Add: exponential backoff for CAS operations on floats (prometheus#1661)

* add: exponential backoff for CAS operations of floats

Signed-off-by: Ivan Goncharov <i.morph@gmail.com>

* add: some more benchmark use cases (higher contention)

Signed-off-by: Ivan Goncharov <i.morph@gmail.com>

* fmt: fumpted some files

Signed-off-by: Ivan Goncharov <i.morph@gmail.com>

* add: license header

Signed-off-by: Ivan Goncharov <i.morph@gmail.com>

* add: comment explaining origin of backoff constants

Signed-off-by: Ivan Goncharov <i.morph@gmail.com>

---------

Signed-off-by: Ivan Goncharov <i.morph@gmail.com>

prometheus/atomic_update.go

@@ -0,0 +1,50 @@
+// Copyright 2014 The Prometheus Authors
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package prometheus
+
+import (
+	"math"
+	"sync/atomic"
+	"time"
+)
+
+// atomicUpdateFloat atomically updates the float64 value pointed to by bits
+// using the provided updateFunc, with an exponential backoff on contention.
+func atomicUpdateFloat(bits *uint64, updateFunc func(float64) float64) {
+	const (
+		// both numbers are derived from empirical observations
+		// documented in this PR: https://github.com/prometheus/client_golang/pull/1661
+		maxBackoff     = 320 * time.Millisecond
+		initialBackoff = 10 * time.Millisecond
+	)
+	backoff := initialBackoff
+
+	for {
+		loadedBits := atomic.LoadUint64(bits)
+		oldFloat := math.Float64frombits(loadedBits)
+		newFloat := updateFunc(oldFloat)
+		newBits := math.Float64bits(newFloat)
+
+		if atomic.CompareAndSwapUint64(bits, loadedBits, newBits) {
+			break
+		} else {
+			// Exponential backoff with sleep and cap to avoid infinite wait
+			time.Sleep(backoff)
+			backoff *= 2
+			if backoff > maxBackoff {
+				backoff = maxBackoff
+			}
+		}
+	}
+}

prometheus/atomic_update_test.go

@@ -0,0 +1,167 @@
+// Copyright 2014 The Prometheus Authors
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package prometheus
+
+import (
+	"math"
+	"sync"
+	"sync/atomic"
+	"testing"
+	"unsafe"
+)
+
+var output float64
+
+func TestAtomicUpdateFloat(t *testing.T) {
+	var val float64 = 0.0
+	bits := (*uint64)(unsafe.Pointer(&val))
+	var wg sync.WaitGroup
+	numGoroutines := 100000
+	increment := 1.0
+
+	for i := 0; i < numGoroutines; i++ {
+		wg.Add(1)
+		go func() {
+			defer wg.Done()
+			atomicUpdateFloat(bits, func(f float64) float64 {
+				return f + increment
+			})
+		}()
+	}
+
+	wg.Wait()
+	expected := float64(numGoroutines) * increment
+	if val != expected {
+		t.Errorf("Expected %f, got %f", expected, val)
+	}
+}
+
+// Benchmark for atomicUpdateFloat with single goroutine (no contention).
+func BenchmarkAtomicUpdateFloat_SingleGoroutine(b *testing.B) {
+	var val float64 = 0.0
+	bits := (*uint64)(unsafe.Pointer(&val))
+
+	for i := 0; i < b.N; i++ {
+		atomicUpdateFloat(bits, func(f float64) float64 {
+			return f + 1.0
+		})
+	}
+
+	output = val
+}
+
+// Benchmark for old implementation with single goroutine (no contention) -> to check overhead of backoff
+func BenchmarkAtomicNoBackoff_SingleGoroutine(b *testing.B) {
+	var val float64 = 0.0
+	bits := (*uint64)(unsafe.Pointer(&val))
+
+	for i := 0; i < b.N; i++ {
+		for {
+			loadedBits := atomic.LoadUint64(bits)
+			newBits := math.Float64bits(math.Float64frombits(loadedBits) + 1.0)
+			if atomic.CompareAndSwapUint64(bits, loadedBits, newBits) {
+				break
+			}
+		}
+	}
+
+	output = val
+}
+
+// Benchmark varying the number of goroutines.
+func benchmarkAtomicUpdateFloatConcurrency(b *testing.B, numGoroutines int) {
+	var val float64 = 0.0
+	bits := (*uint64)(unsafe.Pointer(&val))
+	b.SetParallelism(numGoroutines)
+
+	b.ResetTimer()
+	b.RunParallel(func(pb *testing.PB) {
+		for pb.Next() {
+			atomicUpdateFloat(bits, func(f float64) float64 {
+				return f + 1.0
+			})
+		}
+	})
+
+	output = val
+}
+
+func benchmarkAtomicNoBackoffFloatConcurrency(b *testing.B, numGoroutines int) {
+	var val float64 = 0.0
+	bits := (*uint64)(unsafe.Pointer(&val))
+	b.SetParallelism(numGoroutines)
+
+	b.ResetTimer()
+	b.RunParallel(func(pb *testing.PB) {
+		for pb.Next() {
+			for {
+				loadedBits := atomic.LoadUint64(bits)
+				newBits := math.Float64bits(math.Float64frombits(loadedBits) + 1.0)
+				if atomic.CompareAndSwapUint64(bits, loadedBits, newBits) {
+					break
+				}
+			}
+		}
+	})
+
+	output = val
+}
+
+func BenchmarkAtomicUpdateFloat_1Goroutine(b *testing.B) {
+	benchmarkAtomicUpdateFloatConcurrency(b, 1)
+}
+
+func BenchmarkAtomicNoBackoff_1Goroutine(b *testing.B) {
+	benchmarkAtomicNoBackoffFloatConcurrency(b, 1)
+}
+
+func BenchmarkAtomicUpdateFloat_2Goroutines(b *testing.B) {
+	benchmarkAtomicUpdateFloatConcurrency(b, 2)
+}
+
+func BenchmarkAtomicNoBackoff_2Goroutines(b *testing.B) {
+	benchmarkAtomicNoBackoffFloatConcurrency(b, 2)
+}
+
+func BenchmarkAtomicUpdateFloat_4Goroutines(b *testing.B) {
+	benchmarkAtomicUpdateFloatConcurrency(b, 4)
+}
+
+func BenchmarkAtomicNoBackoff_4Goroutines(b *testing.B) {
+	benchmarkAtomicNoBackoffFloatConcurrency(b, 4)
+}
+
+func BenchmarkAtomicUpdateFloat_8Goroutines(b *testing.B) {
+	benchmarkAtomicUpdateFloatConcurrency(b, 8)
+}
+
+func BenchmarkAtomicNoBackoff_8Goroutines(b *testing.B) {
+	benchmarkAtomicNoBackoffFloatConcurrency(b, 8)
+}
+
+func BenchmarkAtomicUpdateFloat_16Goroutines(b *testing.B) {
+	benchmarkAtomicUpdateFloatConcurrency(b, 16)
+}
+
+func BenchmarkAtomicNoBackoff_16Goroutines(b *testing.B) {
+	benchmarkAtomicNoBackoffFloatConcurrency(b, 16)
+}
+
+func BenchmarkAtomicUpdateFloat_32Goroutines(b *testing.B) {
+	benchmarkAtomicUpdateFloatConcurrency(b, 32)
+}
+
+func BenchmarkAtomicNoBackoff_32Goroutines(b *testing.B) {
+	benchmarkAtomicNoBackoffFloatConcurrency(b, 32)
+}

