blob: 180612fb75ffde410acf005bea48a9a7d2c1d282 [file] [log] [blame]
package graph
import (
"fmt"
"io"
"net"
"net/url"
"strings"
"time"
"github.com/docker/docker/engine"
"github.com/docker/docker/image"
"github.com/docker/docker/pkg/log"
"github.com/docker/docker/registry"
"github.com/docker/docker/utils"
)
func (s *TagStore) CmdPull(job *engine.Job) engine.Status {
if n := len(job.Args); n != 1 && n != 2 {
return job.Errorf("Usage: %s IMAGE [TAG]", job.Name)
}
var (
localName = job.Args[0]
tag string
sf = utils.NewStreamFormatter(job.GetenvBool("json"))
authConfig = &registry.AuthConfig{}
metaHeaders map[string][]string
)
if len(job.Args) > 1 {
tag = job.Args[1]
}
job.GetenvJson("authConfig", authConfig)
job.GetenvJson("metaHeaders", &metaHeaders)
c, err := s.poolAdd("pull", localName+":"+tag)
if err != nil {
if c != nil {
// Another pull of the same repository is already taking place; just wait for it to finish
job.Stdout.Write(sf.FormatStatus("", "Repository %s already being pulled by another client. Waiting.", localName))
<-c
return engine.StatusOK
}
return job.Error(err)
}
defer s.poolRemove("pull", localName+":"+tag)
// Resolve the Repository name from fqn to endpoint + name
hostname, remoteName, err := registry.ResolveRepositoryName(localName)
if err != nil {
return job.Error(err)
}
endpoint, err := registry.ExpandAndVerifyRegistryUrl(hostname)
if err != nil {
return job.Error(err)
}
r, err := registry.NewSession(authConfig, registry.HTTPRequestFactory(metaHeaders), endpoint, true)
if err != nil {
return job.Error(err)
}
if endpoint == registry.IndexServerAddress() {
// If pull "index.docker.io/foo/bar", it's stored locally under "foo/bar"
localName = remoteName
}
if err = s.pullRepository(r, job.Stdout, localName, remoteName, tag, sf, job.GetenvBool("parallel")); err != nil {
return job.Error(err)
}
return engine.StatusOK
}
func (s *TagStore) pullRepository(r *registry.Session, out io.Writer, localName, remoteName, askedTag string, sf *utils.StreamFormatter, parallel bool) error {
out.Write(sf.FormatStatus("", "Pulling repository %s", localName))
repoData, err := r.GetRepositoryData(remoteName)
if err != nil {
if strings.Contains(err.Error(), "HTTP code: 404") {
return fmt.Errorf("Error: image %s not found", remoteName)
} else {
// Unexpected HTTP error
return err
}
}
log.Debugf("Retrieving the tag list")
tagsList, err := r.GetRemoteTags(repoData.Endpoints, remoteName, repoData.Tokens)
if err != nil {
log.Errorf("%v", err)
return err
}
for tag, id := range tagsList {
repoData.ImgList[id] = &registry.ImgData{
ID: id,
Tag: tag,
Checksum: "",
}
}
log.Debugf("Registering tags")
// If no tag has been specified, pull them all
if askedTag == "" {
for tag, id := range tagsList {
repoData.ImgList[id].Tag = tag
}
} else {
// Otherwise, check that the tag exists and use only that one
id, exists := tagsList[askedTag]
if !exists {
return fmt.Errorf("Tag %s not found in repository %s", askedTag, localName)
}
repoData.ImgList[id].Tag = askedTag
}
errors := make(chan error)
for _, image := range repoData.ImgList {
downloadImage := func(img *registry.ImgData) {
if askedTag != "" && img.Tag != askedTag {
log.Debugf("(%s) does not match %s (id: %s), skipping", img.Tag, askedTag, img.ID)
if parallel {
errors <- nil
}
return
}
if img.Tag == "" {
log.Debugf("Image (id: %s) present in this repository but untagged, skipping", img.ID)
if parallel {
errors <- nil
}
return
}
// ensure no two downloads of the same image happen at the same time
if c, err := s.poolAdd("pull", "img:"+img.ID); err != nil {
if c != nil {
out.Write(sf.FormatProgress(utils.TruncateID(img.ID), "Layer already being pulled by another client. Waiting.", nil))
<-c
out.Write(sf.FormatProgress(utils.TruncateID(img.ID), "Download complete", nil))
} else {
log.Debugf("Image (id: %s) pull is already running, skipping: %v", img.ID, err)
}
if parallel {
errors <- nil
}
return
}
defer s.poolRemove("pull", "img:"+img.ID)
out.Write(sf.FormatProgress(utils.TruncateID(img.ID), fmt.Sprintf("Pulling image (%s) from %s", img.Tag, localName), nil))
success := false
var lastErr error
for _, ep := range repoData.Endpoints {
out.Write(sf.FormatProgress(utils.TruncateID(img.ID), fmt.Sprintf("Pulling image (%s) from %s, endpoint: %s", img.Tag, localName, ep), nil))
if err := s.pullImage(r, out, img.ID, ep, repoData.Tokens, sf); err != nil {
// It's not ideal that only the last error is returned, it would be better to concatenate the errors.
