1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477
|
package xfer // import "github.com/docker/docker/distribution/xfer"
import (
"context"
"errors"
"fmt"
"io"
"time"
"github.com/containerd/log"
"github.com/docker/distribution"
"github.com/docker/docker/image"
"github.com/docker/docker/layer"
"github.com/docker/docker/pkg/archive"
"github.com/docker/docker/pkg/ioutils"
"github.com/docker/docker/pkg/progress"
)
const maxDownloadAttempts = 5
// LayerDownloadManager figures out which layers need to be downloaded, then
// registers and downloads those, taking into account dependencies between
// layers.
type LayerDownloadManager struct {
layerStore layer.Store
tm *transferManager
waitDuration time.Duration
maxDownloadAttempts int
}
// SetConcurrency sets the max concurrent downloads for each pull
func (ldm *LayerDownloadManager) SetConcurrency(concurrency int) {
ldm.tm.setConcurrency(concurrency)
}
// NewLayerDownloadManager returns a new LayerDownloadManager.
func NewLayerDownloadManager(layerStore layer.Store, concurrencyLimit int, options ...DownloadOption) *LayerDownloadManager {
manager := LayerDownloadManager{
layerStore: layerStore,
tm: newTransferManager(concurrencyLimit),
waitDuration: time.Second,
maxDownloadAttempts: maxDownloadAttempts,
}
for _, option := range options {
option(&manager)
}
return &manager
}
// DownloadOption set options for the LayerDownloadManager.
type DownloadOption func(*LayerDownloadManager)
// WithMaxDownloadAttempts configures the maximum number of download
// attempts for a download manager.
func WithMaxDownloadAttempts(max int) DownloadOption {
return func(dlm *LayerDownloadManager) {
dlm.maxDownloadAttempts = max
}
}
type downloadTransfer struct {
transfer
layerStore layer.Store
layer layer.Layer
err error
}
// result returns the layer resulting from the download, if the download
// and registration were successful.
func (d *downloadTransfer) result() (layer.Layer, error) {
return d.layer, d.err
}
// A DownloadDescriptor references a layer that may need to be downloaded.
type DownloadDescriptor interface {
// Key returns the key used to deduplicate downloads.
Key() string
// ID returns the ID for display purposes.
ID() string
// DiffID should return the DiffID for this layer, or an error
// if it is unknown (for example, if it has not been downloaded
// before).
DiffID() (layer.DiffID, error)
// Download is called to perform the download.
Download(ctx context.Context, progressOutput progress.Output) (io.ReadCloser, int64, error)
// Close is called when the download manager is finished with this
// descriptor and will not call Download again or read from the reader
// that Download returned.
Close()
}
// DigestRegisterer can be implemented by a DownloadDescriptor, and provides a
// Registered method which gets called after a downloaded layer is registered.
// This allows the user of the download manager to know the DiffID of each
// registered layer. This method is called if a cast to DigestRegisterer is
// successful.
type DigestRegisterer interface {
// TODO existing implementations in distribution and builder-next swallow errors
// when registering the diffID. Consider changing the Registered signature
// to return the error.
Registered(diffID layer.DiffID)
}
// Download is a blocking function which ensures the requested layers are
// present in the layer store. It uses the string returned by the Key method to
// deduplicate downloads. If a given layer is not already known to present in
// the layer store, and the key is not used by an in-progress download, the
// Download method is called to get the layer tar data. Layers are then
// registered in the appropriate order. The caller must call the returned
// release function once it is done with the returned RootFS object.
func (ldm *LayerDownloadManager) Download(ctx context.Context, initialRootFS image.RootFS, layers []DownloadDescriptor, progressOutput progress.Output) (image.RootFS, func(), error) {
var (
topLayer layer.Layer
topDownload *downloadTransfer
watcher *watcher
missingLayer bool
transferKey = ""
downloadsByKey = make(map[string]*downloadTransfer)
)
rootFS := initialRootFS
for _, descriptor := range layers {
key := descriptor.Key()
transferKey += key
if !missingLayer {
missingLayer = true
diffID, err := descriptor.DiffID()
if err == nil {
getRootFS := rootFS
getRootFS.Append(diffID)
l, err := ldm.layerStore.Get(getRootFS.ChainID())
if err == nil {
// Layer already exists.
log.G(ctx).Debugf("Layer already exists: %s", descriptor.ID())
progress.Update(progressOutput, descriptor.ID(), "Already exists")
if topLayer != nil {
layer.ReleaseAndLog(ldm.layerStore, topLayer)
}
topLayer = l
missingLayer = false
rootFS.Append(diffID)
// Register this repository as a source of this layer.
