remove complex blob cache GC code

Containers can contorl if a committed snapshot should
be removed. We don't have to consider if snapshot belongs
to multiple images.

Just respond to containerd a committed snapshot shoud be remove.

Signed-off-by: Changwei Ge <[email protected]>

go.mod

@@ -12,6 +12,7 @@ require (
 	github.com/docker/cli v20.10.0-beta1.0.20201029214301-1d20b15adc38+incompatible
 	github.com/golang/groupcache v0.0.0-20210331224755-41bb18bfe9da
 	github.com/google/go-containerregistry v0.5.1
+	github.com/gorilla/mux v1.8.0
 	github.com/hashicorp/go-retryablehttp v0.7.1
 	github.com/opencontainers/go-digest v1.0.0
 	github.com/opencontainers/image-spec v1.0.3-0.20211202183452-c5a74bcca799
@@ -96,10 +97,10 @@ require (
 )
 
 retract (
-	v0.11.0
-	v0.11.1
-	v0.11.2
 	v0.11.3
+	v0.11.2
+	v0.11.1
+	v0.11.0
 )
 
 replace (

pkg/cache/manager.go

@@ -1,3 +1,9 @@
+/*
+ * Copyright (c) 2022. Nydus Developers. All rights reserved.
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
 package cache
 
 import (
@@ -11,13 +17,17 @@ import (
 	"github.com/containerd/containerd/log"
 	"github.com/containerd/containerd/snapshots"
 	"github.com/containerd/continuity/fs"
-	"github.com/containerd/nydus-snapshotter/config"
 	"github.com/containerd/nydus-snapshotter/pkg/store"
 )
 
+const (
+	chunkMapFileSuffix = ".chunk_map"
+	metaFileSuffix     = ".blob.meta"
+	// Blob cache is suffixed after nydus v2.1
+	dataFileSuffix = ".blob.data"
+)
+
 type Manager struct {
-	db       DB
-	store    *Store
 	cacheDir string
 	period   time.Duration
 	eventCh  chan struct{}
@@ -37,81 +47,21 @@ func NewManager(opt Opt) (*Manager, error) {
 		return nil, errors.Wrapf(err, "failed to create cache dir %s", opt.CacheDir)
 	}
 
-	db, err := store.NewCacheStore(opt.Database)
-	if err != nil {
-		return nil, err
-	}
-	s := NewStore(opt.CacheDir)
-
 	eventCh := make(chan struct{})
 	m := &Manager{
-		db:       db,
-		store:    s,
 		cacheDir: opt.CacheDir,
 		period:   opt.Period,
 		eventCh:  eventCh,
 		fsDriver: opt.FsDriver,
 	}
 
-	// For fscache backend, the cache is maintained by the kernel fscache module,
-	// so here we ignore gc for now, and in the future we need another design
-	// to remove the cache.
-	if opt.FsDriver == config.FsDriverFscache {
-		return m, nil
-	}
-
-	go m.runGC()
-	log.L.Info("gc goroutine start...")
-
 	return m, nil
 }
 
 func (m *Manager) CacheDir() string {
 	return m.cacheDir
 }
 
-func (m *Manager) SchedGC() {
-	if m.fsDriver == config.FsDriverFscache {
-		return
-	}
-	m.eventCh <- struct{}{}
-}
-
-func (m *Manager) runGC() {
-	tick := time.NewTicker(m.period)
-	defer tick.Stop()
-	for {
-		select {
-		case <-m.eventCh:
-			if err := m.gc(); err != nil {
-				log.L.Infof("[event] cache gc err, %v", err)
-			}
-			tick.Reset(m.period)
-		case <-tick.C:
-			if err := m.gc(); err != nil {
-				log.L.Infof("[tick] cache gc err, %v", err)
-			}
-		}
-	}
-}
-
-func (m *Manager) gc() error {
-	delBlobs, err := m.db.GC(m.store.DelBlob)
-	if err != nil {
-		return errors.Wrapf(err, "cache gc err")
-	}
-	log.L.Debugf("remove %d unused blobs successfully", len(delBlobs))
-	return nil
-}
-
-func (m *Manager) AddSnapshot(imageID string, blobs []string) error {
-	return m.db.AddSnapshot(imageID, blobs)
-}
-
-func (m *Manager) DelSnapshot(imageID string) error {
-	return m.db.DelSnapshot(imageID)
-}
-
 // Report each blob disk usage
 func (m *Manager) CacheUsage(ctx context.Context, blobID string) (snapshots.Usage, error) {
 	var usage snapshots.Usage
@@ -136,3 +86,25 @@ func (m *Manager) CacheUsage(ctx context.Context, blobID string) (snapshots.Usag
 
