minutes read

A Dive into Application Architecture - Part 4 - Microservices Architecture

Khoa Pham

March 16, 2023

feature image


Microservices consist of small, independent business services that may be implemented separately or autonomously. The fact that each service can be deployed independently is one of microservices architecture’s greatest benefits. Because each service runs separately, it may be updated, deployed, and expanded to match the demand for specific application functions.

The pros, the cons, and the use case of microservices

Services can be set up on their own. A team can make changes to an existing service without having to build and deploy the whole application again. Services are in charge of making sure that their own data or external state stays around. In the traditional model, where data persistence is handled by a separate data layer, this is not the case.

Well-defined APIs make it possible for services to talk to each other. Details about how each service works on the inside are kept hidden from other services. Services don't have to use the same libraries, frameworks, or tech stack.

There are some things that all microservices architectures have in common:

  • Componentization via services

A component is a piece of software that can be replaced or updated on its own.

  • Organized around business capabilities

The microservices way of dividing is to break things up into services that are set up by what the business can do.

  • Smart endpoints and dumb pipes

Microservices try to be as cohesive and independent as possible. This means that they have their own domain logic and use restful APIs to receive a request, apply logic, and make a response.

  • Decentralized governance

Sharing tested, useful code as libraries makes it easier for other developers to solve similar problems in the same way.

  • Decentralized data management

Microservices also decentralize data storage decisions. This method is sometimes called polyglot persistence or polyglot databases. This means that microservices like to let each service take care of its own database. These databases can be different versions of the same database technology or completely different systems.

  • Infrastructure Automation

That means automating deployment to each new environment and for each microservice separately.

  • Design for failure

Microservices are made to deal with failures and try to handle errors by taking the right steps.

The pros of Microservices Architecture

  • Agility

One of the most important things about microservices is that it's easier to handle bug fixes and new features because the services are smaller and can be deployed on their own. We can update a service without redeploying the whole app, and if something goes wrong, we can roll back an update.

If a bug is found in one part of a monolithic app, it can stop the whole process of releasing the app. So, new features had to wait until a bug was fixed before they could be added, tested, and made public.

  • Small, focused teams

A microservice should be small enough that it can be built, tested, and put into use by a single team. With the microservices model, a company can set up small, cross-functional teams that work on one service or a group of related services in an agile way. When we have a small team, we can move faster. Large teams tend to be less productive because they take longer to communicate and have more work to do. So when that happens, agility goes down.

  • Separated code base

Over time, a monolithic app's code base will grow so big that code dependencies will get tangled. Adding a new feature requires a lot of changes to the code that is already there. Since microservices don't share code or data stores with other services, dependencies are kept to a minimum. This makes it easier to add new features.

  • Right tech stack for the job

In traditional layered architectures, an application usually shares a single stack and is supported by a large relational database. This method has a few problems. For example, every part of an application must use the same stack, data model, and database, even if some modules could do their jobs better with a different tool. It's very frustrating for developers who know that there's a better, faster way to build these parts. Also, it's frustrating for developers when the application stack is too old and they can't use the best practices from newer projects on their own.

But with microservices architecture, small teams can choose the technology that works best for their microservice and use a mix of technology stacks on their services. Because microservices are set up separately and talk to each other using REST, event streaming, and message brokers. It is possible for each service to have a stack that is best for that service.

Technology is always changing, and development libraries and tools are also changing very quickly. Since an application is made up of several smaller services, it is much easier and cheaper to change it into a microservice with better technology.

  • Fault isolation

Loose coupling between microservices also helps to separate faults and make applications more resilient. If one of our microservices stops working, the application as a whole won't be affected. Of course, we should make sure our microservices can handle errors and handle them in the right way, such as by using retry and circuit-breaking patterns. Even when things go wrong, if we fix them in a way that doesn't hurt our business, our customers will always be happy.

  • Scalability

Microservices can be scaled up or down separately, so we can expand sub-services that use more resources without expanding the whole application. So, we can say that microservices need less infrastructure than monolithic apps because they let us scale only the services, we need instead of scaling the whole app.

  • Data isolation

Since microservices use the database-per-service pattern, databases are kept separate. Schema updates become easier because they only affect a single database. Schema updates can be very hard and risky in a single-piece application.

The cons of Microservices Architecture

Microservices have a lot of pros, but they also have many cons. When we switch from a monolith to microservices, we have to manage a lot more things. Here are some of the problems we have to think about before we can use microservices architecture.

