use Kubernetes to handle compute, ELK stack for logging, Prometheus for metrics, OpenTracing for distributed tracing, Redis for caching…”, and so the list goes on. Not to mention there tends to be a bias on build over buy. And we need to somehow provide all of these things as a self-service platform to developers.

While there are ongoing efforts to democratize the cloud and provide reference architectures of sorts, the fact is there are no standards and the proliferation of tools and technologies continues to expand at a rapid pace. On the other hand, as certain tools emerge, such as Kubernetes, the patterns and practices around them have naturally lagged behind. The serverless movement bears this out further. Serverless is the microservice equivalent for PaaS but with a lot less tooling and operations maturity. This is an exciting time, but the cloud has become—without a doubt—an unnavigable wasteland. Even with all the things at your disposal today, it’s still a ton of work to build and operate what is essentially your own PaaS.

But technology is cyclical and the cloud is no different. This evolution, in some sense, parallels what happened with the NoSQL movement. Eric Brewer discusses this in his RICON 2012 talk. When you cut through the hype, NoSQL was about giving developers more control at the expense of less pre-packaged functionality, but it was not intended to be the end game or an alternative to SQL. It’s about two different, equally valid world views: top-down and bottom-up. The top-down view is looking at a model and its semantics and then figuring out what you need to do to implement it. With a relational database, this is using SQL to declaratively construct our model. The bottom-up view is about the layering of primitive components into something more complex. For example, modern databases like CockroachDB present a SQL abstraction on top of a transactional layer on top of a replication layer on top of a simple key-value-store layer. NoSQL gives us a reusable storage component with a lot of flexibility and, over time, as we add more and more pieces on top, we get something that looks more like a database. We start with low-level layers, but the end goal is still the same: nice, user-friendly semantics. I would argue the same thing is happening with PaaS.

What the major cloud providers are doing is unbundling the PaaS. We have our compute, our cluster scheduler, our databases and caches, our message queues, and other components. What’s missing is the glue—the standards and tools that tie these things together into a coherent, manageable unit—a PaaS. Everything old is new again. What we will see is the rebundling of these components gradually happen over time as those standards and tools emerge. Tools like AWS Fargate and Google App Engine Flexible Environment are a step in that direction (Google really screwed up by calling it App Engine Flex because of all the PaaS baggage associated with the App Engine name). The container is just the interface. However, that’s only the start.

PaaS and serverless are great because they truly accelerate application development and reduce operations overhead. However, the trade-off is: we become constrained. For example, with App Engine, we were initially constrained to certain Google Cloud APIs, such as Cloud Datastore and Task Queues, and specific language runtimes. Over time, this has improved, notably, with Cloud SQL, and now today we can use custom runtimes. Similarly, PaaS gives us service autoscaling, high availability, and critical security patches for free, but we lose a degree of control over compute characteristics and workload-processing patterns.

In a sense, what a PaaS offers is an opinionated framework for running applications. Opinionated is good if you want to be productive, but it’s limiting once you have a mature product. What we want are the benefits of PaaS with a bit more flexibility. A PaaS provides us a top-down template from which we can start, but we want to be able to tweak that to our needs. Kubernetes is a key part of that template, but it’s ultimately just a means to an end.

This is why I think no one will be talking about Kubernetes in 10 years. Hopefully by then it’s just not that interesting. If it still is, we’re not done yet.

Posted on

Plant Trees Before You Need the Shade

Like humans, companies go through phases. There’s the early seed and development phase. Founders are so preoccupied with a problem they go crazy. They consider solutions and the feasibility of a business. There’s the startup phase, when a business is actually born, and it stumbles towards product/market fit. There’s the growth and scaling phase, as we try to close more and more deals while, at the same time, hiring the right people. If we’re lucky, we reach the later stages. There’s the expansion phase, as we attempt to land and expand or attack new verticals or geographies. This is when things get really interesting—and hard. Who are the right people to hire? What are the right products to build? The formula that got us here almost certainly won’t get us there. Lastly, there’s maturity, which is when the business has really hit its stride. Maybe there’s an exit, and very likely there’s new leadership involved.

