CSE Roadmap

Core tier · about 4 hours · 8 parts

Part 3 of 8 · 2 min read

The benefits

What the style buys, and the price tag on each item.

BM lists the benefits in plain terms. This section gives each one with its cost, because the cost is what interviewers ask for.

Benefit What it means The cost attached
Technology choice Each service may use the language or database that fits it More tools to learn, patch and run
Resilience A failed service need not bring the others down New failure modes: slow networks, partial outages
Scaling Only the busy service needs more machines Many separate deployments to run
Easy deployment A one-line change ships on its own Needs strong automation and a tested rollback
Team alignment A small team owns each service Needs real team independence (section 5)
Replaceability A small service can be rewritten rather than repaired Interfaces must stay stable

Take the middle ones one at a time.

Resilience rests on the bulkhead idea from ship design: watertight walls keep one leak from sinking the ship. BM calls service boundaries the obvious bulkheads. The warning comes in the same breath. Networks and machines fail, so each caller must decide what to show when a service is gone.

Bulkhead

A bulkhead is a wall that stops a failure in one part of a system from spreading to the rest.

Example: if the recommendations service is down, the product page still loads, just without the "you may also like" row.

Scaling gets a real case in BM. Gilt, an online fashion retailer, started in 2007 with one Rails application. By 2009 the application could not cope with the load. Splitting out the core parts let the company scale only the busy ones. BM reports more than 450 microservices at the time of writing, each on several machines.

A worked example (made up to show the idea). A monolith needs 10 machines, because the search part is busy and the rest is not. Search alone needs 8 machines' worth of work. The other four parts need 2 machines' worth together. As one deployment, all 10 machines run all five parts. Split apart, search runs on 8 machines and the others on 2 to 4. The saving is the difference, but only if the tooling to run five deployments costs less.

Deployment is the benefit BM puts near the top. One change in a million-line monolith means redeploying the whole thing. Because that is risky, releases become rare, and changes pile up. A bigger pile is a riskier release. With a small service, a release is small and a rollback is fast.

Common mistake

Treating the benefits as free. Each one in the table needs work elsewhere: automation for deployment, good monitoring for resilience, and clear ownership for team alignment. Without that work, the system keeps the costs and loses the gains.

Check yourself: 1. Pick one benefit from the table and name the cost that must be paid for it. 2. In the made-up scaling example, why is the saving not guaranteed? 3. Why does BM say large, risky releases become rarer and therefore riskier?