  • Complexity

The services in a microservices application need to work together and make something of value. Since there are many services, there are more things that can go wrong than with a single application. Each service is easier to use, but the whole system is harder to understand.

Even deployments can be hard to manage when hundreds of services deploy at different times. Think about the way they communicate with each other. Components in a monolithic application can easily communicate with each other because it is inter-process communication, which means it can be the same machine and the same process. But communication between microservices is hard, and we need a plan to handle inter-service communications between servers or in different parts of the geo-locations.

  • Network problems and latency

Since microservices communicate with each other through inter-service communication and are small, we should manage network problems. If we call a chain of services for a specific request, this will cause latency issues, and APIs will need to be designed correctly for proper communication. To avoid chatty API calls. We should think about asynchronous communication patterns like message broker systems.

  • Development and Testing

Think about the E2E process, which is one of the most important business needs. If the requirements involve several microservices that need to work as one application, it's harder to build and test these end-to-end (E2E) processes in a microservices architecture than in a monolithic architecture. Not all of the tools out there are made to work with service dependencies. It can be hard to refactor across service boundaries.

  • Data Integrity

Each microservice has its own way of keeping data. So it can be hard to keep data consistent. Most of the time, we should try to be as consistent as possible. Transactional operations, on the other hand, will always be hard.

But these problems aren't stopping people from using microservices. Most organizations are willing to deal with these problems and are adapting their technologies to microservices architecture so they can get benefits from this architecture.

Microservices design patterns

Database per Microservice

When we decide to replace a huge monolithic system with many microservices, the database is the most important thing. In a monolithic architecture, a big database is used in one place. In the beginning, developers might try keeping the database as it is, even when they move to microservice architecture. While it gives some short-term benefits, it is an anti-pattern, especially in a large-scale system, as the microservices will be tightly coupled in the database layer. A better way to do things is to give each microservice its own data store. This way, services don't have to be tightly linked in the database layer.

Database per Microservice


  • Data is completely owned by a Service.
  • Loose coupling among the teams developing the services.
  • With standard database tools, we can make a backup of the whole system that is completely consistent and incremental without any downtime.
  • If our architecture is event-driven, we can store events along with our data. If we store events in the database, we can lose our broker and not lose any data.
  • Data relationships that are fully constrained and correct.


  • Sharing data among services gets harder.
  • It gets much harder to give an application-wide ACID transactional guarantee.
  • Taking the Monolith database apart into smaller pieces requires careful planning and is a hard task.
  • In case of a problem, it could be hard to fix if data-related tasks are spread across multiple microservices.

Event Sourcing

In a microservice architecture, where each microservice has its own database, the microservices need to share information. Systems that are reliable, scalable, and can handle mistakes should communicate with each other asynchronously by exchanging Events.

Use event-based architecture with event sourcing in this scenario. In traditional databases, the current "state" of the business entity is stored directly. In Event Sourcing, events that change the state of an entity or that are otherwise important are stored instead of the entities themselves. It means that changes to a business entity are saved as a series of events that can't be changed. By reprocessing all the Events of a business entity at a given time, the state of that business entity can be found. Since data is stored as a series of events instead of as direct updates to data stores, different services can replay events from the event store to figure out how their own data stores should be set up.

Event Sourcing


  • Excellent for safety. If the source further down fails, the data from it can be reconstituted from the event store.
  • Very easy to change. Messages of any kind can be saved. As long as the right access rights are given, any consumer can use the event store.
  • When used with an event-driven message broker like Kafka, it's great for reporting data in real time.
  • Give atomicity to systems that can grow quickly.


  • It gets hard to read entities from the event store, so we usually need an extra data store (CQRS pattern)
  • The system as a whole gets more complicated, and we usually need Domain-Driven Design.
  • The system needs to deal with duplicate events (idempotent) or missing events.
  • It gets harder to migrate the schema of events.

Command Query Responsibility Segregation

If we use event Sourcing, it will be hard to read data from the Event Store. We need to process all entity events in order to get an entity from the Datastore. Also, sometimes we have different needs for read and write operations when it comes to consistency and speed.

CQRS separates reads and writes into two different models. Commands are used to change data, while queries are used to read data.

