frostfs-s3-gw/api/layer/listing.go

package layer

import (
	"context"
	"errors"
	"fmt"
	"io"
	"sort"
	"strings"
	"sync"

	"git.frostfs.info/TrueCloudLab/frostfs-s3-gw/api/cache"
	"git.frostfs.info/TrueCloudLab/frostfs-s3-gw/api/data"
	s3errors "git.frostfs.info/TrueCloudLab/frostfs-s3-gw/api/errors"
	"git.frostfs.info/TrueCloudLab/frostfs-s3-gw/api/middleware"
	"git.frostfs.info/TrueCloudLab/frostfs-s3-gw/internal/logs"
	"github.com/panjf2000/ants/v2"
	"go.uber.org/zap"
)

type (
	// ListObjectsParamsCommon contains common parameters for ListObjectsV1 and ListObjectsV2.
	ListObjectsParamsCommon struct {
		BktInfo   *data.BucketInfo
		Delimiter string
		Encode    string
		MaxKeys   int
		Prefix    string
	}

	// ListObjectsParamsV1 contains params for ListObjectsV1.
	ListObjectsParamsV1 struct {
		ListObjectsParamsCommon
		Marker string
	}

	// ListObjectsParamsV2 contains params for ListObjectsV2.
	ListObjectsParamsV2 struct {
		ListObjectsParamsCommon
		ContinuationToken string
		StartAfter        string
		FetchOwner        bool
	}

	// ListObjectsInfo contains common fields of data for ListObjectsV1 and ListObjectsV2.
	ListObjectsInfo struct {
		Prefixes    []string
		Objects     []*data.NodeVersion
		IsTruncated bool
	}

	// ListObjectsInfoV1 holds data which ListObjectsV1 returns.
	ListObjectsInfoV1 struct {
		ListObjectsInfo
		NextMarker string
	}

	// ListObjectsInfoV2 holds data which ListObjectsV2 returns.
	ListObjectsInfoV2 struct {
		ListObjectsInfo
		NextContinuationToken string
	}

	// ListObjectVersionsInfo stores info and list of objects versions.
	ListObjectVersionsInfo struct {
		CommonPrefixes      []string
		IsTruncated         bool
		KeyMarker           string
		NextKeyMarker       string
		NextVersionIDMarker string
		Version             []*data.ExtendedNodeVersion
		DeleteMarker        []*data.ExtendedNodeVersion
		VersionIDMarker     string
	}

	allObjectParams struct {
		Bucket            *data.BucketInfo
		Delimiter         string
		Prefix            string
		MaxKeys           int
		Marker            string
		ContinuationToken string
	}
)

// ListObjectsV1 returns objects in a bucket for requests of Version 1.
func (n *layer) ListObjectsV1(ctx context.Context, p *ListObjectsParamsV1) (*ListObjectsInfoV1, error) {
	var result ListObjectsInfoV1

	prm := allObjectParams{
		Bucket:    p.BktInfo,
		Delimiter: p.Delimiter,
		Prefix:    p.Prefix,
		MaxKeys:   p.MaxKeys,
		Marker:    p.Marker,
	}

	objects, next, err := n.getLatestObjectsVersions(ctx, prm)
	if err != nil {
		return nil, err
	}

	if next != nil {
		result.IsTruncated = true
		result.NextMarker = objects[len(objects)-1].FilePath
	}

	result.Prefixes, result.Objects = triageObjects(objects, p.Prefix, p.Delimiter)

	return &result, nil
}

// ListObjectsV2 returns objects in a bucket for requests of Version 2.
func (n *layer) ListObjectsV2(ctx context.Context, p *ListObjectsParamsV2) (*ListObjectsInfoV2, error) {
	var result ListObjectsInfoV2

	prm := allObjectParams{
		Bucket:            p.BktInfo,
		Delimiter:         p.Delimiter,
		Prefix:            p.Prefix,
		MaxKeys:           p.MaxKeys,
		Marker:            p.StartAfter,
		ContinuationToken: p.ContinuationToken,
	}

	objects, next, err := n.getLatestObjectsVersions(ctx, prm)
	if err != nil {
		return nil, err
	}

	if next != nil {
		result.IsTruncated = true
		result.NextContinuationToken = next.OID.EncodeToString()
	}

	result.Prefixes, result.Objects = triageObjects(objects, p.Prefix, p.Delimiter)

