Glossary

Operational definitions — what a term means when someone has to decide with it, not the dictionary entry.

Where two terms are frequently confused, the distinction is stated.

A

Abstraction — A representation that exposes what matters for a purpose and hides the rest. A good abstraction reduces what you need to know; a bad one adds a layer without removing any.

Aggregate — In DDD, a set of objects treated as a single unit of transactional consistency, reached only through its root.

Anti-corruption layer — A translation layer that keeps an external system's model from leaking into yours.

Availability — The fraction of time a system responds within the stated criterion. Expressed as a percentage over a window; without the window and without the criterion, the number means nothing — which is why a system can count as available while returning a wrong result.

B

Backpressure — The mechanism by which an overloaded consumer signals the producer to slow down. Without it, overload turns into message loss or memory exhaustion.

Bottleneck — The resource that saturates first and caps the capacity of the whole. Optimizing any other point does not increase total capacity.

Bounded context — The boundary within which a domain model and its language have a single, consistent meaning. Two contexts may use the same word with different senses, and that is correct.

Bulkhead — Resource isolation so that exhaustion in one part does not bring down the rest. The name comes from the watertight compartments of a ship.

C

CAP — The result stating that, during a network partition, a distributed system cannot be simultaneously consistent and available. It concerns behaviour during the partition only — it is often cited as though it were a permanent choice among the three letters, which it is not.

Circuit breaker — A component that stops calls to a failing service, keeping the caller from wasting resources and propagating the failure.

Cohesion — The degree to which the elements of a module belong together. High cohesion and low coupling are the same decision seen from two sides: what stays together and what stays apart.

Coupling — The degree to which a change in one module forces a change in another. Not a defect to eliminate: a quantity to allocate deliberately. Modules that change together should be coupled.

CQRS — Separation between the model used for writing and the one used for reading. It solves the case where the two have incompatible requirements. The cost of synchronization appears when writes and reads go to separate stores — not before that.

D

DDD — Domain-Driven Design. An approach that structures software around the business domain and the language of the people who understand it.

E

Event sourcing — Persisting the sequence of events that produced the state, rather than the state. State becomes a derivable projection. Powerful and expensive: it demands event versioning and reprocessing.

Eventual consistency — The guarantee that, absent new writes, all replicas converge to the same value. It does not say when, and it is that absence the application has to handle.

F

Fault tolerance — The ability to keep operating correctly despite the failure of components.

Fitness function — An automated check that a desired architectural property still holds. It turns architectural intent into a test.

I

Idempotency — The property of an operation whose effect is the same whether it runs once or many times. It is what makes it safe to repeat a call whose outcome you do not know — and therefore the basis of nearly all recovery in distributed systems.

L

Latency — The time between request and response. An experience requirement is stated in percentiles, because the average hides the tail users feel; the average is for sizing — it is what Little's law needs.

M

Microservices — A style in which the application is a set of independently deployable services, each owning its data. Deployment independence is the goal; small size is a consequence, not a target.

Modular monolith — An application deployed as a single unit, with explicit and enforced internal boundaries. Frequently the right answer in the case where microservices would be adopted on reputation.

Modularity — The degree to which a system is composed of parts with explicit boundaries that can be understood and changed separately.

P

PACELC — An extension of CAP: during a partition (P), you choose between availability and consistency; else (E), you choose between latency and consistency. It describes the everyday dilemma better, because partitions are rare and the second trade-off always applies.

Partitioning — Dividing data into disjoint subsets to distribute load or volume. See also sharding.

Q

Quality attribute — A measurable property of how the system behaves: availability, performance, security. Distinct from a functional requirement, which describes what the system does.

R

Reliability — The probability that a system operates correctly over a period. Distinct from availability: a system can be available and return wrong results.

Replication — Keeping copies of the same data in distinct places. It raises availability and read capacity; it introduces the question of which copy is right.

RPO — Recovery Point Objective. The maximum amount of data you accept losing, measured in time. It sets the replication and backup strategy.

RTO — Recovery Time Objective. The maximum acceptable time to restore service after a failure. It sets the redundancy topology.

S

Saga — A sequence of local transactions with compensations, used when a distributed transaction is not viable. It trades atomicity for availability, and requires every step to have a defined compensation.

Scalability — The ability to absorb growth by adding resources. Distinct from performance: a system can be fast and not scale.

Sharding — Horizontal partitioning in which each partition lives on a separate instance. The choice of key determines whether load spreads or concentrates in a hotspot.

SLA — Service Level Agreement. A contractual commitment about service level, with commercial consequences. It is a business decision.

SLI — Service Level Indicator. The metric itself — for example, the fraction of requests answered under 300 ms.

SLO — Service Level Objective. The internal target on an SLI. It is an engineering decision, and should be stricter than the corresponding SLA.

Strangler fig — A pattern of incremental replacement in which the new system progressively intercepts traffic from the old one until the old one can be shut down.

Strong consistency — The guarantee that every read observes the most recent write. It costs latency even without a network partition — see PACELC.

T

Technical debt — A future cost taken on by a decision that favours speed now. It is only debt when the choice was deliberate and there is intent to repay; the rest is simply poor work.

Trade-off — A choice in which gaining on one dimension implies losing on another. A "trade-off" with no stated loss is not a trade-off, it is a preference.

Transaction — A unit of work with guarantees of atomicity, consistency, isolation and durability. The effective isolation level is rarely what the name suggests; it is worth checking the documentation of the specific database.

U

Ubiquitous language — Vocabulary shared between domain experts and developers, used untranslated in the code and in conversation. It is the mechanism that makes the rest of DDD work.

V

Value object — An object defined by its attributes and not by identity. Two with the same values are interchangeable. Immutable by construction.