minio/cmd/gateway/hdfs/gateway-hdfs.go

886 lines
27 KiB
Go
Raw Normal View History

/*
* MinIO Object Storage (c) 2021 MinIO, Inc.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package hdfs
import (
"context"
2019-09-11 13:21:43 -04:00
"errors"
"fmt"
"io"
"net"
"net/http"
"os"
"os/user"
"path"
"sort"
"strings"
"sync"
"syscall"
"time"
"github.com/colinmarc/hdfs/v2"
"github.com/colinmarc/hdfs/v2/hadoopconf"
krb "github.com/jcmturner/gokrb5/v8/client"
"github.com/jcmturner/gokrb5/v8/config"
"github.com/jcmturner/gokrb5/v8/credentials"
2021-02-07 20:29:47 -05:00
"github.com/jcmturner/gokrb5/v8/keytab"
"github.com/minio/cli"
"github.com/minio/madmin-go"
"github.com/minio/minio-go/v7/pkg/s3utils"
minio "github.com/minio/minio/cmd"
"github.com/minio/minio/internal/logger"
"github.com/minio/pkg/env"
xnet "github.com/minio/pkg/net"
)
const (
hdfsSeparator = minio.SlashSeparator
)
func init() {
const hdfsGatewayTemplate = `NAME:
{{.HelpName}} - {{.Usage}}
USAGE:
{{.HelpName}} {{if .VisibleFlags}}[FLAGS]{{end}} HDFS-NAMENODE [HDFS-NAMENODE...]
{{if .VisibleFlags}}
FLAGS:
{{range .VisibleFlags}}{{.}}
{{end}}{{end}}
HDFS-NAMENODE:
HDFS namenode URI
EXAMPLES:
1. Start minio gateway server for HDFS backend
{{.Prompt}} {{.EnvVarSetCommand}} MINIO_ROOT_USER{{.AssignmentOperator}}accesskey
{{.Prompt}} {{.EnvVarSetCommand}} MINIO_ROOT_PASSWORD{{.AssignmentOperator}}secretkey
{{.Prompt}} {{.HelpName}} hdfs://namenode:8200
2. Start minio gateway server for HDFS with edge caching enabled
{{.Prompt}} {{.EnvVarSetCommand}} MINIO_ROOT_USER{{.AssignmentOperator}}accesskey
{{.Prompt}} {{.EnvVarSetCommand}} MINIO_ROOT_PASSWORD{{.AssignmentOperator}}secretkey
{{.Prompt}} {{.EnvVarSetCommand}} MINIO_CACHE_DRIVES{{.AssignmentOperator}}"/mnt/drive1,/mnt/drive2,/mnt/drive3,/mnt/drive4"
{{.Prompt}} {{.EnvVarSetCommand}} MINIO_CACHE_EXCLUDE{{.AssignmentOperator}}"bucket1/*,*.png"
{{.Prompt}} {{.EnvVarSetCommand}} MINIO_CACHE_QUOTA{{.AssignmentOperator}}90
{{.Prompt}} {{.EnvVarSetCommand}} MINIO_CACHE_AFTER{{.AssignmentOperator}}3
{{.Prompt}} {{.EnvVarSetCommand}} MINIO_CACHE_WATERMARK_LOW{{.AssignmentOperator}}75
{{.Prompt}} {{.EnvVarSetCommand}} MINIO_CACHE_WATERMARK_HIGH{{.AssignmentOperator}}85
{{.Prompt}} {{.HelpName}} hdfs://namenode:8200
`
minio.RegisterGatewayCommand(cli.Command{
Name: minio.HDFSBackendGateway,
Usage: "Hadoop Distributed File System (HDFS)",
Action: hdfsGatewayMain,
CustomHelpTemplate: hdfsGatewayTemplate,
HideHelpCommand: true,
})
}
// Handler for 'minio gateway hdfs' command line.
func hdfsGatewayMain(ctx *cli.Context) {
// Validate gateway arguments.
if ctx.Args().First() == "help" {
cli.ShowCommandHelpAndExit(ctx, minio.HDFSBackendGateway, 1)
}
minio.StartGateway(ctx, &HDFS{args: ctx.Args()})
}
// HDFS implements Gateway.
type HDFS struct {
args []string
}
// Name implements Gateway interface.
func (g *HDFS) Name() string {
return minio.HDFSBackendGateway
}
func getKerberosClient() (*krb.Client, error) {
cfg, err := config.Load(env.Get("KRB5_CONFIG", "/etc/krb5.conf"))
if err != nil {
return nil, err
}
u, err := user.Current()
if err != nil {
return nil, err
}
2021-02-07 20:29:47 -05:00
keytabPath := env.Get("KRB5KEYTAB", "")
if keytabPath != "" {
kt, err := keytab.Load(keytabPath)
if err != nil {
return nil, err
}
username := env.Get("KRB5USERNAME", "")
realm := env.Get("KRB5REALM", "")
if username == "" || realm == "" {
return nil, errors.New("empty KRB5USERNAME or KRB5REALM")
}
return krb.NewWithKeytab(username, realm, kt, cfg), nil
}
// Determine the ccache location from the environment, falling back to the default location.
ccachePath := env.Get("KRB5CCNAME", fmt.Sprintf("/tmp/krb5cc_%s", u.Uid))
if strings.Contains(ccachePath, ":") {
if strings.HasPrefix(ccachePath, "FILE:") {
ccachePath = strings.TrimPrefix(ccachePath, "FILE:")
} else {
return nil, fmt.Errorf("unable to use kerberos ccache: %s", ccachePath)
}
}
ccache, err := credentials.LoadCCache(ccachePath)
if err != nil {
return nil, err
}
return krb.NewFromCCache(ccache, cfg)
}
// NewGatewayLayer returns hdfs gatewaylayer.
func (g *HDFS) NewGatewayLayer(creds madmin.Credentials) (minio.ObjectLayer, error) {
dialFunc := (&net.Dialer{
Timeout: 30 * time.Second,
KeepAlive: 30 * time.Second,
DualStack: true,
}).DialContext
hconfig, err := hadoopconf.LoadFromEnvironment()
if err != nil {
return nil, err
}
opts := hdfs.ClientOptionsFromConf(hconfig)
opts.NamenodeDialFunc = dialFunc
opts.DatanodeDialFunc = dialFunc
// Not addresses found, load it from command line.
