Retry a function with exponential backoff

Wraps f so that on error the call is retried up to max_tries times, with a delay of backoff^attempt seconds between attempts (optionally jittered and capped).

Usage

with_retry(
  f,
  max_tries = 3L,
  backoff = 2,
  ...,
  .on_error = NULL,
  .max_delay = Inf,
  .jitter = TRUE,
  .on_retry = NULL,
  .message = FALSE
)

Arguments

f

A function to wrap.

max_tries

Positive integer. Maximum number of attempts (including the first). Default 3.

backoff

Non-negative number. Base of the exponential delay in seconds; delay for attempt k is backoff^k. Default 2 (i.e. 2, 4, 8, 16 s). Use 0 to disable delays.

...

Passed to the wrapped function on each call. Reserved for future extensions; currently forwarded via the wrapper's ....

.on_error

Optional predicate function(cnd) -> logical, or a character vector of condition classes to retry on (e.g. c("http_error_503")), mirroring httr2's is_transient idiom. When the predicate returns FALSE (or the error matches none of the classes) the error is re-raised immediately. Default NULL retries all errors.

.max_delay

Numeric cap on individual retry delays in seconds. Default Inf.

.jitter

Logical. If TRUE (default), each delay is multiplied by a uniform random factor in [0.5, 1.5] to avoid synchronized retries.

.on_retry

Optional callback function(attempt, delay, cnd) invoked before each sleep. Errors in the callback are swallowed.

.message

Logical. If TRUE, emits a message() before each retry describing the attempt and delay. Default FALSE.

Value

A function that forwards its arguments to f. On persistent failure raises a condition of class "fmisc_retry_exhausted" preserving the original condition's classes.

See also

Other decorators: cache_clear(), decorate(), fmisc_decorators, is_decorated(), print.fmisc_decorated(), undecorate(), with_cache(), with_logging(), with_rate_limit(), with_timing()

Examples

if (FALSE) { # \dontrun{
flaky <- function(x) {
  if (runif(1) < 0.7) stop("transient")
  x * 2
}
robust <- with_retry(flaky, max_tries = 5, backoff = 0.1)
robust(21)
} # }