Modern application development uses container technology more and more, whether this is for running backend or frontend services, or even for developer tooling. Many organizations such as system integrators, independent software vendors, and cloud solution specialists are adopting containerization to keep up with the demands of large scale, distribution, and modern development and operations practices. Many others are jumping on the container bandwagon because everyone seems to be doing it. Instead of blindly following along, it is good to investigate and answer a number of essential questions. What are compelling reasons to use containers or container technology, and what does it mean to use it as a foundation for your application development?
10 compelling reasons to use container technology.
1. Higher degree of hardware abstraction
Using containers to run your software means that you are utilizing host machines at a level where the individual machine doesn’t matter any more. Instead of caring about single machines with well-known names and maintaining and patching these, the operations team can focus on a cluster of host machines. This cluster is a fabric of the host machines, woven together by cluster software to form a contiguous set of resources for computing, memory, and storage. Each machine is built with commodity hardware and is expendable. The cluster can grow and shrink whenever new hosts are added or defective hosts are taken out of commission. The containers will be running on the accumulation of resources managed by the cluster software, without knowing or caring about the individual machines.
It is an important trend in the exploitation of cloud applications to try and pay only for the resources actually consumed. Containers enable more optimal use of the resources in the cluster. The cluster software will allocate and assign the running containers based on the requirements that the container images and application registration indicates. These requirements might include a particular operating system, memory constraints, and the required CPU resources. It can decide what to run and where, while taking into account the constraints that were given. The cluster might also shift containers around whenever it sees fit for a better utilization of the available resources. At the end of the day it is possible to achieve much higher density of running containers and hence applications on the cluster, when compared to running applications on dedicated machines. Fewer machines are needed to run the same workload, which results in lower costs of operations.
The cluster offers a pool of resources that the containers can request for operations. Whenever the need arises for more resources and bigger scale, the cluster can spin up additional container instances from the images and allocate more resources per container, if necessary. As long as there is a surplus of resources it is easy to scale up, and down afterwards. When resources start running out, good capacity planning will make sure more hosts are added to the cluster for provisioning containers. Smarter cluster software can manage scale better. Because containers are very fast to spin up, the scaling process itself is a matter of seconds instead of minutes or hours.
3. Freedom in hosting
Ideally, the cluster software is installed on virtual machines to easily provision a new cluster when necessary, or to add new hosts without having to care about the actual hardware. These virtual machines may reside in the cloud at any cloud provider such as Amazon AWS, Google Cloud or Microsoft Azure, but they can also be located on-premise at your own organization. There is no restriction to run the cluster anywhere specific, although in some situations there is a benefit in choosing for a particular cloud provider. For example, Microsoft offers Azure Container Service which provides automatic provisioning of Kubernetes, Docker Swarm or DC/OS clusters. The various cluster software offerings run on different operating systems, both Linux-based and various Windows Server editions.
4. Container images are immutable
The DevOps team builds new application components by writing and testing code, and compiling it into binaries. In a container world, they will also prepare the container images that have all of the prerequisites for the components, and deploy the new components onto them. The container image is configured to run correctly and then deployed to a central image repository. From that repository, the host can download and run the images. Once the DevOps team releases the container image to the repository it cannot be changed anymore. The images are immutable and can be guaranteed to arrive at the hosts without any changes. Any changes that occur are in the state of a running container, but never the image. The DevOps team has full control over what will be running in the containers, and does not depend on human intervention of configuration. Moreover, it is no longer necessary to wonder whether a container image can still run correctly at a later time. Whenever a compatible cluster with the correct operating system and kernel version is available, container images are guaranteed to run, since it is a frozen set of code and configuration with no alterations since creation. Rolling back deployments and maintaining a running application suddenly becomes a lot easier.
5. Container images are code and configuration
Container images combine the binaries built from code and the configuration of the machine into a single entity. This inseparable set offers better control, stability and predictability than before.
In traditional deployment scenarios, the code is deployed to target machines separate from the configuration. This led to the rise of Desired State Configuration, where script or a declarative language of some form would describe the configuration for the application code and its requirements at a host level. The code and resulting binaries were usually handled by the development team. The team also describes the required configuration in documentation. The operations team would then get scripts from the development team to prepare the host machines and to do final configuration of the software components. This turned out to be error-prone, but also required multiple environments to separate the various stages in an application’s lifecycle, such as testing, acceptance and production. Containers ensure that our applications will always run in exactly the same context over different container hosts.
