services/crunch-run/crunchrun.go

   1 package main
   2
   3 import (
   4         "encoding/json"
   5         "errors"
   6         "flag"
   7         "fmt"
   8         "git.curoverse.com/arvados.git/lib/crunchstat"
   9         "git.curoverse.com/arvados.git/sdk/go/arvados"
  10         "git.curoverse.com/arvados.git/sdk/go/arvadosclient"
  11         "git.curoverse.com/arvados.git/sdk/go/keepclient"
  12         "git.curoverse.com/arvados.git/sdk/go/manifest"
  13         "github.com/curoverse/dockerclient"
  14         "io"
  15         "io/ioutil"
  16         "log"
  17         "os"
  18         "os/exec"
  19         "os/signal"
  20         "path"
  21         "path/filepath"
  22         "sort"
  23         "strings"
  24         "sync"
  25         "syscall"
  26         "time"
  27 )
  28
  29 // IArvadosClient is the minimal Arvados API methods used by crunch-run.
  30 type IArvadosClient interface {
  31         Create(resourceType string, parameters arvadosclient.Dict, output interface{}) error
  32         Get(resourceType string, uuid string, parameters arvadosclient.Dict, output interface{}) error
  33         Update(resourceType string, uuid string, parameters arvadosclient.Dict, output interface{}) error
  34         Call(method, resourceType, uuid, action string, parameters arvadosclient.Dict, output interface{}) error
  35         Discovery(key string) (interface{}, error)
  36 }
  37
  38 // ErrCancelled is the error returned when the container is cancelled.
  39 var ErrCancelled = errors.New("Cancelled")
  40
  41 // IKeepClient is the minimal Keep API methods used by crunch-run.
  42 type IKeepClient interface {
  43         PutHB(hash string, buf []byte) (string, int, error)
  44         ManifestFileReader(m manifest.Manifest, filename string) (keepclient.ReadCloserWithLen, error)
  45 }
  46
  47 // NewLogWriter is a factory function to create a new log writer.
  48 type NewLogWriter func(name string) io.WriteCloser
  49
  50 type RunArvMount func(args []string, tok string) (*exec.Cmd, error)
  51
  52 type MkTempDir func(string, string) (string, error)
  53
  54 // ThinDockerClient is the minimal Docker client interface used by crunch-run.
  55 type ThinDockerClient interface {
  56         StopContainer(id string, timeout int) error
  57         InspectImage(id string) (*dockerclient.ImageInfo, error)
  58         LoadImage(reader io.Reader) error
  59         CreateContainer(config *dockerclient.ContainerConfig, name string, authConfig *dockerclient.AuthConfig) (string, error)
  60         StartContainer(id string, config *dockerclient.HostConfig) error
  61         AttachContainer(id string, options *dockerclient.AttachOptions) (io.ReadCloser, error)
  62         Wait(id string) <-chan dockerclient.WaitResult
  63         RemoveImage(name string, force bool) ([]*dockerclient.ImageDelete, error)
  64 }
  65
  66 // ContainerRunner is the main stateful struct used for a single execution of a
  67 // container.
  68 type ContainerRunner struct {
  69         Docker    ThinDockerClient
  70         ArvClient IArvadosClient
  71         Kc        IKeepClient
  72         arvados.Container
  73         dockerclient.ContainerConfig
  74         dockerclient.HostConfig
  75         token       string
  76         ContainerID string
  77         ExitCode    *int
  78         NewLogWriter
  79         loggingDone   chan bool
  80         CrunchLog     *ThrottledLogger
  81         Stdout        io.WriteCloser
  82         Stderr        *ThrottledLogger
  83         LogCollection *CollectionWriter
  84         LogsPDH       *string
  85         RunArvMount
  86         MkTempDir
  87         ArvMount       *exec.Cmd
  88         ArvMountPoint  string
  89         HostOutputDir  string
  90         CleanupTempDir []string
  91         Binds          []string
  92         OutputPDH      *string
  93         CancelLock     sync.Mutex
  94         Cancelled      bool
  95         SigChan        chan os.Signal
  96         ArvMountExit   chan error
  97         finalState     string
  98         trashLifetime  time.Duration
  99
 100         statLogger   io.WriteCloser
 101         statReporter *crunchstat.Reporter
 102         statInterval time.Duration
 103         cgroupRoot   string
 104         // What we expect the container's cgroup parent to be.
 105         expectCgroupParent string
 106         // What we tell docker to use as the container's cgroup
 107         // parent. Note: Ideally we would use the same field for both
 108         // expectCgroupParent and setCgroupParent, and just make it
 109         // default to "docker". However, when using docker < 1.10 with
 110         // systemd, specifying a non-empty cgroup parent (even the
 111         // default value "docker") hits a docker bug
 112         // (https://github.com/docker/docker/issues/17126). Using two
 113         // separate fields makes it possible to use the "expect cgroup
 114         // parent to be X" feature even on sites where the "specify
 115         // cgroup parent" feature breaks.
 116         setCgroupParent string
 117 }
 118
 119 // SetupSignals sets up signal handling to gracefully terminate the underlying
 120 // Docker container and update state when receiving a TERM, INT or QUIT signal.
 121 func (runner *ContainerRunner) SetupSignals() {
 122         runner.SigChan = make(chan os.Signal, 1)
 123         signal.Notify(runner.SigChan, syscall.SIGTERM)
 124         signal.Notify(runner.SigChan, syscall.SIGINT)
 125         signal.Notify(runner.SigChan, syscall.SIGQUIT)
 126
 127         go func(sig <-chan os.Signal) {
 128                 for range sig {
 129                         if !runner.Cancelled {
 130                                 runner.CancelLock.Lock()
 131                                 runner.Cancelled = true
 132                                 if runner.ContainerID != "" {
 133                                         runner.Docker.StopContainer(runner.ContainerID, 10)
 134                                 }
 135                                 runner.CancelLock.Unlock()
 136                         }
 137                 }
 138         }(runner.SigChan)
 139 }
 140
 141 // LoadImage determines the docker image id from the container record and
 142 // checks if it is available in the local Docker image store.  If not, it loads
 143 // the image from Keep.
