perf: reduce metrics hot-path lock contention (#72)

Author: MattDevy

elastictransport/elastictransport.go

@@ -327,9 +327,7 @@ func (c *Client) Perform(req *http.Request) (*http.Response, error) {
 
 	// Record metrics, when enabled
 	if c.metrics != nil {
-		c.metrics.Lock()
-		c.metrics.requests++
-		c.metrics.Unlock()
+		c.metrics.requests.Add(1)
 	}
 
 	// Update request
@@ -418,9 +416,7 @@ func (c *Client) Perform(req *http.Request) (*http.Response, error) {
 		if err != nil {
 			// Record metrics, when enabled
 			if c.metrics != nil {
-				c.metrics.Lock()
-				c.metrics.failures++
-				c.metrics.Unlock()
+				c.metrics.failures.Add(1)
 			}
 
 			// Report the connection as unsuccessful

elastictransport/metrics.go

@@ -23,6 +23,7 @@ import (
 	"strconv"
 	"strings"
 	"sync"
+	"sync/atomic"
 	"time"
 )
 
@@ -63,8 +64,8 @@ type ConnectionMetric struct {
 type metrics struct {
 	sync.RWMutex
 
-	requests  int
-	failures  int
+	requests  atomic.Uint64
+	failures  atomic.Uint64
 	responses map[int]int
 }
 
@@ -73,20 +74,21 @@ func (c *Client) Metrics() (Metrics, error) {
 	if c.metrics == nil {
 		return Metrics{}, errors.New("transport metrics not enabled")
 	}
-	c.metrics.RLock()
-	defer c.metrics.RUnlock()
 
 	c.poolMu.RLock()
 	defer c.poolMu.RUnlock()
 
-	m := Metrics{
-		Requests:  c.metrics.requests,
-		Failures:  c.metrics.failures,
-		Responses: make(map[int]int, len(c.metrics.responses)),
+	responses := make(map[int]int)
+	c.metrics.RLock()
+	for code, num := range c.metrics.responses {
+		responses[code] = num
 	}
+	c.metrics.RUnlock()
 
-	for code, num := range c.metrics.responses {
-		m.Responses[code] = num
+	m := Metrics{
+		Requests:  int(c.metrics.requests.Load()),
+		Failures:  int(c.metrics.failures.Load()),
+		Responses: responses,
 	}
 
 	if pool, ok := c.pool.(connectionable); ok {

elastictransport/metrics_internal_test.go

@@ -49,8 +49,8 @@ func TestMetrics(t *testing.T) {
 			},
 		)
 
-		tp.metrics.requests = 3
-		tp.metrics.failures = 4
+		tp.metrics.requests.Store(3)
+		tp.metrics.failures.Store(4)
 		tp.metrics.responses[200] = 1
 		tp.metrics.responses[404] = 2