prometheus/counter.go

@@ -134,13 +134,9 @@ func (c *counter) Add(v float64) {
 		return
 	}
 
-	for {
-		oldBits := atomic.LoadUint64(&c.valBits)
-		newBits := math.Float64bits(math.Float64frombits(oldBits) + v)
-		if atomic.CompareAndSwapUint64(&c.valBits, oldBits, newBits) {
-			return
-		}
-	}
+	atomicUpdateFloat(&c.valBits, func(oldVal float64) float64 {
+		return oldVal + v
+	})
 }
 
 func (c *counter) AddWithExemplar(v float64, e Labels) {

prometheus/gauge.go

@@ -120,13 +120,9 @@ func (g *gauge) Dec() {
 }
 
 func (g *gauge) Add(val float64) {
-	for {
-		oldBits := atomic.LoadUint64(&g.valBits)
-		newBits := math.Float64bits(math.Float64frombits(oldBits) + val)
-		if atomic.CompareAndSwapUint64(&g.valBits, oldBits, newBits) {
-			return
-		}
-	}
+	atomicUpdateFloat(&g.valBits, func(oldVal float64) float64 {
+		return oldVal + val
+	})
 }
 
 func (g *gauge) Sub(val float64) {

prometheus/histogram.go

@@ -1646,13 +1646,9 @@ func waitForCooldown(count uint64, counts *histogramCounts) {
 // atomicAddFloat adds the provided float atomically to another float
 // represented by the bit pattern the bits pointer is pointing to.
 func atomicAddFloat(bits *uint64, v float64) {
-	for {
-		loadedBits := atomic.LoadUint64(bits)
-		newBits := math.Float64bits(math.Float64frombits(loadedBits) + v)
-		if atomic.CompareAndSwapUint64(bits, loadedBits, newBits) {
-			break
-		}
-	}
+	atomicUpdateFloat(bits, func(oldVal float64) float64 {
+		return oldVal + v
+	})
 }
 
 // atomicDecUint32 atomically decrements the uint32 p points to.  See

prometheus/summary.go

@@ -471,13 +471,9 @@ func (s *noObjectivesSummary) Observe(v float64) {
 	n := atomic.AddUint64(&s.countAndHotIdx, 1)
 	hotCounts := s.counts[n>>63]
 
-	for {
-		oldBits := atomic.LoadUint64(&hotCounts.sumBits)
-		newBits := math.Float64bits(math.Float64frombits(oldBits) + v)
-		if atomic.CompareAndSwapUint64(&hotCounts.sumBits, oldBits, newBits) {
-			break
-		}
-	}
+	atomicUpdateFloat(&hotCounts.sumBits, func(oldVal float64) float64 {
+		return oldVal + v
+	})
 	// Increment count last as we take it as a signal that the observation
 	// is complete.
 	atomic.AddUint64(&hotCounts.count, 1)
@@ -519,14 +515,13 @@ func (s *noObjectivesSummary) Write(out *dto.Metric) error {
 	// Finally add all the cold counts to the new hot counts and reset the cold counts.
 	atomic.AddUint64(&hotCounts.count, count)
 	atomic.StoreUint64(&coldCounts.count, 0)
-	for {
-		oldBits := atomic.LoadUint64(&hotCounts.sumBits)
-		newBits := math.Float64bits(math.Float64frombits(oldBits) + sum.GetSampleSum())
-		if atomic.CompareAndSwapUint64(&hotCounts.sumBits, oldBits, newBits) {
-			atomic.StoreUint64(&coldCounts.sumBits, 0)
-			break
-		}
-	}
+
+	// Use atomicUpdateFloat to update hotCounts.sumBits atomically.
+	atomicUpdateFloat(&hotCounts.sumBits, func(oldVal float64) float64 {
+		return oldVal + sum.GetSampleSum()
+	})
+	atomic.StoreUint64(&coldCounts.sumBits, 0)
+
 	return nil
 }