 144 func (runner *ContainerRunner) LoadImage() (err error) {
 145
 146         runner.CrunchLog.Printf("Fetching Docker image from collection '%s'", runner.Container.ContainerImage)
 147
 148         var collection arvados.Collection
 149         err = runner.ArvClient.Get("collections", runner.Container.ContainerImage, nil, &collection)
 150         if err != nil {
 151                 return fmt.Errorf("While getting container image collection: %v", err)
 152         }
 153         manifest := manifest.Manifest{Text: collection.ManifestText}
 154         var img, imageID string
 155         for ms := range manifest.StreamIter() {
 156                 img = ms.FileStreamSegments[0].Name
 157                 if !strings.HasSuffix(img, ".tar") {
 158                         return fmt.Errorf("First file in the container image collection does not end in .tar")
 159                 }
 160                 imageID = img[:len(img)-4]
 161         }
 162
 163         runner.CrunchLog.Printf("Using Docker image id '%s'", imageID)
 164
 165         _, err = runner.Docker.InspectImage(imageID)
 166         if err != nil {
 167                 runner.CrunchLog.Print("Loading Docker image from keep")
 168
 169                 var readCloser io.ReadCloser
 170                 readCloser, err = runner.Kc.ManifestFileReader(manifest, img)
 171                 if err != nil {
 172                         return fmt.Errorf("While creating ManifestFileReader for container image: %v", err)
 173                 }
 174
 175                 err = runner.Docker.LoadImage(readCloser)
 176                 if err != nil {
 177                         return fmt.Errorf("While loading container image into Docker: %v", err)
 178                 }
 179         } else {
 180                 runner.CrunchLog.Print("Docker image is available")
 181         }
 182
 183         runner.ContainerConfig.Image = imageID
 184
 185         return nil
 186 }
 187
 188 func (runner *ContainerRunner) ArvMountCmd(arvMountCmd []string, token string) (c *exec.Cmd, err error) {
 189         c = exec.Command("arv-mount", arvMountCmd...)
 190
 191         // Copy our environment, but override ARVADOS_API_TOKEN with
 192         // the container auth token.
 193         c.Env = nil
 194         for _, s := range os.Environ() {
 195                 if !strings.HasPrefix(s, "ARVADOS_API_TOKEN=") {
 196                         c.Env = append(c.Env, s)
 197                 }
 198         }
 199         c.Env = append(c.Env, "ARVADOS_API_TOKEN="+token)
 200
 201         nt := NewThrottledLogger(runner.NewLogWriter("arv-mount"))
 202         c.Stdout = nt
 203         c.Stderr = nt
 204
 205         err = c.Start()
 206         if err != nil {
 207                 return nil, err
 208         }
 209
 210         statReadme := make(chan bool)
 211         runner.ArvMountExit = make(chan error)
 212
 213         keepStatting := true
 214         go func() {
 215                 for keepStatting {
 216                         time.Sleep(100 * time.Millisecond)
 217                         _, err = os.Stat(fmt.Sprintf("%s/by_id/README", runner.ArvMountPoint))
 218                         if err == nil {
 219                                 keepStatting = false
 220                                 statReadme <- true
 221                         }
 222                 }
 223                 close(statReadme)
 224         }()
 225
 226         go func() {
 227                 runner.ArvMountExit <- c.Wait()
 228                 close(runner.ArvMountExit)
 229         }()
 230
 231         select {
 232         case <-statReadme:
 233                 break
 234         case err := <-runner.ArvMountExit:
 235                 runner.ArvMount = nil
 236                 keepStatting = false
 237                 return nil, err
 238         }
 239
 240         return c, nil
 241 }
 242
 243 var tmpBackedOutputDir = false
 244
 245 func (runner *ContainerRunner) SetupArvMountPoint(prefix string) (err error) {
 246         if runner.ArvMountPoint == "" {
 247                 runner.ArvMountPoint, err = runner.MkTempDir("", prefix)
 248         }
 249         return
 250 }
 251
 252 func (runner *ContainerRunner) SetupMounts() (err error) {
 253         err = runner.SetupArvMountPoint("keep")
 254         if err != nil {
 255                 return fmt.Errorf("While creating keep mount temp dir: %v", err)
 256         }
 257
 258         runner.CleanupTempDir = append(runner.CleanupTempDir, runner.ArvMountPoint)
 259
 260         pdhOnly := true
 261         tmpcount := 0
 262         arvMountCmd := []string{"--foreground", "--allow-other", "--read-write"}
 263
 264         if runner.Container.RuntimeConstraints.KeepCacheRAM > 0 {
 265                 arvMountCmd = append(arvMountCmd, "--file-cache", fmt.Sprintf("%d", runner.Container.RuntimeConstraints.KeepCacheRAM))
 266         }
 267
 268         collectionPaths := []string{}
 269         runner.Binds = nil
 270         needCertMount := true
 271
 272         var binds []string
 273         for bind, _ := range runner.Container.Mounts {
 274                 binds = append(binds, bind)
 275         }
 276         sort.Strings(binds)
 277
 278         for _, bind := range binds {
 279                 mnt := runner.Container.Mounts[bind]
 280                 if bind == "stdout" {
 281                         // Is it a "file" mount kind?
 282                         if mnt.Kind != "file" {
 283                                 return fmt.Errorf("Unsupported mount kind '%s' for stdout. Only 'file' is supported.", mnt.Kind)
 284                         }
 285
 286                         // Does path start with OutputPath?