	return &result, nil
}

func (n *layer) ListObjectVersions(ctx context.Context, p *ListObjectVersionsParams) (*ListObjectVersionsInfo, error) {
	versions, err := n.getAllObjectsVersions(ctx, p)
	if err != nil {
		return nil, err
	}

	sortedNames := make([]string, 0, len(versions))
	for k := range versions {
		sortedNames = append(sortedNames, k)
	}
	sort.Strings(sortedNames)

	allObjects := make([]*data.ExtendedNodeVersion, 0, p.MaxKeys)

	for _, name := range sortedNames {
		sortedVersions := versions[name]
		sort.Slice(sortedVersions, func(i, j int) bool {
			return sortedVersions[j].NodeVersion.Timestamp < sortedVersions[i].NodeVersion.Timestamp // sort in reverse order
		})

		for i, version := range sortedVersions {
			version.IsLatest = i == 0
			allObjects = append(allObjects, version)
		}
	}

	if allObjects, err = filterVersionsByMarker(allObjects, p); err != nil {
		return nil, err
	}

	res := &ListObjectVersionsInfo{
		KeyMarker:       p.KeyMarker,
		VersionIDMarker: p.VersionIDMarker,
	}

	res.CommonPrefixes, allObjects = triageExtendedObjects(allObjects, p.Prefix, p.Delimiter)

	if len(allObjects) > p.MaxKeys {
		res.IsTruncated = true
		res.NextKeyMarker = allObjects[p.MaxKeys-1].NodeVersion.FilePath
		res.NextVersionIDMarker = allObjects[p.MaxKeys-1].NodeVersion.OID.EncodeToString()

		allObjects = allObjects[:p.MaxKeys]
	}

	res.Version, res.DeleteMarker = triageVersions(allObjects)
	return res, nil
}

func (n *layer) getLatestObjectsVersions(ctx context.Context, p allObjectParams) (objects []*data.NodeVersion, next *data.NodeVersion, err error) {
	if p.MaxKeys == 0 {
		return nil, nil, nil
	}

	owner := n.BearerOwner(ctx)
	cacheKey := cache.CreateListSessionCacheKey(p.Bucket.CID, p.Prefix, p.ContinuationToken)
	session := n.cache.GetListSession(owner, cacheKey)
	if session != nil {
		// after reading next object from stream in session
		// the current cache value already doesn't match with next token in cache key
		n.cache.DeleteListSession(owner, cacheKey)
	} else {
		session = &data.ListSession{NamesMap: make(map[string]struct{})}
		session.Context, session.Cancel = context.WithCancel(context.Background())

		if bd, err := middleware.GetBoxData(ctx); err == nil {
			session.Context = middleware.SetBoxData(session.Context, bd)
		}

		session.Stream, err = n.treeService.GetLatestVersionsByPrefixStream(session.Context, p.Bucket, p.Prefix)
		if err != nil {
			return nil, nil, err
		}
	}

	poolCtx, cancel := context.WithCancel(ctx)
	defer cancel()

	generator, errorCh := nodesGeneratorStream(poolCtx, p, session)
	objOutCh, err := n.initWorkerPoolStream(poolCtx, 2, p, generator)
	if err != nil {
		return nil, nil, fmt.Errorf("failed to init worker pool: %w", err)
	}

	objects = make([]*data.NodeVersion, 0, p.MaxKeys+1)
	if session.Next != nil {
		objects = append(objects, session.Next)
	}

	for obj := range objOutCh {
		objects = append(objects, obj)
	}

	if err = <-errorCh; err != nil {
		return nil, nil, fmt.Errorf("failed to get next object from stream: %w", err)
	}

	// probably isn't necessary
	sort.Slice(objects, func(i, j int) bool {
		return objects[i].FilePath < objects[j].FilePath
	})

	if len(objects) > p.MaxKeys {
		next = objects[p.MaxKeys]
		objects = objects[:p.MaxKeys]
	}

	if next != nil {
		session.Next = next
		n.cache.PutListSession(owner, cache.CreateListSessionCacheKey(p.Bucket.CID, p.Prefix, next.OID.EncodeToString()), session)
	}

	return
}

func (n *layer) getAllObjectsVersions(ctx context.Context, p *ListObjectVersionsParams) (map[string][]*data.ExtendedNodeVersion, error) {
	nodeVersions, err := n.bucketNodeVersions(ctx, p.BktInfo, p.Prefix)
	if err != nil {
		return nil, err
	}

