storj/cmd/uplink/cmd_cp.go
Jeff Wendling 1cbad0fcab cmd/uplink: add back parallelism
for very large machines (>10Gbit) it is still useful
to have parallelism for uploads because we're actually
bound by getting new pieces from the satellite, so doing
that in parallel provides a big win.

this change adds back that flag to exist for uploads, and
removes the backwards compatibility code for the flag with
the maximum-concurrent-pieces as they are now independent.

the upload code parallelism story is now this:

    - each object is a transfer
    - each transfer happens in N parts (size dynamically
      chosen to avoid having >10000 parts)
    - each part can happen in parallel up to the limit
      specified
    - each parallel part can have up to the limit of
      max concurrent pieces and segments

this change also changes some defaults to be better.

    - the connection pool capacity now takes into acount
      transfers, parallelism and max concurrent pieces
    - the default smallest part size is 1GiB to allow the
      new upload code path to upload multiple segments

Change-Id: Iff6709ae73425fbc2858ed360faa2d3ece297c2d
2023-08-14 20:28:58 -04:00

711 lines
18 KiB
Go

// Copyright (C) 2021 Storj Labs, Inc.
// See LICENSE for copying information.
package main
import (
"bufio"
"context"
"errors"
"fmt"
"io"
"os"
"strconv"
"strings"
"sync"
"time"
"github.com/VividCortex/ewma"
"github.com/vbauerster/mpb/v8"
"github.com/vbauerster/mpb/v8/decor"
"github.com/zeebo/clingy"
"github.com/zeebo/errs"
"storj.io/common/context2"
"storj.io/common/fpath"
"storj.io/common/memory"
"storj.io/common/rpc/rpcpool"
"storj.io/common/sync2"
"storj.io/storj/cmd/uplink/ulext"
"storj.io/storj/cmd/uplink/ulfs"
"storj.io/storj/cmd/uplink/ulloc"
"storj.io/uplink/private/testuplink"
)
type cmdCp struct {
ex ulext.External
access string
recursive bool
transfers int
dryrun bool
progress bool
byteRange string
expires time.Time
metadata map[string]string
parallelism int
parallelismChunkSize memory.Size
uploadConfig testuplink.ConcurrentSegmentUploadsConfig
uploadLogFile string
inmemoryEC bool
locs []ulloc.Location
}
const maxPartCount int64 = 10000
func newCmdCp(ex ulext.External) *cmdCp {
return &cmdCp{ex: ex}
}
func (c *cmdCp) Setup(params clingy.Parameters) {
c.access = params.Flag("access", "Access name or value to use", "").(string)
c.recursive = params.Flag("recursive", "Peform a recursive copy", false,
clingy.Short('r'),
clingy.Transform(strconv.ParseBool), clingy.Boolean,
).(bool)
c.transfers = params.Flag("transfers", "Controls how many uploads/downloads to perform in parallel", 1,
clingy.Short('t'),
clingy.Transform(strconv.Atoi),
clingy.Transform(func(n int) (int, error) {
if n <= 0 {
return 0, errs.New("transfers must be at least 1")
}
return n, nil
}),
).(int)
c.dryrun = params.Flag("dry-run", "Print what operations would happen but don't execute them", false,
clingy.Transform(strconv.ParseBool), clingy.Boolean,
).(bool)
c.progress = params.Flag("progress", "Show a progress bar when possible", true,
clingy.Transform(strconv.ParseBool), clingy.Boolean,
).(bool)
c.byteRange = params.Flag("range", "Downloads the specified range bytes of an object. For more information about the HTTP Range header, see https://www.w3.org/Protocols/rfc2616/rfc2616-sec14.html#sec14.35", "").(string)
c.parallelism = params.Flag("parallelism", "Controls how many parallel parts to upload/download from a file", 1,
clingy.Short('p'),
clingy.Transform(strconv.Atoi),
clingy.Transform(func(n int) (int, error) {
if n <= 0 {
return 0, errs.New("parallelism must be at least 1")
}
return n, nil
}),
).(int)
c.parallelismChunkSize = params.Flag("parallelism-chunk-size", "Set the size of the parts for parallelism, 0 means automatic adjustment", memory.Size(0),
clingy.Transform(memory.ParseString),
clingy.Transform(func(n int64) (memory.Size, error) {
if n < 0 {