var commonPath string
if len(opts.Addresses) == 0 {
var addresses []string
for _, s := range g.args {
u, err := xnet.ParseURL(s)
if err != nil {
return nil, err
}
if u.Scheme != "hdfs" {
return nil, fmt.Errorf("unsupported scheme %s, only supports hdfs://", u)
}
if commonPath != "" && commonPath != u.Path {
return nil, fmt.Errorf("all namenode paths should be same %s", g.args)
}
if commonPath == "" {
commonPath = u.Path
}
addresses = append(addresses, u.Host)
}
opts.Addresses = addresses
}
u, err := user.Current()
if err != nil {
return nil, fmt.Errorf("unable to lookup local user: %s", err)
}
if opts.KerberosClient != nil {
opts.KerberosClient, err = getKerberosClient()
if err != nil {
return nil, fmt.Errorf("unable to initialize kerberos client: %s", err)
}
} else {
opts.User = env.Get("HADOOP_USER_NAME", u.Username)
}
clnt, err := hdfs.NewClient(opts)
if err != nil {
2021-02-07 20:29:47 -05:00
return nil, fmt.Errorf("unable to initialize hdfsClient: %v", err)
}
if err = clnt.MkdirAll(minio.PathJoin(commonPath, hdfsSeparator, minioMetaTmpBucket), os.FileMode(0o755)); err != nil {
return nil, err
}
return &hdfsObjects{clnt: clnt, subPath: commonPath, listPool: minio.NewTreeWalkPool(time.Minute * 30)}, nil
}
func (n *hdfsObjects) Shutdown(ctx context.Context) error {
return n.clnt.Close()
}
func (n *hdfsObjects) LocalStorageInfo(ctx context.Context) (si minio.StorageInfo, errs []error) {
return n.StorageInfo(ctx)
}
func (n *hdfsObjects) StorageInfo(ctx context.Context) (si minio.StorageInfo, errs []error) {
fsInfo, err := n.clnt.StatFs()
if err != nil {
return minio.StorageInfo{}, []error{err}
}
si.Disks = []madmin.Disk{{
UsedSpace: fsInfo.Used,
}}
si.Backend.Type = madmin.Gateway
si.Backend.GatewayOnline = true
return si, nil
}
// hdfsObjects implements gateway for Minio and S3 compatible object storage servers.
type hdfsObjects struct {
minio.GatewayUnsupported
clnt *hdfs.Client
subPath string
listPool *minio.TreeWalkPool
}
func hdfsToObjectErr(ctx context.Context, err error, params ...string) error {
if err == nil {
return nil
}
bucket := ""
object := ""
uploadID := ""
switch len(params) {
case 3:
uploadID = params[2]
fallthrough
case 2:
object = params[1]
fallthrough
case 1:
bucket = params[0]
}
switch {
case os.IsNotExist(err):
if uploadID != "" {
return minio.InvalidUploadID{
UploadID: uploadID,
}
}
if object != "" {
return minio.ObjectNotFound{Bucket: bucket, Object: object}
}
return minio.BucketNotFound{Bucket: bucket}
case os.IsExist(err):
if object != "" {
return minio.PrefixAccessDenied{Bucket: bucket, Object: object}
}
return minio.BucketAlreadyOwnedByYou{Bucket: bucket}
2019-09-11 13:21:43 -04:00
case errors.Is(err, syscall.ENOTEMPTY):
if object != "" {
return minio.PrefixAccessDenied{Bucket: bucket, Object: object}
}
return minio.BucketNotEmpty{Bucket: bucket}
default:
logger.LogIf(ctx, err)
return err
}
}
// hdfsIsValidBucketName verifies whether a bucket name is valid.
func hdfsIsValidBucketName(bucket string) bool {
return s3utils.CheckValidBucketNameStrict(bucket) == nil
}
func (n *hdfsObjects) hdfsPathJoin(args ...string) string {
return minio.PathJoin(append([]string{n.subPath, hdfsSeparator}, args...)...)
}
func (n *hdfsObjects) DeleteBucket(ctx context.Context, bucket string, opts minio.DeleteBucketOptions) error {
if !hdfsIsValidBucketName(bucket) {
return minio.BucketNameInvalid{Bucket: bucket}
}
if opts.Force {
return hdfsToObjectErr(ctx, n.clnt.RemoveAll(n.hdfsPathJoin(bucket)), bucket)
}
return hdfsToObjectErr(ctx, n.clnt.Remove(n.hdfsPathJoin(bucket)), bucket)
}
func (n *hdfsObjects) MakeBucketWithLocation(ctx context.Context, bucket string, opts minio.BucketOptions) error {
if opts.LockEnabled || opts.VersioningEnabled {
return minio.NotImplemented{}
}
if !hdfsIsValidBucketName(bucket) {
return minio.BucketNameInvalid{Bucket: bucket}
}
return hdfsToObjectErr(ctx, n.clnt.Mkdir(n.hdfsPathJoin(bucket), os.FileMode(0o755)), bucket)
}
func (n *hdfsObjects) GetBucketInfo(ctx context.Context, bucket string) (bi minio.BucketInfo, err error) {
fi, err := n.clnt.Stat(n.hdfsPathJoin(bucket))
if err != nil {
return bi, hdfsToObjectErr(ctx, err, bucket)
}
// As hdfs.Stat() doesn't carry anything other than ModTime(), use ModTime() as CreatedTime.
return minio.BucketInfo{
Name: bucket,
Created: fi.ModTime(),
}, nil
}
func (n *hdfsObjects) ListBuckets(ctx context.Context) (buckets []minio.BucketInfo, err error) {
entries, err := n.clnt.ReadDir(n.hdfsPathJoin())
if err != nil {
logger.LogIf(ctx, err)
return nil, hdfsToObjectErr(ctx, err)
}
for _, entry := range entries {
// Ignore all reserved bucket names and invalid bucket names.
if isReservedOrInvalidBucket(entry.Name(), false) {
continue
}
buckets = append(buckets, minio.BucketInfo{
Name: entry.Name(),
// As hdfs.Stat() doesnt carry CreatedTime, use ModTime() as CreatedTime.
Created: entry.ModTime(),
})
}
// Sort bucket infos by bucket name.
sort.Sort(byBucketName(buckets))
return buckets, nil
}
func (n *hdfsObjects) isLeafDir(bucket, leafPath string) bool {
return n.isObjectDir(context.Background(), bucket, leafPath)
}
func (n *hdfsObjects) isLeaf(bucket, leafPath string) bool {
return !strings.HasSuffix(leafPath, hdfsSeparator)
}
func (n *hdfsObjects) listDirFactory() minio.ListDirFunc {
// listDir - lists all the entries at a given prefix and given entry in the prefix.
listDir := func(bucket, prefixDir, prefixEntry string) (emptyDir bool, entries []string, delayIsLeaf bool) {
f, err := n.clnt.Open(n.hdfsPathJoin(bucket, prefixDir))
if err != nil {
if os.IsNotExist(err) {
err = nil
}
logger.LogIf(minio.GlobalContext, err)
return
}
defer f.Close()
fis, err := f.Readdir(0)
if err != nil {
logger.LogIf(minio.GlobalContext, err)
return
}
if len(fis) == 0 {
return true, nil, false
}
for _, fi := range fis {
if fi.IsDir() {
entries = append(entries, fi.Name()+hdfsSeparator)
} else {
entries = append(entries, fi.Name())
}
}
entries, delayIsLeaf = minio.FilterListEntries(bucket, prefixDir, entries, prefixEntry, n.isLeaf)
return false, entries, delayIsLeaf
}
// Return list factory instance.
return listDir
}
// ListObjects lists all blobs in HDFS bucket filtered by prefix.
