Microservices are a software development technique—a variant of the service-oriented architecture (SOA) architectural style that structures an application as a collection of loosely coupled services
Originally published by Tom Huston at https://smartbear.com
Microservice architecture, or simply microservices, is a distinctive method of developing software systems that tries to focus on building single-function modules with well-defined interfaces and operations. The trend has grown popular in recent years as Enterprises look to become more Agile and move towards a DevOps and continuous testing.
Microservices have many benefits for Agile and DevOps teams - as Martin Fowler points out, Netflix, eBay, Amazon, Twitter, PayPal, and other tech stars have all evolved from monolithic to microservices architecture. Unlike microservices, a monolith application is built as a single, autonomous unit. This make changes to the application slow as it affects the entire system. A modification made to a small section of code might require building and deploying an entirely new version of software. Scaling specific functions of an application, also means you have to scale the entire application.
Microservices solve these challenges of monolithic systems by being as modular as possible. In the simplest form, they help build an application as a suite of small services, each running in its own process and are independently deployable. These services may be written in different programming languages and may use different data storage techniques. While this results in the development of systems that are scalable and flexible, it needs a dynamic makeover. Microservices are often connecta via APIs, and can leverage many of the same tools and solutions that have grown in the RESTful and web service ecosystem. Testing these APIs can help validate the flow of data and information throughout your microservice deployment.
Just as there is no formal definition of the term microservices, there’s no standard model that you’ll see represented in every system based on this architectural style. But you can expect most microservice systems to share a few notable characteristics.
Software built as microservices can, by definition, be broken down into multiple component services. Why? So that each of these services can be deployed, tweaked, and then redeployed independently without compromising the integrity of an application. As a result, you might only need to change one or more distinct services instead of having to redeploy entire applications. But this approach does have its downsides, including expensive remote calls (instead of in-process calls), coarser-grained remote APIs, and increased complexity when redistributing responsibilities between components.
The microservices style is usually organized around business capabilities and priorities. Unlike a traditional monolithic development approach—where different teams each have a specific focus on, say, UIs, databases, technology layers, or server-side logic—microservice architecture utilizes cross-functional teams. The responsibilities of each team are to make specific products based on one or more individual services communicating via message bus. In microservices, a team owns the product for its lifetime, as in Amazon’s oft-quoted maxim “You build it, you run it.
Microservices act somewhat like the classical UNIX system: they receive requests, process them, and generate a response accordingly. This is opposite to how many other products such as ESBs (Enterprise Service Buses) work, where high-tech systems for message routing, choreography, and applying business rules are utilized. You could say that microservices have smart endpoints that process info and apply logic, and dumb pipes through which the info flows.
Since microservices involve a variety of technologies and platforms, old-school methods of centralized governance aren’t optimal. Decentralized governance is favored by the microservices community because its developers strive to produce useful tools that can then be used by others to solve the same problems. Just like decentralized governance, microservice architecture also favors decentralized data management. Monolithic systems use a single logical database across different applications. In a microservice application, each service usually manages its unique database.
Like a well-rounded child, microservices are designed to cope with failure. Since several unique and diverse services are communicating together, it’s quite possible that a service could fail, for one reason or another (e.g., when the supplier isn’t available). In these instances, the client should allow its neighboring services to function while it bows out in as graceful a manner as possible. However, monitoring microservices can help prevent the risk of a failure. For obvious reasons, this requirement adds more complexity to microservices as compared to monolithic systems architecture.
Microservices architecture is an evolutionary design and, again, is ideal for evolutionary systems where you can’t fully anticipate the types of devices that may one day be accessing your application.. Many applications start based on monolithic architecture, but as several unforeseen requirements surfaced, can be slowly revamped to microservices that interact over an older monolithic architecture through APIs.
Netflix has a widespread architecture that has evolved from monolithic to SOA. It receives more than one billion calls every day, from more than 800 different types of devices, to its streaming-video API. Each API call then prompts around five additional calls to the backend service.
Amazon has also migrated to microservices. They get countless calls from a variety of applications—including applications that manage the web service API as well as the website itself—which would have been simply impossible for their old, two-tiered architecture to handle.
The auction site eBay is yet another example that has gone through the same transition. Their core application comprises several autonomous applications, with each one executing the business logic for different function areas.
Microservices are not a silver bullet, and by implementing them you will expose communication, teamwork, and other problems that may have been previously implicit but are now forced out into the open. But API Gateways in Microservices can greatly reduce build and qa time and effort.
One common issue involves sharing schema/validation logic across services. What A requires in order to consider some data valid doesn’t always apply to B, if B has different needs. The best recommendation is to apply versioning and distribute schema in shared libraries. Changes to libraries then become discussions between teams. Also, with strong versioning comes dependencies, which can cause more overhead. The best practice to overcome this is planning around backwards compatibility, and accepting regression tests from external services/teams. These prompt you to have a conversation before you disrupt someone else’s business process, not after.
