Files
buildkit/cache/remote.go
Yan Song cf70ab6723 cache: support nydus compression type
Nydus image is a container accelerated image format provided by the
Dragonfly image-service project, which offers the ability to pull
image data on demand, without waiting for the entire image pull to
complete and then start the container. It has been put in production
usage and shown vast improvements over the old OCI image format in
terms of container launching speed, image space, and network bandwidth
efficiency, as well as data integrity. Nydus image can be flexibly
configured as a FUSE-based user-space filesystem or in-kernel
EROFS (from Linux kernel v5.16) with Nydus daemon in user-space,
integrating with VM-based container runtime like KataContainers
is much easier.

Nydus has provided a conversion tool Nydusify for converting OCIv1
image to Nydus image and integrated into Harbor Acceld as a conversion
driver, which assumes that the OCI image is already available in the
registry, but a better way would be to build the Nydus images directly
from the build system instead of using the conversion tool, which would
increase the speed of the image export, so we experimentally integrated
the Nydus export in Buildkit.

Unlike other compression formats (gzip, estargz, etc.) in OCI image,
nydus is divided into two types of layer, blob, and bootstrap, where
blob serves as the data part of each layer of the image, and bootstrap
serves as the metadata of the whole image, the bootstrap is equivalent
to the view of the whole image filesystem after all layers overlay. For
example, for an OCI image with 3 layers, the corresponding nydus image
is 4 layers (3 layers of blob + 1 layer of bootstrap).

The nydus-snapshotter project provides a package to do the actual layer
compression, this commit imports the package to implement the export of
nydus compression type.

