Skip to main content

Repository, Use Case, Mapper, Result, and Circuit Breaker

These are common application/resilience patterns, not universal mandatory layers.

Repository and Use Case​

Repository owns application-data policy and source coordination. Use case owns one business operation/reusable orchestration. Avoid pass-through names that add no contract.

Mapper protects meaning​

fun PriceDto.toDomain(): Money = Money(
    minorUnits = requireNotNull(minorUnits),
    currency = CurrencyCode.parse(currency) ?: error("Unknown currency"),
)

Mapper handles null/default/unit/name/enum differences. Silent fallbacks can corrupt money or permissions; decide strict versus tolerant behavior per field and record schema incidents.

Result: make expected failure explicit​

sealed interface PlaceOrderOutcome {
    data class Accepted(val id: OrderId) : PlaceOrderOutcome
    data class Rejected(val reasons: List<Rejection>) : PlaceOrderOutcome
}

Expected business rejection is data. Programming defects may still throw. Do not catch every Throwable into Unknown; cancellation must propagate and diagnostics must remain.

Kotlin Result is convenient at small boundaries, but a domain sealed type communicates recovery choices better when failures matter.

Circuit Breaker​

A circuit breaker prevents repeated calls to a failing dependency:

Closed -- failures exceed threshold --> Open
Open -- cooldown expires -----------> Half-open
Half-open -- probe succeeds --------> Closed
Half-open -- probe fails -----------> Open

It differs from retry: retry repeats an individual operation; breaker protects system capacity across operations.

Decisions include failure classification, rolling window, threshold, cooldown, probes, per-host/per-operation key, persistence, and user fallback. Do not count 4xx validation failures as infrastructure failure.

On mobile, network loss should usually be handled by offline/source-of-truth policy; an elaborate in-memory breaker resets on process death and may add little. It is useful for expensive unstable SDK/services or shared networking layers when measured failure storms exist.

Pattern interactions​

Retry + timeout + breaker + WorkManager can amplify attempts. Define one attempt budget and observability across layers. Idempotency is required for retried mutations.

Architect takeaway

Patterns must publish behavioral contracts: ownership, error classification, retries, cancellation, freshness, idempotency, and observability.

Exercises​

  1. Map malformed money without silent corruption.
  2. Design a breaker for a flaky ML SDK.
  3. Calculate worst-case attempts across OkHttp, repository, and WorkManager retries.