	versions := make(map[string][]*data.ExtendedNodeVersion, len(nodeVersions))

	sort.Slice(nodeVersions, func(i, j int) bool {
		return nodeVersions[i].FilePath < nodeVersions[j].FilePath
	})

	poolCtx, cancel := context.WithCancel(ctx)
	defer cancel()

	pp := allObjectParams{
		Bucket:            p.BktInfo,
		Delimiter:         p.Delimiter,
		Prefix:            p.Prefix,
		Marker:            p.KeyMarker,
		ContinuationToken: p.VersionIDMarker,
		MaxKeys:           p.MaxKeys,
	}

	objOutCh, err := n.initWorkerPoolVersions(poolCtx, 2, pp, nodesGeneratorVersions(poolCtx, pp, nodeVersions))
	if err != nil {
		return nil, err
	}

	for eoi := range objOutCh {
		objVersions, ok := versions[eoi.NodeVersion.FilePath]
		if !ok {
			objVersions = []*data.ExtendedNodeVersion{eoi}
		} else if dirName := tryDirectoryName(eoi.NodeVersion, p.Prefix, p.Delimiter); dirName == "" {
			objVersions = append(objVersions, eoi)
		} else {
			versions[dirName] = objVersions
		}
	}

	return versions, nil
}

func nodesGeneratorStream(ctx context.Context, p allObjectParams, stream *data.ListSession) (<-chan *data.NodeVersion, <-chan error) {
	nodeCh := make(chan *data.NodeVersion, 1000)
	errCh := make(chan error, 1)
	//existed := make(map[string]struct{}, p.MaxKeys) // to squash the same directories
	existed := stream.NamesMap

	if stream.Next != nil {
		existed[continuationToken] = struct{}{}
	}

	limit := p.MaxKeys
	if stream.Next == nil {
		limit++
	}

	go func() {
		var generated int
		var err error

	LOOP:
		for err == nil {
			node, err := stream.Stream.Next(ctx)
			if err != nil {
				if !errors.Is(err, io.EOF) {
					errCh <- fmt.Errorf("stream next: %w", err)
				}
				break LOOP
			}

			if shouldSkip(node, p, existed) {
				continue
			}

			select {
			case <-ctx.Done():
				break LOOP
			case nodeCh <- node:
				generated++

				if generated == limit { // we use maxKeys+1 to be able to know nextMarker/nextContinuationToken
					break LOOP
				}
			}
		}
		close(nodeCh)
		close(errCh)
	}()

	return nodeCh, errCh
}
func nodesGeneratorVersions(ctx context.Context, p allObjectParams, nodeVersions []*data.NodeVersion) <-chan *data.NodeVersion {
	nodeCh := make(chan *data.NodeVersion)
	existed := make(map[string]struct{}, len(nodeVersions)) // to squash the same directories

	go func() {
		var generated int
	LOOP:
		for _, node := range nodeVersions {
			if shouldSkipVersions(node, p, existed) {
				continue
			}

			select {
			case <-ctx.Done():
				break LOOP
			case nodeCh <- node:
				generated++
				if generated == p.MaxKeys+1 { // we use maxKeys+1 to be able to know nextMarker/nextContinuationToken
					break LOOP
				}
			}
		}
		close(nodeCh)
	}()

	return nodeCh
}

func (n *layer) initWorkerPoolStream(ctx context.Context, size int, p allObjectParams, input <-chan *data.NodeVersion) (<-chan *data.NodeVersion, error) {
	reqLog := n.reqLogger(ctx)
	pool, err := ants.NewPool(size, ants.WithLogger(&logWrapper{reqLog}))
	if err != nil {
		return nil, fmt.Errorf("coudln't init go pool for listing: %w", err)
	}
	objCh := make(chan *data.NodeVersion, size)

	go func() {
		var wg sync.WaitGroup

	LOOP:
		for node := range input {
			select {
			case <-ctx.Done():
				break LOOP
			default:
			}

			if node.IsFilledExtra() || tryDirectoryName(node, p.Prefix, p.Delimiter) != "" { // todo think to not compute twice
				select {
				case <-ctx.Done():
				case objCh <- node:
				}
			} else {
				// We have to make a copy of pointer to data.NodeVersion
				// to get correct value in submitted task function.
				func(node *data.NodeVersion) {
					wg.Add(1)
					err = pool.Submit(func() {
						defer wg.Done()
						oi := n.objectInfoFromObjectsCacheOrFrostFS(ctx, p.Bucket, node)
						if oi == nil {
							// try to get object again
							if oi = n.objectInfoFromObjectsCacheOrFrostFS(ctx, p.Bucket, node); oi == nil {
								// do not process object which are definitely missing in object service
								return
							}
						}