As with anything else, whether or not microservice architecture is right for you depends on your requirements, because they all have their pros and cons. Here’s a quick rundown of some of the good and bad:
To begin with, let’s explore Conway’s Law, which states: “Organizations which design systems…are constrained to produce designs which are copies of the communication structures of these organizations.”
Imagine Company X with two teams: Support and Accounting. Instinctively, we separate out the high risk activities; it’s only difficult deciding responsibilities like customer refunds. Consider how we might answer questions like “Does the Accounting team have enough people to process both customer refunds and credits?” or “Wouldn’t it be a better outcome to have our Support people be able to apply credits and deal with frustrated customers?” The answers get resolved by Company X’s new policy: Support can apply a credit, but Accounting has to process a refund to return money to a customer. The roles and responsibilities in this interconnected system have been successfully split, while gaining customer satisfaction and minimizing risks.
Likewise, at the beginning of designing any software application, companies typically assemble a team and create a project. Over time, the team grows, and multiple projects on the same codebase are completed. More often than not, this leads to competing projects: two people will find it difficult to work at cross purposes in the same area of code without introducing tradeoffs. And adding more people to the equation only makes the problem worse. As Fred Brooks puts it, nine women can’t make a baby in one month.
Moreover, in Company X or in any dev team, priorities frequently shift, resulting in management and communication issues. Last month’s highest priority item may have caused our team to push hard to ship code, but now a user is reporting an issue, and we no longer have time to resolve it because of this month’s priority. This is the most compelling reason to adopt SOA, including the microservices variety. Service-oriented approaches recognize the frictions involved between change management, domain knowledge, and business priorities, allowing dev teams to explicitly separate and address them. Of course, this in itself is a tradeoff—it requires coordination—but it allows you to centralize friction and introduce efficiency, as opposed to suffering from a large number of small inefficiencies.
Most importantly, smartly implementing an SOA or microservice architecture forces you to apply the Interface Separation Principle. Due to the connected nature of mature systems, when isolating issues of concern, the typical approach is to find a seam or communication point and then draw a dotted line between two halves of the system. Without careful thought, however, this can lead to accidentally creating two smaller but growing monoliths, now connected with some kind of bridge. The consequence of this can be marooning important code on the wrong side of a barrier: Team A doesn’t bother to look after it, while Team B needs it, so they reinvent it.
We’ve named some problems that commonly emerge; now let’s begin to look at some solutions.
How do you deploy relatively independent yet integrated services without spawning accidental monoliths? Well, suppose you have a large application, as in the sample from our Company X below, and are splitting up the codebase and teams to scale. Instead of finding an entire section of an application to split off, you can look for something on the edge of the application graph. You can tell which sections these are because nothing depends on them. In our example, the arrows pointing to Printer and Storage suggest they’re two things that can be easily removed from our main application and abstracted away. Printing either a Job or Invoice is irrelevant; a Printer just wants printable data. Turning these—Printer and Storage—into external services avoids the monoliths problem alluded to before. It also makes sense as they are used multiple times, and there’s little that can be reinvented. Use cases are well known from past experience, so you can avoid accidentally removing key functionality.
So how do we go from monoliths to services? One way is through service objects. Without removing code from your application, you effectively just begin to structure it as though it were completely external. To do that, you’ll first need to differentiate the actions that can be done and the data that is present as inputs and outputs of those actions. Consider the code below, with a notion of doing something useful and a status of that task.
# A class to model a core transaction and execute it
@status = 'Queued'
@status = 'Finished'
return @status == 'Finished'
return @status == 'Queued'
To prepare this to begin looking like a microservice, what’s next?
# Service to do useful work and modify a status
# A model of our Job's status
@status = 'Queued'
return @status == 'Finished'
return @status == 'Queued'
@status = 'Finished'
Now we’ve distinguished two distinct classes: one that models the data, and one that performs the operations. Importantly, our JobService class has little or no state—you can call the same actions over and over, changing only the data, and expect to get consistent results. If JobService somehow started taking place over a network, our otherwise monolithic application wouldn’t care. Shifting these types of classes into a library, and substituting a network client for the previous implementation, would allow you to transform the existing code into a scalable external service.
This is Hexagonal Architecture, where the core of your application and the coordination is in the center, and the external components are orchestrated around it to achieve your goals.
Now let’s take a closer look at what makes something a microservice as opposed to a traditional SOA.
Perhaps the most important distinction is side effects. Microservices avoid them. To see why, let’s look at an older approach: Unix pipes.
ls | wc -l
Above, two programs are chained together: the first lists all of the files in a directory, the second reads the number of lines in a stream of input. Imagine writing a comparable program, then having to modify it into the below:
ls | less
Composing small pieces of functionality relies on repeatable results, a standard mechanism for input and output, and an exit code for a program to indicate success or lack thereof. We know this works from observational evidence, and we also know that a Unix pipe is a “dumb” interface because it has no control statements. The pipe applies SRP by pushing data from A to B, and it’s up to members of the pipeline to decide if the input is unacceptable.
