scrape: Add metrics to track bytes and entries in the metadata cache (#6675)

Signed-off-by: gotjosh <josue@grafana.com>

scrape/manager.go

@@ -27,12 +27,79 @@ import (
 	"github.com/go-kit/kit/log/level"
 	"github.com/pkg/errors"
 
+	"github.com/prometheus/client_golang/prometheus"
 	"github.com/prometheus/prometheus/config"
 	"github.com/prometheus/prometheus/discovery/targetgroup"
 	"github.com/prometheus/prometheus/pkg/labels"
 	"github.com/prometheus/prometheus/storage"
 )
 
+var targetMetadataCache = newMetadataMetricsCollector()
+
+// MetadataMetricsCollector is a Custom Collector for the metadata cache metrics.
+type MetadataMetricsCollector struct {
+	CacheEntries *prometheus.Desc
+	CacheBytes   *prometheus.Desc
+
+	scrapeManager *Manager
+}
+
+func newMetadataMetricsCollector() *MetadataMetricsCollector {
+	return &MetadataMetricsCollector{
+		CacheEntries: prometheus.NewDesc(
+			"prometheus_target_metadata_cache_entries",
+			"Total number of metric metadata entries in the cache",
+			[]string{"scrape_job"},
+			nil,
+		),
+		CacheBytes: prometheus.NewDesc(
+			"prometheus_target_metadata_cache_bytes",
+			"The number of bytes that are currently used for storing metric metadata in the cache",
+			[]string{"scrape_job"},
+			nil,
+		),
+	}
+}
+
+func (mc *MetadataMetricsCollector) registerManager(m *Manager) {
+	mc.scrapeManager = m
+}
+
+// Describe sends the metrics descriptions to the channel.
+func (mc *MetadataMetricsCollector) Describe(ch chan<- *prometheus.Desc) {
+	ch <- mc.CacheEntries
+	ch <- mc.CacheBytes
+}
+
+// Collect creates and sends the metrics for the metadata cache.
+func (mc *MetadataMetricsCollector) Collect(ch chan<- prometheus.Metric) {
+	if mc.scrapeManager == nil {
+		return
+	}
+
+	for tset, targets := range mc.scrapeManager.TargetsActive() {
+		var size, length int
+		for _, t := range targets {
+			size += t.MetadataSize()
+			length += t.MetadataLength()
+		}
+
+		ch <- prometheus.MustNewConstMetric(
+			mc.CacheEntries,
+			prometheus.GaugeValue,
+			float64(length),
+			tset,
+		)
+
+		ch <- prometheus.MustNewConstMetric(
+			mc.CacheBytes,
+			prometheus.GaugeValue,
+			float64(size),
+			tset,
+		)
+	}
+}
+
 // Appendable returns an Appender.
 type Appendable interface {
 	Appender() (storage.Appender, error)
@@ -43,7 +110,7 @@ func NewManager(logger log.Logger, app Appendable) *Manager {
 	if logger == nil {
 		logger = log.NewNopLogger()
 	}
-	return &Manager{
+	m := &Manager{
 		append:        app,
 		logger:        logger,
 		scrapeConfigs: make(map[string]*config.ScrapeConfig),
@@ -51,6 +118,9 @@ func NewManager(logger log.Logger, app Appendable) *Manager {
 		graceShut:     make(chan struct{}),
 		triggerReload: make(chan struct{}, 1),
 	}
+	targetMetadataCache.registerManager(m)
+
+	return m
 }
 