Consistent in all of this are two things: culture and capabilities. Culture is the invisible hand inside your organization. It’s your company’s autopilot. Specifically, culture is the unique combination of processes and values within an organization. These processes and values are what enable us to replicate our success. They allow people to make decisions which are in alignment with the goals of the company without having to constantly coordinate with one another.

This also means your culture is derived from your capabilities, what your organization can and cannot do. Clayton Christensen groups these factors into three buckets: resources, processes, and values. Resources are the (mostly) tangible things a company has—people, capital, brands, intellectual property, relationships with customers, manufacturers, distributors, and so forth. Processes are what we do with resources to accomplish the organization’s goals, such as developing products, developing employees, hiring, firing, doing market research, and allocating resources. They take in resources and produce value. Processes help us protect and scale our values by providing a means of documenting and codifying them. These are predominantly intangible things. Finally, values define how a company makes decisions. What goes to the top of the list, and what gets ignored. What gets investment, and what doesn’t. These are our priorities that guide us.

There are a few problems with how leadership tends to view capabilities. There is typically an overemphasis on resources. This happens because in the startup phase, success is largely governed by resources. Notably, people. This is especially true of software startups. I quote this section from The Innovator’s Dilemma frequently:

In the start-up stages of an organization, much of what gets done is attributable to resources—people, in particular. The addition or departure of a few key people can profoundly influence its success. Over time, however, the locus of the organization’s capabilities shifts toward its processes and values. As people address recurrent tasks, processes become defined. And as the business model takes shape and it becomes clear which types of business need to be accorded highest priority, values coalesce. In fact, one reason that many soaring young companies flame out after an IPO based on a single hot product is that their initial success is grounded in resources—often the founding engineers—and they fail to develop processes that can create a sequence of hot products.

In the beginning stages, people drive success. Early engineers and founders (the two are not mutually exclusive) have an itch to scratch that they are so passionate about, they evangelize this crazy grand vision and people get excited. The company is small, focused, passionate, and everyone is working closely together to solve a customer problem they are obsessed with. And, sometimes, it pays off.

Next, leadership starts asking itself, “OK, we shipped this crazy successful product, now how do we grow?” This is where the wheels start to come off. There are three problems that occur.

First, the focus shifts away from solving a problem you’re obsessed with to finding the next big product. These companies fail to find a new product because they are searching for one without passion. They are seeking top-line revenue growth, not pursuing a vision. They are looking at markets through the lens of “here’s where we can make money” and not “here’s where we can solve problems,” and in doing so, they lose sight of the customer. At this stage, they basically have forgotten what got them here.

Second, their processes get in their own way. This seems contradictory given that processes, by their very nature, are meant to facilitate repeatability. If we have good processes in place, we should be able to apply them time and time again, even with different people, and end up with consistent results. But things break down when we apply the same processes to different problems. In essence, they try to find success using what brought them success to begin with, but with a warped perspective. You can look at every startup and they will all have wildly different stories about how they found success, but the one thing they will all have in common is that relentless itch. When we attack a new market, we need to do a reset on the processes and values, not a recycle. Christensen suggests if your company is so deeply entrenched in its processes that this isn’t viable, as is often the case with large enterprises, spin it out into a new venture. Processes are as dynamic as the company itself. They are not a one-size-fits-all deal. Christensen calls this the migration of capabilities.

The factors that define an organization’s capabilities and disabilities evolve over time—they start in resources; then move to visible, articulated processes and values; and migrate finally to culture. As long as the organization continues to face the same sorts of problems that its processes and values were designed to address, managing the organization can be straightforward. But because those factors also define what an organization cannot do, they constitute disabilities when the problems facing the company change fundamentally.

Third, they don’t fully appreciate the nature of dynamic priorities. Software startups in particular often mistakenly attribute their initial success to technology when, in reality, it’s because of people and timing. Aside from the passion, you need people early on who can ship and ship often. These might not be the most technically capable engineers, but they get things done when it matters. Later, you need people who can still ship but while cleaning things up. Lastly, you need maintainers—people who can refine without breaking anything. That’s not to say you have these archetypes exclusively at each stage—you want a balance of people—but it’s similar to how companies commonly need a different type of CTO at different phases or an Interim CTO.