return 0, errs.New("parallelism-chunk-size cannot be below 0")
}
return memory.Size(n), nil
}),
).(memory.Size)
c.uploadConfig = testuplink.DefaultConcurrentSegmentUploadsConfig()
c.uploadConfig.SchedulerOptions.MaximumConcurrent = params.Flag(
"maximum-concurrent-pieces",
"Maximum concurrent pieces to upload at once per part",
c.uploadConfig.SchedulerOptions.MaximumConcurrent,
clingy.Transform(strconv.Atoi),
clingy.Advanced,
).(int)
c.uploadConfig.SchedulerOptions.MaximumConcurrentHandles = params.Flag(
"maximum-concurrent-segments",
"Maximum concurrent segments to upload at once per part",
c.uploadConfig.SchedulerOptions.MaximumConcurrentHandles,
clingy.Transform(strconv.Atoi),
clingy.Advanced,
).(int)
c.uploadConfig.LongTailMargin = params.Flag(
"long-tail-margin",
"How many extra pieces to upload and cancel per segment",
c.uploadConfig.LongTailMargin,
clingy.Transform(strconv.Atoi),
clingy.Advanced,
).(int)
c.uploadLogFile = params.Flag("upload-log-file", "File to write upload logs to", "",
clingy.Advanced,
).(string)
c.inmemoryEC = params.Flag("inmemory-erasure-coding", "Keep erasure-coded pieces in-memory instead of writing them on the disk during upload", false,
clingy.Transform(strconv.ParseBool),
clingy.Boolean,
clingy.Advanced,
).(bool)
c.expires = params.Flag("expires",
"Schedule removal after this time (e.g. '+2h', 'now', '2020-01-02T15:04:05Z0700')",
time.Time{}, clingy.Transform(parseHumanDate), clingy.Type("relative_date")).(time.Time)
c.metadata = params.Flag("metadata",
"optional metadata for the object. Please use a single level JSON object of string to string only",
nil, clingy.Transform(parseJSON), clingy.Type("string")).(map[string]string)
c.locs = params.Arg("locations", "Locations to copy (at least one source and one destination). Use - for standard input/output",
clingy.Transform(ulloc.Parse),
clingy.Repeated,
).([]ulloc.Location)
}
func (c *cmdCp) Execute(ctx context.Context) error {
if len(c.locs) < 2 {
return errs.New("must have at least one source and destination path")
}
if c.uploadLogFile != "" {
fh, err := os.OpenFile(c.uploadLogFile, os.O_WRONLY|os.O_CREATE|os.O_EXCL, 0644)
if err != nil {
return errs.Wrap(err)
}
defer func() { _ = fh.Close() }()
bw := bufio.NewWriter(fh)
defer func() { _ = bw.Flush() }()
ctx = testuplink.WithLogWriter(ctx, bw)
}
fs, err := c.ex.OpenFilesystem(ctx, c.access,
ulext.ConcurrentSegmentUploadsConfig(c.uploadConfig),
ulext.ConnectionPoolOptions(rpcpool.Options{
// Allow at least as many connections as the maximum concurrent pieces per
// parallel part per transfer, plus a few extra for the satellite.
Capacity: c.transfers*c.parallelism*c.uploadConfig.SchedulerOptions.MaximumConcurrent + 5,
KeyCapacity: 2,
IdleExpiration: 2 * time.Minute,
}))
if err != nil {
return err
}
defer func() { _ = fs.Close() }()
if c.inmemoryEC {
ctx = fpath.WithTempData(ctx, "", true)
}
var eg errs.Group
for _, source := range c.locs[:len(c.locs)-1] {
eg.Add(c.dispatchCopy(ctx, fs, source, c.locs[len(c.locs)-1]))
}
return combineErrs(eg)
}
func (c *cmdCp) dispatchCopy(ctx context.Context, fs ulfs.Filesystem, source, dest ulloc.Location) error {
if !source.Remote() && !dest.Remote() {
return errs.New("at least one location must be a remote sj:// location")
}
// we ensure the source and destination are lexically directoryish
// if they map to directories. the destination is always converted to be
// directoryish if the copy is recursive or if there are multiple source paths
if fs.IsLocalDir(ctx, source) {
source = source.AsDirectoryish()
}
if c.recursive || len(c.locs) > 2 || fs.IsLocalDir(ctx, dest) {
dest = dest.AsDirectoryish()
}
if c.recursive {
if c.byteRange != "" {
return errs.New("unable to do recursive copy with byte range")
}
return c.copyRecursive(ctx, fs, source, dest)
}
// if the destination is directoryish, we add the basename of the source
// to the end of the destination to pick a filename.