Signed-off-by: Yan Song <imeoer@linux.alibaba.com>
2022-11-01 03:20:23 +00:00

357 lines
12 KiB
Go

package cache
import (
"context"
"fmt"
"net/url"
"strings"
"github.com/containerd/containerd/content"
"github.com/containerd/containerd/errdefs"
"github.com/containerd/containerd/reference"
"github.com/moby/buildkit/cache/config"
"github.com/moby/buildkit/session"
"github.com/moby/buildkit/solver"
"github.com/moby/buildkit/util/bklog"
"github.com/moby/buildkit/util/compression"
"github.com/moby/buildkit/util/contentutil"
"github.com/moby/buildkit/util/leaseutil"
"github.com/moby/buildkit/util/progress/logs"
"github.com/moby/buildkit/util/pull/pullprogress"
ocispecs "github.com/opencontainers/image-spec/specs-go/v1"
"github.com/pkg/errors"
"golang.org/x/sync/errgroup"
)
type Unlazier interface {
Unlazy(ctx context.Context) error
}
// GetRemotes gets []*solver.Remote from content store for this ref (potentially pulling lazily).
// Compressionopt can be used to specify the compression type of blobs. If Force is true, the compression
// type is applied to all blobs in the chain. If Force is false, it's applied only to the newly created
// layers. If all is true, all available chains that has the specified compression type of topmost blob are
// appended to the result.
// Note: Use WorkerRef.GetRemotes instead as moby integration requires custom GetRemotes implementation.
func (sr *immutableRef) GetRemotes(ctx context.Context, createIfNeeded bool, refCfg config.RefConfig, all bool, s session.Group) ([]*solver.Remote, error) {
ctx, done, err := leaseutil.WithLease(ctx, sr.cm.LeaseManager, leaseutil.MakeTemporary)
if err != nil {
return nil, err
}
defer done(ctx)
// fast path if compression variants aren't required
// NOTE: compressionopt is applied only to *newly created layers* if Force != true.
remote, err := sr.getRemote(ctx, createIfNeeded, refCfg, s)
if err != nil {
return nil, err
}
if !all || refCfg.Compression.Force || len(remote.Descriptors) == 0 {
return []*solver.Remote{remote}, nil // early return if compression variants aren't required
}
// Search all available remotes that has the topmost blob with the specified
// compression with all combination of copmressions
res := []*solver.Remote{remote}
topmost, parentChain := remote.Descriptors[len(remote.Descriptors)-1], remote.Descriptors[:len(remote.Descriptors)-1]
vDesc, err := getBlobWithCompression(ctx, sr.cm.ContentStore, topmost, refCfg.Compression.Type)
if err != nil {
return res, nil // compression variant doesn't exist. return the main blob only.
}
var variants []*solver.Remote
if len(parentChain) == 0 {
variants = append(variants, &solver.Remote{
Descriptors: []ocispecs.Descriptor{vDesc},
Provider: sr.cm.ContentStore,
})
} else {
// get parents with all combination of all available compressions.
parents, err := getAvailableBlobs(ctx, sr.cm.ContentStore, &solver.Remote{
Descriptors: parentChain,
Provider: remote.Provider,
})
if err != nil {
return nil, err
}
variants = appendRemote(parents, vDesc, sr.cm.ContentStore)
}
// Return the main remote and all its compression variants.
// NOTE: Because compressionopt is applied only to *newly created layers* in the main remote (i.e. res[0]),
// it's possible that the main remote doesn't contain any blobs of the compressionopt.Type.
// The topmost blob of the variants (res[1:]) is guaranteed to be the compressionopt.Type.
res = append(res, variants...)
return res, nil
}
func appendRemote(parents []*solver.Remote, desc ocispecs.Descriptor, p content.Provider) (res []*solver.Remote) {
for _, pRemote := range parents {
provider := contentutil.NewMultiProvider(pRemote.Provider)
provider.Add(desc.Digest, p)
res = append(res, &solver.Remote{
Descriptors: append(pRemote.Descriptors, desc),
Provider: provider,
})
}
return
}
func getAvailableBlobs(ctx context.Context, cs content.Store, chain *solver.Remote) ([]*solver.Remote, error) {
if len(chain.Descriptors) == 0 {
return nil, nil
}
target, parentChain := chain.Descriptors[len(chain.Descriptors)-1], chain.Descriptors[:len(chain.Descriptors)-1]
parents, err := getAvailableBlobs(ctx, cs, &solver.Remote{
Descriptors: parentChain,
Provider: chain.Provider,
})
if err != nil {
return nil, err
}
var descs []ocispecs.Descriptor
if err := walkBlob(ctx, cs, target, func(desc ocispecs.Descriptor) bool {
descs = append(descs, desc)
return true
}); err != nil {
bklog.G(ctx).WithError(err).Warn("failed to walk variant blob") // is not a critical error at this moment.
}
var res []*solver.Remote
for _, desc := range descs {
desc := desc
if len(parents) == 0 { // bottommost ref
res = append(res, &solver.Remote{
Descriptors: []ocispecs.Descriptor{desc},
Provider: cs,
})
continue
}
res = append(res, appendRemote(parents, desc, cs)...)
}
if len(res) == 0 {
// no available compression blobs for this blob. return the original blob.
if len(parents) == 0 { // bottommost ref
return []*solver.Remote{chain}, nil
}
return appendRemote(parents, target, chain.Provider), nil
}
return res, nil
}
func (sr *immutableRef) getRemote(ctx context.Context, createIfNeeded bool, refCfg config.RefConfig, s session.Group) (*solver.Remote, error) {
err := sr.computeBlobChain(ctx, createIfNeeded, refCfg.Compression, s)
if err != nil {
return nil, err
}
chain := sr.layerChain()
mproviderBase := contentutil.NewMultiProvider(nil)
mprovider := &lazyMultiProvider{mprovider: mproviderBase}
remote := &solver.Remote{
Provider: mprovider,
}
for _, ref := range chain {
desc, err := ref.ociDesc(ctx, sr.descHandlers, refCfg.PreferNonDistributable)
if err != nil {
return nil, err
