Unnecessary contention in stats implementation

[diegommm]

In file stats.go, all code paths accessing (*counter).prev are protected with a mutex, yet every time the field is being accessed with an atomic Load/Store operation.
These are the possible code paths accessing the field, in reverse order:

• (*counter).prev > (*counter).value [ReadWrite] >
  • (*counter).report > (*scope).report, protected by (*scope).cm.RLock
  • (*counter).cachedReport > (*scope).cachedReport, protected by (*scope).cm.RLock
  • (*histogram).report > (*scope).report, protected by (*scope).hm.RLock
  • (*histogram).cachedReport > (*scope).cachedReport, protected by (*scope).hm.RLock
• (*counter).prev > (*counter).snapshot [Read] >
  • (*scope).Snapshot, protected by (*scope).cm.RLock
  • (*histogram).snapshotValues > (*scope).Snapshot, protected by (*scope).hm.RLock
  • (*histogram).snapshotDurations > (*scope).Snapshot, protected by (*scope).hm.RLock

Note that:

• A (*histogram) can have multiple (*counter) but only its methods access them, i.e. clients of (*histogram) only use its methods, not its fields.
• There's no background forking in any of the methods, so e.g. (*scope).Snapshot would work sequentially inside the (R)Lock/(R)Unlock blocks of each mutex.

So (*counter).prev is always protected by a mutex, there's no need for atomic operations to access it.