Add failed_download_count prometheus metric (#309) (#395)

Prometheus metric "failed_download_count" for failed downloads

docs/prometheus_grafana.md

@@ -20,6 +20,7 @@ Following metrics will be exported:
 | blocky_cache_hit_count / blocky_cache_miss_count | Cache hit/miss counters |
 | blocky_prefetch_count | Amount of prefetched DNS responses |
 | blocky_prefetch_domain_name_cache_count | Amount of domain names being prefetched |
+| blocky_failed_download_count      | Number of failed list downloads |
 
 ### Grafana dashboard
 

evt/events.go

@@ -29,6 +29,9 @@ const (
 	// CachingDomainsToPrefetchCountChanged fires, if a number of domains being prefetched changed, Parameter: new count
 	CachingDomainsToPrefetchCountChanged = "caching:domainsToPrefetchCountChanged"
 
+	// CachingFailedDownloadChanged fires, if a download of a blocking list fails
+	CachingFailedDownloadChanged = "caching:failedDownload"
+
 	// ApplicationStarted fires on start of the application. Parameter: version number, build time
 	ApplicationStarted = "application:started"
 )

lists/list_cache.go

@@ -272,6 +272,8 @@ func (b *ListCache) downloadFile(link string) (io.ReadCloser, error) {
 			default:
 				logger.Warnf("Can't download file: %s", err)
 			}
+
+			evt.Bus().Publish(evt.CachingFailedDownloadChanged, link)
 		}))
 
 	return body, err

lists/list_cache_test.go

@@ -7,7 +7,7 @@ import (
 	"sync/atomic"
 	"time"
 
-	"github.com/0xERR0R/blocky/evt"
+	. "github.com/0xERR0R/blocky/evt"
 
 	. "github.com/0xERR0R/blocky/helpertest"
 	. "github.com/onsi/ginkgo"
@@ -69,6 +69,11 @@ var _ = Describe("ListCache", func() {
 		When("If timeout occurs", func() {
 			var attempt uint64 = 1
 			It("Should perform a retry", func() {
+				failedDownloadCount := 0
+				_ = Bus().SubscribeOnce(CachingFailedDownloadChanged, func(_ string) {
+					failedDownloadCount++
+				})
+
 				// should produce a timeout on first attempt
 				s := httptest.NewServer(http.HandlerFunc(func(rw http.ResponseWriter, req *http.Request) {
 					a := atomic.LoadUint64(&attempt)
@@ -92,6 +97,7 @@ var _ = Describe("ListCache", func() {
 					g.Expect(group).Should(Equal("gr1"))
 				}, "1s").Should(Succeed())
 
+				Expect(failedDownloadCount).Should(Equal(1))
 			})
 		})
 		When("a temporary err occurs on download", func() {
@@ -214,7 +220,7 @@ var _ = Describe("ListCache", func() {
 
 				resultCnt := 0
 
-				_ = evt.Bus().SubscribeOnce(evt.BlockingCacheGroupChanged, func(listType ListCacheType, group string, cnt int) {
+				_ = Bus().SubscribeOnce(BlockingCacheGroupChanged, func(listType ListCacheType, group string, cnt int) {
 					resultCnt = cnt
 				})
 

metrics/metrics_event_publisher.go

@@ -120,6 +120,7 @@ func registerCachingEventListeners() {
 	missCount := cacheMissCount()
 	prefetchCount := domainPrefetchCount()
 	prefetchHitCount := domainPrefetchHitCount()
+	failedDownloadCount := failedDownloadCount()
 
 	RegisterMetric(entryCount)
 	RegisterMetric(prefetchDomainCount)
@@ -127,6 +128,7 @@ func registerCachingEventListeners() {
 	RegisterMetric(missCount)
 	RegisterMetric(prefetchCount)
 	RegisterMetric(prefetchHitCount)
+	RegisterMetric(failedDownloadCount)
 
 	subscribe(evt.CachingDomainsToPrefetchCountChanged, func(cnt int) {
 		prefetchDomainCount.Set(float64(cnt))
@@ -151,6 +153,17 @@ func registerCachingEventListeners() {
 	subscribe(evt.CachingResultCacheChanged, func(cnt int) {
 		entryCount.Set(float64(cnt))
 	})
+
+	subscribe(evt.CachingFailedDownloadChanged, func(_ string) {
+		failedDownloadCount.Inc()
+	})
+}
+
+func failedDownloadCount() prometheus.Counter {
+	return prometheus.NewCounter(prometheus.CounterOpts{
+		Name: "blocky_failed_download_count",
+		Help: "Failed download counter",
+	})
 }
 
 func cacheHitCount() prometheus.Counter {