if withRegistered, ok := descriptor.(DigestRegisterer); ok { // As layerstore may set the driver
withRegistered.Registered(diffID)
}
continue
}
}
}
// Does this layer have the same data as a previous layer in
// the stack? If so, avoid downloading it more than once.
var topDownloadUncasted transfer
if existingDownload, ok := downloadsByKey[key]; ok {
xferFunc := ldm.makeDownloadFuncFromDownload(descriptor, existingDownload, topDownload)
defer topDownload.transfer.release(watcher)
topDownloadUncasted, watcher = ldm.tm.transfer(transferKey, xferFunc, progressOutput)
topDownload = topDownloadUncasted.(*downloadTransfer)
continue
}
// Layer is not known to exist - download and register it.
progress.Update(progressOutput, descriptor.ID(), "Pulling fs layer")
var xferFunc doFunc
if topDownload != nil {
xferFunc = ldm.makeDownloadFunc(descriptor, "", topDownload)
defer topDownload.transfer.release(watcher)
} else {
xferFunc = ldm.makeDownloadFunc(descriptor, rootFS.ChainID(), nil)
}
topDownloadUncasted, watcher = ldm.tm.transfer(transferKey, xferFunc, progressOutput)
topDownload = topDownloadUncasted.(*downloadTransfer)
downloadsByKey[key] = topDownload
}
if topDownload == nil {
return rootFS, func() {
if topLayer != nil {
layer.ReleaseAndLog(ldm.layerStore, topLayer)
}
}, nil
}
// Won't be using the list built up so far - will generate it
// from downloaded layers instead.
rootFS.DiffIDs = []layer.DiffID{}
defer func() {
if topLayer != nil {
layer.ReleaseAndLog(ldm.layerStore, topLayer)
}
}()
select {
case <-ctx.Done():
topDownload.transfer.release(watcher)
return rootFS, func() {}, ctx.Err()
case <-topDownload.done():
break
}
l, err := topDownload.result()
if err != nil {
topDownload.transfer.release(watcher)
return rootFS, func() {}, err
}
// Must do this exactly len(layers) times, so we don't include the
// base layer on Windows.
for range layers {
if l == nil {
topDownload.transfer.release(watcher)
return rootFS, func() {}, errors.New("internal error: too few parent layers")
}
rootFS.DiffIDs = append([]layer.DiffID{l.DiffID()}, rootFS.DiffIDs...)
l = l.Parent()
}
return rootFS, func() { topDownload.transfer.release(watcher) }, err
}
// makeDownloadFunc returns a function that performs the layer download and
// registration. If parentDownload is non-nil, it waits for that download to
// complete before the registration step, and registers the downloaded data
// on top of parentDownload's resulting layer. Otherwise, it registers the
// layer on top of the ChainID given by parentLayer.
func (ldm *LayerDownloadManager) makeDownloadFunc(descriptor DownloadDescriptor, parentLayer layer.ChainID, parentDownload *downloadTransfer) doFunc {
return func(progressChan chan<- progress.Progress, start <-chan struct{}, inactive chan<- struct{}) transfer {
d := &downloadTransfer{
transfer: newTransfer(),
layerStore: ldm.layerStore,
}
go func() {
defer func() {
close(progressChan)
}()
progressOutput := progress.ChanOutput(progressChan)
select {
case <-start:
default:
progress.Update(progressOutput, descriptor.ID(), "Waiting")
<-start
}
if parentDownload != nil {
// Did the parent download already fail or get
// cancelled?
select {
case <-parentDownload.done():
_, err := parentDownload.result()
if err != nil {
d.err = err
return
}
default:
}
}
var (
downloadReader io.ReadCloser
size int64
err error
attempt int = 1
)
defer descriptor.Close()
for {
downloadReader, size, err = descriptor.Download(d.transfer.context(), progressOutput)
if err == nil {
break
}
// If an error was returned because the context
// was cancelled, we shouldn't retry.
