Cloud native models using containerized software in a continuous delivery approach could benefit from serverless computing where the cloud vendor generates the exact amount of resources required to run a workload on the fly. While the major cloud vendors have recognized this and are already creating products to abstract away the infrastructure, it may not work for every situation in spite of the benefits.
Cloud native, put simply, involves using containerized applications and Kubernetes to deliver software in small packages called microservices. This enables developers to build and deliver software faster and more efficiently in a continuous delivery model. In the cloud native world, you should be able to develop code once and run it anywhere, on prem or any public cloud, or at least that is the ideal.
Serverless is actually a bit of a misnomer. There are servers underlying the model, but instead of dedicated virtual machines, the cloud vendor delivers exactly the right number of resources to run a particular workload for the right amount of time and no more.
Nothing is perfect
Such an arrangement would seem to be perfectly suited to a continuous delivery model, and while vendors have recognized the beauty of such an approach, as one engineer pointed out, there is never a free lunch in processes that are this complex, and it won’t be a perfect solution for every situation.
Arpana Sinha, director of product management at Google, says the Kubernetes community has really embraced the serverless idea, but she says that it is limited in its current implementation, delivered in the form of functions with products like AWS Lambda, Google Cloud Functions and Azure Functions.
“Actually, I think the functions concept is a limited concept. It is unfortunate that that is the only thing that people associate with serverless,” she said.
She says that Google has tried to be more expansive in its definition. “It’s basically a concept for developers where you are able to seamlessly go from writing code to deployment and the infrastructure takes care of all of the rest, making sure your code is deployed in the appropriate way across the appropriate, most resilient parts of the infrastructure, scaling it as your app needs additional resources, scaling it down as your traffic goes down, and charging you only for what you’re consuming,” she explained.
But Matt Whittington, senior engineer on the Kubernetes Team at Atlassian says, while it sounds good in theory, in practice, fully automated infrastructure could be unrealistic in some instances. “Serverless could be promising for certain workloads because it really allows developers to focus on the code, but it’s not a perfect solution. There is still some underlying tuning.”
He says you may not be able to leave it completely up to the vendor unless there is a way to specify the requirements for each container, such as instructing them you need a minimum container load time, a certain container kill time or perhaps you need to deliver it a specific location. He says in reality it won’t be fully automated, at least while developers fiddle with the settings to make sure they are getting the resources they need without over-provisioning and paying for more than they need.
Vendors bringing solutions
The vendors are putting in their two cents trying to create tools that bring this ideal together. For instance, Google announced a service called Google Cloud Run at Google Cloud Next last month. It’s based on the open-source Knative project, and in essence combines the goodness of serverless for developers running containers. Other similar services include AWS Fargate and Azure Container Instances, both of which are attempting to bring together these two technologies in a similar package.
In fact, Gabe Monroy, partner program manager at Microsoft, says Azure Container Instances is designed to solve this problem without being dependent on a functions-driven programming approach. “What Azure Container Instances does is it allows you to run containers directly on the Azure compute fabric, no virtual machines, hypervisor isolated, pay-per-second billing. We call it serverless containers,” he said.
While serverless and containers might seem like a good fit, as Monroy points out, there isn’t a one-size-fits-all approach to cloud-native technologies, whatever the approach may be. Some people will continue to use a function-driven serverless approach like AWS Lambda or Azure Functions and others will shift to containers and look for other ways to bring these technologies together. Whatever happens, as developer needs change, it is clear the open-source community and vendors will respond with tools to help them. Bringing serverless and containers together is just one example of that.