func (n *hdfsObjects) ListObjects(ctx context.Context, bucket, prefix, marker, delimiter string, maxKeys int) (loi minio.ListObjectsInfo, err error) {
var mutex sync.Mutex
Use hdfs.Readdir() to optimize HDFS directory listings (#10121) Currently, listing directories on HDFS incurs a per-entry remote Stat() call penalty, the cost of which can really blow up on directories with many entries (+1,000) especially when considered in addition to peripheral calls (such as validation) and the fact that minio is an intermediary to the client (whereas other clients listed below can query HDFS directly). Because listing directories this way is expensive, the Golang HDFS library provides the [`Client.Open()`] function which creates a [`FileReader`] that is able to batch multiple calls together through the [`Readdir()`] function. This is substantially more efficient for very large directories. In one case we were witnessing about +20 seconds to list a directory with 1,500 entries, admittedly large, but the Java hdfs ls utility as well as the HDFS library sample ls utility were much faster. Hadoop HDFS DFS (4.02s): λ ~/code/minio → use-readdir » time hdfs dfs -ls /directory/with/1500/entries/ … hdfs dfs -ls 5.81s user 0.49s system 156% cpu 4.020 total Golang HDFS library (0.47s): λ ~/code/hdfs → master » time ./hdfs ls -lh /directory/with/1500/entries/ … ./hdfs ls -lh 0.13s user 0.14s system 56% cpu 0.478 total mc and minio **without** optimization (16.96s): λ ~/code/minio → master » time mc ls myhdfs/directory/with/1500/entries/ … ./mc ls 0.22s user 0.29s system 3% cpu 16.968 total mc and minio **with** optimization (0.40s): λ ~/code/minio → use-readdir » time mc ls myhdfs/directory/with/1500/entries/ … ./mc ls 0.13s user 0.28s system 102% cpu 0.403 total [`Client.Open()`]: https://godoc.org/github.com/colinmarc/hdfs#Client.Open [`FileReader`]: https://godoc.org/github.com/colinmarc/hdfs#FileReader [`Readdir()`]: https://godoc.org/github.com/colinmarc/hdfs#FileReader.Readdir
2020-07-24 14:31:51 -04:00
fileInfos := make(map[string]os.FileInfo)
targetPath := n.hdfsPathJoin(bucket, prefix)
Use hdfs.Readdir() to optimize HDFS directory listings (#10121) Currently, listing directories on HDFS incurs a per-entry remote Stat() call penalty, the cost of which can really blow up on directories with many entries (+1,000) especially when considered in addition to peripheral calls (such as validation) and the fact that minio is an intermediary to the client (whereas other clients listed below can query HDFS directly). Because listing directories this way is expensive, the Golang HDFS library provides the [`Client.Open()`] function which creates a [`FileReader`] that is able to batch multiple calls together through the [`Readdir()`] function. This is substantially more efficient for very large directories. In one case we were witnessing about +20 seconds to list a directory with 1,500 entries, admittedly large, but the Java hdfs ls utility as well as the HDFS library sample ls utility were much faster. Hadoop HDFS DFS (4.02s): λ ~/code/minio → use-readdir » time hdfs dfs -ls /directory/with/1500/entries/ … hdfs dfs -ls 5.81s user 0.49s system 156% cpu 4.020 total Golang HDFS library (0.47s): λ ~/code/hdfs → master » time ./hdfs ls -lh /directory/with/1500/entries/ … ./hdfs ls -lh 0.13s user 0.14s system 56% cpu 0.478 total mc and minio **without** optimization (16.96s): λ ~/code/minio → master » time mc ls myhdfs/directory/with/1500/entries/ … ./mc ls 0.22s user 0.29s system 3% cpu 16.968 total mc and minio **with** optimization (0.40s): λ ~/code/minio → use-readdir » time mc ls myhdfs/directory/with/1500/entries/ … ./mc ls 0.13s user 0.28s system 102% cpu 0.403 total [`Client.Open()`]: https://godoc.org/github.com/colinmarc/hdfs#Client.Open [`FileReader`]: https://godoc.org/github.com/colinmarc/hdfs#FileReader [`Readdir()`]: https://godoc.org/github.com/colinmarc/hdfs#FileReader.Readdir
2020-07-24 14:31:51 -04:00
var targetFileInfo os.FileInfo
if targetFileInfo, err = n.populateDirectoryListing(targetPath, fileInfos); err != nil {
return loi, hdfsToObjectErr(ctx, err, bucket)
}
// If the user is trying to list a single file, bypass the entire directory-walking code below
// and just return the single file's information.
if !targetFileInfo.IsDir() {
return minio.ListObjectsInfo{
IsTruncated: false,
NextMarker: "",
Objects: []minio.ObjectInfo{
fileInfoToObjectInfo(bucket, prefix, targetFileInfo),
},
Prefixes: []string{},
}, nil
}
getObjectInfo := func(ctx context.Context, bucket, entry string) (minio.ObjectInfo, error) {
mutex.Lock()
defer mutex.Unlock()
Use hdfs.Readdir() to optimize HDFS directory listings (#10121) Currently, listing directories on HDFS incurs a per-entry remote Stat() call penalty, the cost of which can really blow up on directories with many entries (+1,000) especially when considered in addition to peripheral calls (such as validation) and the fact that minio is an intermediary to the client (whereas other clients listed below can query HDFS directly). Because listing directories this way is expensive, the Golang HDFS library provides the [`Client.Open()`] function which creates a [`FileReader`] that is able to batch multiple calls together through the [`Readdir()`] function. This is substantially more efficient for very large directories. In one case we were witnessing about +20 seconds to list a directory with 1,500 entries, admittedly large, but the Java hdfs ls utility as well as the HDFS library sample ls utility were much faster. Hadoop HDFS DFS (4.02s): λ ~/code/minio → use-readdir » time hdfs dfs -ls /directory/with/1500/entries/ … hdfs dfs -ls 5.81s user 0.49s system 156% cpu 4.020 total Golang HDFS library (0.47s): λ ~/code/hdfs → master » time ./hdfs ls -lh /directory/with/1500/entries/ … ./hdfs ls -lh 0.13s user 0.14s system 56% cpu 0.478 total mc and minio **without** optimization (16.96s): λ ~/code/minio → master » time mc ls myhdfs/directory/with/1500/entries/ … ./mc ls 0.22s user 0.29s system 3% cpu 16.968 total mc and minio **with** optimization (0.40s): λ ~/code/minio → use-readdir » time mc ls myhdfs/directory/with/1500/entries/ … ./mc ls 0.13s user 0.28s system 102% cpu 0.403 total [`Client.Open()`]: https://godoc.org/github.com/colinmarc/hdfs#Client.Open [`FileReader`]: https://godoc.org/github.com/colinmarc/hdfs#FileReader [`Readdir()`]: https://godoc.org/github.com/colinmarc/hdfs#FileReader.Readdir
2020-07-24 14:31:51 -04:00
filePath := path.Clean(n.hdfsPathJoin(bucket, entry))
fi, ok := fileInfos[filePath]
// If the file info is not known, this may be a recursive listing and filePath is a
// child of a sub-directory. In this case, obtain that sub-directory's listing.