}
// NOTE: The media type might be missing for some migrated ones
// from before lease based storage. If so, we should detect
// the media type from blob data.
//
// Discussion: https://github.com/moby/buildkit/pull/1277#discussion_r352795429
if desc.MediaType == "" {
desc.MediaType, err = compression.DetectLayerMediaType(ctx, sr.cm.ContentStore, desc.Digest, false)
if err != nil {
return nil, err
}
}
// update distribution source annotation for lazy-refs (non-lazy refs
// will already have their dsl stored in the content store, which is
// used by the push handlers)
var addAnnotations []string
isLazy, err := ref.isLazy(ctx)
if err != nil {
return nil, err
} else if isLazy {
imageRefs := ref.getImageRefs()
for _, imageRef := range imageRefs {
refspec, err := reference.Parse(imageRef)
if err != nil {
return nil, err
}
u, err := url.Parse("dummy://" + refspec.Locator)
if err != nil {
return nil, err
}
source, repo := u.Hostname(), strings.TrimPrefix(u.Path, "/")
if desc.Annotations == nil {
desc.Annotations = make(map[string]string)
}
dslKey := fmt.Sprintf("%s.%s", "containerd.io/distribution.source", source)
var existingRepos []string
if existings, ok := desc.Annotations[dslKey]; ok {
existingRepos = strings.Split(existings, ",")
}
addNewRepo := true
for _, existing := range existingRepos {
if existing == repo {
addNewRepo = false
break
}
}
if addNewRepo {
existingRepos = append(existingRepos, repo)
}
desc.Annotations[dslKey] = strings.Join(existingRepos, ",")
addAnnotations = append(addAnnotations, dslKey)
}
}
if needsForceCompression(ctx, sr.cm.ContentStore, desc, refCfg) {
if needs, err := refCfg.Compression.Type.NeedsConversion(ctx, sr.cm.ContentStore, desc); err != nil {
return nil, err
} else if needs {
// ensure the compression type.
// compressed blob must be created and stored in the content store.
blobDesc, err := getBlobWithCompressionWithRetry(ctx, ref, refCfg.Compression, s)
if err != nil {
return nil, errors.Wrapf(err, "failed to get compression blob %q", refCfg.Compression.Type)
}
newDesc := desc
newDesc.MediaType = blobDesc.MediaType
newDesc.Digest = blobDesc.Digest
newDesc.Size = blobDesc.Size
newDesc.URLs = blobDesc.URLs
newDesc.Annotations = nil
for _, k := range addAnnotations {
newDesc.Annotations[k] = desc.Annotations[k]
}
for k, v := range blobDesc.Annotations {
if newDesc.Annotations == nil {
newDesc.Annotations = make(map[string]string)
}
newDesc.Annotations[k] = v
}
desc = newDesc
}
}
remote.Descriptors = append(remote.Descriptors, desc)
mprovider.Add(lazyRefProvider{
ref: ref,
desc: desc,
dh: sr.descHandlers[desc.Digest],
session: s,
})
}
return remote, nil
}
func getBlobWithCompressionWithRetry(ctx context.Context, ref *immutableRef, comp compression.Config, s session.Group) (ocispecs.Descriptor, error) {
if blobDesc, err := ref.getBlobWithCompression(ctx, comp.Type); err == nil {
return blobDesc, nil
}
if err := ensureCompression(ctx, ref, comp, s); err != nil {
return ocispecs.Descriptor{}, errors.Wrapf(err, "failed to get and ensure compression type of %q", comp.Type)
}
return ref.getBlobWithCompression(ctx, comp.Type)
}
type lazyMultiProvider struct {
mprovider *contentutil.MultiProvider
plist []lazyRefProvider
}
func (mp *lazyMultiProvider) Add(p lazyRefProvider) {
mp.mprovider.Add(p.desc.Digest, p)
mp.plist = append(mp.plist, p)
}
func (mp *lazyMultiProvider) ReaderAt(ctx context.Context, desc ocispecs.Descriptor) (content.ReaderAt, error) {
return mp.mprovider.ReaderAt(ctx, desc)
}
func (mp *lazyMultiProvider) Unlazy(ctx context.Context) error {
eg, egctx := errgroup.WithContext(ctx)
for _, p := range mp.plist {
p := p
eg.Go(func() error {
return p.Unlazy(egctx)
})
}
return eg.Wait()
}
type lazyRefProvider struct {
ref *immutableRef
desc ocispecs.Descriptor
dh *DescHandler
session session.Group
}
func (p lazyRefProvider) ReaderAt(ctx context.Context, desc ocispecs.Descriptor) (content.ReaderAt, error) {
if desc.Digest != p.desc.Digest {
return nil, errdefs.ErrNotFound
}
if err := p.Unlazy(ctx); err != nil {
return nil, err
}
return p.ref.cm.ContentStore.ReaderAt(ctx, desc)
}
func (p lazyRefProvider) Unlazy(ctx context.Context) error {
_, err := p.ref.cm.unlazyG.Do(ctx, string(p.desc.Digest), func(ctx context.Context) (_ interface{}, rerr error) {
if isLazy, err := p.ref.isLazy(ctx); err != nil {
return nil, err
} else if !isLazy {
return nil, nil
}
defer func() {
if rerr == nil {
rerr = p.ref.linkBlob(ctx, p.desc)
}
}()
if p.dh == nil {
// shouldn't happen, if you have a lazy immutable ref it already should be validated
// that descriptor handlers exist for it
return nil, errors.New("unexpected nil descriptor handler")
}
if p.dh.Progress != nil {
var stopProgress func(error)
ctx, stopProgress = p.dh.Progress.Start(ctx)
defer stopProgress(rerr)
}
// For now, just pull down the whole content and then return a ReaderAt from the local content
// store. If efficient partial reads are desired in the future, something more like a "tee"
// that caches remote partial reads to a local store may need to replace this.
err := contentutil.Copy(ctx, p.ref.cm.ContentStore, &pullprogress.ProviderWithProgress{
Provider: p.dh.Provider(p.session),
Manager: p.ref.cm.ContentStore,
}, p.desc, p.dh.Ref, logs.LoggerFromContext(ctx))
if err != nil {
return nil, err
}
if imageRefs := p.ref.getImageRefs(); len(imageRefs) > 0 {
// just use the first image ref, it's arbitrary
imageRef := imageRefs[0]
if p.ref.GetDescription() == "" {
if err := p.ref.SetDescription("pulled from " + imageRef); err != nil {
return nil, err
}
}
}
return nil, nil
})
return err
}