Fixup random interval jittering

CHANGELOG.md

@@ -14,8 +14,7 @@ even interval. This means that `interval="10s"` will collect every :00, :10, etc
 To ease scale concerns, flushing will be "jittered" by a random amount so that
 all Telegraf instances do not flush at the same time. Both of these options can
 be controlled via the `round_interval` and `flush_jitter` config options.
-- Telegraf will now retry metric flushes, twice by default. This can be configued
-via the `flush_retries` agent config option.
+- Telegraf will now retry metric flushes twice
 
 ### Features
 - [#205](https://github.com/influxdb/telegraf/issues/205): Include per-db redis keyspace info

agent.go

@@ -1,9 +1,10 @@
 package telegraf
 
 import (
+	"crypto/rand"
 	"fmt"
 	"log"
-	"math/rand"
+	"math/big"
 	"os"
 	"sort"
 	"sync"
@@ -381,9 +382,10 @@ func (a *Agent) flusher(shutdown chan struct{}, pointChan chan *client.Point) er
 	// Inelegant, but this sleep is to allow the Gather threads to run, so that
 	// the flusher will flush after metrics are collected.
 	time.Sleep(time.Millisecond * 100)
+
 	ticker := time.NewTicker(a.FlushInterval.Duration)
 	points := make([]*client.Point, 0)
-	jitter := rand.Int63n(int64(a.FlushJitter.Duration))
+
 	for {
 		select {
 		case <-shutdown:
@@ -391,15 +393,7 @@ func (a *Agent) flusher(shutdown chan struct{}, pointChan chan *client.Point) er
 			a.flush(points, shutdown, true)
 			return nil
 		case <-ticker.C:
-			timer := time.NewTimer(time.Duration(jitter))
-			select {
-			case <-timer.C:
-				a.flush(points, shutdown, false)
-			case <-shutdown:
-				log.Println("Hang on, flushing any cached points before shutdown")
-				a.flush(points, shutdown, true)
-				return nil
-			}
+			a.flush(points, shutdown, false)
 			points = make([]*client.Point, 0)
 		case pt := <-pointChan:
 			points = append(points, pt)
@@ -407,10 +401,38 @@ func (a *Agent) flusher(shutdown chan struct{}, pointChan chan *client.Point) er
 	}
 }
 
+// jitterInterval applies the the interval jitter to the flush interval using
+// crypto/rand number generator
+func jitterInterval(ininterval, injitter time.Duration) time.Duration {
+	var jitter int64
+	outinterval := ininterval
+	if injitter.Nanoseconds() != 0 {
+		maxjitter := big.NewInt(injitter.Nanoseconds())
+		if j, err := rand.Int(rand.Reader, maxjitter); err == nil {
+			jitter = j.Int64()
+		}
+		outinterval = time.Duration(jitter + ininterval.Nanoseconds())
+	}
+
+	if outinterval.Nanoseconds() < time.Duration(500*time.Millisecond).Nanoseconds() {
+		log.Printf("Flush interval %s too low, setting to 500ms\n", outinterval)
+		outinterval = time.Duration(500 * time.Millisecond)
+	}
+
+	return outinterval
+}
+
 // Run runs the agent daemon, gathering every Interval
 func (a *Agent) Run(shutdown chan struct{}) error {
 	var wg sync.WaitGroup
 
+	a.FlushInterval.Duration = jitterInterval(a.FlushInterval.Duration,
+		a.FlushJitter.Duration)
+
+	log.Printf("Agent Config: Interval:%s, Debug:%#v, Hostname:%#v, "+
+		"Flush Interval:%s\n",
+		a.Interval, a.Debug, a.Hostname, a.FlushInterval)
+
 	// channel shared between all plugin threads for accumulating points
 	pointChan := make(chan *client.Point, 1000)
 

agent_test.go

@@ -3,6 +3,9 @@ package telegraf
 import (
 	"github.com/stretchr/testify/assert"
 	"testing"
+	"time"
+
+	"github.com/influxdb/telegraf/duration"
 