if !ok {
parentPath := path.Dir(filePath)
if _, err := n.populateDirectoryListing(parentPath, fileInfos); err != nil {
Use hdfs.Readdir() to optimize HDFS directory listings (#10121) Currently, listing directories on HDFS incurs a per-entry remote Stat() call penalty, the cost of which can really blow up on directories with many entries (+1,000) especially when considered in addition to peripheral calls (such as validation) and the fact that minio is an intermediary to the client (whereas other clients listed below can query HDFS directly). Because listing directories this way is expensive, the Golang HDFS library provides the [`Client.Open()`] function which creates a [`FileReader`] that is able to batch multiple calls together through the [`Readdir()`] function. This is substantially more efficient for very large directories. In one case we were witnessing about +20 seconds to list a directory with 1,500 entries, admittedly large, but the Java hdfs ls utility as well as the HDFS library sample ls utility were much faster. Hadoop HDFS DFS (4.02s): λ ~/code/minio → use-readdir » time hdfs dfs -ls /directory/with/1500/entries/ … hdfs dfs -ls 5.81s user 0.49s system 156% cpu 4.020 total Golang HDFS library (0.47s): λ ~/code/hdfs → master » time ./hdfs ls -lh /directory/with/1500/entries/ … ./hdfs ls -lh 0.13s user 0.14s system 56% cpu 0.478 total mc and minio **without** optimization (16.96s): λ ~/code/minio → master » time mc ls myhdfs/directory/with/1500/entries/ … ./mc ls 0.22s user 0.29s system 3% cpu 16.968 total mc and minio **with** optimization (0.40s): λ ~/code/minio → use-readdir » time mc ls myhdfs/directory/with/1500/entries/ … ./mc ls 0.13s user 0.28s system 102% cpu 0.403 total [`Client.Open()`]: https://godoc.org/github.com/colinmarc/hdfs#Client.Open [`FileReader`]: https://godoc.org/github.com/colinmarc/hdfs#FileReader [`Readdir()`]: https://godoc.org/github.com/colinmarc/hdfs#FileReader.Readdir
2020-07-24 14:31:51 -04:00
return minio.ObjectInfo{}, hdfsToObjectErr(ctx, err, bucket)
}
fi, ok = fileInfos[filePath]
if !ok {
err = fmt.Errorf("could not get FileInfo for path '%s'", filePath)
return minio.ObjectInfo{}, hdfsToObjectErr(ctx, err, bucket, entry)
}
}
Use hdfs.Readdir() to optimize HDFS directory listings (#10121) Currently, listing directories on HDFS incurs a per-entry remote Stat() call penalty, the cost of which can really blow up on directories with many entries (+1,000) especially when considered in addition to peripheral calls (such as validation) and the fact that minio is an intermediary to the client (whereas other clients listed below can query HDFS directly). Because listing directories this way is expensive, the Golang HDFS library provides the [`Client.Open()`] function which creates a [`FileReader`] that is able to batch multiple calls together through the [`Readdir()`] function. This is substantially more efficient for very large directories. In one case we were witnessing about +20 seconds to list a directory with 1,500 entries, admittedly large, but the Java hdfs ls utility as well as the HDFS library sample ls utility were much faster. Hadoop HDFS DFS (4.02s): λ ~/code/minio → use-readdir » time hdfs dfs -ls /directory/with/1500/entries/ … hdfs dfs -ls 5.81s user 0.49s system 156% cpu 4.020 total Golang HDFS library (0.47s): λ ~/code/hdfs → master » time ./hdfs ls -lh /directory/with/1500/entries/ … ./hdfs ls -lh 0.13s user 0.14s system 56% cpu 0.478 total mc and minio **without** optimization (16.96s): λ ~/code/minio → master » time mc ls myhdfs/directory/with/1500/entries/ … ./mc ls 0.22s user 0.29s system 3% cpu 16.968 total mc and minio **with** optimization (0.40s): λ ~/code/minio → use-readdir » time mc ls myhdfs/directory/with/1500/entries/ … ./mc ls 0.13s user 0.28s system 102% cpu 0.403 total [`Client.Open()`]: https://godoc.org/github.com/colinmarc/hdfs#Client.Open [`FileReader`]: https://godoc.org/github.com/colinmarc/hdfs#FileReader [`Readdir()`]: https://godoc.org/github.com/colinmarc/hdfs#FileReader.Readdir
2020-07-24 14:31:51 -04:00
objectInfo := fileInfoToObjectInfo(bucket, entry, fi)
Use hdfs.Readdir() to optimize HDFS directory listings (#10121) Currently, listing directories on HDFS incurs a per-entry remote Stat() call penalty, the cost of which can really blow up on directories with many entries (+1,000) especially when considered in addition to peripheral calls (such as validation) and the fact that minio is an intermediary to the client (whereas other clients listed below can query HDFS directly). Because listing directories this way is expensive, the Golang HDFS library provides the [`Client.Open()`] function which creates a [`FileReader`] that is able to batch multiple calls together through the [`Readdir()`] function. This is substantially more efficient for very large directories. In one case we were witnessing about +20 seconds to list a directory with 1,500 entries, admittedly large, but the Java hdfs ls utility as well as the HDFS library sample ls utility were much faster. Hadoop HDFS DFS (4.02s): λ ~/code/minio → use-readdir » time hdfs dfs -ls /directory/with/1500/entries/ … hdfs dfs -ls 5.81s user 0.49s system 156% cpu 4.020 total Golang HDFS library (0.47s): λ ~/code/hdfs → master » time ./hdfs ls -lh /directory/with/1500/entries/ … ./hdfs ls -lh 0.13s user 0.14s system 56% cpu 0.478 total mc and minio **without** optimization (16.96s): λ ~/code/minio → master » time mc ls myhdfs/directory/with/1500/entries/ … ./mc ls 0.22s user 0.29s system 3% cpu 16.968 total mc and minio **with** optimization (0.40s): λ ~/code/minio → use-readdir » time mc ls myhdfs/directory/with/1500/entries/ … ./mc ls 0.13s user 0.28s system 102% cpu 0.403 total [`Client.Open()`]: https://godoc.org/github.com/colinmarc/hdfs#Client.Open [`FileReader`]: https://godoc.org/github.com/colinmarc/hdfs#FileReader [`Readdir()`]: https://godoc.org/github.com/colinmarc/hdfs#FileReader.Readdir
2020-07-24 14:31:51 -04:00
delete(fileInfos, filePath)
return objectInfo, nil
}
return minio.ListObjects(ctx, n, bucket, prefix, marker, delimiter, maxKeys, n.listPool, n.listDirFactory(), n.isLeaf, n.isLeafDir, getObjectInfo, getObjectInfo)
}
func fileInfoToObjectInfo(bucket string, entry string, fi os.FileInfo) minio.ObjectInfo {
return minio.ObjectInfo{
Bucket: bucket,
Name: entry,
ModTime: fi.ModTime(),
Size: fi.Size(),
IsDir: fi.IsDir(),
AccTime: fi.(*hdfs.FileInfo).AccessTime(),
}
}
Use hdfs.Readdir() to optimize HDFS directory listings (#10121) Currently, listing directories on HDFS incurs a per-entry remote Stat() call penalty, the cost of which can really blow up on directories with many entries (+1,000) especially when considered in addition to peripheral calls (such as validation) and the fact that minio is an intermediary to the client (whereas other clients listed below can query HDFS directly). Because listing directories this way is expensive, the Golang HDFS library provides the [`Client.Open()`] function which creates a [`FileReader`] that is able to batch multiple calls together through the [`Readdir()`] function. This is substantially more efficient for very large directories. In one case we were witnessing about +20 seconds to list a directory with 1,500 entries, admittedly large, but the Java hdfs ls utility as well as the HDFS library sample ls utility were much faster. Hadoop HDFS DFS (4.02s): λ ~/code/minio → use-readdir » time hdfs dfs -ls /directory/with/1500/entries/ … hdfs dfs -ls 5.81s user 0.49s system 156% cpu 4.020 total Golang HDFS library (0.47s): λ ~/code/hdfs → master » time ./hdfs ls -lh /directory/with/1500/entries/ … ./hdfs ls -lh 0.13s user 0.14s system 56% cpu 0.478 total mc and minio **without** optimization (16.96s): λ ~/code/minio → master » time mc ls myhdfs/directory/with/1500/entries/ … ./mc ls 0.22s user 0.29s system 3% cpu 16.968 total mc and minio **with** optimization (0.40s): λ ~/code/minio → use-readdir » time mc ls myhdfs/directory/with/1500/entries/ … ./mc ls 0.13s user 0.28s system 102% cpu 0.403 total [`Client.Open()`]: https://godoc.org/github.com/colinmarc/hdfs#Client.Open [`FileReader`]: https://godoc.org/github.com/colinmarc/hdfs#FileReader [`Readdir()`]: https://godoc.org/github.com/colinmarc/hdfs#FileReader.Readdir
2020-07-24 14:31:51 -04:00
// Lists a path's direct, first-level entries and populates them in the `fileInfos` cache which maps
// a path entry to an `os.FileInfo`. It also saves the listed path's `os.FileInfo` in the cache.
