Summary

During my overall IT carreer I came across different architectural design patterns where oppinions differ on the question if they’re the right ones for the problems/challanges teams are dealing with. Before reading this book I was familiar with some of the microservices concepts but it was some article (on modern architectures) that rouse my attention and introduced me to Sam Newman. I was reading the (German) iX Magazin about modern architectures. At the same time I was surprised to read about the similarities between Hexagonal Architecture and microservices. But also topics like DDD, CD, CI are bound together in way that you need to take a hollistic approach to (building) microservices.

I recommend this book Along with Martin Fowlers extensive arcticles on microservices. to anyone willing to spend some time (book has ca. 500 pages) learning about Information hiding, communication between microservices, proper teams setup, role of (IT) architects and much more. Fair enough the author emphasizes multiple time the complexity of decoupling existing services (monoliths) into smaller, independent ones (microservices). The book recommendations in each chapter also give a great hint where you can enlarge upon a specific topic.

What follows next is an ORG mode / outline style collection of notes, thoughts and quotes from the book.

Ch. 1: What are Microservices?

Definition

Microservices are independently releasable services that are modeled around a business domain. A service encapsulates functionality and makes it accessible to other services via networks.

• Microservices are a type of SOA architecture
  • service boundaries are important
  • independent deployability is key
• Microservices embrace the concept of Information hiding Introduced by David Parnas in Information Distribution Aspects of Design Methodology

Key Concepts

• Independent deployability

  • criteria for this: make sure microservices are loosely coupled
    • be able to change one service without having to change anything else

• Modeled around a business domain

  • definition of service boundaries (see DDD)

• Owning their own state

  • hide internal state (same as encapsulation in OOP)
  • clean delineation between internal implementation details and external contract
    • be backward-compatible
  • hide database that backs the service
    • avoid DB showing

• Size

  “A microservice should be as big as my head” - James Lewis

  • keep it to the size at which it can be easily understood

• Flexibility

  “Microservices buy you options” - James Lewis

  • they have a cost and you must decide whether the cost is worth the options you want to take up

• Alignment of architecture and organization

  “Organizations which design systems are constrained to produce designs which are copies of the communication structures of the organization” - Melvin Conway

  • have team vertically organized
    • same team owns front-end, business logic, data, back-end, security
      • a so called stream-aligned team

Advantages

• Technology Heterogeneity

• Robustness

  • a component in a system can fail but as long as the problem doesn’t cascade, the rest of the system still works

• Scaling Book recommendation: The art of scalability

• Ease of Deployment

  • fast delivery of features
  • decreases fear of deployment (see Accelerate)
  • change a single service and deploy it independently of the system

• Organizational alignment

  • small teams working on small code bases tend to be more productive
  • microservices allow us to better align our architecture to our organization
    • minimize the number of people working in the team
      • helps to find the sweet spot of team size and productivity

• Composability

  • reusable components
  • allow functionality to be consumed in different ways
    • for different purposes: website, desktop, mobile application etc.

Pain Points

• Developer Experience

  • new technologies are options not requirements
  • when adopting microservices
    • you’ll have to understand issues around data consistency, latency, service modeling
    • and how these ideas change the way you think about software development
    • it takes time to understand new concepts
      • this leads to less time developing new features

• Technology overload

• Costs

• Reporting

  • data and logs are scattered across multiple components

• Monitoring and troubleshooting Book recommendation: Distributed Systems Observability

• Security

• Latency

• Data consistency

Ch. 2: How to model microservices

Information hiding

• hide as many details as possible behind a module / microservice boundary
• Parnas identified following benefits:
  • improved development time
  • comprehensibility
    • each module is isolated and therefore better to understand
  • flexibility

Cohesion

• code that changes together, stays together
• strong cohesion
  • ensure related behavior is at one place
• weak cohesion
  • related functionality is spread across the system

Coupling

• loosely coupled
  • change to one service should not require a change to another
• a loosely coupled services knows as little as it needs about the services it communicates with
  • limitation of number of different types of calls is important

Interplay of coupling and cohesion

A structure is stable if cohesion is strong and coupling is low.

• cohesion applies to the relationship between things inside a boundary
• coupling describes relationship between things across a boundary
• still: there is no best way how to organize code

Types of coupling

Domain coupling

• one microservice interacts with another microservice because it needs the functionality the other microservice provides

• considered as a loose form of coupling
• again, information hiding: Share only what you absolutely have to, and send only the absolute minimum amount of data that you need

Pass-through coupling

• one microservice passes data to some other microservice because data is needed by another microservice

Common coupling

• when 1 or 2 microservices make use of a common set of data
  • use of shared DB
  • use of shared memory/filesystem
• problem: changes to data can impact multiple microservices at once
• better solution would be to implement CRUD operations and let only 1 microservice handle shared DB operations

Content coupling

• when an upstream service reaches into internals of a downstream service anc changes its internal state

DDD

DDD stands for Domain-Driven Design.

