The microservice premium is the added cost and complexity of running services independently. Beyond writing business logic, teams need delivery pipelines, centralized logging, network monitoring, and container orchestration. That overhead trades simplicity and cost for independent scalability; for a small team or simple application, it can consume time better spent building features. section 12.3.3
Consider paying the premium when a concrete business need outweighs that overhead:
- A large team is slowed by developers interfering with one another’s deployments. section 12.3.3
- A module needs substantially different scaling hardware from the rest of the application. section 12.3.3
- Independent deployments, different technology choices, or strong security isolation matter enough to justify separate services. section 12.11 section 12.4.4
A practical default is to start with a well-structured modular monolith, then revisit the decision as the business and engineering organization grow. Splitting services also brings a network tax: calls can be slow or fail, and distributed systems add security, consistency, and observability concerns. A service boundary is not free just because it looks tidy on a diagram. section 12.3.3 section 12.11
Table 12.2 The 8 Fallacies & The Network Tax
|
The Fallacy (The Myth)
|
The Architectural Reality (The Tax)
|
The Technical Solution
|
|
1. The network is reliable.
|
The Reliability & Consistency Tax:
Network calls will randomly drop, and servers will crash mid-transaction,
leaving corrupted data.
|
Timeouts, Circuit Breakers, Exponential Backoff, and the
Saga Pattern.
|
|
2. Latency is zero.
|
The Latency Tax: A blazing-fast
5ms local database query turns into a slow 150ms HTTP network chain.
|
Asynchronous workflows, Event-Driven Architecture, and
aggressive Caching.
|
|
3. Bandwidth is infinite.
|
The Serialization Tax: Sending
massive "God Objects" across the wire hogs memory and chokes the
network pipeline.
|
Pagination, GraphQL (fetching only what you need), and
efficient binary serialization (like gRPC/Protobuf).
|
|
4. The network is secure.
|
The Security Tax: Internal
microservice traffic traveling across cloud infrastructure can be easily
intercepted by bad actors.
|
Zero Trust Architecture, Mutual TLS (mTLS), and Service
Meshes.
|
|
5. Topology doesn't change.
|
The Discovery Tax: Server IP
addresses change constantly as auto-scalers spin up and then spin down cloud
containers.
|
API Gateways, dynamic Service Discovery (e.g., Consul),
and Load Balancers.
|
|
6. There is one administrator.
|
The Versioning Tax: Autonomous
teams deploy updates at different times, accidentally breaking API contracts.
|
Strict API Versioning strategies and Consumer-Driven
Contract Testing.
|
|
7. Transport cost is zero.
|
The Financial Tax: Cloud
providers charge real money for network egress (data leaving the server) and
CPU cycles spent parsing JSON.
|
Payload compression, Local L1 caching, and minimizing
unnecessary cross-region network calls.
|
|
8. The network is homogeneous.
|
The Interoperability Tax: Different
services will inevitably be written in different languages and run on
different hardware.
|
Standardized communication protocols (REST/JSON) and
language-agnostic Message Brokers.
|
Use the fallacies-and-taxes table as a checklist for costs to account for before extracting a feature into a service. section 12.6.2