func (n *hdfsObjects) populateDirectoryListing(filePath string, fileInfos map[string]os.FileInfo) (os.FileInfo, error) {
Use hdfs.Readdir() to optimize HDFS directory listings (#10121) Currently, listing directories on HDFS incurs a per-entry remote Stat() call penalty, the cost of which can really blow up on directories with many entries (+1,000) especially when considered in addition to peripheral calls (such as validation) and the fact that minio is an intermediary to the client (whereas other clients listed below can query HDFS directly). Because listing directories this way is expensive, the Golang HDFS library provides the [`Client.Open()`] function which creates a [`FileReader`] that is able to batch multiple calls together through the [`Readdir()`] function. This is substantially more efficient for very large directories. In one case we were witnessing about +20 seconds to list a directory with 1,500 entries, admittedly large, but the Java hdfs ls utility as well as the HDFS library sample ls utility were much faster. Hadoop HDFS DFS (4.02s): λ ~/code/minio → use-readdir » time hdfs dfs -ls /directory/with/1500/entries/ … hdfs dfs -ls 5.81s user 0.49s system 156% cpu 4.020 total Golang HDFS library (0.47s): λ ~/code/hdfs → master » time ./hdfs ls -lh /directory/with/1500/entries/ … ./hdfs ls -lh 0.13s user 0.14s system 56% cpu 0.478 total mc and minio **without** optimization (16.96s): λ ~/code/minio → master » time mc ls myhdfs/directory/with/1500/entries/ … ./mc ls 0.22s user 0.29s system 3% cpu 16.968 total mc and minio **with** optimization (0.40s): λ ~/code/minio → use-readdir » time mc ls myhdfs/directory/with/1500/entries/ … ./mc ls 0.13s user 0.28s system 102% cpu 0.403 total [`Client.Open()`]: https://godoc.org/github.com/colinmarc/hdfs#Client.Open [`FileReader`]: https://godoc.org/github.com/colinmarc/hdfs#FileReader [`Readdir()`]: https://godoc.org/github.com/colinmarc/hdfs#FileReader.Readdir
2020-07-24 14:31:51 -04:00
dirReader, err := n.clnt.Open(filePath)
if err != nil {
return nil, err
Use hdfs.Readdir() to optimize HDFS directory listings (#10121) Currently, listing directories on HDFS incurs a per-entry remote Stat() call penalty, the cost of which can really blow up on directories with many entries (+1,000) especially when considered in addition to peripheral calls (such as validation) and the fact that minio is an intermediary to the client (whereas other clients listed below can query HDFS directly). Because listing directories this way is expensive, the Golang HDFS library provides the [`Client.Open()`] function which creates a [`FileReader`] that is able to batch multiple calls together through the [`Readdir()`] function. This is substantially more efficient for very large directories. In one case we were witnessing about +20 seconds to list a directory with 1,500 entries, admittedly large, but the Java hdfs ls utility as well as the HDFS library sample ls utility were much faster. Hadoop HDFS DFS (4.02s): λ ~/code/minio → use-readdir » time hdfs dfs -ls /directory/with/1500/entries/ … hdfs dfs -ls 5.81s user 0.49s system 156% cpu 4.020 total Golang HDFS library (0.47s): λ ~/code/hdfs → master » time ./hdfs ls -lh /directory/with/1500/entries/ … ./hdfs ls -lh 0.13s user 0.14s system 56% cpu 0.478 total mc and minio **without** optimization (16.96s): λ ~/code/minio → master » time mc ls myhdfs/directory/with/1500/entries/ … ./mc ls 0.22s user 0.29s system 3% cpu 16.968 total mc and minio **with** optimization (0.40s): λ ~/code/minio → use-readdir » time mc ls myhdfs/directory/with/1500/entries/ … ./mc ls 0.13s user 0.28s system 102% cpu 0.403 total [`Client.Open()`]: https://godoc.org/github.com/colinmarc/hdfs#Client.Open [`FileReader`]: https://godoc.org/github.com/colinmarc/hdfs#FileReader [`Readdir()`]: https://godoc.org/github.com/colinmarc/hdfs#FileReader.Readdir
2020-07-24 14:31:51 -04:00
}
dirStat := dirReader.Stat()
key := path.Clean(filePath)
if !dirStat.IsDir() {
return dirStat, nil
}
Use hdfs.Readdir() to optimize HDFS directory listings (#10121) Currently, listing directories on HDFS incurs a per-entry remote Stat() call penalty, the cost of which can really blow up on directories with many entries (+1,000) especially when considered in addition to peripheral calls (such as validation) and the fact that minio is an intermediary to the client (whereas other clients listed below can query HDFS directly). Because listing directories this way is expensive, the Golang HDFS library provides the [`Client.Open()`] function which creates a [`FileReader`] that is able to batch multiple calls together through the [`Readdir()`] function. This is substantially more efficient for very large directories. In one case we were witnessing about +20 seconds to list a directory with 1,500 entries, admittedly large, but the Java hdfs ls utility as well as the HDFS library sample ls utility were much faster. Hadoop HDFS DFS (4.02s): λ ~/code/minio → use-readdir » time hdfs dfs -ls /directory/with/1500/entries/ … hdfs dfs -ls 5.81s user 0.49s system 156% cpu 4.020 total Golang HDFS library (0.47s): λ ~/code/hdfs → master » time ./hdfs ls -lh /directory/with/1500/entries/ … ./hdfs ls -lh 0.13s user 0.14s system 56% cpu 0.478 total mc and minio **without** optimization (16.96s): λ ~/code/minio → master » time mc ls myhdfs/directory/with/1500/entries/ … ./mc ls 0.22s user 0.29s system 3% cpu 16.968 total mc and minio **with** optimization (0.40s): λ ~/code/minio → use-readdir » time mc ls myhdfs/directory/with/1500/entries/ … ./mc ls 0.13s user 0.28s system 102% cpu 0.403 total [`Client.Open()`]: https://godoc.org/github.com/colinmarc/hdfs#Client.Open [`FileReader`]: https://godoc.org/github.com/colinmarc/hdfs#FileReader [`Readdir()`]: https://godoc.org/github.com/colinmarc/hdfs#FileReader.Readdir
2020-07-24 14:31:51 -04:00
fileInfos[key] = dirStat
infos, err := dirReader.Readdir(0)
if err != nil {
return nil, err