Concepts

• Ubiquitous language

  • use the same terms in code as the user use
  • have common language between delivery team and actual people (aka customers)
    • helps to understand business by logic
    • helps with communication
  • use real-world language in code

• Aggregates

  • a representation of real domain concept

    • something like an Order, an Invoice, Stock Item

  • aggregates typically have an information cycle around them

  • in general

    • aggregate as something that has

      • state
      • identity
      • information cycle

      that will be managed as part of the system

  • aggregates can have relationships to other aggregates

    

    Relationship between aggregates

• Bounded context

  • a larger organizational boundary
    • within it explicit responsibilities need to be carried out
  • bounded contexts hide implementation details (Information hiding)
  • bounded contexts contain 1-n aggregates
    • some aggregates may be exposed outside the bounded context
    • others may be hidden internally

Event Storming

• collaborative brainstorming exercise designed to help design a domain model
• invented by Alberto Brandolini

Boundaries between microservices

There are some factors when defining clear boundaries between microservice

• volatility
• data
  • also with concern to security
• technology
• organizational
  • Layering Inside vs Layering Outside

Ch. 3: Split the monolith

• you need to have a goal before moving to microservices

  • should be a conscious decision

  • without clear understanding of what you want to achieve, you could fall into the trap of confusing activity with outcome

    Spinning up a few more copies of your existing monolith system behind a load balancer may well help you scale your system much more efficiently than going through a complex and length decomposition to microservices.

Decomposition patterns

• Strangler fig pattern By Martin Fowler

  

  Strangler fig pattern

• Parallel run

• Feature toggles

Data Decomposition concerns

• performance
• data integrity
• transactions
• Tooling
• Reporting DB

Ch. 4: Communication styles

• styles for IPC communications
  • synchronous blocking
  • asynchronous blocking
  • request-response
  • Event-Driven Architecture
  • Common data

EDA

• events vs messages
  • event: is a fact
  • message: is a thing
  • a message contains an event

Ch. 5: Implementing communication

Criterias for ideal technology

• backward compatibility

• make your interface(s) explicit

  • use of explicit schemas Like OpenAPI

• keep your APIs technology-agnostic

• make your service simple for the consumers

• hide internal implementation details

Technology choices

• RPC

  • SOAP
  • gRPC

• REST Book recommendation: REST in Practice: Hypermedia and Systems Architecture (by Jim Webber, Savas Parastatidis, Ian Robinson)

• GraphQL

  • alternative: BFF (Backend-For Frontend) pattern This article provides a quite good introduction.
    

    Backend for Frontend (BFF)

• Message brokers

  • use queues/topics

API Gateway

• built on top on existing HTTP proxy products
• main function: reverse proxy
  • but also authentication, logging, rate limiting
• Examples:
  • AWS API Gateway
  • GCP API Gateway

Ch. 6: Workflow

Distributed Transactions

Two-phase Commits (2PC)

• a commit algorithm to make transactional changes in a distributed system, where multiple separate parts need to be updated

Sagas

• coordinate multiple changes in state
• but without locking resources for a long period
• involves
  • backward recovery
  • forward recovery
• allows to recover from business failures not technical ones
• when rollback is involved, maybe a compensating transaction is needed

Books

• Enterprise Integration Patterns: Designing, Building, and Deploying Messaging Solutions
• Practical Process Automation

Ch. 7: Build

• on Continuous Integration (CI)
• how to organize artifacts
  • monorepo
  • multirepo

Ch. 8: Deployment

Principles of Microservices Deployment

• isolated execution

  • own computing resources
  • don’t impact other microservices instances

• focus on automation

  • adopt automation as core part of your culture

• Infrastructure as a Code Book: Infrastructure as Code: Managing Servers in the Cloud

• zero-downtime deployment

  • independent deployability
    • new deployment of microservices can be done without downtime to users/clients of microservices

• desired state management

  • maintain microservices in a defined state
    • allocate new instances if needed
  • GitOps
    • brings together desired state management and IaC (Infrastructure as Code)

• progressive delivery

  • implement many of the ideeas in Accelerate
  • separate deployment from release
  • feature releases
    • use as part of trunk-based development
    • not yet finished functionality can be deployed and hidden from users (e.g. feature toggles)
    • functionality can still be turned on/off
  • canary releases
  • parallel runs

Ch. 10: From monitoring to obersavability

The observability of a system

• is the extenct to which you can understand the internal state of the system from external output

• monitoring is something we do

  • it’s an activity

• observability

  • rather a property of a system

• pillars of observability

  • metrics
  • logging/logs
  • events
  • traces
  

  Correlation IDs in logs

Alert fatigue

Alert fatigue—also known as alarm fatigue—is when an overwhelming number of alerts desensitizes the people tasked with responding to them, leading to missed or ignored alerts or delayed responses – Source Also a good reading: Alarm design: From nuclear power to WebOps.

