CSE Roadmap

Common tier · about 4 hours · 7 parts

Part 6 of 7 · 3 min read

Comparing the three styles

Each style trades simplicity against scale, control and agility.

Comparing the three styles

Think of choosing a vehicle: a van, a convoy of trucks, or a fleet of couriers. Each moves goods, with different cost, speed and complexity.

FSA rates each style on a five-star scale for the characteristics from Level 0. The ratings appear in charts, so the table below gives the qualities the text states in words, not star counts.

Pipeline Service-based Event-driven
Kind Monolithic Distributed Distributed
Partitioning Technical (by filter type) Domain Mainly technical: a domain is spread across processors, mediators and topics
Quanta Always one One or more (a shared UI or database makes one) One to many, set by shared databases and by request-reply
Typical size A few filters 4 to 12 services Many processors
Strengths Simple, cheap, easy to extend with filters Pragmatic: good agility, testability, deployability, fault tolerance, with moderate cost Five stars in the book for performance, scalability, fault tolerance and evolvability
Weak points Scale and deploy as one unit Shared database coupling, coarse deployment Simplicity and testability rate low: event flows are dynamic and the number of scenarios can reach thousands
Data One store Shared database with logical partitions Often per processor, eventual consistency

Two points the books stress. First, service-based has no five-star rating, yet rates four stars in many areas, which explains its popularity for business applications. Second, request and reply can tie two event processors into one quantum: if the first must wait for an order id from the second, they share characteristics even though the messages are asynchronous.

A worked example (made up). A bank's loan application:

Part Best fit Reason
Nightly credit data import and cleaning Pipeline One-way steps, easy to extend
Customer portal, loan officer tools, back office Service-based A few domain services, shared customer data, ACID transactions
Credit check, fraud scoring, notification Event-driven Parallel reactions to "application submitted"

One organisation can use all three. The styles also combine: a service-based system may use events between its parts.

Interview angle

Answer in your head first, then open it.

What is the pipeline style?

One-way filters joined by pipes: producer, transformer, tester, consumer. Simple and extensible.

What is service-based architecture?

A few coarse domain services with one shared database. Distributed, but cheaper than microservices.

Broker or mediator?

Broker: no control, highest scale and decoupling. Mediator: central control and better error handling, at a cost to speed and coupling.

How does request-reply work over queues?

A request queue and a reply queue with a correlation id.

When would you avoid event-driven?

When the flow needs strict control and immediate answers, or the team cannot handle async debugging.

Check yourself:

  • Which style would you pick for a nightly ETL job, and why?
  • What is the main weakness of the broker topology?

Words you will meet

Word Plain meaning
Pipeline architecture One-way filters joined by pipes
Pipe A one-way channel between filters
Filter A single-task step: producer, transformer, tester or consumer
Architecture quantum An independently deployable part with its own database and characteristics
Service-based architecture A few large services sharing one database
Domain service A coarse-grained service for one business area
Entity library Shared code that maps database tables to objects
API layer A gateway between the UI and the services
Event-driven architecture Components react to events asynchronously
Initiating event The event that starts a flow
Processing event An event announcing what a processor did
Broker topology Event flow without a central coordinator
Mediator topology A central mediator drives the flow with commands
Request-reply A request queue and a reply queue used for answers
Correlation id A header that ties a reply to its request
Responsiveness How fast the user is told the action was accepted