Use hdfs.Readdir() to optimize HDFS directory listings (#10121) Currently, listing directories on HDFS incurs a per-entry remote Stat() call penalty, the cost of which can really blow up on directories with many entries (+1,000) especially when considered in addition to peripheral calls (such as validation) and the fact that minio is an intermediary to the client (whereas other clients listed below can query HDFS directly). Because listing directories this way is expensive, the Golang HDFS library provides the [`Client.Open()`] function which creates a [`FileReader`] that is able to batch multiple calls together through the [`Readdir()`] function. This is substantially more efficient for very large directories. In one case we were witnessing about +20 seconds to list a directory with 1,500 entries, admittedly large, but the Java hdfs ls utility as well as the HDFS library sample ls utility were much faster. Hadoop HDFS DFS (4.02s): λ ~/code/minio → use-readdir » time hdfs dfs -ls /directory/with/1500/entries/ … hdfs dfs -ls 5.81s user 0.49s system 156% cpu 4.020 total Golang HDFS library (0.47s): λ ~/code/hdfs → master » time ./hdfs ls -lh /directory/with/1500/entries/ … ./hdfs ls -lh 0.13s user 0.14s system 56% cpu 0.478 total mc and minio **without** optimization (16.96s): λ ~/code/minio → master » time mc ls myhdfs/directory/with/1500/entries/ … ./mc ls 0.22s user 0.29s system 3% cpu 16.968 total mc and minio **with** optimization (0.40s): λ ~/code/minio → use-readdir » time mc ls myhdfs/directory/with/1500/entries/ … ./mc ls 0.13s user 0.28s system 102% cpu 0.403 total [`Client.Open()`]: https://godoc.org/github.com/colinmarc/hdfs#Client.Open [`FileReader`]: https://godoc.org/github.com/colinmarc/hdfs#FileReader [`Readdir()`]: https://godoc.org/github.com/colinmarc/hdfs#FileReader.Readdir
2020-07-24 14:31:51 -04:00
}
for _, fileInfo := range infos {
filePath := minio.PathJoin(filePath, fileInfo.Name())
Use hdfs.Readdir() to optimize HDFS directory listings (#10121) Currently, listing directories on HDFS incurs a per-entry remote Stat() call penalty, the cost of which can really blow up on directories with many entries (+1,000) especially when considered in addition to peripheral calls (such as validation) and the fact that minio is an intermediary to the client (whereas other clients listed below can query HDFS directly). Because listing directories this way is expensive, the Golang HDFS library provides the [`Client.Open()`] function which creates a [`FileReader`] that is able to batch multiple calls together through the [`Readdir()`] function. This is substantially more efficient for very large directories. In one case we were witnessing about +20 seconds to list a directory with 1,500 entries, admittedly large, but the Java hdfs ls utility as well as the HDFS library sample ls utility were much faster. Hadoop HDFS DFS (4.02s): λ ~/code/minio → use-readdir » time hdfs dfs -ls /directory/with/1500/entries/ … hdfs dfs -ls 5.81s user 0.49s system 156% cpu 4.020 total Golang HDFS library (0.47s): λ ~/code/hdfs → master » time ./hdfs ls -lh /directory/with/1500/entries/ … ./hdfs ls -lh 0.13s user 0.14s system 56% cpu 0.478 total mc and minio **without** optimization (16.96s): λ ~/code/minio → master » time mc ls myhdfs/directory/with/1500/entries/ … ./mc ls 0.22s user 0.29s system 3% cpu 16.968 total mc and minio **with** optimization (0.40s): λ ~/code/minio → use-readdir » time mc ls myhdfs/directory/with/1500/entries/ … ./mc ls 0.13s user 0.28s system 102% cpu 0.403 total [`Client.Open()`]: https://godoc.org/github.com/colinmarc/hdfs#Client.Open [`FileReader`]: https://godoc.org/github.com/colinmarc/hdfs#FileReader [`Readdir()`]: https://godoc.org/github.com/colinmarc/hdfs#FileReader.Readdir
2020-07-24 14:31:51 -04:00
fileInfos[filePath] = fileInfo
}
return dirStat, nil
Use hdfs.Readdir() to optimize HDFS directory listings (#10121) Currently, listing directories on HDFS incurs a per-entry remote Stat() call penalty, the cost of which can really blow up on directories with many entries (+1,000) especially when considered in addition to peripheral calls (such as validation) and the fact that minio is an intermediary to the client (whereas other clients listed below can query HDFS directly). Because listing directories this way is expensive, the Golang HDFS library provides the [`Client.Open()`] function which creates a [`FileReader`] that is able to batch multiple calls together through the [`Readdir()`] function. This is substantially more efficient for very large directories. In one case we were witnessing about +20 seconds to list a directory with 1,500 entries, admittedly large, but the Java hdfs ls utility as well as the HDFS library sample ls utility were much faster. Hadoop HDFS DFS (4.02s): λ ~/code/minio → use-readdir » time hdfs dfs -ls /directory/with/1500/entries/ … hdfs dfs -ls 5.81s user 0.49s system 156% cpu 4.020 total Golang HDFS library (0.47s): λ ~/code/hdfs → master » time ./hdfs ls -lh /directory/with/1500/entries/ … ./hdfs ls -lh 0.13s user 0.14s system 56% cpu 0.478 total mc and minio **without** optimization (16.96s): λ ~/code/minio → master » time mc ls myhdfs/directory/with/1500/entries/ … ./mc ls 0.22s user 0.29s system 3% cpu 16.968 total mc and minio **with** optimization (0.40s): λ ~/code/minio → use-readdir » time mc ls myhdfs/directory/with/1500/entries/ … ./mc ls 0.13s user 0.28s system 102% cpu 0.403 total [`Client.Open()`]: https://godoc.org/github.com/colinmarc/hdfs#Client.Open [`FileReader`]: https://godoc.org/github.com/colinmarc/hdfs#FileReader [`Readdir()`]: https://godoc.org/github.com/colinmarc/hdfs#FileReader.Readdir
2020-07-24 14:31:51 -04:00
}
// deleteObject deletes a file path if its empty. If it's successfully deleted,
// it will recursively move up the tree, deleting empty parent directories
// until it finds one with files in it. Returns nil for a non-empty directory.
func (n *hdfsObjects) deleteObject(basePath, deletePath string) error {
if basePath == deletePath {
return nil
}
// Attempt to remove path.
if err := n.clnt.Remove(deletePath); err != nil {
2019-09-11 13:21:43 -04:00
if errors.Is(err, syscall.ENOTEMPTY) {
// Ignore errors if the directory is not empty. The server relies on
// this functionality, and sometimes uses recursion that should not
// error on parent directories.
return nil
}
2019-09-11 13:21:43 -04:00
return err
}
// Trailing slash is removed when found to ensure
// slashpath.Dir() to work as intended.
deletePath = strings.TrimSuffix(deletePath, hdfsSeparator)
deletePath = path.Dir(deletePath)
// Delete parent directory. Errors for parent directories shouldn't trickle down.
n.deleteObject(basePath, deletePath)
return nil
}
// ListObjectsV2 lists all blobs in HDFS bucket filtered by prefix
func (n *hdfsObjects) ListObjectsV2(ctx context.Context, bucket, prefix, continuationToken, delimiter string, maxKeys int,
fetchOwner bool, startAfter string) (loi minio.ListObjectsV2Info, err error) {
// fetchOwner is not supported and unused.