						node.FillExtra(oi)

						select {
						case <-ctx.Done():
						case objCh <- node:
						}
					})
					if err != nil {
						wg.Done()
						reqLog.Warn(logs.FailedToSubmitTaskToPool, zap.Error(err))
					}
				}(node)
			}
		}
		wg.Wait()
		close(objCh)
		pool.Release()
	}()

	return objCh, nil
}

func (n *layer) initWorkerPoolVersions(ctx context.Context, size int, p allObjectParams, input <-chan *data.NodeVersion) (<-chan *data.ExtendedNodeVersion, error) {
	reqLog := n.reqLogger(ctx)
	pool, err := ants.NewPool(size, ants.WithLogger(&logWrapper{reqLog}))
	if err != nil {
		return nil, fmt.Errorf("coudln't init go pool for listing: %w", err)
	}
	objCh := make(chan *data.ExtendedNodeVersion)

	go func() {
		var wg sync.WaitGroup

	LOOP:
		for node := range input {
			select {
			case <-ctx.Done():
				break LOOP
			default:
			}

			if node.IsFilledExtra() {
				select {
				case <-ctx.Done():
				case objCh <- &data.ExtendedNodeVersion{NodeVersion: node}:
				}
			} else {
				// We have to make a copy of pointer to data.NodeVersion
				// to get correct value in submitted task function.
				func(node *data.NodeVersion) {
					wg.Add(1)
					err = pool.Submit(func() {
						defer wg.Done()

						oi := n.objectInfoFromObjectsCacheOrFrostFS(ctx, p.Bucket, node)
						if oi == nil {
							// try to get object again
							if oi = n.objectInfoFromObjectsCacheOrFrostFS(ctx, p.Bucket, node); oi == nil {
								// do not process object which are definitely missing in object service
								return
							}
						}

						node.FillExtra(oi)

						select {
						case <-ctx.Done():
						case objCh <- &data.ExtendedNodeVersion{NodeVersion: node}:
						}
					})
					if err != nil {
						wg.Done()
						reqLog.Warn(logs.FailedToSubmitTaskToPool, zap.Error(err))
					}
				}(node)
			}
		}
		wg.Wait()
		close(objCh)
		pool.Release()
	}()

	return objCh, nil
}

func (n *layer) bucketNodeVersions(ctx context.Context, bkt *data.BucketInfo, prefix string) ([]*data.NodeVersion, error) {
	var err error

	owner := n.BearerOwner(ctx)
	cacheKey := cache.CreateObjectsListCacheKey(bkt.CID, prefix, false)
	nodeVersions := n.cache.GetList(owner, cacheKey)

	if nodeVersions == nil {
		nodeVersions, err = n.treeService.GetAllVersionsByPrefix(ctx, bkt, prefix)
		if err != nil {
			return nil, fmt.Errorf("get all versions from tree service: %w", err)
		}

		n.cache.PutList(owner, cacheKey, nodeVersions)
	}

	return nodeVersions, nil
}

func shouldSkip(node *data.NodeVersion, p allObjectParams, existed map[string]struct{}) bool {
	if node.IsDeleteMarker {
		return true
	}

	filePath := node.FilePath
	if dirName := tryDirectoryName(node, p.Prefix, p.Delimiter); len(dirName) != 0 {
		filePath = dirName
	}
	if _, ok := existed[filePath]; ok {
		return true
	}

	if filePath <= p.Marker {
		return true
	}

	if p.ContinuationToken != "" {
		if _, ok := existed[continuationToken]; !ok {
			if p.ContinuationToken != node.OID.EncodeToString() {
				return true
			}
			existed[continuationToken] = struct{}{}
		}
	}

	existed[filePath] = struct{}{}
	return false
}

func shouldSkipVersions(node *data.NodeVersion, p allObjectParams, existed map[string]struct{}) bool {
	filePath := node.FilePath
	if dirName := tryDirectoryName(node, p.Prefix, p.Delimiter); len(dirName) != 0 {
		filePath = dirName
		if _, ok := existed[filePath]; ok {
			return true
		}
	}