While resources are an essential part of early success, it’s processes and values that will sustain you. However, we tend to overfit on resources because we become biased from that success—investing heavily in technology and innovation and grounding the company’s success in a few key individuals. We also overfit because resources are more visible and measurable. Being deliberate about establishing processes and values—which are derived from vision—helps to overcome this bias, but it needs to happen early and be continually reinforced. We also need to ensure our processes adapt to new problems. The larger and more complex an organization becomes, the harder this gets.

Moreover, we need to be conscientious about which processes matter to us the most. Often the most important capabilities aren’t reflected by the most visible processes—product development or customer service, for example—but in the less visible, background processes that support decisions about where to invest resources. These might include determining how market research is done, how financial and sales projections are drawn from this analysis, how products are conceived, how planning and budgets are negotiated, and so on. These processes are where many companies get their inability to cope with change.

And this is where the breakdown happens: a company has highly capable people—people who have helped shape its success as a startup—but arms them with the wrong processes and values. The result is often a boom followed by a bunch of fizzles as they try to catch lightning in a bottle once more. A compelling vision plants a seed. Strong processes and clear values help that seed to grow. But the shade produced by that seed—our capabilities—is stationary, so when we approach a new challenge, we need to recognize when to start tilling.

Posted on

Building a Distributed Log from Scratch, Part 5: Sketching a New System

In part four of this series we looked at some key trade-offs involved with a distributed log implementation and discussed a few lessons learned while building NATS Streaming. In this fifth and final installment, we’ll conclude by outlining the design for a new log-based system that draws from the previous entries in the series.

The Context

For context, NATS and NATS Streaming are two different things. NATS Streaming is a log-based streaming system built on top of NATS, and NATS is a lightweight pub/sub messaging system. NATS was originally built (and then open sourced) as the control plane for Cloud Foundry. NATS Streaming was built in response to the community’s ask for higher-level guarantees—durability, at-least-once delivery, and so forth—beyond what NATS provided. It was built as a separate layer on top of NATS. I tend to describe NATS as a dial tone—ubiquitous and always on—perfect for “online” communications. NATS Streaming is the voicemail—leave a message after the beep and someone will get to it later. There are, of course, more nuances than this, but that’s the gist.

The key point here is that NATS and NATS Streaming are distinct systems with distinct protocols, distinct APIs, and distinct client libraries. In fact, NATS Streaming was designed to essentially act as a client to NATS. As such, clients don’t talk to NATS Streaming directly, rather all communication goes through NATS. However, the NATS Streaming binary can be configured to either embed NATS or point to a standalone deployment. The architecture is shown below in a diagram borrowed from the NATS website.

Architecturally, this makes a lot of sense. It supports the end-to-end principle in that we layer on additional functionality rather than bake it in to the underlying infrastructure. After all, we can always build stronger guarantees on top, but we can’t always remove them from below. This particular architecture, however, introduces a few challenges (disclosure: while I’m still a fan, I’m no longer involved with the NATS project and the NATS team is aware of these problems and no doubt working to address many of them).

First, there is no “cross-talk” between NATS and NATS Streaming, meaning messages published to NATS are not visible in NATS Streaming and vice versa. Again, they are two completely separate systems that just share the same infrastructure. This means we’re not really layering on message durability to NATS, we’re just exposing a new system which provides these semantics.

Second, because NATS Streaming runs as a “sidecar” to NATS and all of its communication runs through NATS, there is an inherent bottleneck at the NATS connection. This may only be a theoretical limit, but it precludes certain optimizations like using sendfile to do zero-copy reads of the log. It also means we rely on timeouts even in cases where the server could send a response immediately, such as when there is no leader elected for the cluster.

Third, NATS Streaming currently lacks a compelling story around linear scaling other than running multiple clusters and partitioning channels among them at the application level. With respect to scaling a single channel, the only alternative at the moment is to partition it into multiple channels at the application level. My hope is that as clustering matures, this will too.