 	return usage, nil
 }
+
+func (m *Manager) RemoveBlobCache(blobID string) error {
+	blobCachePath := path.Join(m.cacheDir, blobID)
+	blobCacheSuffixedPath := path.Join(m.cacheDir, blobID+dataFileSuffix)
+	blobChunkMap := path.Join(m.cacheDir, blobID+chunkMapFileSuffix)
+	blobMeta := path.Join(m.cacheDir, blobID+metaFileSuffix)
+
+	// NOTE: Delete chunk bitmap file before data blob
+	stuffs := []string{blobChunkMap, blobMeta, blobCachePath, blobCacheSuffixedPath}
+
+	for _, f := range stuffs {
+		err := os.Remove(f)
+		if err != nil {
+			if errors.Is(err, os.ErrNotExist) {
+				log.L.Debugf("file %s doest not exist.", f)
+				continue
+			}
+			return err
+		}
+	}
+	return nil
+}

pkg/filesystem/fs/fs.go

@@ -581,6 +581,15 @@ func (fs *Filesystem) CacheUsage(ctx context.Context, blobDigest string) (snapsh
 	return fs.cacheMgr.CacheUsage(ctx, blobID)
 }
 
+func (fs *Filesystem) RemoveCache(blobDigest string) error {
+	digest := digest.Digest(blobDigest)
+	if err := digest.Validate(); err != nil {
+		return errors.Wrapf(err, "invalid blob digest from label %s", label.CRILayerDigest)
+	}
+	blobID := digest.Hex()
+	return fs.cacheMgr.RemoveBlobCache(blobID)
+}
+
 // WaitUntilReady wait until daemon ready by snapshotID, it will wait until nydus domain socket established
 // and the status of nydusd daemon must be ready
 func (fs *Filesystem) WaitUntilReady(snapshotID string) error {
@@ -597,19 +606,6 @@ func (fs *Filesystem) WaitUntilReady(snapshotID string) error {
 	return d.WaitUntilReady()
 }
 
-func (fs *Filesystem) DelSnapshot(imageID string) error {
-	if fs.cacheMgr == nil {
-		return nil
-	}
-
-	if err := fs.cacheMgr.DelSnapshot(imageID); err != nil {
-		return errors.Wrap(err, "del snapshot err")
-	}
-	log.L.Debugf("remove snapshot %s\n", imageID)
-	fs.cacheMgr.SchedGC()
-	return nil
-}
-
 func (fs *Filesystem) Umount(ctx context.Context, mountPoint string) error {
 	if !fs.hasDaemon() {
 		return nil
@@ -709,21 +705,6 @@ func (fs *Filesystem) mount(d *daemon.Daemon, labels map[string]string) error {
 	return nil
 }
 
-func (fs *Filesystem) AddSnapshot(labels map[string]string) error {
-	// Do nothing if there's no cacheMgr
-	if fs.cacheMgr == nil {
-		return nil
-	}
-
-	imageID, _ := registry.ParseLabels(labels)
-	blobs, err := fs.getBlobIDs(labels)
-	if err != nil {
-		return err
-	}
-	log.L.Infof("image %s with blob caches %v", imageID, blobs)
-	return fs.cacheMgr.AddSnapshot(imageID, blobs)
-}
-
 // 1. Create a daemon instance
 // 2. Build command line
 // 3. Start daemon