	if filePath < p.Marker {
		return true
	}

	if p.ContinuationToken != "" {
		if _, ok := existed[continuationToken]; !ok {
			if p.ContinuationToken != node.OID.EncodeToString() {
				return true
			}
			existed[continuationToken] = struct{}{}
		}
	}

	existed[filePath] = struct{}{}
	return false
}

func triageObjects(allObjects []*data.NodeVersion, prefix, delimiter string) (prefixes []string, objects []*data.NodeVersion) {
	objects = make([]*data.NodeVersion, 0, len(allObjects))
	for _, ov := range allObjects {
		if dirName := tryDirectoryName(ov, prefix, delimiter); dirName != "" {
			prefixes = append(prefixes, dirName)
		} else {
			objects = append(objects, ov)
		}
	}

	return
}

func triageExtendedObjects(allObjects []*data.ExtendedNodeVersion, prefix, delimiter string) (prefixes []string, objects []*data.ExtendedNodeVersion) {
	for _, ov := range allObjects {
		if dirName := tryDirectoryName(ov.NodeVersion, prefix, delimiter); dirName != "" {
			prefixes = append(prefixes, dirName)
		} else {
			objects = append(objects, ov)
		}
	}

	return
}

func (n *layer) objectInfoFromObjectsCacheOrFrostFS(ctx context.Context, bktInfo *data.BucketInfo, node *data.NodeVersion) (oi *data.ObjectInfo) {
	owner := n.BearerOwner(ctx)
	if extInfo := n.cache.GetObject(owner, newAddress(bktInfo.CID, node.OID)); extInfo != nil {
		return extInfo.ObjectInfo
	}

	meta, err := n.objectHead(ctx, bktInfo, node.OID)
	if err != nil {
		n.reqLogger(ctx).Warn(logs.CouldNotFetchObjectMeta, zap.Error(err))
		return nil
	}

	oi = objectInfoFromMeta(bktInfo, meta)
	oi.MD5Sum = node.MD5
	n.cache.PutObject(owner, &data.ExtendedObjectInfo{ObjectInfo: oi, NodeVersion: node})

	return oi
}

// tryDirectoryName forms directory name by prefix and delimiter.
// If node isn't a directory empty string is returned.
// This function doesn't check if node has a prefix. It must do a caller.
func tryDirectoryName(node *data.NodeVersion, prefix, delimiter string) string {
	if len(delimiter) == 0 {
		return ""
	}

	tail := strings.TrimPrefix(node.FilePath, prefix)
	index := strings.Index(tail, delimiter)
	if index >= 0 {
		return prefix + tail[:index+1]
	}

	return ""
}

func filterVersionsByMarker(objects []*data.ExtendedNodeVersion, p *ListObjectVersionsParams) ([]*data.ExtendedNodeVersion, error) {
	if p.KeyMarker == "" {
		return objects, nil
	}

	for i, obj := range objects {
		if obj.NodeVersion.FilePath == p.KeyMarker {
			for j := i; j < len(objects); j++ {
				if objects[j].NodeVersion.FilePath != obj.NodeVersion.FilePath {
					if p.VersionIDMarker == "" {
						return objects[j:], nil
					}
					break
				}
				if objects[j].NodeVersion.OID.EncodeToString() == p.VersionIDMarker {
					return objects[j+1:], nil
				}
			}
			return nil, s3errors.GetAPIError(s3errors.ErrInvalidVersion)
		} else if obj.NodeVersion.FilePath > p.KeyMarker {
			if p.VersionIDMarker != "" {
				return nil, s3errors.GetAPIError(s3errors.ErrInvalidVersion)
			}
			return objects[i:], nil
		}
	}

	// don't use nil as empty slice to be consistent with `return objects[j+1:], nil` above
	// that can be empty
	return []*data.ExtendedNodeVersion{}, nil
}

func triageVersions(objVersions []*data.ExtendedNodeVersion) ([]*data.ExtendedNodeVersion, []*data.ExtendedNodeVersion) {
	if len(objVersions) == 0 {
		return nil, nil
	}

	var resVersion []*data.ExtendedNodeVersion
	var resDelMarkVersions []*data.ExtendedNodeVersion

	for _, version := range objVersions {
		if version.NodeVersion.IsDeleteMarker {
			resDelMarkVersions = append(resDelMarkVersions, version)
		} else {
			resVersion = append(resVersion, version)
		}
	}

	return resVersion, resDelMarkVersions
}