 287                         prefix := runner.Container.OutputPath
 288                         if !strings.HasSuffix(prefix, "/") {
 289                                 prefix += "/"
 290                         }
 291                         if !strings.HasPrefix(mnt.Path, prefix) {
 292                                 return fmt.Errorf("Stdout path does not start with OutputPath: %s, %s", mnt.Path, prefix)
 293                         }
 294                 }
 295
 296                 if bind == "/etc/arvados/ca-certificates.crt" {
 297                         needCertMount = false
 298                 }
 299
 300                 if strings.HasPrefix(bind, runner.Container.OutputPath+"/") && bind != runner.Container.OutputPath+"/" {
 301                         if mnt.Kind != "collection" {
 302                                 return fmt.Errorf("Only mount points of kind 'collection' are supported underneath the output_path: %v", bind)
 303                         }
 304                 }
 305
 306                 switch {
 307                 case mnt.Kind == "collection":
 308                         var src string
 309                         if mnt.UUID != "" && mnt.PortableDataHash != "" {
 310                                 return fmt.Errorf("Cannot specify both 'uuid' and 'portable_data_hash' for a collection mount")
 311                         }
 312                         if mnt.UUID != "" {
 313                                 if mnt.Writable {
 314                                         return fmt.Errorf("Writing to existing collections currently not permitted.")
 315                                 }
 316                                 pdhOnly = false
 317                                 src = fmt.Sprintf("%s/by_id/%s", runner.ArvMountPoint, mnt.UUID)
 318                         } else if mnt.PortableDataHash != "" {
 319                                 if mnt.Writable {
 320                                         return fmt.Errorf("Can never write to a collection specified by portable data hash")
 321                                 }
 322                                 src = fmt.Sprintf("%s/by_id/%s", runner.ArvMountPoint, mnt.PortableDataHash)
 323                         } else {
 324                                 src = fmt.Sprintf("%s/tmp%d", runner.ArvMountPoint, tmpcount)
 325                                 arvMountCmd = append(arvMountCmd, "--mount-tmp")
 326                                 arvMountCmd = append(arvMountCmd, fmt.Sprintf("tmp%d", tmpcount))
 327                                 tmpcount += 1
 328                         }
 329                         if mnt.Writable {
 330                                 if bind == runner.Container.OutputPath {
 331                                         runner.HostOutputDir = src
 332                                 } else if strings.HasPrefix(bind, runner.Container.OutputPath+"/") {
 333                                         return fmt.Errorf("Writable mount points are not permitted underneath the output_path: %v", bind)
 334                                 }
 335                                 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s", src, bind))
 336                         } else {
 337                                 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s:ro", src, bind))
 338                         }
 339                         collectionPaths = append(collectionPaths, src)
 340
 341                 case mnt.Kind == "tmp" && bind == runner.Container.OutputPath:
 342                         runner.HostOutputDir, err = runner.MkTempDir("", "")
 343                         if err != nil {
 344                                 return fmt.Errorf("While creating mount temp dir: %v", err)
 345                         }
 346                         st, staterr := os.Stat(runner.HostOutputDir)
 347                         if staterr != nil {
 348                                 return fmt.Errorf("While Stat on temp dir: %v", staterr)
 349                         }
 350                         err = os.Chmod(runner.HostOutputDir, st.Mode()|os.ModeSetgid|0777)
 351                         if staterr != nil {
 352                                 return fmt.Errorf("While Chmod temp dir: %v", err)
 353                         }
 354                         runner.CleanupTempDir = append(runner.CleanupTempDir, runner.HostOutputDir)
 355                         runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s", runner.HostOutputDir, bind))
 356                         tmpBackedOutputDir = true
 357
 358                 case mnt.Kind == "tmp":
 359                         runner.Binds = append(runner.Binds, bind)
 360
 361                 case mnt.Kind == "json":
 362                         jsondata, err := json.Marshal(mnt.Content)
 363                         if err != nil {
 364                                 return fmt.Errorf("encoding json data: %v", err)
 365                         }
 366                         // Create a tempdir with a single file
 367                         // (instead of just a tempfile): this way we
 368                         // can ensure the file is world-readable
 369                         // inside the container, without having to
 370                         // make it world-readable on the docker host.
