Randomized delays for retrying failed calls.
When many callers fail at the same moment and all retry after the same delay, they hit the dependency again at the same moment too. Adding randomness to the delay, called jitter, spreads the retries out so the dependency is not flooded when it comes back.
Each function implements one strategy. full/3 and equal/3 grow the
delay exponentially with each attempt, starting from base_delay;
decorrelated/3 bases each delay on the previous one. None goes above
max_delay. Delays are in whatever unit base_delay and max_delay use,
usually milliseconds.
Summary
Functions
Returns a random delay between base_delay and three times
previous_delay, capped at max_delay.
Returns half of the exponential backoff for attempt, plus a random amount
up to the other half.
Returns a random delay between 0 and the exponential backoff for
attempt.
Functions
@spec equal( base_delay :: non_neg_integer(), max_delay :: non_neg_integer(), attempt :: non_neg_integer() ) :: non_neg_integer()
Returns half of the exponential backoff for attempt, plus a random amount
up to the other half.
The backoff is computed as in full/3. Every delay is at least half of it,
so retries are spread out while never following each other too closely.
Examples
Rexsilient.Jitter.equal(100, 10_000, 3)
# a random value from 400 to 800@spec full( base_delay :: non_neg_integer(), max_delay :: non_neg_integer(), attempt :: non_neg_integer() ) :: non_neg_integer()
Returns a random delay between 0 and the exponential backoff for
attempt.
The backoff is base_delay * 2 ** attempt, capped at max_delay, and
attempt starts at 0. This spreads retries the most, but a delay may be
very short.
Examples
Rexsilient.Jitter.full(100, 10_000, 3)
# a random integer from 0 to 800