Merge pull request #883 from yeya24/feature/cdn-metrics

feature: add some cdn metrics
dragonflyoss · Sep 9, 2019 · 536bc97 · 536bc97
2 parents c9927ed + c6a29e0
commit 536bc97
Show file tree

Hide file tree

Showing 5 changed files with 39 additions and 8 deletions.
diff --git a/docs/user_guide/metrics.md b/docs/user_guide/metrics.md
@@ -16,8 +16,11 @@ This doc contains all the metrics that Dragonfly components currently support. N
 - dragonfly_supernode_dfgettasks_registered_total{callsystem} - total times of registering new dfgettasks. counter type.
 - dragonfly_supernode_dfgettasks_failed_total{callsystem} - total times of failed dfgettasks. counter type.
 - dragonfly_supernode_schedule_duration_milliseconds{peer} - duration for task scheduling in milliseconds
-- dragonfly_supernode_trigger_cdn_total{} - total times of triggering cdn.
-- dragonfly_supernode_trigger_cdn_failed_total{} - total failed times of triggering cdn.
+- dragonfly_supernode_cdn_trigger_total{} - total times of triggering cdn. counter type.
+- dragonfly_supernode_cdn_trigger_total{} - total failed times of triggering cdn. counter type.
+- dragonfly_supernode_cdn_cache_hit_total{} - total times of hitting cdn cache. counter type.
+- dragonfly_supernode_cdn_download_total{} - total times of cdn downloading. counter type.
+- dragonfly_supernode_cdn_download_failed_total{} - total failure times of cdn downloading. counter type.
 
 ## Dfdaemon
 

diff --git a/supernode/daemon/mgr/cdn/downloader_test.go b/supernode/daemon/mgr/cdn/downloader_test.go
@@ -31,6 +31,7 @@ import (
 	"github.com/dragonflyoss/Dragonfly/supernode/httpclient"
 
 	"github.com/go-check/check"
+	"github.com/prometheus/client_golang/prometheus"
 )
 
 func Test(t *testing.T) {
@@ -45,7 +46,7 @@ func init() {
 }
 
 func (s *CDNDownloadTestSuite) TestDownload(c *check.C) {
-	cm, _ := NewManager(config.NewConfig(), nil, nil, httpclient.NewOriginClient())
+	cm, _ := NewManager(config.NewConfig(), nil, nil, httpclient.NewOriginClient(), prometheus.DefaultRegisterer)
 	bytes := []byte("hello world")
 	bytesLength := int64(len(bytes))
 

diff --git a/supernode/daemon/mgr/cdn/manager.go b/supernode/daemon/mgr/cdn/manager.go
@@ -23,20 +23,41 @@ import (
 
 	"github.com/dragonflyoss/Dragonfly/apis/types"
 	"github.com/dragonflyoss/Dragonfly/pkg/limitreader"
+	"github.com/dragonflyoss/Dragonfly/pkg/metricsutils"
 	"github.com/dragonflyoss/Dragonfly/pkg/netutils"
 	"github.com/dragonflyoss/Dragonfly/pkg/ratelimiter"
 	"github.com/dragonflyoss/Dragonfly/pkg/stringutils"
 	"github.com/dragonflyoss/Dragonfly/supernode/config"
 	"github.com/dragonflyoss/Dragonfly/supernode/daemon/mgr"
+	"github.com/dragonflyoss/Dragonfly/supernode/httpclient"
 	"github.com/dragonflyoss/Dragonfly/supernode/store"
 	"github.com/dragonflyoss/Dragonfly/supernode/util"
 
-	"github.com/dragonflyoss/Dragonfly/supernode/httpclient"
+	"github.com/prometheus/client_golang/prometheus"
 	"github.com/sirupsen/logrus"
 )
 
 var _ mgr.CDNMgr = &Manager{}
 
+type metrics struct {
+	cdnCacheHitCount     *prometheus.CounterVec
+	cdnDownloadCount     *prometheus.CounterVec
+	cdnDownloadFailCount *prometheus.CounterVec
+}
+
+func newMetrics(register prometheus.Registerer) *metrics {
+	return &metrics{
+		cdnCacheHitCount: metricsutils.NewCounter(config.SubsystemSupernode, "cdn_cache_hit_total",
+			"Total times of hitting cdn cache", []string{}, register),
+
+		cdnDownloadCount: metricsutils.NewCounter(config.SubsystemSupernode, "cdn_download_total",
+			"Total times of cdn download", []string{}, register),
+
+		cdnDownloadFailCount: metricsutils.NewCounter(config.SubsystemSupernode, "cdn_download_failed_total",
+			"Total failure times of cdn download", []string{}, register),
+	}
+}
+
 // Manager is an implementation of the interface of CDNMgr.
 type Manager struct {
 	cfg             *config.Config
@@ -51,10 +72,12 @@ type Manager struct {
 	originClient    httpclient.OriginHTTPClient
 	pieceMD5Manager *pieceMD5Mgr
 	writer          *superWriter
+	metrics         *metrics
 }
 