 371                         tmpdir, err := runner.MkTempDir("", "")
 372                         if err != nil {
 373                                 return fmt.Errorf("creating temp dir: %v", err)
 374                         }
 375                         runner.CleanupTempDir = append(runner.CleanupTempDir, tmpdir)
 376                         tmpfn := filepath.Join(tmpdir, "mountdata.json")
 377                         err = ioutil.WriteFile(tmpfn, jsondata, 0644)
 378                         if err != nil {
 379                                 return fmt.Errorf("writing temp file: %v", err)
 380                         }
 381                         runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s:ro", tmpfn, bind))
 382                 }
 383         }
 384
 385         if runner.HostOutputDir == "" {
 386                 return fmt.Errorf("Output path does not correspond to a writable mount point")
 387         }
 388
 389         if wantAPI := runner.Container.RuntimeConstraints.API; needCertMount && wantAPI != nil && *wantAPI {
 390                 for _, certfile := range arvadosclient.CertFiles {
 391                         _, err := os.Stat(certfile)
 392                         if err == nil {
 393                                 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:/etc/arvados/ca-certificates.crt:ro", certfile))
 394                                 break
 395                         }
 396                 }
 397         }
 398
 399         if pdhOnly {
 400                 arvMountCmd = append(arvMountCmd, "--mount-by-pdh", "by_id")
 401         } else {
 402                 arvMountCmd = append(arvMountCmd, "--mount-by-id", "by_id")
 403         }
 404         arvMountCmd = append(arvMountCmd, runner.ArvMountPoint)
 405
 406         token, err := runner.ContainerToken()
 407         if err != nil {
 408                 return fmt.Errorf("could not get container token: %s", err)
 409         }
 410
 411         runner.ArvMount, err = runner.RunArvMount(arvMountCmd, token)
 412         if err != nil {
 413                 return fmt.Errorf("While trying to start arv-mount: %v", err)
 414         }
 415
 416         for _, p := range collectionPaths {
 417                 _, err = os.Stat(p)
 418                 if err != nil {
 419                         return fmt.Errorf("While checking that input files exist: %v", err)
 420                 }
 421         }
 422
 423         return nil
 424 }
 425
 426 func (runner *ContainerRunner) ProcessDockerAttach(containerReader io.Reader) {
 427         // Handle docker log protocol
 428         // https://docs.docker.com/engine/reference/api/docker_remote_api_v1.15/#attach-to-a-container
 429
 430         header := make([]byte, 8)
 431         for {
 432                 _, readerr := io.ReadAtLeast(containerReader, header, 8)
 433
 434                 if readerr == nil {
 435                         readsize := int64(header[7]) | (int64(header[6]) << 8) | (int64(header[5]) << 16) | (int64(header[4]) << 24)
 436                         if header[0] == 1 {
 437                                 // stdout
 438                                 _, readerr = io.CopyN(runner.Stdout, containerReader, readsize)
 439                         } else {
 440                                 // stderr
 441                                 _, readerr = io.CopyN(runner.Stderr, containerReader, readsize)
 442                         }
 443                 }
 444
 445                 if readerr != nil {
 446                         if readerr != io.EOF {
 447                                 runner.CrunchLog.Printf("While reading docker logs: %v", readerr)
 448                         }
 449
 450                         closeerr := runner.Stdout.Close()
 451                         if closeerr != nil {
 452                                 runner.CrunchLog.Printf("While closing stdout logs: %v", closeerr)
 453                         }
 454
 455                         closeerr = runner.Stderr.Close()
 456                         if closeerr != nil {
 457                                 runner.CrunchLog.Printf("While closing stderr logs: %v", closeerr)
 458                         }
 459
 460                         if runner.statReporter != nil {
 461                                 runner.statReporter.Stop()
 462                                 closeerr = runner.statLogger.Close()
 463                                 if closeerr != nil {
 464                                         runner.CrunchLog.Printf("While closing crunchstat logs: %v", closeerr)
 465                                 }
 466                         }
 467
 468                         runner.loggingDone <- true
 469                         close(runner.loggingDone)
 470                         return
 471                 }
 472         }
 473 }
 474
 475 func (runner *ContainerRunner) StartCrunchstat() {
 476         runner.statLogger = NewThrottledLogger(runner.NewLogWriter("crunchstat"))
 477         runner.statReporter = &crunchstat.Reporter{
 478                 CID:          runner.ContainerID,
 479                 Logger:       log.New(runner.statLogger, "", 0),
 480                 CgroupParent: runner.expectCgroupParent,
 481                 CgroupRoot:   runner.cgroupRoot,
 482                 PollPeriod:   runner.statInterval,
 483         }
 484         runner.statReporter.Start()
 485 }
 486
 487 // AttachLogs connects the docker container stdout and stderr logs to the
 488 // Arvados logger which logs to Keep and the API server logs table.
 489 func (runner *ContainerRunner) AttachStreams() (err error) {
 490
 491         runner.CrunchLog.Print("Attaching container streams")
 492
 493         var containerReader io.Reader
 494         containerReader, err = runner.Docker.AttachContainer(runner.ContainerID,
 495                 &dockerclient.AttachOptions{Stream: true, Stdout: true, Stderr: true})
 496         if err != nil {
 497                 return fmt.Errorf("While attaching container stdout/stderr streams: %v", err)
 498         }
 499
 500         runner.loggingDone = make(chan bool)
 501
 502         if stdoutMnt, ok := runner.Container.Mounts["stdout"]; ok {
 503                 stdoutPath := stdoutMnt.Path[len(runner.Container.OutputPath):]
 504                 index := strings.LastIndex(stdoutPath, "/")
 505                 if index > 0 {
 506                         subdirs := stdoutPath[:index]
 507                         if subdirs != "" {
 508                                 st, err := os.Stat(runner.HostOutputDir)
 509                                 if err != nil {
 510                                         return fmt.Errorf("While Stat on temp dir: %v", err)
 511                                 }
 512                                 stdoutPath := path.Join(runner.HostOutputDir, subdirs)
 513                                 err = os.MkdirAll(stdoutPath, st.Mode()|os.ModeSetgid|0777)
 514                                 if err != nil {
 515                                         return fmt.Errorf("While MkdirAll %q: %v", stdoutPath, err)
 516                                 }
 517                         }
 518                 }
 519                 stdoutFile, err := os.Create(path.Join(runner.HostOutputDir, stdoutPath))
 520                 if err != nil {
 521                         return fmt.Errorf("While creating stdout file: %v", err)
 522                 }
 523                 runner.Stdout = stdoutFile
 524         } else {
 525                 runner.Stdout = NewThrottledLogger(runner.NewLogWriter("stdout"))
 526         }
 527         runner.Stderr = NewThrottledLogger(runner.NewLogWriter("stderr"))
 528
 529         go runner.ProcessDockerAttach(containerReader)
 530
 531         return nil
 532 }
 533
 534 // CreateContainer creates the docker container.