Fourth, without extending its protocol, NATS Streaming’s authorization is intrinsically limited to the authorization provided by NATS since all communication goes through it. In and of itself, this isn’t a problem. NATS supports multi-user authentication and subject-level permissions, but since NATS Streaming uses an opaque protocol atop NATS, it’s difficult to setup proper ACLs at the streaming level. Of course, many layered protocols support authentication, e.g. HTTP atop TCP. For example, the NATS Streaming protocol could carry authentication tokens or session keys, but it currently does not do this.

Fifth, NATS Streaming does not support wildcard semantics, which—at least in my opinion—is a large selling point of NATS and, as a result, something users have come to expect. Specifically, NATS supports two wildcards in subject subscriptions: asterisk (*) which matches any token in the subject (e.g. foo.* matches foo.bar, foo.baz, etc.) and full wildcard (>) which matches one or more tokens at the tail of the subject (e.g. foo.> matches foo.bar, foo.bar.baz, etc.). Note that this limitation in NATS Streaming is not directly related to the overall architecture but more in how we design the log.

More generally, clustering and data replication was more of an afterthought in NATS Streaming. As we discussed in part four, it’s hard to add this after the fact. Combined with the APIs NATS Streaming provides (which do flow control and track consumer state), this creates a lot of complexity in the server.

A New System

I wasn’t involved much with NATS Streaming beyond the clustering implementation. However, from that work—and through my own use of NATS and from discussions I’ve had with the community—I’ve thought about how I would build something like it if I were to start over. It would look a bit different from NATS Streaming and Kafka, yet also share some similarities. I’ve dubbed this theoretical system Jetstream (update: this is now Liftbridge), though I’ve yet to actually build anything beyond small prototypes. It’s a side project of mine I hope to get to at some point.

Core NATS has a strong community with solid mindshare, but NATS Streaming doesn’t fully leverage this since it’s a new silo. Jetstream aims to address the above problems starting from a simple proposition: many people are already using NATS today and simply want streaming semantics for what they already have. However, we must also acknowledge that other users are happy with NATS as it currently is and have no need for additional features that might compromise simplicity or performance. This was a deciding factor in choosing not to build NATS Streaming’s functionality directly into NATS.

Like NATS Streaming, Jetstream is a separate component which acts as a NATS client. Unlike NATS Streaming, it augments NATS as opposed to implementing a wholly new protocol. More succinctly, Jetstream is a durable stream augmentation for NATS. Next, we’ll talk about how it accomplishes this by sketching out a design.

Cross-Talk

In NATS Streaming, the log is modeled as a channel. Clients create channels implicitly by publishing or subscribing to a topic (called a subject in NATS). A channel might be foo but internally this is translated to a NATS pub/sub subject such as _STAN.pub.foo. Therefore, while NATS Streaming is technically a client of NATS, it’s done so just to dispatch communication between the client and server. The log is implemented on top of plain pub/sub messaging.

Jetstream is merely a consumer of NATS. In it, the log is modeled as a stream. Clients create streams explicitly, which are subscriptions to NATS subjects that are sequenced, replicated, and durably stored. Thus, there is no “cross-talk” or internal subjects needed because Jetstream messages are NATS messages. Clients just publish their messages to NATS as usual and, behind the scenes, Jetstream will handle storing them in a log. In some sense, it’s just an audit log of messages flowing through NATS.

With this, we get wildcards for free since streams are bound to NATS subjects. There are some trade-offs to this, however, which we will discuss in a bit.

Performance

Jetstream does not track subscription positions. It is up to consumers to track their position in a stream or, optionally, store their position in a stream (more on this later). This means we treat a stream as a simple log, allowing us to do fast, sequential disk I/O and minimize replication and protocol chatter as well as code complexity.

Consumers connect directly to Jetstream using a pull-based socket API. The log is stored in the manner described in part one. This enables us to do zero-copy reads from a stream and other important optimizations which NATS Streaming is precluded from doing. It also simplifies things around flow control and batching as we discussed in part three.

Scalability

Jetstream is designed to be clustered and horizontally scalable from the start. We make the observation that NATS is already efficient at routing messages, particularly with high consumer fan-out, and provides clustering of the interest graph. Streams provide the unit of storage and scalability in Jetstream.

A stream is a named log attached to a NATS subject. Akin to a partition in Kafka, each stream has a replicationFactor, which controls the number of nodes in the Jetstream cluster that participate in replicating the stream, and each stream has a leader. The leader is responsible for receiving messages from NATS, sequencing them, and performing replication (NATS provides per-publisher message ordering).

