From 7dcbf69536eba2d7133067c56e959873f00b4d0b Mon Sep 17 00:00:00 2001
From: Jacob Hoffman-Andrews <jsha+github@letsencrypt.org>
Date: Fri, 20 May 2022 19:44:13 -0700
Subject: [PATCH] Add sendDelay metric (#6130)

Fixes #6125
---
 cmd/expiration-mailer/main.go      | 16 ++++++++++++++++
 cmd/expiration-mailer/main_test.go |  2 ++
 2 files changed, 18 insertions(+)

diff --git a/cmd/expiration-mailer/main.go b/cmd/expiration-mailer/main.go
index 48ca31349..a286ed05f 100644
--- a/cmd/expiration-mailer/main.go
+++ b/cmd/expiration-mailer/main.go
@@ -58,6 +58,7 @@ type mailer struct {
 }
 
 type mailerStats struct {
+	sendDelay                         *prometheus.GaugeVec
 	nagsAtCapacity                    *prometheus.GaugeVec
 	errorCount                        *prometheus.CounterVec
 	sendLatency                       prometheus.Histogram
@@ -389,6 +390,12 @@ func (m *mailer) findExpiringCertificates(ctx context.Context) error {
 			continue // nothing to do
 		}
 
+		// Report the send delay metric. Note: this is the worst-case send delay
+		// of any certificate in this batch because it's based on the first (oldest).
+		sendDelay := expiresIn - certs[0].Expires.Sub(m.clk.Now())
+		m.stats.sendDelay.With(prometheus.Labels{"nag_group": expiresIn.String()}).Set(
+			float64(sendDelay.Truncate(time.Second).Seconds()))
+
 		processingStarted := m.clk.Now()
 		err = m.processCerts(ctx, certs)
 		if err != nil {
@@ -455,6 +462,14 @@ type Config struct {
 }
 
 func initStats(stats prometheus.Registerer) mailerStats {
+	sendDelay := prometheus.NewGaugeVec(
+		prometheus.GaugeOpts{
+			Name: "send_delay",
+			Help: "For the last batch of certificates, difference between the idealized send time and actual send time. Will always be nonzero, bigger numbers are worse",
+		},
+		[]string{"nag_group"})
+	stats.MustRegister(sendDelay)
+
 	nagsAtCapacity := prometheus.NewGaugeVec(
 		prometheus.GaugeOpts{
 			Name: "nags_at_capacity",
@@ -510,6 +525,7 @@ func initStats(stats prometheus.Registerer) mailerStats {
 	stats.MustRegister(accountsNeedingMail)
 
 	return mailerStats{
+		sendDelay:                         sendDelay,
 		nagsAtCapacity:                    nagsAtCapacity,
 		errorCount:                        errorCount,
 		sendLatency:                       sendLatency,
diff --git a/cmd/expiration-mailer/main_test.go b/cmd/expiration-mailer/main_test.go
index 8b2ee87ea..be4362990 100644
--- a/cmd/expiration-mailer/main_test.go
+++ b/cmd/expiration-mailer/main_test.go
@@ -338,6 +338,8 @@ func TestFindExpiringCertificates(t *testing.T) {
 	err = testCtx.m.findExpiringCertificates(context.Background())
 	test.AssertNotError(t, err, "Failed to find expiring certs")
 	test.AssertEquals(t, len(testCtx.mc.Messages), 0)
+	test.AssertMetricWithLabelsEquals(t, testCtx.m.stats.sendDelay, prometheus.Labels{"nag_group": "48h0m0s"}, 90000)
+	test.AssertMetricWithLabelsEquals(t, testCtx.m.stats.sendDelay, prometheus.Labels{"nag_group": "192h0m0s"}, 82800)
 }
 
 func makeRegistration(sac sapb.StorageAuthorityClient, id int64, jsonKey []byte, contacts []string) (*corepb.Registration, error) {