approximate histogram stats to save cpu

Summary:
sounds like we're willing to tradeoff minor inaccuracy in stats for speed. start with histogram stats. ticker stats will be harder (and, IMO, we shouldn't change them in this manner) as many test cases rely on them being exactly correct.
Closes https://github.com/facebook/rocksdb/pull/2720

Differential Revision: D5607884

Pulled By: ajkr

fbshipit-source-id: 1b754cda35ea6b252d1fdd5aa3cfb58866506372

monitoring/histogram.cc

@@ -105,17 +105,26 @@ void HistogramStat::Add(uint64_t value) {
   // by concurrent threads is tolerable.
   const size_t index = bucketMapper.IndexForValue(value);
   assert(index < num_buckets_);
-  buckets_[index].fetch_add(1, std::memory_order_relaxed);
+  buckets_[index].store(buckets_[index].load(std::memory_order_relaxed) + 1,
+                        std::memory_order_relaxed);
 
   uint64_t old_min = min();
-  while (value < old_min && !min_.compare_exchange_weak(old_min, value)) {}
+  if (value < old_min) {
+    min_.store(value, std::memory_order_relaxed);
+  }
 
   uint64_t old_max = max();
-  while (value > old_max && !max_.compare_exchange_weak(old_max, value)) {}
+  if (value > old_max) {
+    max_.store(value, std::memory_order_relaxed);
+  }
 
-  num_.fetch_add(1, std::memory_order_relaxed);
-  sum_.fetch_add(value, std::memory_order_relaxed);
-  sum_squares_.fetch_add(value * value, std::memory_order_relaxed);
+  num_.store(num_.load(std::memory_order_relaxed) + 1,
+             std::memory_order_relaxed);
+  sum_.store(sum_.load(std::memory_order_relaxed) + value,
+             std::memory_order_relaxed);
+  sum_squares_.store(
+      sum_squares_.load(std::memory_order_relaxed) + value * value,
+      std::memory_order_relaxed);
 }
 
 void HistogramStat::Merge(const HistogramStat& other) {