fix: more accurate storage metrics after zot restart

Signed-off-by: Alexei Dodon <[email protected]>

pkg/api/controller.go

@@ -372,6 +372,10 @@ func (c *Controller) StartBackgroundTasks(reloadCtx context.Context) {
 		ext.EnableMetricsExtension(c.Config, c.Log, c.Config.Storage.RootDirectory)
 		ext.EnableSearchExtension(c.Config, c.StoreController, c.MetaDB, taskScheduler, c.CveScanner, c.Log)
 	}
+	// runs once if metrics are enabled
+	if c.Config.IsMetricsEnabled() {
+		c.StoreController.DefaultStore.PopulateStorageMetrics(time.Duration(0), taskScheduler)
+	}
 
 	if c.Config.Storage.SubPaths != nil {
 		for route, storageConfig := range c.Config.Storage.SubPaths {
@@ -396,6 +400,10 @@ func (c *Controller) StartBackgroundTasks(reloadCtx context.Context) {
 			substore := c.StoreController.SubStore[route]
 			if substore != nil {
 				substore.RunDedupeBlobs(time.Duration(0), taskScheduler)
+
+				if c.Config.IsMetricsEnabled() {
+					substore.PopulateStorageMetrics(time.Duration(0), taskScheduler)
+				}
 			}
 		}
 	}

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 +1055,76 @@ 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
+}
+
+func (gen *StorageMetricsInitGenerator) getRandomDelay() int {
+	maxDelay := 30
+
+	return gen.rand.Intn(maxDelay)
+}
+
+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.getRandomDelay()
+
+	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 true
+}
+
+func (gen *StorageMetricsInitGenerator) Reset() {
+	gen.lastRepo = ""
+	gen.done = false
+}
+
+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

@@ -1929,6 +1929,16 @@ 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,
+	}
+
+	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

pkg/test/mocks/image_store_mock.go

@@ -55,6 +55,7 @@ type MockedImageStore struct {
 	GetAllBlobsFn                func(repo string) ([]string, error)
 	CleanupRepoFn                func(repo string, blobs []godigest.Digest, removeRepo bool) (int, error)
 	PutIndexContentFn            func(repo string, index ispec.Index) error
+	RunStorageMetricsFn          func(interval time.Duration, sch *scheduler.Scheduler)
 }
 
 func (is MockedImageStore) Lock(t *time.Time) {
@@ -405,3 +406,9 @@ func (is MockedImageStore) PutIndexContent(repo string, index ispec.Index) error
 
 	return nil
 }
+
+func (is MockedImageStore) PopulateStorageMetrics(interval time.Duration, sch *scheduler.Scheduler) {
+	if is.RunStorageMetricsFn != nil {
+		is.RunStorageMetricsFn(interval, sch)
+	}
+}