Like Kafka’s controller, there is a single metadata leader for a Jetstream cluster which is responsible for processing requests to create or delete streams. If a request is sent to a follower, it’s automatically forwarded to the leader. When a stream is created, the metadata leader selects replicationFactor nodes to participate in the stream (initially, this selection is random but could be made more intelligent, e.g. selecting based on current load) and replicates the stream to all nodes in the cluster. Once this replication completes, the stream has been created and its leader begins processing messages. This means NATS messages are not stored unless there is a stream matching their subject (this is the trade-off to support wildcards, but it also means we don’t waste resources storing messages we might not care about). This can be mitigated by having publishers ensure a stream exists before publishing, e.g. at startup.

There can exist multiple streams attached to the same NATS subject or even subjects that are semantically equivalent, e.g. foo.bar and foo.*. Each of these streams will receive a copy of the message as NATS handles this fan-out. However, the stream name is unique within a given subject. For example, creating two streams for the subject foo.bar named foo and bar, respectively, will create two streams which will independently sequence all of the messages on the NATS subject foo.bar, but attempting to create two streams for the same subject both named foo will result in creating just a single stream (creation is idempotent).

With this in mind, we can scale linearly with respect to consumers—covered in part three—by adding more nodes to the Jetstream cluster and creating more streams which will be distributed among the cluster. This has the advantage that we don’t need to worry about partitioning so long as NATS is able to withstand the load (there is also an assumption that we can ensure reasonable balance of stream leaders across the cluster). We’ve basically split out message routing from storage and consumption, which allows us to scale independently.

Additionally, streams can join a named consumer group. This, in effect, partitions a NATS subject among the streams in the group, again covered in part three, allowing us to create competing consumers for load-balancing purposes. This works by using NATS queue subscriptions, so the downside is partitioning is effectively random. The upside is consumer groups don’t affect normal streams.

Compaction and Offset Tracking

Streams support multiple log-compaction rules: time-based, message-based, and size-based. As in Kafka, we also support a fourth kind: key compaction. This is how offset storage will work, which was described in part three, but it also enables some other interesting use cases like KTables in Kafka Streams.

As discussed above, messages in Jetstream are simply NATS messages. There is no special protocol needed for Jetstream to process messages. However, publishers can choose to optionally “enhance” their messages by providing additional metadata and serializing their messages into envelopes. The envelope includes a special cookie Jetstream uses to detect if a message is an envelope or a simple NATS message (if the cookie is present by coincidence and envelope deserialization fails, we fall back to treating it as a normal message).

One of the metadata fields on the envelope is an optional message key. A stream can be configured to compact by key. In this case, it retains only the last message for each key (if no key is present, the message is always retained).

Consumers can optionally store their offsets in Jetstream (this can also be transparently managed by a client library similar to Kafka’s high-level consumer). This works by storing offsets in a stream keyed by consumer. A consumer (or consumer library) publishes their latest offset. This allows them to later retrieve their offset from the stream, and key compaction means Jetstream will only retain the latest offset for each consumer. For improved performance, the client library should only periodically checkpoint this offset.

Authorization

Because Jetstream is a separate server which is merely a consumer of NATS, it can provide ACLs or other authorization mechanisms on streams. A simple configuration might be to restrict NATS access to Jetstream and configure Jetstream to only allow access to certain subjects. There is more work involved because there is a separate access-control system, but this gives greater flexibility by separating out the systems.

At-Least Once Delivery

To ensure at-least-once delivery of messages, Jetstream relies on replication and publisher acks. When a message is received on a stream, it’s assigned an offset by the leader and then replicated. Upon a successful replication, the stream publishes an ack to NATS on the reply subject of the message, if present (the reply subject is a part of the NATS message protocol).