 535 func (runner *ContainerRunner) CreateContainer() error {
 536         runner.CrunchLog.Print("Creating Docker container")
 537
 538         runner.ContainerConfig.Cmd = runner.Container.Command
 539         if runner.Container.Cwd != "." {
 540                 runner.ContainerConfig.WorkingDir = runner.Container.Cwd
 541         }
 542
 543         for k, v := range runner.Container.Environment {
 544                 runner.ContainerConfig.Env = append(runner.ContainerConfig.Env, k+"="+v)
 545         }
 546         if wantAPI := runner.Container.RuntimeConstraints.API; wantAPI != nil && *wantAPI {
 547                 tok, err := runner.ContainerToken()
 548                 if err != nil {
 549                         return err
 550                 }
 551                 runner.ContainerConfig.Env = append(runner.ContainerConfig.Env,
 552                         "ARVADOS_API_TOKEN="+tok,
 553                         "ARVADOS_API_HOST="+os.Getenv("ARVADOS_API_HOST"),
 554                         "ARVADOS_API_HOST_INSECURE="+os.Getenv("ARVADOS_API_HOST_INSECURE"),
 555                 )
 556                 runner.ContainerConfig.NetworkDisabled = false
 557         } else {
 558                 runner.ContainerConfig.NetworkDisabled = true
 559         }
 560
 561         var err error
 562         runner.ContainerID, err = runner.Docker.CreateContainer(&runner.ContainerConfig, "", nil)
 563         if err != nil {
 564                 return fmt.Errorf("While creating container: %v", err)
 565         }
 566
 567         runner.HostConfig = dockerclient.HostConfig{
 568                 Binds:        runner.Binds,
 569                 CgroupParent: runner.setCgroupParent,
 570                 LogConfig: dockerclient.LogConfig{
 571                         Type: "none",
 572                 },
 573         }
 574
 575         return runner.AttachStreams()
 576 }
 577
 578 // StartContainer starts the docker container created by CreateContainer.
 579 func (runner *ContainerRunner) StartContainer() error {
 580         runner.CrunchLog.Printf("Starting Docker container id '%s'", runner.ContainerID)
 581         err := runner.Docker.StartContainer(runner.ContainerID, &runner.HostConfig)
 582         if err != nil {
 583                 return fmt.Errorf("could not start container: %v", err)
 584         }
 585         return nil
 586 }
 587
 588 // WaitFinish waits for the container to terminate, capture the exit code, and
 589 // close the stdout/stderr logging.
 590 func (runner *ContainerRunner) WaitFinish() error {
 591         runner.CrunchLog.Print("Waiting for container to finish")
 592
 593         result := runner.Docker.Wait(runner.ContainerID)
 594         wr := <-result
 595         if wr.Error != nil {
 596                 return fmt.Errorf("While waiting for container to finish: %v", wr.Error)
 597         }
 598         runner.ExitCode = &wr.ExitCode
 599
 600         // wait for stdout/stderr to complete
 601         <-runner.loggingDone
 602
 603         return nil
 604 }
 605
 606 // HandleOutput sets the output, unmounts the FUSE mount, and deletes temporary directories
 607 func (runner *ContainerRunner) CaptureOutput() error {
 608         if runner.finalState != "Complete" {
 609                 return nil
 610         }
 611
 612         if wantAPI := runner.Container.RuntimeConstraints.API; wantAPI != nil && *wantAPI {
 613                 // Output may have been set directly by the container, so
 614                 // refresh the container record to check.
 615                 err := runner.ArvClient.Get("containers", runner.Container.UUID,
 616                         nil, &runner.Container)
 617                 if err != nil {
 618                         return err
 619                 }
 620                 if runner.Container.Output != "" {
 621                         // Container output is already set.
 622                         runner.OutputPDH = &runner.Container.Output
 623                         return nil
 624                 }
 625         }
 626
 627         if runner.HostOutputDir == "" {
 628                 return nil
 629         }
 630
 631         _, err := os.Stat(runner.HostOutputDir)
 632         if err != nil {
 633                 return fmt.Errorf("While checking host output path: %v", err)
 634         }
 635
 636         var manifestText string
 637
 638         collectionMetafile := fmt.Sprintf("%s/.arvados#collection", runner.HostOutputDir)
 639         _, err = os.Stat(collectionMetafile)
 640         if err != nil {
 641                 // Regular directory
 642                 cw := CollectionWriter{runner.Kc, nil, sync.Mutex{}}
 643                 manifestText, err = cw.WriteTree(runner.HostOutputDir, runner.CrunchLog.Logger)
 644                 if err != nil {
 645                         return fmt.Errorf("While uploading output files: %v", err)
 646                 }
 647         } else {
 648                 // FUSE mount directory
 649                 file, openerr := os.Open(collectionMetafile)
 650                 if openerr != nil {
 651                         return fmt.Errorf("While opening FUSE metafile: %v", err)
 652                 }
 653                 defer file.Close()
 654
 655                 var rec arvados.Collection
 656                 err = json.NewDecoder(file).Decode(&rec)
 657                 if err != nil {
 658                         return fmt.Errorf("While reading FUSE metafile: %v", err)
 659                 }
 660                 manifestText = rec.ManifestText
 661         }
 662
 663         // Pre-populate output from the configured mount points
 664         var binds []string
 665         for bind, _ := range runner.Container.Mounts {
 666                 binds = append(binds, bind)
 667         }
 668         sort.Strings(binds)
 669
 670         for _, bind := range binds {
 671                 mnt := runner.Container.Mounts[bind]
 672
 673                 bindSuffix := strings.TrimPrefix(bind, runner.Container.OutputPath)
 674
 675                 if bindSuffix == bind || len(bindSuffix) <= 0 {
 676                         // either does not start with OutputPath or is OutputPath itself
 677                         continue
 678                 }
 679
 680                 if strings.Index(bindSuffix, "/") != 0 {
 681                         return fmt.Errorf("Expected bind to be of the format '%v/*' but found: %v", runner.Container.OutputPath, bind)
 682                 }
 683
 684                 jsondata, err := json.Marshal(mnt.Content)
 685                 if err != nil {
 686                         return fmt.Errorf("While marshal of mount content: %v", err)
 687                 }
 688                 var content map[string]interface{}
 689                 err = json.Unmarshal(jsondata, &content)
 690                 if err != nil {
 691                         return fmt.Errorf("While unmarshal of mount content: %v", err)
 692                 }
 693
 694                 if content["exclude_from_output"] == true {
 695                         continue
 696                 }
 697
 698                 idx := strings.Index(mnt.PortableDataHash, "/")
 699                 if idx > 0 {
 700                         mnt.Path = mnt.PortableDataHash[idx:]
 701                         mnt.PortableDataHash = mnt.PortableDataHash[0:idx]
 702                 }
 703
 704                 // append to manifest_text
 705                 m, err := runner.getCollectionManifestForPath(mnt, bindSuffix)
 706                 if err != nil {
 707                         return err
 708                 }
 709
 710                 manifestText = manifestText + m
 711         }
 712
 713         // Save output
 714         var response arvados.Collection
 715         err = runner.ArvClient.Create("collections",
 716                 arvadosclient.Dict{
 717                         "collection": arvadosclient.Dict{
 718                                 "trash_at":      time.Now().Add(runner.trashLifetime).Format(time.RFC3339),
 719                                 "name":          "output for " + runner.Container.UUID,
 720                                 "manifest_text": manifestText}},
 721                 &response)
 722         if err != nil {
 723                 return fmt.Errorf("While creating output collection: %v", err)
 724         }
 725         runner.OutputPDH = &response.PortableDataHash
 726         return nil
 727 }
 728
 729 // Fetch the collection for the mnt.PortableDataHash
 730 // Return the manifest_text fragment corresponding to the specified mnt.Path
 731 //  after making any required updates.