var base string
if dest.Directoryish() && !source.Std() {
// we undirectoryish the source so that we ignore any trailing slashes
// when finding the base name.
var ok bool
base, ok = source.Undirectoryish().Base()
if !ok {
return errs.New("destination is a directory and cannot find base name for source %q", source)
}
}
dest = joinDestWith(dest, base)
if !dest.Std() {
fmt.Fprintln(clingy.Stdout(ctx), copyVerb(source, dest), source, "to", dest)
}
var bar *mpb.Bar
if c.progress && !dest.Std() {
progress := mpb.New(mpb.WithOutput(clingy.Stdout(ctx)))
defer progress.Wait()
var namer barNamer
bar = newProgressBar(progress, namer.NameFor(source, dest), 1, 1)
defer func() {
bar.Abort(true)
bar.Wait()
}()
}
return c.copyFile(ctx, fs, source, dest, bar)
}
func (c *cmdCp) copyRecursive(ctx context.Context, fs ulfs.Filesystem, source, dest ulloc.Location) error {
if source.Std() || dest.Std() {
return errs.New("cannot recursively copy to stdin/stdout")
}
iter, err := fs.List(ctx, source, &ulfs.ListOptions{
Recursive: true,
})
if err != nil {
return err
}
var (
limiter = sync2.NewLimiter(c.transfers)
es errs.Group
mu sync.Mutex
)
fprintln := func(w io.Writer, args ...interface{}) {
mu.Lock()
defer mu.Unlock()
fmt.Fprintln(w, args...)
}
fprintf := func(w io.Writer, format string, args ...interface{}) {
mu.Lock()
defer mu.Unlock()
fmt.Fprintf(w, format, args...)
}
addError := func(err error) {
if err == nil {
return
}
mu.Lock()
defer mu.Unlock()
es.Add(err)
}
type copyOp struct {
src ulloc.Location
dest ulloc.Location
}
var verb string
var verbing string
var namer barNamer
var ops []copyOp
for iter.Next() {
item := iter.Item().Loc
rel, err := source.RelativeTo(item)
if err != nil {
return err
}
dest := joinDestWith(dest, rel)
verb = copyVerb(item, dest)
verbing = copyVerbing(item, dest)
namer.Preview(item, dest)
ops = append(ops, copyOp{src: item, dest: dest})
}
if err := iter.Err(); err != nil {
return errs.Wrap(err)
}
fprintln(clingy.Stdout(ctx), verbing, len(ops), "files...")
var progress *mpb.Progress
if c.progress {
progress = mpb.New(mpb.WithOutput(clingy.Stdout(ctx)))
defer progress.Wait()
}
for i, op := range ops {
i := i
op := op
ok := limiter.Go(ctx, func() {
var bar *mpb.Bar
if progress != nil {
bar = newProgressBar(progress, namer.NameFor(op.src, op.dest), i+1, len(ops))
defer func() {
bar.Abort(true)
bar.Wait()
}()
} else {
fprintf(clingy.Stdout(ctx), "%s %s to %s (%d of %d)\n", verb, op.src, op.dest, i+1, len(ops))
}
if err := c.copyFile(ctx, fs, op.src, op.dest, bar); err != nil {
addError(errs.New("%s %s to %s failed: %w", verb, op.src, op.dest, err))
}
})
if !ok {
break
}
}
limiter.Wait()
if progress != nil {
// Wait for all of the progress bar output to complete before doing
// additional output.
