From 380a3a083469f3950bf9bdd0a03ca8199fdb91dd Mon Sep 17 00:00:00 2001
From: Fran Bull <fran@tailscale.com>
Date: Fri, 12 Jul 2024 07:08:47 -0700
Subject: [PATCH] appc: track metrics for route info storing

Track how often we're writing state and how many routes we're writing.

Updates #11008

Signed-off-by: Fran Bull <fran@tailscale.com>
---
 appc/appconnector.go      | 39 +++++++++++++++++++++++++++++++++++++++
 appc/appconnector_test.go | 33 +++++++++++++++++++++++++++++++++
 2 files changed, 72 insertions(+)

diff --git a/appc/appconnector.go b/appc/appconnector.go
index 9dab4764c..671ced953 100644
--- a/appc/appconnector.go
+++ b/appc/appconnector.go
@@ -11,6 +11,7 @@
 
 import (
 	"context"
+	"fmt"
 	"net/netip"
 	"slices"
 	"strings"
@@ -21,6 +22,7 @@
 	"golang.org/x/net/dns/dnsmessage"
 	"tailscale.com/types/logger"
 	"tailscale.com/types/views"
+	"tailscale.com/util/clientmetric"
 	"tailscale.com/util/dnsname"
 	"tailscale.com/util/execqueue"
 	"tailscale.com/util/mak"
@@ -78,6 +80,42 @@ type RouteAdvertiser interface {
 	UnadvertiseRoute(...netip.Prefix) error
 }
 
+var (
+	metricStoreRoutesRateBuckets = []int64{1, 2, 3, 4, 5, 10, 100, 1000}
+	metricStoreRoutesNBuckets    = []int64{1, 2, 3, 4, 5, 10, 100, 1000, 10000}
+	metricStoreRoutesRate        []*clientmetric.Metric
+	metricStoreRoutesN           []*clientmetric.Metric
+)
+
+func initMetricStoreRoutes() {
+	for _, n := range metricStoreRoutesRateBuckets {
+		metricStoreRoutesRate = append(metricStoreRoutesRate, clientmetric.NewCounter(fmt.Sprintf("appc_store_routes_rate_%d", n)))
+	}
+	metricStoreRoutesRate = append(metricStoreRoutesRate, clientmetric.NewCounter("appc_store_routes_rate_over"))
+	for _, n := range metricStoreRoutesNBuckets {
+		metricStoreRoutesN = append(metricStoreRoutesN, clientmetric.NewCounter(fmt.Sprintf("appc_store_routes_n_routes_%d", n)))
+	}
+	metricStoreRoutesN = append(metricStoreRoutesN, clientmetric.NewCounter("appc_store_routes_n_routes_over"))
+}
+
+func recordMetric(val int64, buckets []int64, metrics []*clientmetric.Metric) {
+	if len(buckets) < 1 {
+		return
+	}
+	// finds the first bucket where val <=, or len(buckets) if none match
+	// for bucket values of 1, 10, 100; 0-1 goes to [0], 2-10 goes to [1], 11-100 goes to [2], 101+ goes to [3]
+	bucket, _ := slices.BinarySearch(buckets, val)
+	metrics[bucket].Add(1)
+}
+
+func metricStoreRoutes(rate, nRoutes int64) {
+	if len(metricStoreRoutesRate) == 0 {
+		initMetricStoreRoutes()
+	}
+	recordMetric(rate, metricStoreRoutesRateBuckets, metricStoreRoutesRate)
+	recordMetric(nRoutes, metricStoreRoutesNBuckets, metricStoreRoutesN)
+}
+
 // RouteInfo is a data structure used to persist the in memory state of an AppConnector
 // so that we can know, even after a restart, which routes came from ACLs and which were
 // learned from domains.
@@ -141,6 +179,7 @@ func NewAppConnector(logf logger.Logf, routeAdvertiser RouteAdvertiser, routeInf
 	}
 	ac.writeRateMinute = newRateLogger(time.Now, time.Minute, func(c int64, s time.Time, l int64) {
 		ac.logf("routeInfo write rate: %d in minute starting at %v (%d routes)", c, s, l)
+		metricStoreRoutes(c, l)
 	})
 	ac.writeRateDay = newRateLogger(time.Now, 24*time.Hour, func(c int64, s time.Time, l int64) {
 		ac.logf("routeInfo write rate: %d in 24 hours starting at %v (%d routes)", c, s, l)
diff --git a/appc/appconnector_test.go b/appc/appconnector_test.go
index cf5efeff5..7dba8cebd 100644
--- a/appc/appconnector_test.go
+++ b/appc/appconnector_test.go
@@ -15,6 +15,7 @@
 	"golang.org/x/net/dns/dnsmessage"
 	"tailscale.com/appc/appctest"
 	"tailscale.com/tstest"
+	"tailscale.com/util/clientmetric"
 	"tailscale.com/util/mak"
 	"tailscale.com/util/must"
 )
@@ -569,3 +570,35 @@ func TestRateLogger(t *testing.T) {
 		t.Fatalf("wasCalled: got false, want true")
 	}
 }
+
+func TestRouteStoreMetrics(t *testing.T) {
+	metricStoreRoutes(1, 1)
+	metricStoreRoutes(1, 1)         // the 1 buckets value should be 2
+	metricStoreRoutes(5, 5)         // the 5 buckets value should be 1
+	metricStoreRoutes(6, 6)         // the 10 buckets value should be 1
+	metricStoreRoutes(10001, 10001) // the over buckets value should be 1
+	wanted := map[string]int64{
+		"appc_store_routes_n_routes_1":    2,
+		"appc_store_routes_rate_1":        2,
+		"appc_store_routes_n_routes_5":    1,
+		"appc_store_routes_rate_5":        1,
+		"appc_store_routes_n_routes_10":   1,
+		"appc_store_routes_rate_10":       1,
+		"appc_store_routes_n_routes_over": 1,
+		"appc_store_routes_rate_over":     1,
+	}
+	for _, x := range clientmetric.Metrics() {
+		if x.Value() != wanted[x.Name()] {
+			t.Errorf("%s: want: %d, got: %d", x.Name(), wanted[x.Name()], x.Value())
+		}
+	}
+}
+
+func TestMetricBucketsAreSorted(t *testing.T) {
+	if !slices.IsSorted(metricStoreRoutesRateBuckets) {
+		t.Errorf("metricStoreRoutesRateBuckets must be in order")
+	}
+	if !slices.IsSorted(metricStoreRoutesNBuckets) {
+		t.Errorf("metricStoreRoutesNBuckets must be in order")
+	}
+}