internal/telemetry: support DD_TELEMETRY_HEARTBEAT_INTERVAL

internal/telemetry/client.go

@@ -48,6 +48,8 @@ var (
 	}
 	// TODO: Default telemetry URL?
 	hostname string
+
+	defaultHeartbeatInterval = 60 // seconds
 )
 
 func init() {
@@ -73,6 +75,10 @@ type Client struct {
 	// How often to send batched requests. Defaults to 60s
 	SubmissionInterval time.Duration
 
+	// The interval for sending a heartbeat signal to the backend.
+	// Configurable with DD_TELEMETRY_HEARTBEAT_INTERVAL. Default 60s.
+	heartbeatInterval time.Duration
+
 	// e.g. "tracers", "profilers", "appsec"
 	Namespace Namespace
 
@@ -114,8 +120,10 @@ type Client struct {
 	// seqID is a sequence number used to order telemetry messages by
 	// the back end.
 	seqID int64
-	// t is used to schedule flushing outstanding messages
-	t *time.Timer
+	// flushT is used to schedule flushing outstanding messages
+	flushT *time.Timer
+	// heartbeatT is used to schedule heartbeat messages
+	heartbeatT *time.Timer
 	// requests hold all messages which don't need to be immediately sent
 	requests []*Request
 	// metrics holds un-sent metrics that will be aggregated the next time
@@ -195,7 +203,14 @@ func (c *Client) Start(integrations []Integration, configuration []Configuration
 	if c.SubmissionInterval == 0 {
 		c.SubmissionInterval = 60 * time.Second
 	}
-	c.t = time.AfterFunc(c.SubmissionInterval, c.backgroundFlush)
+	c.flushT = time.AfterFunc(c.SubmissionInterval, c.backgroundFlush)
+
+	heartbeat := internal.IntEnv("DD_TELEMETRY_HEARTBEAT_INTERVAL", defaultHeartbeatInterval)
+	if heartbeat < 1 || heartbeat > 3600 {
+		heartbeat = defaultHeartbeatInterval
+	}
+	c.heartbeatInterval = time.Duration(heartbeat) * time.Second
+	c.heartbeatT = time.AfterFunc(c.heartbeatInterval, c.backgroundHeartbeat)
 }
 
 // Stop notifies the telemetry endpoint that the app is closing. All outstanding
@@ -208,7 +223,8 @@ func (c *Client) Stop() {
 		return
 	}
 	c.started = false
-	c.t.Stop()
+	c.flushT.Stop()
+	c.heartbeatT.Stop()
 	// close request types have no body
 	r := c.newRequest(RequestTypeAppClosing)
 	c.scheduleSubmit(r)
@@ -425,14 +441,25 @@ func (c *Client) scheduleSubmit(r *Request) {
 	c.requests = append(c.requests, r)
 }
 
+func (c *Client) backgroundHeartbeat() {
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	if !c.started {
+		return
+	}
+	err := c.submit(c.newRequest(RequestTypeAppHeartbeat))
+	if err != nil {
+		c.log("heartbeat failed: %s", err)
+	}
+	c.heartbeatT.Reset(c.heartbeatInterval)
+}
+
 func (c *Client) backgroundFlush() {
 	c.mu.Lock()
 	defer c.mu.Unlock()
 	if !c.started {
 		return
 	}
-	r := c.newRequest(RequestTypeAppHeartbeat)
-	c.scheduleSubmit(r)
 	c.flush()
-	c.t.Reset(c.SubmissionInterval)
+	c.flushT.Reset(c.SubmissionInterval)
 }

internal/telemetry/client_test.go

@@ -9,6 +9,7 @@ import (
 	"encoding/json"
 	"net/http"
 	"net/http/httptest"
+	"os"
 	"reflect"
 	"sort"
 	"sync"
@@ -19,6 +20,9 @@ import (
 )
 
 func TestClient(t *testing.T) {
+	os.Setenv("DD_TELEMETRY_HEARTBEAT_INTERVAL", "1")
+	defer os.Unsetenv("DD_TELEMETRY_HEARTBEAT_INTERVAL")
+
 	heartbeat := make(chan struct{})
 
 	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
@@ -43,7 +47,13 @@ func TestClient(t *testing.T) {
 	client.Start(nil, nil) // test idempotence
 	defer client.Stop()
 
-	<-heartbeat
+	timeout := time.After(30 * time.Second)
+	select {
+	case <-timeout:
+		t.Fatal("Heartbeat took more than 30 seconds. Should have been ~1 second")
+	case <-heartbeat:
+	}
+
 }
 
 func TestMetrics(t *testing.T) {