There are two implications with this. First, if the publisher doesn’t care about ensuring its message is stored, it need not set a reply subject. Second, because there are potentially multiple (or no) streams attached to a subject (and creation/deletion of streams is dynamic), it’s not possible for the publisher to know how many acks to expect. This is a trade-off we make for enabling subject fan-out and wildcards while remaining scalable and fast. We make the assertion that if guaranteed delivery is important, the publisher should be responsible for determining the destination streams a priori. This allows attaching streams to a subject for use cases that do not require strong guarantees without the publisher having to be aware. Note that this might be an area for future improvement to increase usability, such as storing streams in a registry. However, this is akin to other similar systems, like Kafka, where you must first create a topic and then you publish to that topic.

One caveat to this is if there are existing application-level uses of the reply subject on NATS messages. That is, if other systems are already publishing replies, then Jetstream will overload this. The alternative would be to require the envelope, which would include a canonical reply subject for acks, for at-least-once delivery. Otherwise we would need a change to the NATS protocol itself.

Replication Protocol

For metadata replication and leadership election, we rely on Raft. However, for replication of streams, rather than using Raft or other quorum-based techniques, we use a technique similar to Kafka as described in part two.

For each stream, we maintain an in-sync replica set (ISR), which is all of the replicas currently up to date (at stream creation time, this is all of the replicas). During replication, the leader writes messages to a WAL, and we only wait on replicas in the ISR before committing. If a replica falls behind or fails, it’s removed from the ISR. If the leader fails, any replica in the ISR can take its place. If a failed replica catches back up, it rejoins the ISR. The general stream replication process is as follows:

  1. Client creates a stream with a replicationFactor of n.
  2. Metadata leader selects n replicas to participate and one leader at random (this comprises the initial ISR).
  3. Metadata leader replicates the stream via Raft to the entire cluster.
  4. The nodes participating in the stream initialize it, and the leader subscribes to the NATS subject.
  5. The leader initializes the high-water mark (HW) to 0. This is the offset of the last committed message in the stream.
  6. The leader begins sequencing messages from NATS and writes them to the log uncommitted.
  7. Replicas consume from the leader’s log to replicate messages to their own log. We piggyback the leader’s HW on these responses, and replicas periodically checkpoint the HW to stable storage.
  8. Replicas acknowledge they’ve replicated the message.
  9. Once the leader has heard from the ISR, the message is committed and the HW is updated.

Note that clients only see committed messages in the log. There are a variety of failures that can occur in the replication process. A few of them are described below along with how they are mitigated.

If a follower suspects that the leader has failed, it will notify the metadata leader. If the metadata leader receives a notification from the majority of the ISR within a bounded period, it will select a new leader for the stream, apply this update to the Raft group, and notify the replica set. These notifications need to go through Raft as well in the event of a metadata leader failover occurring at the same time as a stream leader failure. Committed messages are always preserved during a leadership change, but uncommitted messages could be lost.

If the stream leader detects that a replica has failed or fallen too far behind, it removes the replica from the ISR by notifying the metadata leader. The metadata leader replicates this fact via Raft. The stream leader continues to commit messages with fewer replicas in the ISR, entering an under-replicated state.

When a failed replica is restarted, it recovers the latest HW from stable storage and truncates its log up to the HW. This removes any potentially uncommitted messages in the log. The replica then begins fetching messages from the leader starting at the HW. Once the replica has caught up, it’s added back into the ISR and the system resumes its fully replicated state.

If the metadata leader fails, Raft will handle electing a new leader. The metadata Raft group stores the leader and ISR for every stream, so failover of the metadata leader is not a problem.

There are a few other corner cases and nuances to handle, but this covers replication in broad strokes. We also haven’t discussed how to implement failure detection (Kafka uses ZooKeeper for this), but we won’t prescribe that here.

Wrapping Up

This concludes our series on building a distributed log that is fast, highly available, and scalable. In part one, we introduced the log abstraction and talked about the storage mechanics behind it. In part two, we covered high availability and data replication. In part three, we we discussed scaling message delivery. In part four, we looked at some trade-offs and lessons learned. Lastly, in part five, we outlined the design for a new log-based system that draws from the previous entries in the series.

The goal of this series was to learn a bit about the internals of a log abstraction, to learn how it can achieve the three priorities described earlier, and to learn some applied distributed systems theory. Hopefully you found it useful or, at the very least, interesting.

If you or your company are looking for help with system architecture, performance, or scalability, contact Real Kinetic.