marker := continuationToken
if marker == "" {
marker = startAfter
}
resultV1, err := n.ListObjects(ctx, bucket, prefix, marker, delimiter, maxKeys)
if err != nil {
return loi, err
}
return minio.ListObjectsV2Info{
Objects: resultV1.Objects,
Prefixes: resultV1.Prefixes,
ContinuationToken: continuationToken,
NextContinuationToken: resultV1.NextMarker,
IsTruncated: resultV1.IsTruncated,
}, nil
}
func (n *hdfsObjects) DeleteObject(ctx context.Context, bucket, object string, opts minio.ObjectOptions) (minio.ObjectInfo, error) {
err := hdfsToObjectErr(ctx, n.deleteObject(n.hdfsPathJoin(bucket), n.hdfsPathJoin(bucket, object)), bucket, object)
return minio.ObjectInfo{
Bucket: bucket,
Name: object,
}, err
}
func (n *hdfsObjects) DeleteObjects(ctx context.Context, bucket string, objects []minio.ObjectToDelete, opts minio.ObjectOptions) ([]minio.DeletedObject, []error) {
errs := make([]error, len(objects))
dobjects := make([]minio.DeletedObject, len(objects))
for idx, object := range objects {
_, errs[idx] = n.DeleteObject(ctx, bucket, object.ObjectName, opts)
if errs[idx] == nil {
dobjects[idx] = minio.DeletedObject{
ObjectName: object.ObjectName,
}
}
}
return dobjects, errs
}
func (n *hdfsObjects) GetObjectNInfo(ctx context.Context, bucket, object string, rs *minio.HTTPRangeSpec, h http.Header, lockType minio.LockType, opts minio.ObjectOptions) (gr *minio.GetObjectReader, err error) {
objInfo, err := n.GetObjectInfo(ctx, bucket, object, opts)
if err != nil {
return nil, err
}
var startOffset, length int64
startOffset, length, err = rs.GetOffsetLength(objInfo.Size)
if err != nil {
return nil, err
}
pr, pw := io.Pipe()
go func() {
nerr := n.getObject(ctx, bucket, object, startOffset, length, pw, objInfo.ETag, opts)
pw.CloseWithError(nerr)
}()
// Setup cleanup function to cause the above go-routine to
// exit in case of partial read
pipeCloser := func() { pr.Close() }
return minio.NewGetObjectReaderFromReader(pr, objInfo, opts, pipeCloser)
}
func (n *hdfsObjects) CopyObject(ctx context.Context, srcBucket, srcObject, dstBucket, dstObject string, srcInfo minio.ObjectInfo, srcOpts, dstOpts minio.ObjectOptions) (minio.ObjectInfo, error) {
cpSrcDstSame := minio.IsStringEqual(n.hdfsPathJoin(srcBucket, srcObject), n.hdfsPathJoin(dstBucket, dstObject))
if cpSrcDstSame {
return n.GetObjectInfo(ctx, srcBucket, srcObject, minio.ObjectOptions{})
}
return n.PutObject(ctx, dstBucket, dstObject, srcInfo.PutObjReader, minio.ObjectOptions{
ServerSideEncryption: dstOpts.ServerSideEncryption,
UserDefined: srcInfo.UserDefined,
})
}
func (n *hdfsObjects) getObject(ctx context.Context, bucket, key string, startOffset, length int64, writer io.Writer, etag string, opts minio.ObjectOptions) error {
if _, err := n.clnt.Stat(n.hdfsPathJoin(bucket)); err != nil {
return hdfsToObjectErr(ctx, err, bucket)
}
rd, err := n.clnt.Open(n.hdfsPathJoin(bucket, key))
if err != nil {
return hdfsToObjectErr(ctx, err, bucket, key)
}
defer rd.Close()
_, err = io.Copy(writer, io.NewSectionReader(rd, startOffset, length))
if err == io.ErrClosedPipe {
// hdfs library doesn't send EOF correctly, so io.Copy attempts
// to write which returns io.ErrClosedPipe - just ignore
// this for now.
err = nil
}
return hdfsToObjectErr(ctx, err, bucket, key)
}
func (n *hdfsObjects) isObjectDir(ctx context.Context, bucket, object string) bool {
f, err := n.clnt.Open(n.hdfsPathJoin(bucket, object))
if err != nil {
if os.IsNotExist(err) {
return false
}
logger.LogIf(ctx, err)
return false
}
defer f.Close()
fis, err := f.Readdir(1)
if err != nil && err != io.EOF {
logger.LogIf(ctx, err)
return false
}
// Readdir returns an io.EOF when len(fis) == 0.
return len(fis) == 0
}
// GetObjectInfo reads object info and replies back ObjectInfo.
func (n *hdfsObjects) GetObjectInfo(ctx context.Context, bucket, object string, opts minio.ObjectOptions) (objInfo minio.ObjectInfo, err error) {
_, err = n.clnt.Stat(n.hdfsPathJoin(bucket))
if err != nil {
return objInfo, hdfsToObjectErr(ctx, err, bucket)
}
if strings.HasSuffix(object, hdfsSeparator) && !n.isObjectDir(ctx, bucket, object) {
return objInfo, hdfsToObjectErr(ctx, os.ErrNotExist, bucket, object)
}
fi, err := n.clnt.Stat(n.hdfsPathJoin(bucket, object))
if err != nil {
return objInfo, hdfsToObjectErr(ctx, err, bucket, object)
}
return minio.ObjectInfo{
Bucket: bucket,
Name: object,
ModTime: fi.ModTime(),
Size: fi.Size(),
IsDir: fi.IsDir(),
AccTime: fi.(*hdfs.FileInfo).AccessTime(),
}, nil
}
func (n *hdfsObjects) PutObject(ctx context.Context, bucket string, object string, r *minio.PutObjReader, opts minio.ObjectOptions) (objInfo minio.ObjectInfo, err error) {
_, err = n.clnt.Stat(n.hdfsPathJoin(bucket))
if err != nil {
return objInfo, hdfsToObjectErr(ctx, err, bucket)
}
name := n.hdfsPathJoin(bucket, object)
// If its a directory create a prefix {
if strings.HasSuffix(object, hdfsSeparator) && r.Size() == 0 {
if err = n.clnt.MkdirAll(name, os.FileMode(0o755)); err != nil {
n.deleteObject(n.hdfsPathJoin(bucket), name)
return objInfo, hdfsToObjectErr(ctx, err, bucket, object)
}
} else {
tmpname := n.hdfsPathJoin(minioMetaTmpBucket, minio.MustGetUUID())
var w *hdfs.FileWriter
w, err = n.clnt.Create(tmpname)
if err != nil {
return objInfo, hdfsToObjectErr(ctx, err, bucket, object)
}
defer n.deleteObject(n.hdfsPathJoin(minioMetaTmpBucket), tmpname)
if _, err = io.Copy(w, r); err != nil {
w.Close()
return objInfo, hdfsToObjectErr(ctx, err, bucket, object)
}
dir := path.Dir(name)
if dir != "" {
if err = n.clnt.MkdirAll(dir, os.FileMode(0o755)); err != nil {
w.Close()
n.deleteObject(n.hdfsPathJoin(bucket), dir)
return objInfo, hdfsToObjectErr(ctx, err, bucket, object)
}
}
w.Close()
if err = n.clnt.Rename(tmpname, name); err != nil {
return objInfo, hdfsToObjectErr(ctx, err, bucket, object)
}
}
fi, err := n.clnt.Stat(name)
if err != nil {
return objInfo, hdfsToObjectErr(ctx, err, bucket, object)
}
return minio.ObjectInfo{
Bucket: bucket,
Name: object,
ETag: r.MD5CurrentHexString(),
ModTime: fi.ModTime(),
Size: fi.Size(),
IsDir: fi.IsDir(),
AccTime: fi.(*hdfs.FileInfo).AccessTime(),
}, nil
}
func (n *hdfsObjects) NewMultipartUpload(ctx context.Context, bucket string, object string, opts minio.ObjectOptions) (uploadID string, err error) {
_, err = n.clnt.Stat(n.hdfsPathJoin(bucket))
if err != nil {
return uploadID, hdfsToObjectErr(ctx, err, bucket)
}
uploadID = minio.MustGetUUID()
if err = n.clnt.CreateEmptyFile(n.hdfsPathJoin(minioMetaTmpBucket, uploadID)); err != nil {
return uploadID, hdfsToObjectErr(ctx, err, bucket)
}
return uploadID, nil
}
func (n *hdfsObjects) ListMultipartUploads(ctx context.Context, bucket string, prefix string, keyMarker string, uploadIDMarker string, delimiter string, maxUploads int) (lmi minio.ListMultipartsInfo, err error) {
_, err = n.clnt.Stat(n.hdfsPathJoin(bucket))
if err != nil {
return lmi, hdfsToObjectErr(ctx, err, bucket)
}
// It's decided not to support List Multipart Uploads, hence returning empty result.