 // Manager maintains a set of scrape pools and manages start/stop cycles

scrape/scrape.go

@@ -136,19 +136,22 @@ var (
 )
 
 func init() {
-	prometheus.MustRegister(targetIntervalLength)
-	prometheus.MustRegister(targetReloadIntervalLength)
-	prometheus.MustRegister(targetScrapePools)
-	prometheus.MustRegister(targetScrapePoolsFailed)
-	prometheus.MustRegister(targetScrapePoolReloads)
-	prometheus.MustRegister(targetScrapePoolReloadsFailed)
-	prometheus.MustRegister(targetSyncIntervalLength)
-	prometheus.MustRegister(targetScrapePoolSyncsCounter)
-	prometheus.MustRegister(targetScrapeSampleLimit)
-	prometheus.MustRegister(targetScrapeSampleDuplicate)
-	prometheus.MustRegister(targetScrapeSampleOutOfOrder)
-	prometheus.MustRegister(targetScrapeSampleOutOfBounds)
-	prometheus.MustRegister(targetScrapeCacheFlushForced)
+	prometheus.MustRegister(
+		targetIntervalLength,
+		targetReloadIntervalLength,
+		targetScrapePools,
+		targetScrapePoolsFailed,
+		targetScrapePoolReloads,
+		targetScrapePoolReloadsFailed,
+		targetSyncIntervalLength,
+		targetScrapePoolSyncsCounter,
+		targetScrapeSampleLimit,
+		targetScrapeSampleDuplicate,
+		targetScrapeSampleOutOfOrder,
+		targetScrapeSampleOutOfBounds,
+		targetScrapeCacheFlushForced,
+		targetMetadataCache,
+	)
 }
 
 // scrapePool manages scrapes for sets of targets.
@@ -658,6 +661,11 @@ type metaEntry struct {
 	unit     string
 }
 
+func (m *metaEntry) size() int {
+	// The attribute lastIter although part of the struct it is not metadata.
+	return len(m.help) + len(m.unit) + len(m.typ)
+}
+
 func newScrapeCache() *scrapeCache {
 	return &scrapeCache{
 		series:        map[string]*cacheEntry{},
@@ -842,6 +850,25 @@ func (c *scrapeCache) ListMetadata() []MetricMetadata {
 	return res
 }
 
+// MetadataSize returns the size of the metadata cache.
+func (c *scrapeCache) SizeMetadata() (s int) {
+	c.metaMtx.Lock()
+	defer c.metaMtx.Unlock()
+	for _, e := range c.metadata {
+		s += e.size()
+	}
+
+	return s
+}
+
+// MetadataLen returns the number of metadata entries in the cache.
+func (c *scrapeCache) LengthMetadata() int {
+	c.metaMtx.Lock()
+	defer c.metaMtx.Unlock()
+
+	return len(c.metadata)
+}
+
 func newScrapeLoop(ctx context.Context,
 	sc scraper,
 	l log.Logger,

scrape/target.go

@@ -78,6 +78,8 @@ func (t *Target) String() string {
 type MetricMetadataStore interface {
 	ListMetadata() []MetricMetadata
 	GetMetadata(metric string) (MetricMetadata, bool)
+	SizeMetadata() int
+	LengthMetadata() int
 }
 
 // MetricMetadata is a piece of metadata for a metric.
@@ -98,6 +100,28 @@ func (t *Target) MetadataList() []MetricMetadata {
 	return t.metadata.ListMetadata()
 }
 
+func (t *Target) MetadataSize() int {
+	t.mtx.RLock()
+	defer t.mtx.RUnlock()
+
+	if t.metadata == nil {
+		return 0
+	}
+
+	return t.metadata.SizeMetadata()
+}
+
+func (t *Target) MetadataLength() int {
+	t.mtx.RLock()
+	defer t.mtx.RUnlock()
+
+	if t.metadata == nil {
+		return 0
+	}
+
+	return t.metadata.LengthMetadata()
+}
+
 // Metadata returns type and help metadata for the given metric.
 func (t *Target) Metadata(metric string) (MetricMetadata, bool) {
 	t.mtx.RLock()

web/api/v1/api_test.go

@@ -77,6 +77,9 @@ func (s *testMetaStore) GetMetadata(metric string) (scrape.MetricMetadata, bool)
 	return scrape.MetricMetadata{}, false
 }
 
+func (s *testMetaStore) SizeMetadata() int   { return 0 }
+func (s *testMetaStore) LengthMetadata() int { return 0 }
+
 // testTargetRetriever represents a list of targets to scrape.
 // It is used to represent targets as part of test cases.
 type testTargetRetriever struct {