 732 //  Ex:
 733 //    If mnt.Path is not specified,
 734 //      return the entire manifest_text after replacing any "." with bindSuffix
 735 //    If mnt.Path corresponds to one stream,
 736 //      return the manifest_text for that stream after replacing that stream name with bindSuffix
 737 //    Otherwise, check if a filename in any one stream is being sought. Return the manifest_text
 738 //      for that stream after replacing stream name with bindSuffix minus the last word
 739 //      and the file name with last word of the bindSuffix
 740 //  Allowed path examples:
 741 //    "path":"/"
 742 //    "path":"/subdir1"
 743 //    "path":"/subdir1/subdir2"
 744 //    "path":"/subdir/filename" etc
 745 func (runner *ContainerRunner) getCollectionManifestForPath(mnt arvados.Mount, bindSuffix string) (string, error) {
 746         var collection arvados.Collection
 747         err := runner.ArvClient.Get("collections", mnt.PortableDataHash, nil, &collection)
 748         if err != nil {
 749                 return "", fmt.Errorf("While getting collection for %v: %v", mnt.PortableDataHash, err)
 750         }
 751
 752         manifestText := ""
 753         if mnt.Path == "" || mnt.Path == "/" {
 754                 // no path specified; return the entire manifest text
 755                 manifestText = collection.ManifestText
 756                 manifestText = strings.Replace(manifestText, "./", "."+bindSuffix+"/", -1)
 757                 manifestText = strings.Replace(manifestText, ". ", "."+bindSuffix+" ", -1)
 758         } else {
 759                 // either a single stream or file from a stream is being sought
 760                 bindIdx := strings.LastIndex(bindSuffix, "/")
 761                 var bindSubdir, bindFileName string
 762                 if bindIdx >= 0 {
 763                         bindSubdir = "." + bindSuffix[0:bindIdx]
 764                         bindFileName = bindSuffix[bindIdx+1:]
 765                 }
 766                 pathIdx := strings.LastIndex(mnt.Path, "/")
 767                 var pathSubdir, pathFileName string
 768                 if pathIdx >= 0 {
 769                         pathSubdir = "." + mnt.Path[0:pathIdx]
 770                         pathFileName = mnt.Path[pathIdx+1:]
 771                 }
 772                 streams := strings.Split(collection.ManifestText, "\n")
 773                 for _, stream := range streams {
 774                         tokens := strings.Split(stream, " ")
 775                         if tokens[0] == "."+mnt.Path {
 776                                 // path refers to this complete stream
 777                                 adjustedStream := strings.Replace(stream, "."+mnt.Path, "."+bindSuffix, -1)
 778                                 manifestText = adjustedStream + "\n"
 779                                 break
 780                         } else {
 781                                 // look for a matching file in this stream
 782                                 if tokens[0] == pathSubdir {
 783                                         // path refers to a file in this stream
 784                                         for _, token := range tokens {
 785                                                 if strings.Index(token, ":"+pathFileName) > 0 {
 786                                                         // found the file in the stream; discard all other file tokens
 787                                                         for _, t := range tokens {
 788                                                                 if strings.Index(t, ":") == -1 {
 789                                                                         manifestText += (" " + t)
 790                                                                 } else {
 791                                                                         break // done reading all non-file tokens
 792                                                                 }
 793                                                         }
 794                                                         manifestText = strings.Trim(manifestText, " ")
 795                                                         token = strings.Replace(token, ":"+pathFileName, ":"+bindFileName, -1)
 796                                                         manifestText += (" " + token + "\n")
 797                                                         manifestText = strings.Replace(manifestText, pathSubdir, bindSubdir, -1)
 798                                                         break
 799                                                 }
 800                                         }
 801                                 }
 802                         }
 803                 }
 804         }
 805
 806         return manifestText, nil
 807 }
 808
 809 func (runner *ContainerRunner) loadDiscoveryVars() {
 810         tl, err := runner.ArvClient.Discovery("defaultTrashLifetime")
 811         if err != nil {
 812                 log.Fatalf("getting defaultTrashLifetime from discovery document: %s", err)
 813         }
 814         runner.trashLifetime = time.Duration(tl.(float64)) * time.Second
 815 }
 816
 817 func (runner *ContainerRunner) CleanupDirs() {
 818         if runner.ArvMount != nil {
 819                 umount := exec.Command("fusermount", "-z", "-u", runner.ArvMountPoint)
 820                 umnterr := umount.Run()
 821                 if umnterr != nil {
 822                         runner.CrunchLog.Printf("While running fusermount: %v", umnterr)
 823                 }
 824
 825                 mnterr := <-runner.ArvMountExit
 826                 if mnterr != nil {
 827                         runner.CrunchLog.Printf("Arv-mount exit error: %v", mnterr)
 828                 }
 829         }
 830
 831         for _, tmpdir := range runner.CleanupTempDir {
 832                 rmerr := os.RemoveAll(tmpdir)
 833                 if rmerr != nil {
 834                         runner.CrunchLog.Printf("While cleaning up temporary directory %s: %v", tmpdir, rmerr)
 835                 }
 836         }
 837 }
 838
 839 // CommitLogs posts the collection containing the final container logs.
