fix: more accurate storage metrics after zot restart (#1972)

Signed-off-by: Alexei Dodon <adodon@cisco.com>

pkg/storage/common/common.go

@@ -6,8 +6,10 @@ import (
 	"encoding/json"
 	"errors"
 	"fmt"
+	"math/rand"
 	"path"
 	"strings"
+	"time"
 
 	"github.com/docker/distribution/registry/storage/driver"
 	godigest "github.com/opencontainers/go-digest"
@@ -18,6 +20,7 @@ import (
 
 	zerr "zotregistry.io/zot/errors"
 	zcommon "zotregistry.io/zot/pkg/common"
+	"zotregistry.io/zot/pkg/extensions/monitoring"
 	zlog "zotregistry.io/zot/pkg/log"
 	"zotregistry.io/zot/pkg/scheduler"
 	storageConstants "zotregistry.io/zot/pkg/storage/constants"
@@ -1052,3 +1055,72 @@ func (dt *dedupeTask) DoWork(ctx context.Context) error {
 
 	return err
 }
+
+type StorageMetricsInitGenerator struct {
+	ImgStore storageTypes.ImageStore
+	done     bool
+	Metrics  monitoring.MetricServer
+	lastRepo string
+	nextRun  time.Time
+	rand     *rand.Rand
+	Log      zlog.Logger
+	MaxDelay int
+}
+
+func (gen *StorageMetricsInitGenerator) Next() (scheduler.Task, error) {
+	if gen.lastRepo == "" && gen.nextRun.IsZero() {
+		gen.rand = rand.New(rand.NewSource(time.Now().UTC().UnixNano())) //nolint: gosec
+	}
+
+	delay := gen.rand.Intn(gen.MaxDelay)
+
+	gen.nextRun = time.Now().Add(time.Duration(delay) * time.Second)
+
+	repo, err := gen.ImgStore.GetNextRepository(gen.lastRepo)
+	if err != nil {
+		return nil, err
+	}
+
+	gen.Log.Debug().Str("repo", repo).Int("randomDelay", delay).Msg("StorageMetricsInitGenerator")
+
+	if repo == "" {
+		gen.done = true
+
+		return nil, nil
+	}
+	gen.lastRepo = repo
+
+	return NewStorageMetricsTask(gen.ImgStore, gen.Metrics, repo), nil
+}
+
+func (gen *StorageMetricsInitGenerator) IsDone() bool {
+	return gen.done
+}
+
+func (gen *StorageMetricsInitGenerator) IsReady() bool {
+	return time.Now().After(gen.nextRun)
+}
+
+func (gen *StorageMetricsInitGenerator) Reset() {
+	gen.lastRepo = ""
+	gen.done = false
+	gen.nextRun = time.Time{}
+}
+
+type smTask struct {
+	imgStore storageTypes.ImageStore
+	metrics  monitoring.MetricServer
+	repo     string
+}
+
+func NewStorageMetricsTask(imgStore storageTypes.ImageStore, metrics monitoring.MetricServer, repo string,
+) *smTask {
+	return &smTask{imgStore, metrics, repo}
+}
+
+func (smt *smTask) DoWork(ctx context.Context) error {
+	// run task
+	monitoring.SetStorageUsage(smt.metrics, smt.imgStore.RootDir(), smt.repo)
+
+	return nil
+}

pkg/storage/imagestore/imagestore.go

@@ -488,7 +488,10 @@ func (is *ImageStore) PutImageManifest(repo, reference, mediaType string, //noli
 		is.Unlock(&lockLatency)
 
 		if err == nil {
-			monitoring.SetStorageUsage(is.metrics, is.rootDir, repo)
+			if is.storeDriver.Name() == storageConstants.LocalStorageDriverName {
+				monitoring.SetStorageUsage(is.metrics, is.rootDir, repo)
+			}
+
 			monitoring.IncUploadCounter(is.metrics, repo)
 		}
 	}()
@@ -621,7 +624,11 @@ func (is *ImageStore) DeleteImageManifest(repo, reference string, detectCollisio
 }
 
 func (is *ImageStore) deleteImageManifest(repo, reference string, detectCollisions bool) error {
-	defer monitoring.SetStorageUsage(is.metrics, is.rootDir, repo)
+	defer func() {
+		if is.storeDriver.Name() == storageConstants.LocalStorageDriverName {
+			monitoring.SetStorageUsage(is.metrics, is.rootDir, repo)
+		}
+	}()
 
 	index, err := common.GetIndex(is, repo, is.log)
 	if err != nil {
@@ -1929,6 +1936,17 @@ func (is *ImageStore) RunDedupeBlobs(interval time.Duration, sch *scheduler.Sche
 	sch.SubmitGenerator(generator, interval, scheduler.MediumPriority)
 }
 
+func (is *ImageStore) PopulateStorageMetrics(interval time.Duration, sch *scheduler.Scheduler) {
+	generator := &common.StorageMetricsInitGenerator{
+		ImgStore: is,
+		Metrics:  is.metrics,
+		Log:      is.log,
+		MaxDelay: 15, //nolint:gomnd
+	}
+
+	sch.SubmitGenerator(generator, interval, scheduler.LowPriority)
+}
+
 type blobStream struct {
 	reader io.Reader
 	closer io.Closer

pkg/storage/types/types.go

@@ -61,6 +61,7 @@ type ImageStore interface { //nolint:interfacebloat
 	RunDedupeForDigest(digest godigest.Digest, dedupe bool, duplicateBlobs []string) error
 	GetNextDigestWithBlobPaths(repos []string, lastDigests []godigest.Digest) (godigest.Digest, []string, error)
 	GetAllBlobs(repo string) ([]string, error)
+	PopulateStorageMetrics(interval time.Duration, sch *scheduler.Scheduler)
 }
 
 type Driver interface { //nolint:interfacebloat