This is a mix of Lewis and Fowler article and my own experience on microservices.
Services Vs Monoliths
| Monolith (Cons) | Services (Pros) |
|---|---|
| Build and deploy the whole application after any change | Build and deploy only affected components |
| Slow release cycle to integrate all code in monolith | Faster and uncorrelated releases of smaller services |
| Built using a single stack | Each service can use its own stack (ex: language) |
| Scale by deploying multiple copies of the monolith | Scale by deploying only many instances of bottle neck services |
| Programming interface to separate components | Service contracts to separate components (better isolation) |
| Monolith (Pros) | Services (Cons) |
|---|---|
| Simple to deploy and need few infrastructure | Complex to deploy and relies on infrastructure (service routing, application containers …) |
| Allow fine grained coop between components (call overhead low) | Coarsed grained collaboration between services (network latency) |
| Consistent versioning between components (more predictable interactions) | Business flows got through different services and versions (many untested interactions) |
SOA Vs Micro services
| Service Oriented Architecture | Microservices |
|---|---|
| More strict contracts (ex: based on webservices) | Contracts designed to be easily changed (tolerant reader pattern, ex: adding optional fields to protocol buffer) |
| Orchestration delegated to service Bus | Services decide on their dependencies |
| N/A | Heavy user of virtualisation and container technologies |
Dumb pipes and Smart end-points ?!
I think there are a number of advanced features any service bus should have.
- Service routing : route messages between services to the right node.
- Load aware routing : spread load across service instances
- Session affinity : queries from a client go to the same node to avoid over-replicating state.
- Flow tracing : monitor the service call tree for a given external client query
Some other can be implemented as services.
- Service discovery : this can be implemented as a meta-service
- Circuit breakers
Technical architecture should be a mirror of human organisation
Services are a way to break Conway adage : “Large systems designed by companies will mirror their communication structures”
| Stack layers | Team organisation |
|---|---|
| Decentralized source control and data stores | Release when you are ready, do not wait for the rest of the company |
| Full stack (presentation/backend/database) | Less team involved (more agility) to develop new features. Team are dedicated to a business product, not a layer of the system |
| Release and deploy automation | You build it, you run it. Developers get experience on operation constraints |
Concept map
