From 847cde8aa7580a6507b38b22fa59499b7dfbcc1e Mon Sep 17 00:00:00 2001
From: Rowan Seymour <rowanseymour@gmail.com>
Date: Thu, 19 Dec 2024 16:35:51 -0500
Subject: [PATCH] Send metrics without channel type as well

---
 backends/rapidpro/stats.go      | 41 +++++++++++++++++++++++----------
 backends/rapidpro/stats_test.go | 19 +++++++++++++--
 2 files changed, 46 insertions(+), 14 deletions(-)

diff --git a/backends/rapidpro/stats.go b/backends/rapidpro/stats.go
index 4c412470..10c71282 100644
--- a/backends/rapidpro/stats.go
+++ b/backends/rapidpro/stats.go
@@ -11,17 +11,42 @@ import (
 
 type CountByType map[courier.ChannelType]int
 
-// converts per channel counts into cloudwatch metrics with type as a dimension
+// converts per channel counts into a set of cloudwatch metrics with type as a dimension, and a total count without type
 func (c CountByType) metrics(name string) []types.MetricDatum {
-	m := make([]types.MetricDatum, 0, len(c))
+	m := make([]types.MetricDatum, 0, len(c)+1)
+	total := 0
 	for typ, count := range c {
 		m = append(m, cwatch.Datum(name, float64(count), types.StandardUnitCount, cwatch.Dimension("ChannelType", string(typ))))
+
+		total += count
 	}
+
+	m = append(m, cwatch.Datum(name, float64(total), types.StandardUnitCount))
 	return m
 }
 
 type DurationByType map[courier.ChannelType]time.Duration
 
+func (c DurationByType) metrics(name string, avgDenom func(courier.ChannelType) int) []types.MetricDatum {
+	m := make([]types.MetricDatum, 0, len(c)+1)
+	totalDuration := time.Duration(0)
+	totalDenom := 0
+	for typ, d := range c { // convert to averages
+		denom := avgDenom(typ)
+		avgTime := d / time.Duration(denom)
+		m = append(m, cwatch.Datum(name, float64(avgTime)/float64(time.Second), types.StandardUnitSeconds, cwatch.Dimension("ChannelType", string(typ))))
+
+		totalDuration += d
+		totalDenom += denom
+	}
+
+	if totalDenom > 0 {
+		overallAvg := float64(totalDuration) / float64(totalDenom)
+		m = append(m, cwatch.Datum(name, overallAvg/float64(time.Second), types.StandardUnitSeconds))
+	}
+	return m
+}
+
 type Stats struct {
 	IncomingRequests CountByType    // number of handler requests
 	IncomingMessages CountByType    // number of messages received
@@ -61,19 +86,11 @@ func (s *Stats) ToMetrics() []types.MetricDatum {
 	metrics = append(metrics, s.IncomingStatuses.metrics("IncomingStatuses")...)
 	metrics = append(metrics, s.IncomingEvents.metrics("IncomingEvents")...)
 	metrics = append(metrics, s.IncomingIgnored.metrics("IncomingIgnored")...)
-
-	for typ, d := range s.IncomingDuration { // convert to averages
-		avgTime := d / time.Duration(s.IncomingRequests[typ])
-		metrics = append(metrics, cwatch.Datum("IncomingDuration", float64(avgTime)/float64(time.Second), types.StandardUnitSeconds, cwatch.Dimension("ChannelType", string(typ))))
-	}
+	metrics = append(metrics, s.IncomingDuration.metrics("IncomingDuration", func(typ courier.ChannelType) int { return s.IncomingRequests[typ] })...)
 
 	metrics = append(metrics, s.OutgoingSends.metrics("OutgoingSends")...)
 	metrics = append(metrics, s.OutgoingErrors.metrics("OutgoingErrors")...)
-
-	for typ, d := range s.OutgoingDuration { // convert to averages
-		avgTime := d / time.Duration(s.OutgoingSends[typ]+s.OutgoingErrors[typ])
-		metrics = append(metrics, cwatch.Datum("OutgoingDuration", float64(avgTime)/float64(time.Second), types.StandardUnitSeconds, cwatch.Dimension("ChannelType", string(typ))))
-	}
+	metrics = append(metrics, s.OutgoingDuration.metrics("OutgoingDuration", func(typ courier.ChannelType) int { return s.OutgoingSends[typ] + s.OutgoingErrors[typ] })...)
 
 	metrics = append(metrics, cwatch.Datum("ContactsCreated", float64(s.ContactsCreated), types.StandardUnitCount))
 	return metrics
diff --git a/backends/rapidpro/stats_test.go b/backends/rapidpro/stats_test.go
index caf92fd0..12d5477f 100644
--- a/backends/rapidpro/stats_test.go
+++ b/backends/rapidpro/stats_test.go
@@ -4,8 +4,10 @@ import (
 	"testing"
 	"time"
 
+	"github.com/aws/aws-sdk-go-v2/service/cloudwatch/types"
 	"github.com/nyaruka/courier"
 	"github.com/nyaruka/courier/backends/rapidpro"
+	"github.com/nyaruka/gocommon/aws/cwatch"
 	"github.com/stretchr/testify/assert"
 )
 
@@ -33,7 +35,7 @@ func TestStats(t *testing.T) {
 	assert.Equal(t, rapidpro.DurationByType{"T": time.Second * 2, "FBA": time.Second * 3}, stats.OutgoingDuration)
 
 	metrics := stats.ToMetrics()
-	assert.Len(t, metrics, 8)
+	assert.Len(t, metrics, 17)
 
 	sc.RecordOutgoing("FBA", true, time.Second)
 	sc.RecordOutgoing("FBA", true, time.Second)
@@ -51,5 +53,18 @@ func TestStats(t *testing.T) {
 	assert.Equal(t, rapidpro.DurationByType{"FBA": time.Second * 2}, stats.OutgoingDuration)
 
 	metrics = stats.ToMetrics()
-	assert.Len(t, metrics, 3)
+	assert.Len(t, metrics, 11)
+	assert.Equal(t, []types.MetricDatum{
+		cwatch.Datum("IncomingRequests", 0, "Count"),
+		cwatch.Datum("IncomingMessages", 0, "Count"),
+		cwatch.Datum("IncomingStatuses", 0, "Count"),
+		cwatch.Datum("IncomingEvents", 0, "Count"),
+		cwatch.Datum("IncomingIgnored", 0, "Count"),
+		cwatch.Datum("OutgoingSends", 2, "Count", cwatch.Dimension("ChannelType", "FBA")),
+		cwatch.Datum("OutgoingSends", 2, "Count"),
+		cwatch.Datum("OutgoingErrors", 0, "Count"),
+		cwatch.Datum("OutgoingDuration", 1, "Seconds", cwatch.Dimension("ChannelType", "FBA")),
+		cwatch.Datum("OutgoingDuration", 1, "Seconds"),
+		cwatch.Datum("ContactsCreated", 0, "Count"),
+	}, metrics)
 }