  • Commands should be based on tasks instead of data.
  • Instead of being done at the same time, commands can be put on a queue and done at a later time.
  • The database is never changed by a query. A query gives back a DTO that does not encapsulate any domain knowledge.

We can physically separate the read data from the write data to keep them even more separate. In that case, the read database can use its own, query-optimized data schema. For example, it can store a materialized view of the data so that complex joins or ORM mappings don't have to be done. It might even store data in a different way. For example, the write database could be a relational database, while the read database could be a document database.

Command Query Responsibility Segregation


  • Independent scaling. CQRS makes it possible for the read-and-write workloads to grow at different rates, which may lead to fewer lock conflicts.
  • Optimized data schemas. The read side can use a schema that is optimized for queries, while the write side uses a schema that is optimized for changes.
  • Security. It's easier to make sure that only the right domain entities are writing to the data.
  • Keeping things separate. Keeping the read and write sides of a model separate can make it easier to fix and change. Most of the business logic that is hard to explain goes into the write model. The read model can be easy to understand.
  • Simpler queries. By saving a materialized view in the read database, the application can avoid complex joins when querying.


  • To support the CQRS pattern, we need to know a lot about different database technologies.
  • When using CQRS patterns, more database technologies are needed. This means that there are more costs, either in terms of hardware or, if a cloud provider is used, in terms of the cost of using the cloud.
  • Service level agreements need to take special care to make sure that data is consistent.
  • Using a lot of databases means there are more places where something could go wrong, so companies need to have thorough monitoring and fail-safe mechanisms in place.
  • Code duplication is unavoidable since read and write models need to be independent.


If we use microservice architecture with a database per microservice, it can be hard to keep everything consistent through distributed transactions. We can't use the traditional Two-phase commit protocol because SQL databases don't support it or it doesn't scale (many NoSQL Databases).

In Microservice Architecture, we can use the Saga pattern for distributed transactions. Saga is an old pattern that was created in 1987 as an alternative way to think about database transactions that take a long time to finish in SQL databases. But a newer version of this pattern works great for distributed transactions as well.

The Saga pattern is a local transaction sequence in which each transaction updates data in the Data Store within a single Microservice and publishes an Event or Message. The first transaction in a saga is initiated by an external request (Event or Action). Once the local transaction is done (data is stored in the Data Store and a message or event is published), the published message or event triggers the next local transaction in the Saga.

Most Saga transactions are coordinated in one of two ways:

  1. Choreography: Decentralized coordination in which each Microservice sends and receives events and messages from other Microservices and decides whether or not an action should be taken.
  2. Orchestration: Centralized coordination in which an Orchestrator tells the participating Microservices which local transaction needs to be done.


  • The main benefit of the Saga Pattern is that it helps maintain data consistency across multiple services without tight coupling. In a microservice architecture, this is a very important part.
  • Orchestrator makes it easy to keep track of the whole workflow in a single place.
  • Avoid having services that depend on each other in a loop.
  • When more steps are added, the complexity of the transaction stays the same.
  • The orchestrator works well for complicated workflows with many participants.
  • The choreography, on the other hand, works well for simple workflows with a small number of participants.


  • If there are too many local transactions, it will be hard to keep track of which services are listening to which events.
  • If the orchestrator has too much logic, it could be hard to keep maintaining

Backends for Frontends

BFF Architecture ensures the Backend and the Frontend applications are separated and decoupled into Services. It can communicate together via either API or GraphQL. If the app has both a mobile app and a web app, it would be hard to use the same backend Microservice for both of them. Most of the time, the API requirements for mobile apps and web apps are different because mobile apps have different screen sizes, displays, performances, energy sources, and network bandwidths.

BFF also has other benefits, like acting as a facade for downstream Microservices. This makes communication between the UI and downstream Microservices less chatty. Also, BFFs are used to increase security when downstream microservices are set up in a DMZ network, which is very safe.

Backends for Frontends


  • We can make them work best with a certain UI.
  • Make sure people are safer.
  • Give the UIs and Microservices that come after them a less chatty way to communicate with each other.
  • It can have code for things like authentication and authorization, logging, caching, retry policies, load balancing, etc. that affect more than one area. This can make the code for services further down much simpler because they won't have to worry about these things.
  • It can make handling errors more consistent.
  • BFFs make it possible for development teams to be more in charge of their projects by giving them more freedom and flexibility.