progress.Wait()
}
if len(es) > 0 {
for _, e := range es {
fprintln(clingy.Stdout(ctx), e)
}
return errs.New("recursive %s failed (%d of %d)", verb, len(es), len(ops))
}
return nil
}
func (c *cmdCp) copyFile(ctx context.Context, fs ulfs.Filesystem, source, dest ulloc.Location, bar *mpb.Bar) (err error) {
if c.dryrun {
return nil
}
if dest.Remote() && source.Remote() {
return fs.Copy(ctx, source, dest)
}
offset, length, err := parseRange(c.byteRange)
if err != nil {
return errs.Wrap(err)
}
mrh, err := fs.Open(ctx, source)
if err != nil {
return err
}
defer func() { _ = mrh.Close() }()
mwh, err := fs.Create(ctx, dest, &ulfs.CreateOptions{
Expires: c.expires,
Metadata: c.metadata,
})
if err != nil {
return err
}
defer func() { _ = mwh.Abort(ctx) }()
partSize, err := c.calculatePartSize(mrh.Length(), c.parallelismChunkSize.Int64())
if err != nil {
return err
}
return errs.Wrap(c.parallelCopy(
ctx,
source, dest,
mrh, mwh,
c.parallelism, partSize,
offset, length,
bar,
))
}
// calculatePartSize returns the needed part size in order to upload the file with size of 'length'.
// It hereby respects if the client requests/prefers a certain size and only increases if needed.
func (c *cmdCp) calculatePartSize(length, preferredSize int64) (requiredSize int64, err error) {
segC := (length / maxPartCount / memory.GiB.Int64()) + 1
requiredSize = segC * memory.GiB.Int64()
switch {
case preferredSize == 0:
return requiredSize, nil
case requiredSize <= preferredSize:
return preferredSize, nil
case length < 0: // let the user pick their size if we don't have a length to know better
return preferredSize, nil
default:
return 0, errs.New(fmt.Sprintf("the specified chunk size %s is too small, requires %s or larger",
memory.FormatBytes(preferredSize), memory.FormatBytes(requiredSize)))
}
}
func copyVerbing(source, dest ulloc.Location) (verb string) {
return copyVerb(source, dest) + "ing"
}
func copyVerb(source, dest ulloc.Location) (verb string) {
switch {
case dest.Remote():
return "upload"
case source.Remote():
return "download"
default:
return "copy"
}
}
func joinDestWith(dest ulloc.Location, suffix string) ulloc.Location {
dest = dest.AppendKey(suffix)
// if the destination is local and directoryish, remove any
// trailing slashes that it has. this makes it so that if
// a remote file is name "foo/", then we copy it down as
// just "foo".
if dest.Local() && dest.Directoryish() {
dest = dest.Undirectoryish()
}
return dest
}
func (c *cmdCp) parallelCopy(
ctx context.Context,
source, dest ulloc.Location,
src ulfs.MultiReadHandle,
dst ulfs.MultiWriteHandle,
p int, chunkSize int64,
offset, length int64,
bar *mpb.Bar) error {
if offset != 0 {
if err := src.SetOffset(offset); err != nil {
return err
}
}
var (
limiter = sync2.NewLimiter(p)
es errs.Group
mu sync.Mutex
)
ctx, cancel := context.WithCancel(ctx)
defer func() { _ = src.Close() }()
defer func() {
nocancel := context2.WithoutCancellation(ctx)
timedctx, cancel := context.WithTimeout(nocancel, 5*time.Second)
defer cancel()
_ = dst.Abort(timedctx)
}()
defer cancel()
addError := func(err error) {
if err == nil {
return
}
mu.Lock()
defer mu.Unlock()
es.Add(err)
// abort all other concurrenty copies
cancel()
}
var readBufs *ulfs.BytesPool
if p > 1 && chunkSize > 0 && source.Std() {
// Create the read buffer pool only for uploads from stdin with parallelism > 1.
readBufs = ulfs.NewBytesPool(int(chunkSize))
}
for i := 0; length != 0; i++ {
i := i
chunk := chunkSize
if length > 0 && chunkSize > length {
chunk = length
}
length -= chunk
rh, err := src.NextPart(ctx, chunk)
if err != nil {
if !errors.Is(err, io.EOF) {
addError(errs.New("error getting reader for part %d: %v", i, err))
}
break
}
wh, err := dst.NextPart(ctx, chunk)
if err != nil {
_ = rh.Close()
addError(errs.New("error getting writer for part %d: %v", i, err))
break
}
ok := limiter.Go(ctx, func() {
defer func() { _ = rh.Close() }()
defer func() { _ = wh.Abort() }()
if readBufs != nil {
buf := readBufs.Get()
defer readBufs.Put(buf)
rh = ulfs.NewBufferedReadHandle(ctx, rh, buf)
}
var w io.Writer = wh
if bar != nil {
bar.SetTotal(rh.Info().ContentLength, false)
bar.EnableTriggerComplete()
pw := bar.ProxyWriter(w)
defer func() {
_ = pw.Close()
}()
w = pw
}
_, err := sync2.Copy(ctx, w, rh)
if err == nil {
err = wh.Commit()
}
if err != nil {
// TODO: it would be also nice to use wh.Abort and rh.Close directly
// to avoid some of the waiting that's caused by sync2.Copy.