6. Stages instead of environments
Having Development, Testing, Acceptance and Production environments (DTAP for short) might be a thing of the past when containers and clusters come into the picture. Traditionally there are separate environments to indicate the level of quality and trust in your application and its components. Now you can think of the various environment stages that an application goes through as being reflected by the versions of your container images. You could run the versions for DTAP in a single cluster, at different endpoints managed by the cluster software. If necessary, one can still separate a production from a non-production cluster, but this will mostly be from a security perspective, rather than for reasons of stability and availability of the stages in the application’s lifecycle.
The entire environment can even be described as code by composition files describing the various elements and containers. For example, the tool Docker Compose uses YML files to describe the various parts in an environment for one or more applications. Transitioning from one environment to the next, say staging to production, can be accomplished by promoting these files.
7. Fit and alignment with distributed architectures
Transitioning to container-based applications fits well with modern architecture styles such as microservices. Large scale and heavily distributed applications benefit from such an architecture, and consequently the use of container technology. In the aforementioned architectures, individual containers usually contain single processes, such as a service for Web API, website hosting or background processing. This requires one to rethink the design of the application in small, isolated and manageable pieces.
Typically, each of the container images will be running multiple times for failover, robustness and scalability. In addition, there will be many separate container images that together form the complete application. Each of the running container instances is network-separated to build an intrinsically distributed application. The application architecture must take this into account. You should carefully consider the fact that other components use the network for communication. This comes at a price and a risk. The call to another component is not cheap in terms of time, when made over HTTP or TCP compared to an in-process or single-machine call. Also, there is a chance that the process in another container might not be available or reachable. Countermeasures to deal with outages need to be built into the application. Implementation patterns such as Circuit Breaker and Retry become part of the developer’s portfolio. It is likely that such an architecture has its effect on the way the teams and company are organized. The benefits of this new architecture type are a very nice alignment with the container approach and with the DevOps teams that build and run the applications. This alignment comes from the isolation of multiple smaller parts in the system, which is designed to deal with potentially short lifetimes or unavailability of the services. The cost of such an architecture and different implementation patterns lies in their complexity and learning curve.
8. Support in application frameworks and tooling
There is an abundance of frameworks to facilitate developing code for components and services to run in containers. The architecture and development model allow a freedom of technology choice for each of the containers, as these are loosely coupled and have no binary or technical dependencies on one another. There is no specific vendor lock-in when choosing the framework for development. However, a fit-for-purpose selection is highly recommended.
As far as tooling is concerned, the world of containers has converged on Docker as the de facto standard for interacting with container instances, images and hosts. With the standardization around Docker as the container interaction protocol and Docker as a company to deliver the low-level tooling, other companies and software have chosen to align. Development environments offer tooling and integration with the Docker ecosystem. Docker images are used as the image format for container clusters to deploy new components into the cluster. To an increasing extent, web application services are being deployed from Docker images, instead of doing installer or file copy based deployments.
3 Reasons to not use containers
The above-mentioned reasons might be convincing to start with containers tomorrow. As almost always, there is another side to take into account. Containers might not be what is best for you, your applications, or your organization.
1. Container technology is not a silver bullet
Even though containers are popular, especially among developers and operations, it is not the solution for all problems in application delivery and architectures. Switching to containers as a basis for your application will not necessarily solve or prevent performance, stability or scaling issues. Containers do not fit every part of an application or the bigger landscape that it is part of. Careful consideration is required to decide when using containers makes sense and when it doesn’t.
2. The benefits of containers come at a price
Creating an application that utilizes containers requires learning the technology, investing time to change the architecture and the deployment strategies. Various design patterns need to be implemented by the developers in order to make a resilient, robust distributed application. The build and release pipelines have to be changed for the creation and deployment of container images. This is more complicated than a traditional monolithic application and its lifecycle. The new way of combining development and operations into DevOps involves changing the teams, and aligning the organization with the business domains.
So, while the benefits that an application and the landscape will have from using containers, they do not come free of charge. It means investing before the benefits can be reaped, and there is no guarantee that in the end, the benefits outweigh the costs, particularly for simpler and smaller applications.
3. The application paradigm is moving further
Containers are fairly new and many are still getting their heads around applying them in application development. The world of modern application development is moving on in a rapid pace and other, newer ways of creating distributed applications are emerging. Serverless computing is a nascent form of computing that allows near-infinite scaling and a cost-model that charges by the millisecond consumed. It introduces an even higher level of hardware abstraction, because it does not even require hosts like a cluster would need. Additionally, the way serverless computing is implemented focuses on the actual logic instead of much of the plumbing that container implementations still need. This implies that the costs of building and running applications may be lower when choosing a serverless model.
Container technology is taking a prominent place in today’s approach to application development and hosting. There is a lot to be said and considered for switching to containers as technology for modern cloud-scale applications.
Choose wisely and modernize your IT with containers, … if it makes sense. </>