Let’s go back to Company X’s Job and Invoice systems. Each controls a transaction and can be used together or separately: Invoices can be created for jobs, jobs can be created without an invoice, and invoices can be created without a job. Unlike Unix shell commands, the systems that own jobs and invoices have their own users working independently. But without falling back to a policy, it’s impossible to enforce rules for either system globally.
Say we want to extract out the key operations that can be repeatedly executed—the services for sending an invoice, mutating a job status and mutating an invoice status. These are completely separate from the task of persisting data.
Here this allows us to wire together the discrete components into two pipelines:
The invoice calculation related steps are idempotent, and it’s then trivial to compose a draft invoice or preview the amounts payable by the customer by leveraging our new dedicated microservices.
Unlike traditional SOA, the difference here is that we have low-level details exposed via a simple interface, as compared to a high-level API call that might execute an entire business action. With a high-level API, in fact, it becomes difficult to rewire small components together, since the service designer has removed many of the seams or choices we can take by providing a one-shot interface.
By this point, the repetition of business logic, policy and rules leads many to traditionally push this complexity into a service bus or singular, centralized workflow orchestration tool. However, the crucial advantage of microservice architecture is not that we never share business rules/processes/policies, but that we push them into discrete packages, aligned to business needs. Not only does this mean that policy is distributed, but it also means that you can change your business processes without risk.
“Wait a minute,” some of you may be murmuring over your morning coffee, “isn’t this just another name for SOA?” Service-Oriented Architecture (SOA) sprung up during the first few years of this century, and microservice architecture (abbreviated by some as MSA) bears a number of similarities. Traditional SOA, however, is a broader framework and can mean a wide variety of things. Some microservices advocates reject the SOA tag altogether, while others consider microservices to be simply an ideal, refined form of SOA. In any event, we think there are clear enough differences to justify a distinct “microservice” concept (at least as a special form of SOA, as we’ll illustrate later).
The typical SOA model, for example, usually has more dependent ESBs, with microservices using faster messaging mechanisms. SOA also focuses on imperative programming, whereas microservices architecture focuses on a responsive-actor programming style. Moreover, SOA models tend to have an outsized relational database, while microservices frequently use NoSQL or micro-SQL databases (which can be connected to conventional databases). But the real difference has to do with the architecture methods used to arrive at an integrated set of services in the first place.
Since everything changes in the digital world, agile development techniques that can keep up with the demands of software evolution are invaluable. Most of the practices used in microservices architecture come from developers who have created software applications for large enterprise organizations, and who know that today’s end users expect dynamic yet consistent experiences across a wide range of devices. Scalable, adaptable, modular, and quickly accessible cloud-based applications are in high demand. And this has led many developers to change their approach.
Whether or not microservice architecture becomes the preferred style of developers in future, it’s clearly a potent idea that offers serious benefits for designing and implementing enterprise applications. Many developers and organizations, without ever using the name or even labeling their practice as SOA, have been using an approach toward leveraging APIs that could be classified as microservices.
We’ve also seen a number of existing technologies try to address parts of the segmentation and communication problems that microservices aim to resolve. SOAP does well at describing the operations available on a given endpoint and where to discover it via WSDLs. UDDI is theoretically a good step toward advertising what a service can do and where it can be found. But these technologies have been compromised by a relatively complex implementation, and tend not to be adopted in newer projects. REST-based services face the same issues, and although you can use WSDLs with REST, it is not widely done.
Assuming discovery is a solved problem, sharing schema and meaning across unrelated applications still remains a difficult proposition for anything other than microservices and other SOA systems. Technologies such as RDFS, OWL, and RIF exist and are standardized, but are not commonly used. JSON-LD and Schema.org offer a glimpse of what an entire open web that shares definitions looks like, but these aren’t yet adopted in large private enterprises.
The power of shared, standardized definitions are making inroads within government, though. Tim Berners Lee has been widely advocating Linked Data. The results are visible through in data.gov and data.gov.uk, and you can explore the large number of data sets available as well-described linked data here. If a large number of standardized definitions can be agreed upon, the next steps are most likely toward agents: small programs that orchestrate microservices from a large number of vendors to achieve certain goals. When you add the increasing complexity and communication requirements of SaaS apps, wearables, and the Internet of Things into the overall picture, it’s clear that microservice architecture probably has a very bright future ahead.
Thanks for reading ❤
If you liked this post, share it with all of your programming buddies!
We explore Serverless with Java and Amazon Web Services (AWS). We'll create a simple serverless function using Amazon Web Services, Java, and Gradle. Learn how to simply integrate AWS deployment with a Gradle script. Learn how to use Okta to secure the serverless function with JSON Web Tokens, OAuth, and OpenID Connect.
An introduction to using AWS and AWS Lambda with the Serverless Framework.
Setting up Automated Continuous Delivery and Continuous Integration DevOps Pipeline AWS with AWS CodePipeline and AWS DataPipeline
How to Build and Manage Infrastructure as Code with Amazon Web Services - AWS CloudFormation, AWS CloudWatch and Amazon EC2
In this article, we'll better understand the features of monolithic versus microservices architectures, and what issues microservices help to address.