What makes a good alert

An alert has to be:

• relevant
• unique
• timely
• prioritized
  • give enough information to decide in which order alerts should be dealth with
• understandable
  • information has to be clear and readable
• diagnostic
  • it needs to be clear what is wrong
• advisory
  • help the operator understand what actions need to taken
• focussed
  • draw attention to the most important issues

On the importance of testing quote

“Not testing in production is like not practitioning with the full orchestra because your solo sounded fine at home”

Semantic monitoring

• compare against normal conditions
• you could use synthetic transactions
• other options
  • A/B testing
  • canary releases
  • Chaos engineering
  • parallel runs
  • smoke tests

Tools

• opentelemetry.io

Ch. 11: Security

Lifecycle of secrets

• Creation
  • How we create the secret
• Distribution
  • How do we make sure the secrets get to the right place?
• Storage
  • Is the secret stored in a way only authorized parties can access it?
• Monitoring
  • Do we know how secret is used?
• Rotations
  • Are we able to change the secret without causing problems?

Ch. 12: Resiliency

Resiliency

• defined by David D. Woods Book: Resilience Engineering: Concepts and Precepts

• aspects

  • robustness
    • ability to absorb perturbation
  • rebound
    • recover after a traumatic event
  • graceful extensibility
    • how to deal with an unexpected situation
  • sustained adaptability
    • adapt to changing environments, stakeholders and demands

Ch. 14: User interfaces

Stream-aligned teams

• topologies how to build organizations, teams Book recommendation: Team Topologies: Organizing Business and Technology Teams for Fast Flow

• aka “full-stack teams”

• a team aligned to a single, valuable stream of work

• the team is empowered to build and deliver customer or user value as quickly and independently as possible, without requiring hand-offs to other teams to perform parts of the work

Microfrontends

• architectural style where independently deliverable frontend applications are composed into a greater whole Check out Martin Fowler’s article.

• possible implementations

  • widget-based decomposition
  • page-based decomposition

SCS

• stands for Self-Contained Systems Read more on the official site

• highlights

  • each SCS is an autonomous web application with no shared UI
  • each SCS is owned by one team
  • asynchronous communication should be used whenever possible
  • no business code can be shared between multiple SCSs

Ch. 15: Organizational structures

• Stream-aligned teams
  • concept aligns with loosely-coupled organizations (as in Accelerate)

Conways Law

“Any organization that designs a system will inevitably produce a design whose structure is a copy of the organizations communication structure” - Melvin Conway

All about people

“Whatever industry you operate in, it is all about your people, and catching them doing things right, and providing them with the confidence, the motivation, the freedom and desire to achieve their true potential” - John Timpson Also interesting is the concept of paved roads, where best-practices are available but deviations are also allowed.

Ch. 16: The evolutionary architect

Role of architects

• We should think of the role of IT architects more as town planners than architects for the built environment Sam uses the Seagram Building (designed by Mies van der Rohe) as an universal place to visualize the role of an (IT) architect.

Buildings and software

The comparison with software should be obvious. As our users use our software, we need to react and change. We cannot foresee everything that will happen, and so rather than plan for any eventuality, we should plan to allow for change by avoiding the urge to overspecify every last thing. Our city (the system) needs to be a good, happy place for everyone who uses it. One thing that people often forget is that our system doesn’t just accommodate users; it also accommodates developers and operations people who also have to work there, and who have the job of making sure it can change as required.

Governance

Governance ensures that enterprise objectives are achieved by evaluating stakeholder needs, conditions and options; setting direction through prioritisation and decision making; and monitoring performance, compliance and progress against agreed-on direction and objectives. – Cobit 5

Responsibilities of the evolutionary architect

• Vision
  • clearly communicated technical vision for the system that will help meet requirements of customers and organization
• Empathy
  • understand impact of decissions on customers and colleagues
• Collaboration
  • engage with as many of your pears and colleagues as possible to help define, refine and execute the vision
• Adaptability
  • tech vision changes as required by customers/organization
• Autonomy
  • balance between standardizing and enabling autonomy for your teams
• Governance
  • system being implemented fits the tech vision
  • make sure it’s easy for people to do the right thing

Book recommendations

• Building evolutionary architectures
• The software architect elevator