//
// However, some of the source / destination implementations don't seem
// to have concurrent safe API with that regards.
//
// Also, we may want to check that it actually helps, before implementing it.
addError(errs.New("failed to %s part %d: %v", copyVerb(source, dest), i, err))
}
})
if !ok {
break
}
}
limiter.Wait()
// don't try to commit if any error occur
if len(es) == 0 {
es.Add(dst.Commit(ctx))
}
return errs.Wrap(combineErrs(es))
}
func newProgressBar(progress *mpb.Progress, name string, which, total int) *mpb.Bar {
const counterFmt = " % .2f / % .2f"
const percentageFmt = "%.2f "
const speedFmt = "% .2f"
movingAverage := ewma.NewMovingAverage()
prepends := []decor.Decorator{decor.Name(name + " ")}
if total > 1 {
prepends = append(prepends, decor.Name(fmt.Sprintf("(%d of %d)", which, total)))
}
prepends = append(prepends, decor.CountersKiloByte(counterFmt))
appends := []decor.Decorator{
decor.NewPercentage(percentageFmt),
decor.MovingAverageSpeed(decor.SizeB1024(1024), speedFmt, movingAverage),
}
return progress.AddBar(0,
mpb.PrependDecorators(prepends...),
mpb.AppendDecorators(appends...),
)
}
func parseRange(r string) (offset, length int64, err error) {
r = strings.TrimPrefix(strings.TrimSpace(r), "bytes=")
if r == "" {
return 0, -1, nil
}
if strings.Contains(r, ",") {
return 0, 0, errs.New("invalid range: must be single range")
}
idx := strings.Index(r, "-")
if idx < 0 {
return 0, 0, errs.New(`invalid range: no "-"`)
}
start, end := strings.TrimSpace(r[:idx]), strings.TrimSpace(r[idx+1:])
var starti, endi int64
if start != "" {
starti, err = strconv.ParseInt(start, 10, 64)
if err != nil {
return 0, 0, errs.New("invalid range: %w", err)
}
}
if end != "" {
endi, err = strconv.ParseInt(end, 10, 64)
if err != nil {
return 0, 0, errs.New("invalid range: %w", err)
}
}
switch {
case start == "" && end == "":
return 0, 0, errs.New("invalid range")
case start == "":
return -endi, -1, nil
case end == "":
return starti, -1, nil
case starti < 0:
return 0, 0, errs.New("invalid range: negative start: %q", start)
case starti > endi:
return 0, 0, errs.New("invalid range: %v > %v", starti, endi)
default:
return starti, endi - starti + 1, nil
}
}
// combineErrs makes a more readable error message from the errors group.
func combineErrs(group errs.Group) error {
if len(group) == 0 {
return nil
}
errstrings := make([]string, len(group))
for i, err := range group {
errstrings[i] = err.Error()
}
return fmt.Errorf("%s", strings.Join(errstrings, "\n"))
}
type barNamer struct {
longestTotalLen int
}
func (n *barNamer) Preview(src, dst ulloc.Location) {
if src.Local() {
n.preview(src)
return
}
n.preview(dst)
}
func (n *barNamer) NameFor(src, dst ulloc.Location) string {
if src.Local() {
return n.nameFor(src)
}
return n.nameFor(dst)
}
func (n *barNamer) preview(loc ulloc.Location) {
locLen := len(loc.String())
if locLen > n.longestTotalLen {
n.longestTotalLen = locLen
}
}
func (n *barNamer) nameFor(loc ulloc.Location) string {
name := loc.String()
if n.longestTotalLen > 0 {
pad := n.longestTotalLen - len(loc.String())
name += strings.Repeat(" ", pad)
}
return name
}