emqx / esockd / 580

67%

master: 67%

Pull #225

github

fix: drain stale countdown ticks before re-arming a dead timer

Address review feedback on #225:

- when the countdown timer is found dead, drain the {timeout, _, countdown}
  messages already in the mailbox before re-arming: a leaked tick that is
  processed in the window right after the re-armed timer has fired (e.g. a
  1ms tick) but before its own message is handled would see the timer as
  dead again, arm yet another one and drop the real tick, delaying every
  bucket's refill. Consuming the whole batch up front keeps a single re-arm
  from being followed by such a cascade.
- correct the doc comments: arm_countdown_timer/2 arms no NEW timer while
  no buckets exist (a still-running one is left to wind down after the last
  delete), and ensure_countdown_timer/1 is not the endpoint for the
  countdown tick, which recomputes the schedule and arms the next timer
  directly.

Tests: inject a burst of stale countdown ticks against a dead timer and
verify the limiter re-arms once and buckets keep refilling on cadence.

15 of 15 new or added lines in 1 file covered. (100.0%)

720 of 1061 relevant lines covered (67.86%)

205.96 hits per line