  • Code duplication among BFFs.
  • There are a lot of BFFs in case a lot of other UIs are used (e.g., Smart TV, Web, Mobile, Desktop).
  • BFFs should not have any business logic. Instead, they should only have client-specific logic and behavior.
  • By using BFFs, we're adding more moving parts that need to be managed, maintained, monitored, and deployed.

API Gateway

In Microservice Architecture, the user interface (UI) usually connects to more than one Microservice. If the Microservices are small (FaaS), the Client may need to connect to a lot of them, which can be noisy and difficult. Also, both the services and their APIs can change over time. Large businesses will have other issues that affect them all (SSL termination, authentication, authorization, throttling, logging, etc.).

API Gateway is such a way that might solve these problems. API Gateway is a facade that sits between the Client APP and the Backend Microservices. It can act as a reverse proxy and send the request from the Client to the right Backend Microservice. It can also send the client's request to multiple Microservices and return the responses from all of them to the Client. It also helps with important issues that cut across different areas.

API Gateway


  • Cut down on the number of calls back between the client and the microservices.
  • SSL termination, authentication, and authorization provide high security.
  • Cross-cutting issues, such as logging and monitoring, throttling, and load balancing are handled centrally.
  • Offer loose coupling between frontend and backend microservices.


  • Can cause Microservice Architecture to have a single point of failure.
  • Due to the extra network call, the delay got worse.
  • If it isn't scaled, it's easy for them to become the slowest part of the whole Enterprise.
  • Extra costs for development and maintenance.


If we want to use Microservice Architecture in a "brownfield project", we need to migrate the existing Monolithic applications to Microservices. Moving an existing, large, production-ready monolithic application to microservices is not easy because it could affect the availability of the application.

The Strangler pattern is a way to solve the problem. Strangler pattern means to move a Monolithic application to a Microservice Architecture by replacing certain functions with new Microservices over time. Also, new features are only added to Microservices, while the old Monolithic application stays the same. The requests are then sent between the legacy Monolith and the Microservices by setting up a Facade (API Gateway). Once the functionality is moved from the Monolith to the Microservices, the Facade will pick up the client request and send it to the new Microservices. Once all the functions of the legacy monolith have been moved, the legacy monolith application is "strangled".



  • Move monolithic applications to microservices in a safe way.
  • Both the migration and the development of new features can happen at the same time.
  • The process of moving can happen at its own pace.


  • It gets hard for the existing Monolith and the new Microservices to share the Data Store.
  • By adding a Facade (API Gateway), the system's latency will go up.
  • It gets harder to test E2E.

Circuit Breaker

In Microservice Architecture, where Microservices communicate to each other Synchronously, a Microservice will usually call other services to meet business needs. Calls to other services can fail because of temporary problems (slow network connection, timeouts, or temporal unavailability). In this case, retrying the call can fix the problem. But if there is a major problem (such as the Microservice not working at all), the Microservice will be down for a longer time. In these situations, retrying is useless and wastes valuable resources (the thread is blocked, which wastes CPU cycles). Also, if one service fails, it could cause failures in other parts of the application. In these situations, it's better to fail immediately.

In these kinds of situations, the Circuit Breaker pattern can save the day. A proxy that works like an electrical circuit breaker should be used for a Microservice to request another Microservice. The proxy should keep track of how many recent failures there have been and use that number to decide whether to let the operation continue or just throw an error immediately.

There are three possible states for the Circuit Breaker:

  • Closed: The Circuit Breaker sends requests to the Microservice and keeps track of how many times they fail over a certain amount of time. If a certain number of failures happen in a certain amount of time, it trips and goes to Open State.
  • Open: The Microservice request fails right away, and an exception is sent back. The Circuit Breaker goes to a Half-Open state after a timeout.
  • Half-Open: Only a certain number of requests from the Microservice are allowed to go through and run the operation. If these requests work, the circuit breaker goes to a state called "Closed." If any request fails, the Circuit Breaker goes to the Open state.
Circuit Breaker


  • Make the Microservice Architecture more resilient and able to handle faults.
  • Stops failures from spreading to other Microservices.


  • Need sophisticated handling of exceptions.
  • Logging and Monitoring.
  • Should be able to be reset manually


Because each microservice can be developed, deployed, and scaled independently by a business, microservices architecture enables increased agility and pluggability. However, microservices monitoring and management can be particularly challenging, given the need to track and maintain each service component of an application and its interactions.