wagoe.platform.resilience — com.wagoe/wagoe-platform 1.1.0-rc-1

Retry and circuit breaker for outbound calls — the RPC client and the libs/external adapters both go through here.

Outside shell/ for the reason wagoe.platform.database is: other modules call it, and reaching into platform's shell is what bb check:ports refuses.

Both take predicates over a result rather than deciding what counts as a failure. Only the caller knows whether a second attempt could run something twice.

call clj

(call {:keys [policy cache breaker circuit]}
      {:keys [retryable? failure? on-open] :as opts}
      f)

with-retry inside with-breaker, under settings: (f attempt) is retried while retryable?, and the last result counts against the circuit when failure?. Returns that result, or (on-open retry-after-ms).

settings clj

(settings {:keys [cache circuit-breaker]} policy circuit timeout-ms)

Retry and breaker settings for calls to circuit.

timeout-ms is one attempt's longest wait: the probe lease must outlive a whole run of attempts, or it expires mid-probe and a second caller sends a second probe. Throws on an invalid :circuit-breaker — an inert breaker looks exactly like a working one.

with-breaker clj

(with-breaker cache config circuit failure? on-open f)

Call (f) unless the breaker for circuit is open; record the outcome.

With no cache there is no breaker and f always runs. When open, returns (on-open retry-after-ms) without calling f. failure? decides whether a result counts against the circuit — count each call once, after its retries, or a threshold of 5 trips on the second call.

config is the :breaker of settings.

with-retry clj

(with-retry policy retryable? f & [{:keys [on-retry on-give-up]}])

Call (f attempt) (1-based) until retryable? rejects its result or policy runs out of attempts, sleeping retry/delay-ms in between. Returns the last result.

:on-retry gets [result attempt delay-ms] before each sleep, and :on-give-up [result attempts] when a retryable result is the last.