 840 func (runner *ContainerRunner) CommitLogs() error {
 841         runner.CrunchLog.Print(runner.finalState)
 842         runner.CrunchLog.Close()
 843
 844         // Closing CrunchLog above allows it to be committed to Keep at this
 845         // point, but re-open crunch log with ArvClient in case there are any
 846         // other further (such as failing to write the log to Keep!) while
 847         // shutting down
 848         runner.CrunchLog = NewThrottledLogger(&ArvLogWriter{runner.ArvClient, runner.Container.UUID,
 849                 "crunch-run", nil})
 850
 851         if runner.LogsPDH != nil {
 852                 // If we have already assigned something to LogsPDH,
 853                 // we must be closing the re-opened log, which won't
 854                 // end up getting attached to the container record and
 855                 // therefore doesn't need to be saved as a collection
 856                 // -- it exists only to send logs to other channels.
 857                 return nil
 858         }
 859
 860         mt, err := runner.LogCollection.ManifestText()
 861         if err != nil {
 862                 return fmt.Errorf("While creating log manifest: %v", err)
 863         }
 864
 865         var response arvados.Collection
 866         err = runner.ArvClient.Create("collections",
 867                 arvadosclient.Dict{
 868                         "collection": arvadosclient.Dict{
 869                                 "trash_at":      time.Now().Add(runner.trashLifetime).Format(time.RFC3339),
 870                                 "name":          "logs for " + runner.Container.UUID,
 871                                 "manifest_text": mt}},
 872                 &response)
 873         if err != nil {
 874                 return fmt.Errorf("While creating log collection: %v", err)
 875         }
 876         runner.LogsPDH = &response.PortableDataHash
 877         return nil
 878 }
 879
 880 // UpdateContainerRunning updates the container state to "Running"
 881 func (runner *ContainerRunner) UpdateContainerRunning() error {
 882         runner.CancelLock.Lock()
 883         defer runner.CancelLock.Unlock()
 884         if runner.Cancelled {
 885                 return ErrCancelled
 886         }
 887         return runner.ArvClient.Update("containers", runner.Container.UUID,
 888                 arvadosclient.Dict{"container": arvadosclient.Dict{"state": "Running"}}, nil)
 889 }
 890
 891 // ContainerToken returns the api_token the container (and any
 892 // arv-mount processes) are allowed to use.
 893 func (runner *ContainerRunner) ContainerToken() (string, error) {
 894         if runner.token != "" {
 895                 return runner.token, nil
 896         }
 897
 898         var auth arvados.APIClientAuthorization
 899         err := runner.ArvClient.Call("GET", "containers", runner.Container.UUID, "auth", nil, &auth)
 900         if err != nil {
 901                 return "", err
 902         }
 903         runner.token = auth.APIToken
 904         return runner.token, nil
 905 }
 906
 907 // UpdateContainerComplete updates the container record state on API
 908 // server to "Complete" or "Cancelled"
 909 func (runner *ContainerRunner) UpdateContainerFinal() error {
 910         update := arvadosclient.Dict{}
 911         update["state"] = runner.finalState
 912         if runner.LogsPDH != nil {
 913                 update["log"] = *runner.LogsPDH
 914         }
 915         if runner.finalState == "Complete" {
 916                 if runner.ExitCode != nil {
 917                         update["exit_code"] = *runner.ExitCode
 918                 }
 919                 if runner.OutputPDH != nil {
 920                         update["output"] = *runner.OutputPDH
 921                 }
 922         }
 923         return runner.ArvClient.Update("containers", runner.Container.UUID, arvadosclient.Dict{"container": update}, nil)
 924 }
 925
 926 // IsCancelled returns the value of Cancelled, with goroutine safety.
 927 func (runner *ContainerRunner) IsCancelled() bool {
 928         runner.CancelLock.Lock()
 929         defer runner.CancelLock.Unlock()
 930         return runner.Cancelled
 931 }
 932
 933 // NewArvLogWriter creates an ArvLogWriter
 934 func (runner *ContainerRunner) NewArvLogWriter(name string) io.WriteCloser {
 935         return &ArvLogWriter{runner.ArvClient, runner.Container.UUID, name, runner.LogCollection.Open(name + ".txt")}
 936 }
 937
 938 // Run the full container lifecycle.
 939 func (runner *ContainerRunner) Run() (err error) {
 940         runner.CrunchLog.Printf("Executing container '%s'", runner.Container.UUID)
 941
 942         hostname, hosterr := os.Hostname()
 943         if hosterr != nil {
 944                 runner.CrunchLog.Printf("Error getting hostname '%v'", hosterr)
 945         } else {
 946                 runner.CrunchLog.Printf("Executing on host '%s'", hostname)
 947         }
 948
 949         // Clean up temporary directories _after_ finalizing
 950         // everything (if we've made any by then)
 951         defer runner.CleanupDirs()
 952
 953         runner.finalState = "Queued"
 954
 955         defer func() {
 956                 // checkErr prints e (unless it's nil) and sets err to
 957                 // e (unless err is already non-nil). Thus, if err
 958                 // hasn't already been assigned when Run() returns,
 959                 // this cleanup func will cause Run() to return the
 960                 // first non-nil error that is passed to checkErr().