// As the error is also given to the output stream the user will see the error.
lastErr = err
out.Write(sf.FormatProgress(utils.TruncateID(img.ID), fmt.Sprintf("Error pulling image (%s) from %s, endpoint: %s, %s", img.Tag, localName, ep, err), nil))
continue
}
success = true
break
}
if !success {
err := fmt.Errorf("Error pulling image (%s) from %s, %v", img.Tag, localName, lastErr)
out.Write(sf.FormatProgress(utils.TruncateID(img.ID), err.Error(), nil))
if parallel {
errors <- err
return
}
}
out.Write(sf.FormatProgress(utils.TruncateID(img.ID), "Download complete", nil))
if parallel {
errors <- nil
}
}
if parallel {
go downloadImage(image)
} else {
downloadImage(image)
}
}
if parallel {
var lastError error
for i := 0; i < len(repoData.ImgList); i++ {
if err := <-errors; err != nil {
lastError = err
}
}
if lastError != nil {
return lastError
}
}
for tag, id := range tagsList {
if askedTag != "" && tag != askedTag {
continue
}
if err := s.Set(localName, tag, id, true); err != nil {
return err
}
}
return nil
}
func (s *TagStore) pullImage(r *registry.Session, out io.Writer, imgID, endpoint string, token []string, sf *utils.StreamFormatter) error {
history, err := r.GetRemoteHistory(imgID, endpoint, token)
if err != nil {
return err
}
out.Write(sf.FormatProgress(utils.TruncateID(imgID), "Pulling dependent layers", nil))
// FIXME: Try to stream the images?
// FIXME: Launch the getRemoteImage() in goroutines
for i := len(history) - 1; i >= 0; i-- {
id := history[i]
// ensure no two downloads of the same layer happen at the same time
if c, err := s.poolAdd("pull", "layer:"+id); err != nil {
log.Debugf("Image (id: %s) pull is already running, skipping: %v", id, err)
<-c
}
defer s.poolRemove("pull", "layer:"+id)
if !s.graph.Exists(id) {
out.Write(sf.FormatProgress(utils.TruncateID(id), "Pulling metadata", nil))
var (
imgJSON []byte
imgSize int
err error
img *image.Image
)
retries := 5
for j := 1; j <= retries; j++ {
imgJSON, imgSize, err = r.GetRemoteImageJSON(id, endpoint, token)
if err != nil && j == retries {
out.Write(sf.FormatProgress(utils.TruncateID(id), "Error pulling dependent layers", nil))
return err
} else if err != nil {
time.Sleep(time.Duration(j) * 500 * time.Millisecond)
continue
}
img, err = image.NewImgJSON(imgJSON)
if err != nil && j == retries {
out.Write(sf.FormatProgress(utils.TruncateID(id), "Error pulling dependent layers", nil))
return fmt.Errorf("Failed to parse json: %s", err)
} else if err != nil {
time.Sleep(time.Duration(j) * 500 * time.Millisecond)
continue
} else {
break
}
}
for j := 1; j <= retries; j++ {
// Get the layer
status := "Pulling fs layer"
if j > 1 {
status = fmt.Sprintf("Pulling fs layer [retries: %d]", j)
}
out.Write(sf.FormatProgress(utils.TruncateID(id), status, nil))
layer, err := r.GetRemoteImageLayer(img.ID, endpoint, token, int64(imgSize))
if uerr, ok := err.(*url.Error); ok {
err = uerr.Err
}
if terr, ok := err.(net.Error); ok && terr.Timeout() && j < retries {
time.Sleep(time.Duration(j) * 500 * time.Millisecond)
continue
} else if err != nil {
out.Write(sf.FormatProgress(utils.TruncateID(id), "Error pulling dependent layers", nil))
return err
}
defer layer.Close()
err = s.graph.Register(imgJSON,
utils.ProgressReader(layer, imgSize, out, sf, false, utils.TruncateID(id), "Downloading"),
img)
if terr, ok := err.(net.Error); ok && terr.Timeout() && j < retries {
time.Sleep(time.Duration(j) * 500 * time.Millisecond)
continue
} else if err != nil {
out.Write(sf.FormatProgress(utils.TruncateID(id), "Error downloading dependent layers", nil))
return err
} else {
break
}
}
}
out.Write(sf.FormatProgress(utils.TruncateID(id), "Download complete", nil))
}
return nil
}