// Run runs a command in the container, it assumes that `branch` has already
// been created.
// Notice that any failure in this function leads to the branch having
// uncommitted dirty changes. This state needs to be cleaned up before the
// pipeline is rerun. The reason we don't do it here is that even if we try our
// best the process crashing at the wrong time could still leave it in an
// inconsistent state.
func (p *pipeline) run(cmd []string) error {
// this function always increments counter
defer func() { p.counter++ }()
// Check if the commit already exists
exists, err := btrfs.FileExists(path.Join(p.outRepo, p.runCommit()))
if err != nil {
return err
}
// if the commit exists there's no work to be done
if exists {
return nil
}
// Set the command
p.config.Config.Cmd = []string{"sh"}
//p.config.Config.Volumes["/out"] = emptyStruct()
// Map the out directory in as a bind
hostPath := btrfs.HostPath(path.Join(p.outRepo, p.branch))
bind := fmt.Sprintf("%s:/out", hostPath)
p.config.HostConfig.Binds = append(p.config.HostConfig.Binds, bind)
log.Print(p.config.HostConfig.Binds)
// Make sure this bind is only visible for the duration of run
defer func() { p.config.HostConfig.Binds = p.config.HostConfig.Binds[:len(p.config.HostConfig.Binds)-1] }()
// Start the container
p.container, err = startContainer(p.config)
if err != nil {
return err
}
if err := pipeToStdin(p.container, strings.NewReader(strings.Join(cmd, " ")+"\n")); err != nil {
return err
}
// Create a place to put the logs
f, err := btrfs.CreateAll(path.Join(p.outRepo, p.branch, ".log"))
if err != nil {
return err
}
defer f.Close()
// Copy the logs from the container in to the file.
if err = containerLogs(p.container, f); err != nil {
return err
}
// Wait for the command to finish:
exit, err := waitContainer(p.container)
if err != nil {
return err
}
if exit != 0 {
// The command errored
return fmt.Errorf("Command:\n\t%s\nhad exit code: %d.\n",
strings.Join(cmd, " "), exit)
}
return btrfs.Commit(p.outRepo, p.runCommit(), p.branch)
}
// inject injects data from an external source into the output directory
func (p *pipeline) inject(name string, public bool) error {
switch {
case strings.HasPrefix(name, "s3://"):
bucket, err := s3utils.GetBucket(name)
if err != nil {
return err
}
client := s3utils.NewClient(public)
var wg sync.WaitGroup
s3utils.ForEachFile(name, public, "", func(file string, modtime time.Time) error {
// Grab the path, it's handy later
_path, err := s3utils.GetPath(name)
if err != nil {
return err
}
if err != nil {
return err
}
// Check if the file belongs on shit shard
match, err := route.Match(file, p.shard)
if err != nil {
return err
}
if !match {
return nil
}
// Check if the file has changed
changed, err := btrfs.Changed(path.Join(p.outRepo, p.branch,
strings.TrimPrefix(file, _path)), modtime)
if err != nil {
return err
}
if !changed {
return nil
}
// TODO match the on disk timestamps to s3's timestamps and make
// sure we only pull data that has changed
wg.Add(1)
go func() {
defer wg.Done()
response, err := client.GetObject(&s3.GetObjectInput{
Bucket: &bucket,
Key: &file,
})
if err != nil {
return
}
src := response.Body
dst, err := btrfs.CreateAll(path.Join(p.outRepo, p.branch, strings.TrimPrefix(file, _path)))
if err != nil {
return
}
defer dst.Close()
_, err = io.Copy(dst, src)
if err != nil {
return
}
err = btrfs.Chtimes(path.Join(p.outRepo, p.branch, strings.TrimPrefix(file, _path)), modtime, modtime)
if err != nil {
return
}
}()
return nil
})
wg.Wait()
default:
log.Print("Unknown protocol: ", name)
return ErrUnknownProtocol
}
return nil
}