 // NewManager returns a new Manager.
-func NewManager(cfg *config.Config, cacheStore *store.Store, progressManager mgr.ProgressMgr, originClient httpclient.OriginHTTPClient) (*Manager, error) {
+func NewManager(cfg *config.Config, cacheStore *store.Store, progressManager mgr.ProgressMgr,
+	originClient httpclient.OriginHTTPClient, register prometheus.Registerer) (*Manager, error) {
 	rateLimiter := ratelimiter.NewRateLimiter(ratelimiter.TransRate(config.TransLimit(cfg.MaxBandwidth-cfg.SystemReservedBandwidth)), 2)
 	metaDataManager := newFileMetaDataManager(cacheStore)
 	pieceMD5Manager := newpieceMD5Mgr()
@@ -71,6 +94,7 @@ func NewManager(cfg *config.Config, cacheStore *store.Store, progressManager mgr
 		detector:        newCacheDetector(cacheStore, metaDataManager, originClient),
 		originClient:    originClient,
 		writer:          newSuperWriter(cacheStore, cdnReporter),
+		metrics:         newMetrics(register),
 	}, nil
 }
 
@@ -95,6 +119,7 @@ func (cm *Manager) TriggerCDN(ctx context.Context, task *types.TaskInfo) (*types
 
 	if startPieceNum == -1 {
 		logrus.Infof("cache full hit for taskId:%s on local", task.ID)
+		cm.metrics.cdnCacheHitCount.WithLabelValues().Inc()
 		return updateTaskInfo, nil
 	}
 
@@ -107,7 +132,9 @@ func (cm *Manager) TriggerCDN(ctx context.Context, task *types.TaskInfo) (*types
 
 	// start to download the source file
 	resp, err := cm.download(ctx, task.ID, task.RawURL, task.Headers, startPieceNum, httpFileLength, pieceContSize)
+	cm.metrics.cdnDownloadCount.WithLabelValues().Inc()
 	if err != nil {
+		cm.metrics.cdnDownloadFailCount.WithLabelValues().Inc()
 		return getUpdateTaskInfoWithStatusOnly(types.TaskInfoCdnStatusFAILED), err
 	}
 	defer resp.Body.Close()

diff --git a/supernode/daemon/mgr/task/manager.go b/supernode/daemon/mgr/task/manager.go
@@ -58,10 +58,10 @@ func newMetrics(register prometheus.Registerer) *metrics {
 		tasksRegisterCount: metricsutils.NewCounter(config.SubsystemSupernode, "tasks_registered_total",
 			"Total times of registering tasks", []string{}, register),
 
-		triggerCdnCount: metricsutils.NewCounter(config.SubsystemSupernode, "trigger_cdn_total",
+		triggerCdnCount: metricsutils.NewCounter(config.SubsystemSupernode, "cdn_trigger_total",
 			"Total times of triggering cdn", []string{}, register),
 
-		triggerCdnFailCount: metricsutils.NewCounter(config.SubsystemSupernode, "trigger_cdn_failed_total",
+		triggerCdnFailCount: metricsutils.NewCounter(config.SubsystemSupernode, "cdn_trigger_failed_total",
 			"Total failure times of triggering cdn", []string{}, register),
 
 		scheduleDurationMilliSeconds: metricsutils.NewHistogram(config.SubsystemSupernode, "schedule_duration_milliseconds",

diff --git a/supernode/server/server.go b/supernode/server/server.go
@@ -87,7 +87,7 @@ func New(cfg *config.Config, register prometheus.Registerer) (*Server, error) {
 		return nil, err
 	}
 
-	cdnMgr, err := cdn.NewManager(cfg, storeLocal, progressMgr, originClient)
+	cdnMgr, err := cdn.NewManager(cfg, storeLocal, progressMgr, originClient, register)
 	if err != nil {
 		return nil, err
 	}