select {
case <-d.transfer.context().Done():
d.err = err
return
default:
}
if _, isDNR := err.(DoNotRetry); isDNR || attempt >= ldm.maxDownloadAttempts {
log.G(context.TODO()).Errorf("Download failed after %d attempts: %v", attempt, err)
d.err = err
return
}
log.G(context.TODO()).Infof("Download failed, retrying (%d/%d): %v", attempt, ldm.maxDownloadAttempts, err)
delay := attempt * 5
ticker := time.NewTicker(ldm.waitDuration)
attempt++
selectLoop:
for {
progress.Updatef(progressOutput, descriptor.ID(), "Retrying in %d second%s", delay, (map[bool]string{true: "s"})[delay != 1])
select {
case <-ticker.C:
delay--
if delay == 0 {
ticker.Stop()
break selectLoop
}
case <-d.transfer.context().Done():
ticker.Stop()
d.err = errors.New("download cancelled during retry delay")
return
}
}
}
close(inactive)
if parentDownload != nil {
select {
case <-d.transfer.context().Done():
d.err = errors.New("layer registration cancelled")
downloadReader.Close()
return
case <-parentDownload.done():
}
l, err := parentDownload.result()
if err != nil {
d.err = err
downloadReader.Close()
return
}
parentLayer = l.ChainID()
}
reader := progress.NewProgressReader(ioutils.NewCancelReadCloser(d.transfer.context(), downloadReader), progressOutput, size, descriptor.ID(), "Extracting")
defer reader.Close()
inflatedLayerData, err := archive.DecompressStream(reader)
if err != nil {
d.err = fmt.Errorf("could not get decompression stream: %v", err)
return
}
defer inflatedLayerData.Close()
var src distribution.Descriptor
if fs, ok := descriptor.(distribution.Describable); ok {
src = fs.Descriptor()
}
if ds, ok := d.layerStore.(layer.DescribableStore); ok {
d.layer, err = ds.RegisterWithDescriptor(inflatedLayerData, parentLayer, src)
} else {
d.layer, err = d.layerStore.Register(inflatedLayerData, parentLayer)
}
if err != nil {
select {
case <-d.transfer.context().Done():
d.err = errors.New("layer registration cancelled")
default:
d.err = fmt.Errorf("failed to register layer: %v", err)
}
return
}
progress.Update(progressOutput, descriptor.ID(), "Pull complete")
if withRegistered, ok := descriptor.(DigestRegisterer); ok {
withRegistered.Registered(d.layer.DiffID())
}
// Doesn't actually need to be its own goroutine, but
// done like this so we can defer close(c).
go func() {
<-d.transfer.released()
if d.layer != nil {
layer.ReleaseAndLog(d.layerStore, d.layer)
}
}()
}()
return d
}
}
// makeDownloadFuncFromDownload returns a function that performs the layer
// registration when the layer data is coming from an existing download. It
// waits for sourceDownload and parentDownload to complete, and then
// reregisters the data from sourceDownload's top layer on top of
// parentDownload. This function does not log progress output because it would
// interfere with the progress reporting for sourceDownload, which has the same
// Key.
func (ldm *LayerDownloadManager) makeDownloadFuncFromDownload(descriptor DownloadDescriptor, sourceDownload *downloadTransfer, parentDownload *downloadTransfer) doFunc {
return func(progressChan chan<- progress.Progress, start <-chan struct{}, inactive chan<- struct{}) transfer {
d := &downloadTransfer{
transfer: newTransfer(),
layerStore: ldm.layerStore,
}
go func() {
defer func() {
close(progressChan)
}()
<-start
close(inactive)
select {
case <-d.transfer.context().Done():
d.err = errors.New("layer registration cancelled")
return
case <-parentDownload.done():
}
l, err := parentDownload.result()
if err != nil {
d.err = err
return
}
parentLayer := l.ChainID()
// sourceDownload should have already finished if
// parentDownload finished, but wait for it explicitly
// to be sure.
select {
case <-d.transfer.context().Done():
d.err = errors.New("layer registration cancelled")
return
case <-sourceDownload.done():
}
l, err = sourceDownload.result()
if err != nil {
d.err = err
return
}
layerReader, err := l.TarStream()
if err != nil {
d.err = err
return
}
defer layerReader.Close()
var src distribution.Descriptor
if fs, ok := l.(distribution.Describable); ok {
src = fs.Descriptor()
}
if ds, ok := d.layerStore.(layer.DescribableStore); ok {
d.layer, err = ds.RegisterWithDescriptor(layerReader, parentLayer, src)
} else {
d.layer, err = d.layerStore.Register(layerReader, parentLayer)
}
if err != nil {
d.err = fmt.Errorf("failed to register layer: %v", err)
return
}
if withRegistered, ok := descriptor.(DigestRegisterer); ok {
withRegistered.Registered(d.layer.DiffID())
}
// Doesn't actually need to be its own goroutine, but
// done like this so we can defer close(c).
go func() {
<-d.transfer.released()
if d.layer != nil {
layer.ReleaseAndLog(d.layerStore, d.layer)
}
}()
}()
return d
}
}
|