 	// needing to load the plugins
 	_ "github.com/influxdb/telegraf/plugins/all"
@@ -55,3 +58,97 @@ func TestAgent_LoadOutput(t *testing.T) {
 	outputsEnabled, _ = a.LoadOutputs([]string{"influxdb", "foo", "kafka", "bar"}, config)
 	assert.Equal(t, 2, len(outputsEnabled))
 }
+
+func TestAgent_ZeroJitter(t *testing.T) {
+	a := &Agent{
+		FlushInterval: duration.Duration{10 * time.Second},
+		FlushJitter:   duration.Duration{0 * time.Second},
+	}
+	flushinterval := jitterInterval(a.FlushInterval.Duration,
+		a.FlushJitter.Duration)
+
+	actual := flushinterval.Nanoseconds()
+	exp := time.Duration(10 * time.Second).Nanoseconds()
+
+	if actual != exp {
+		t.Errorf("Actual %v, expected %v", actual, exp)
+	}
+}
+
+func TestAgent_ZeroInterval(t *testing.T) {
+	min := time.Duration(500 * time.Millisecond).Nanoseconds()
+	max := time.Duration(5 * time.Second).Nanoseconds()
+
+	for i := 0; i < 1000; i++ {
+		a := &Agent{
+			FlushInterval: duration.Duration{0 * time.Second},
+			FlushJitter:   duration.Duration{5 * time.Second},
+		}
+
+		flushinterval := jitterInterval(a.FlushInterval.Duration,
+			a.FlushJitter.Duration)
+		actual := flushinterval.Nanoseconds()
+
+		if actual > max {
+			t.Errorf("Didn't expect interval %d to be > %d", actual, max)
+			break
+		}
+		if actual < min {
+			t.Errorf("Didn't expect interval %d to be < %d", actual, min)
+			break
+		}
+	}
+}
+
+func TestAgent_ZeroBoth(t *testing.T) {
+	a := &Agent{
+		FlushInterval: duration.Duration{0 * time.Second},
+		FlushJitter:   duration.Duration{0 * time.Second},
+	}
+
+	flushinterval := jitterInterval(a.FlushInterval.Duration,
+		a.FlushJitter.Duration)
+
+	actual := flushinterval
+	exp := time.Duration(500 * time.Millisecond)
+
+	if actual != exp {
+		t.Errorf("Actual %v, expected %v", actual, exp)
+	}
+}
+
+func TestAgent_JitterMax(t *testing.T) {
+	max := time.Duration(32 * time.Second).Nanoseconds()
+
+	for i := 0; i < 1000; i++ {
+		a := &Agent{
+			FlushInterval: duration.Duration{30 * time.Second},
+			FlushJitter:   duration.Duration{2 * time.Second},
+		}
+		flushinterval := jitterInterval(a.FlushInterval.Duration,
+			a.FlushJitter.Duration)
+		actual := flushinterval.Nanoseconds()
+		if actual > max {
+			t.Errorf("Didn't expect interval %d to be > %d", actual, max)
+			break
+		}
+	}
+}
+
+func TestAgent_JitterMin(t *testing.T) {
+	min := time.Duration(30 * time.Second).Nanoseconds()
+
+	for i := 0; i < 1000; i++ {
+		a := &Agent{
+			FlushInterval: duration.Duration{30 * time.Second},
+			FlushJitter:   duration.Duration{2 * time.Second},
+		}
+		flushinterval := jitterInterval(a.FlushInterval.Duration,
+			a.FlushJitter.Duration)
+		actual := flushinterval.Nanoseconds()
+		if actual < min {
+			t.Errorf("Didn't expect interval %d to be < %d", actual, min)
+			break
+		}
+	}
+}

cmd/telegraf/telegraf.go

@@ -143,9 +143,6 @@ func main() {
 	log.Printf("Starting Telegraf (version %s)\n", Version)
 	log.Printf("Loaded outputs: %s", strings.Join(outputs, " "))
 	log.Printf("Loaded plugins: %s", strings.Join(plugins, " "))
-	log.Printf("Agent Config: Interval:%s, Debug:%#v, Hostname:%#v, "+
-		"Flush Interval:%s\n",
-		ag.Interval, ag.Debug, ag.Hostname, ag.FlushInterval)
 	log.Printf("Tags enabled: %s", config.ListTags())
 
 	if *fPidfile != "" {

config.go

@@ -230,13 +230,13 @@ var header = `# Telegraf configuration
   # ie, if interval="10s" then always collect on :00, :10, :20, etc.
   round_interval = true
 
-  # Default data flushing interval for all outputs
+  # Default data flushing interval for all outputs. You should not set this below
+  # interval. Maximum flush_interval will be flush_interval + flush_jitter
   flush_interval = "10s"
-  # Jitter the flush interval by a random range
-  # ie, a jitter of 5s and interval 10s means flush will happen every 10-15s
-  flush_jitter = "5s"
-  # Number of times to retry each data flush
-  flush_retries = 2
+  # Jitter the flush interval by a random amount. This is primarily to avoid
+  # large write spikes for users running a large number of telegraf instances.
+  # ie, a jitter of 5s and interval 10s means flushes will happen every 10-15s
+  flush_jitter = "0s"
 
   # Run telegraf in debug mode
   debug = false