GitLab

regression test for issue-65

fix bug in meter traversal logic

Add methods to clear stats

The previous attempt actually make the problem worse:

1. It didn't update the "last update" time.
2. It updated the accumulator but then didn't set the total. That meant that the
_next_ update would count the bandwidth from two time periods.

This change also reverts the changes to the test (the test was right, the code
was wrong).

1. The last update is when we last saw activity. We don't need to update it when
we reset the meter.
2. Defer isn't _quite_ as fast as it could be.

1. Handle `registered` inside the main loop. This way we can avoid atomics.
2. Don't start calculating bandwidth for newly active meters until they've been
active for a round. This:
  1. Ensures we only write to the snapshot from within the main loop.
  2. Gives us a better estimated bandwidth usage.

add a feature for tracking idle timers

Also renames a confusing mutex to make it clear what it protects. It protects
snapshots, not the meter slice.

add gomod support // tag v0.0.1.

Just repair spelling mistake

Signed-off-by: gukq <gukaiqiang@gmail.com>

fixes #2

Use range and then trim the meters slice later. It's *much* easier to read.

The go schedule isn't that accurate.