refactor: configurable cache metric (#1317)

Allows adding caches for different hcloud resources.

See #1313

Author: lukasmetzner

internal/cache/cache.go

@@ -6,9 +6,8 @@ import (
 	"sync"
 	"time"
 
+	"github.com/prometheus/client_golang/prometheus"
 	"k8s.io/klog/v2"
-
-	"github.com/hetznercloud/hcloud-cloud-controller-manager/internal/metrics"
 )
 
 type Mode string
@@ -63,9 +62,9 @@ type Cache[T any] struct {
 	fetchAll       func(ctx context.Context) ([]*T, error)
 	getID          func(value *T) int64
 	getName        func(value *T) string
-
-	defaultMaxAge time.Duration
-	defaultMode   Mode
+	metric         *prometheus.CounterVec
+	defaultMaxAge  time.Duration
+	defaultMode    Mode
 
 	byID   map[int64]*entry[T]
 	byName map[string]*entry[T]
@@ -79,6 +78,7 @@ func newCache[T any](
 	fetchAll func(ctx context.Context) ([]*T, error),
 	getID func(value *T) int64,
 	getName func(value *T) string,
+	metric *prometheus.CounterVec,
 	defaultMode Mode,
 	defaultMaxAge time.Duration,
 ) *Cache[T] {
@@ -88,6 +88,7 @@ func newCache[T any](
 		fetchAll:       fetchAll,
 		getID:          getID,
 		getName:        getName,
+		metric:         metric,
 
 		defaultMode:   defaultMode,
 		defaultMaxAge: defaultMaxAge,
@@ -140,12 +141,12 @@ func (c *Cache[T]) All(ctx context.Context, opts ...RefreshOption) ([]*T, error)
 	}
 
 	if now.Sub(refreshedAllAt) > refreshOpts.maxAge {
-		metrics.CacheRequests.WithLabelValues(subsystem, string(refreshOpts.mode), "miss").Inc()
+		c.metric.WithLabelValues(subsystem, string(refreshOpts.mode), "miss").Inc()
 		if err := c.refreshAll(ctx); err != nil {
 			return nil, err
 		}
 	} else {
-		metrics.CacheRequests.WithLabelValues(subsystem, string(refreshOpts.mode), "hit").Inc()
+		c.metric.WithLabelValues(subsystem, string(refreshOpts.mode), "hit").Inc()
 	}
 
 	values := make([]*T, 0, len(c.byID))
@@ -166,7 +167,7 @@ func (c *Cache[T]) getFromCache(
 	refreshOpts := newCacheRefreshOpts(c, opts...)
 
 	if refreshOpts.mode == ModeOff {
-		metrics.CacheRequests.WithLabelValues(subsystem, string(refreshOpts.mode), "miss").Inc()
+		c.metric.WithLabelValues(subsystem, string(refreshOpts.mode), "miss").Inc()
 		klog.V(4).InfoS("cache mode is off: fetching entry from api", "subsystem", subsystem)
 		return fetch()
 	}
@@ -177,7 +178,7 @@ func (c *Cache[T]) getFromCache(
 	now := time.Now()
 
 	if e := lookup(); e != nil && now.Sub(e.refreshedAt) <= refreshOpts.maxAge {
-		metrics.CacheRequests.WithLabelValues(subsystem, string(refreshOpts.mode), "hit").Inc()
+		c.metric.WithLabelValues(subsystem, string(refreshOpts.mode), "hit").Inc()
 		klog.V(4).InfoS(
 			"cache hit",
 			"subsystem", subsystem,
@@ -201,7 +202,7 @@ func (c *Cache[T]) getFromCache(
 		// Handled above through early return
 	}
 
-	metrics.CacheRequests.WithLabelValues(subsystem, string(refreshOpts.mode), "miss").Inc()
+	c.metric.WithLabelValues(subsystem, string(refreshOpts.mode), "miss").Inc()
 
 	// When the value is not found in the API, the entry is not removed from the cache.
 	// Expired entries are only evicted after an hour and when a value is found.

internal/cache/cache_test.go

@@ -7,6 +7,7 @@ import (
 	"testing/synctest"
 	"time"
 
+	"github.com/prometheus/client_golang/prometheus"
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
 
@@ -34,6 +35,10 @@ func newTestCache(mode Mode) *Cache[hcloud.Server] {
 		nil,
 		func(value *hcloud.Server) int64 { return value.ID },
 		func(value *hcloud.Server) string { return value.Name },
+		prometheus.NewCounterVec(
+			prometheus.CounterOpts{Name: "test_cache_requests_total"},
+			[]string{"subsystem", "mode", "result"},
+		),
 		mode,
 		10*time.Second,
 	)

internal/cache/servercache.go

@@ -4,9 +4,21 @@ import (
 	"context"
 	"time"
 
+	"github.com/prometheus/client_golang/prometheus"
+
+	"github.com/hetznercloud/hcloud-cloud-controller-manager/internal/metrics"
 	"github.com/hetznercloud/hcloud-go/v2/hcloud"
 )
 
+var serverCacheRequests = prometheus.NewCounterVec(prometheus.CounterOpts{
+	Name: "cloud_controller_manager_server_cache_requests_total",
+	Help: "Total cache requests to the Servers API partitioned by subsystem, mode and result.",
+}, []string{"subsystem", "mode", "result"})
+
+func init() {
+	metrics.GetRegistry().MustRegister(serverCacheRequests)
+}
+
 func NewServerCache(client *hcloud.Client, defaultMode Mode, defaultMaxAge time.Duration) *Cache[hcloud.Server] {
 	return newCache[hcloud.Server](
 		func(ctx context.Context, id int64) (*hcloud.Server, error) {
@@ -23,6 +35,7 @@ func NewServerCache(client *hcloud.Client, defaultMode Mode, defaultMaxAge time.
 		},
 		func(value *hcloud.Server) int64 { return value.ID },
 		func(value *hcloud.Server) string { return value.Name },
+		serverCacheRequests,
 		defaultMode,
 		defaultMaxAge,
 	)

internal/cache/servercache_test.go

@@ -51,6 +51,7 @@ func TestNewServerCache(t *testing.T) {
 			require.NotNil(t, cache.fetchAll)
 			require.NotNil(t, cache.getID)
 			require.NotNil(t, cache.getName)
+			require.NotNil(t, cache.metric)
 
 			ctx := t.Context()
 

internal/metrics/metrics.go

@@ -32,20 +32,13 @@ const (
 	writeTimeout   = 20 * time.Second
 )
 
-var (
-	OperationCalled = prometheus.NewCounterVec(prometheus.CounterOpts{
-		Name: "cloud_controller_manager_operations_total",
-		Help: "The total number of operation was called",
-	}, []string{"op"})
-
-	CacheRequests = prometheus.NewCounterVec(prometheus.CounterOpts{
-		Name: "cloud_controller_manager_server_cache_requests_total",
-		Help: "Total cache requests partitioned by subsystem, mode and result.",
-	}, []string{"subsystem", "mode", "result"})
-)
+var OperationCalled = prometheus.NewCounterVec(prometheus.CounterOpts{
+	Name: "cloud_controller_manager_operations_total",
+	Help: "The total number of operation was called",
+}, []string{"op"})
 
 func init() {
-	GetRegistry().MustRegister(OperationCalled, CacheRequests)
+	GetRegistry().MustRegister(OperationCalled)
 }
 
 func GetRegistry() prometheus.Registerer {