 961                 checkErr := func(e error) {
 962                         if e == nil {
 963                                 return
 964                         }
 965                         runner.CrunchLog.Print(e)
 966                         if err == nil {
 967                                 err = e
 968                         }
 969                 }
 970
 971                 // Log the error encountered in Run(), if any
 972                 checkErr(err)
 973
 974                 if runner.finalState == "Queued" {
 975                         runner.CrunchLog.Close()
 976                         runner.UpdateContainerFinal()
 977                         return
 978                 }
 979
 980                 if runner.IsCancelled() {
 981                         runner.finalState = "Cancelled"
 982                         // but don't return yet -- we still want to
 983                         // capture partial output and write logs
 984                 }
 985
 986                 checkErr(runner.CaptureOutput())
 987                 checkErr(runner.CommitLogs())
 988                 checkErr(runner.UpdateContainerFinal())
 989
 990                 // The real log is already closed, but then we opened
 991                 // a new one in case we needed to log anything while
 992                 // finalizing.
 993                 runner.CrunchLog.Close()
 994         }()
 995
 996         err = runner.ArvClient.Get("containers", runner.Container.UUID, nil, &runner.Container)
 997         if err != nil {
 998                 err = fmt.Errorf("While getting container record: %v", err)
 999                 return
1000         }
1001
1002         // setup signal handling
1003         runner.SetupSignals()
1004
1005         // check for and/or load image
1006         err = runner.LoadImage()
1007         if err != nil {
1008                 runner.finalState = "Cancelled"
1009                 err = fmt.Errorf("While loading container image: %v", err)
1010                 return
1011         }
1012
1013         // set up FUSE mount and binds
1014         err = runner.SetupMounts()
1015         if err != nil {
1016                 runner.finalState = "Cancelled"
1017                 err = fmt.Errorf("While setting up mounts: %v", err)
1018                 return
1019         }
1020
1021         err = runner.CreateContainer()
1022         if err != nil {
1023                 return
1024         }
1025
1026         runner.StartCrunchstat()
1027
1028         if runner.IsCancelled() {
1029                 return
1030         }
1031
1032         err = runner.UpdateContainerRunning()
1033         if err != nil {
1034                 return
1035         }
1036         runner.finalState = "Cancelled"
1037
1038         err = runner.StartContainer()
1039         if err != nil {
1040                 return
1041         }
1042
1043         err = runner.WaitFinish()
1044         if err == nil {
1045                 runner.finalState = "Complete"
1046         }
1047         return
1048 }
1049
1050 // NewContainerRunner creates a new container runner.
1051 func NewContainerRunner(api IArvadosClient,
1052         kc IKeepClient,
1053         docker ThinDockerClient,
1054         containerUUID string) *ContainerRunner {
1055
1056         cr := &ContainerRunner{ArvClient: api, Kc: kc, Docker: docker}
1057         cr.NewLogWriter = cr.NewArvLogWriter
1058         cr.RunArvMount = cr.ArvMountCmd
1059         cr.MkTempDir = ioutil.TempDir
1060         cr.LogCollection = &CollectionWriter{kc, nil, sync.Mutex{}}
1061         cr.Container.UUID = containerUUID
1062         cr.CrunchLog = NewThrottledLogger(cr.NewLogWriter("crunch-run"))
1063         cr.CrunchLog.Immediate = log.New(os.Stderr, containerUUID+" ", 0)
1064         cr.loadDiscoveryVars()
1065         return cr
1066 }
1067
1068 func main() {
1069         statInterval := flag.Duration("crunchstat-interval", 10*time.Second, "sampling period for periodic resource usage reporting")
1070         cgroupRoot := flag.String("cgroup-root", "/sys/fs/cgroup", "path to sysfs cgroup tree")
1071         cgroupParent := flag.String("cgroup-parent", "docker", "name of container's parent cgroup (ignored if -cgroup-parent-subsystem is used)")
1072         cgroupParentSubsystem := flag.String("cgroup-parent-subsystem", "", "use current cgroup for given subsystem as parent cgroup for container")
1073         caCertsPath := flag.String("ca-certs", "", "Path to TLS root certificates")
1074         flag.Parse()
1075
1076         containerId := flag.Arg(0)
1077
1078         if *caCertsPath != "" {
1079                 arvadosclient.CertFiles = []string{*caCertsPath}
1080         }
1081
1082         api, err := arvadosclient.MakeArvadosClient()
1083         if err != nil {
1084                 log.Fatalf("%s: %v", containerId, err)
1085         }
1086         api.Retries = 8
1087
1088         var kc *keepclient.KeepClient
1089         kc, err = keepclient.MakeKeepClient(api)
1090         if err != nil {
1091                 log.Fatalf("%s: %v", containerId, err)
1092         }
1093         kc.Retries = 4
1094
1095         var docker *dockerclient.DockerClient
1096         docker, err = dockerclient.NewDockerClient("unix:///var/run/docker.sock", nil)
1097         if err != nil {
1098                 log.Fatalf("%s: %v", containerId, err)
1099         }
1100
1101         cr := NewContainerRunner(api, kc, docker, containerId)
1102         cr.statInterval = *statInterval
1103         cr.cgroupRoot = *cgroupRoot
1104         cr.expectCgroupParent = *cgroupParent
1105         if *cgroupParentSubsystem != "" {
1106                 p := findCgroup(*cgroupParentSubsystem)
1107                 cr.setCgroupParent = p
1108                 cr.expectCgroupParent = p
1109         }
1110
1111         err = cr.Run()
1112         if err != nil {
1113                 log.Fatalf("%s: %v", containerId, err)
1114         }
1115
1116 }