return lmi, nil
}
func (n *hdfsObjects) checkUploadIDExists(ctx context.Context, bucket, object, uploadID string) (err error) {
_, err = n.clnt.Stat(n.hdfsPathJoin(minioMetaTmpBucket, uploadID))
if err != nil {
return hdfsToObjectErr(ctx, err, bucket, object, uploadID)
}
return nil
}
// GetMultipartInfo returns multipart info of the uploadId of the object
func (n *hdfsObjects) GetMultipartInfo(ctx context.Context, bucket, object, uploadID string, opts minio.ObjectOptions) (result minio.MultipartInfo, err error) {
_, err = n.clnt.Stat(n.hdfsPathJoin(bucket))
if err != nil {
return result, hdfsToObjectErr(ctx, err, bucket)
}
if err = n.checkUploadIDExists(ctx, bucket, object, uploadID); err != nil {
return result, err
}
result.Bucket = bucket
result.Object = object
result.UploadID = uploadID
return result, nil
}
func (n *hdfsObjects) ListObjectParts(ctx context.Context, bucket, object, uploadID string, partNumberMarker int, maxParts int, opts minio.ObjectOptions) (result minio.ListPartsInfo, err error) {
_, err = n.clnt.Stat(n.hdfsPathJoin(bucket))
if err != nil {
return result, hdfsToObjectErr(ctx, err, bucket)
}
if err = n.checkUploadIDExists(ctx, bucket, object, uploadID); err != nil {
return result, err
}
// It's decided not to support List parts, hence returning empty result.
return result, nil
}
func (n *hdfsObjects) CopyObjectPart(ctx context.Context, srcBucket, srcObject, dstBucket, dstObject, uploadID string, partID int,
startOffset int64, length int64, srcInfo minio.ObjectInfo, srcOpts, dstOpts minio.ObjectOptions) (minio.PartInfo, error) {
return n.PutObjectPart(ctx, dstBucket, dstObject, uploadID, partID, srcInfo.PutObjReader, dstOpts)
}
func (n *hdfsObjects) PutObjectPart(ctx context.Context, bucket, object, uploadID string, partID int, r *minio.PutObjReader, opts minio.ObjectOptions) (info minio.PartInfo, err error) {
_, err = n.clnt.Stat(n.hdfsPathJoin(bucket))
if err != nil {
return info, hdfsToObjectErr(ctx, err, bucket)
}
var w *hdfs.FileWriter
w, err = n.clnt.Append(n.hdfsPathJoin(minioMetaTmpBucket, uploadID))
if err != nil {
return info, hdfsToObjectErr(ctx, err, bucket, object, uploadID)
}
defer w.Close()
_, err = io.Copy(w, r.Reader)
if err != nil {
return info, hdfsToObjectErr(ctx, err, bucket, object, uploadID)
}
info.PartNumber = partID
info.ETag = r.MD5CurrentHexString()
info.LastModified = minio.UTCNow()
info.Size = r.Reader.Size()
return info, nil
}
func (n *hdfsObjects) CompleteMultipartUpload(ctx context.Context, bucket, object, uploadID string, parts []minio.CompletePart, opts minio.ObjectOptions) (objInfo minio.ObjectInfo, err error) {
_, err = n.clnt.Stat(n.hdfsPathJoin(bucket))
if err != nil {
return objInfo, hdfsToObjectErr(ctx, err, bucket)
}
if err = n.checkUploadIDExists(ctx, bucket, object, uploadID); err != nil {
return objInfo, err
}
name := n.hdfsPathJoin(bucket, object)
dir := path.Dir(name)
if dir != "" {
if err = n.clnt.MkdirAll(dir, os.FileMode(0o755)); err != nil {
return objInfo, hdfsToObjectErr(ctx, err, bucket, object)
}
}
err = n.clnt.Rename(n.hdfsPathJoin(minioMetaTmpBucket, uploadID), name)
// Object already exists is an error on HDFS
// remove it and then create it again.
if os.IsExist(err) {
if err = n.clnt.Remove(name); err != nil {
if dir != "" {
n.deleteObject(n.hdfsPathJoin(bucket), dir)
}
return objInfo, hdfsToObjectErr(ctx, err, bucket, object)
}
if err = n.clnt.Rename(n.hdfsPathJoin(minioMetaTmpBucket, uploadID), name); err != nil {
if dir != "" {
n.deleteObject(n.hdfsPathJoin(bucket), dir)
}
return objInfo, hdfsToObjectErr(ctx, err, bucket, object)
}
}
fi, err := n.clnt.Stat(name)
if err != nil {
return objInfo, hdfsToObjectErr(ctx, err, bucket, object)
}
// Calculate s3 compatible md5sum for complete multipart.
s3MD5 := minio.ComputeCompleteMultipartMD5(parts)
return minio.ObjectInfo{
Bucket: bucket,
Name: object,
ETag: s3MD5,
ModTime: fi.ModTime(),
Size: fi.Size(),
IsDir: fi.IsDir(),
AccTime: fi.(*hdfs.FileInfo).AccessTime(),
}, nil
}
func (n *hdfsObjects) AbortMultipartUpload(ctx context.Context, bucket, object, uploadID string, opts minio.ObjectOptions) (err error) {
_, err = n.clnt.Stat(n.hdfsPathJoin(bucket))
if err != nil {
return hdfsToObjectErr(ctx, err, bucket)
}
return hdfsToObjectErr(ctx, n.clnt.Remove(n.hdfsPathJoin(minioMetaTmpBucket, uploadID)), bucket, object, uploadID)
}