services/crunch-run/crunchrun.go

   1 package main
   2
   3 import (
   4         "bytes"
   5         "context"
   6         "encoding/json"
   7         "errors"
   8         "flag"
   9         "fmt"
  10         "io"
  11         "io/ioutil"
  12         "log"
  13         "os"
  14         "os/exec"
  15         "os/signal"
  16         "path"
  17         "path/filepath"
  18         "sort"
  19         "strings"
  20         "sync"
  21         "syscall"
  22         "time"
  23
  24         "git.curoverse.com/arvados.git/lib/crunchstat"
  25         "git.curoverse.com/arvados.git/sdk/go/arvados"
  26         "git.curoverse.com/arvados.git/sdk/go/arvadosclient"
  27         "git.curoverse.com/arvados.git/sdk/go/keepclient"
  28         "git.curoverse.com/arvados.git/sdk/go/manifest"
  29
  30         dockertypes "github.com/docker/docker/api/types"
  31         dockercontainer "github.com/docker/docker/api/types/container"
  32         dockernetwork "github.com/docker/docker/api/types/network"
  33         dockerclient "github.com/docker/docker/client"
  34 )
  35
  36 // IArvadosClient is the minimal Arvados API methods used by crunch-run.
  37 type IArvadosClient interface {
  38         Create(resourceType string, parameters arvadosclient.Dict, output interface{}) error
  39         Get(resourceType string, uuid string, parameters arvadosclient.Dict, output interface{}) error
  40         Update(resourceType string, uuid string, parameters arvadosclient.Dict, output interface{}) error
  41         Call(method, resourceType, uuid, action string, parameters arvadosclient.Dict, output interface{}) error
  42         CallRaw(method string, resourceType string, uuid string, action string, parameters arvadosclient.Dict) (reader io.ReadCloser, err error)
  43         Discovery(key string) (interface{}, error)
  44 }
  45
  46 // ErrCancelled is the error returned when the container is cancelled.
  47 var ErrCancelled = errors.New("Cancelled")
  48
  49 // IKeepClient is the minimal Keep API methods used by crunch-run.
  50 type IKeepClient interface {
  51         PutHB(hash string, buf []byte) (string, int, error)
  52         ManifestFileReader(m manifest.Manifest, filename string) (arvados.File, error)
  53 }
  54
  55 // NewLogWriter is a factory function to create a new log writer.
  56 type NewLogWriter func(name string) io.WriteCloser
  57
  58 type RunArvMount func(args []string, tok string) (*exec.Cmd, error)
  59
  60 type MkTempDir func(string, string) (string, error)
  61
  62 // ThinDockerClient is the minimal Docker client interface used by crunch-run.
  63 type ThinDockerClient interface {
  64         ContainerAttach(ctx context.Context, container string, options dockertypes.ContainerAttachOptions) (dockertypes.HijackedResponse, error)
  65         ContainerCreate(ctx context.Context, config *dockercontainer.Config, hostConfig *dockercontainer.HostConfig,
  66                 networkingConfig *dockernetwork.NetworkingConfig, containerName string) (dockercontainer.ContainerCreateCreatedBody, error)
  67         ContainerStart(ctx context.Context, container string, options dockertypes.ContainerStartOptions) error
  68         ContainerStop(ctx context.Context, container string, timeout *time.Duration) error
  69         ContainerWait(ctx context.Context, container string, condition dockercontainer.WaitCondition) (<-chan dockercontainer.ContainerWaitOKBody, <-chan error)
  70         ImageInspectWithRaw(ctx context.Context, image string) (dockertypes.ImageInspect, []byte, error)
  71         ImageLoad(ctx context.Context, input io.Reader, quiet bool) (dockertypes.ImageLoadResponse, error)
  72         ImageRemove(ctx context.Context, image string, options dockertypes.ImageRemoveOptions) ([]dockertypes.ImageDeleteResponseItem, error)
  73 }
  74
  75 // ThinDockerClientProxy is a proxy implementation of ThinDockerClient
  76 // that executes the docker requests on dockerclient.Client
  77 type ThinDockerClientProxy struct {
  78         Docker *dockerclient.Client
  79 }
  80
  81 // ContainerAttach invokes dockerclient.Client.ContainerAttach
  82 func (proxy ThinDockerClientProxy) ContainerAttach(ctx context.Context, container string, options dockertypes.ContainerAttachOptions) (dockertypes.HijackedResponse, error) {
  83         return proxy.Docker.ContainerAttach(ctx, container, options)
  84 }
  85
  86 // ContainerCreate invokes dockerclient.Client.ContainerCreate
  87 func (proxy ThinDockerClientProxy) ContainerCreate(ctx context.Context, config *dockercontainer.Config, hostConfig *dockercontainer.HostConfig,
  88         networkingConfig *dockernetwork.NetworkingConfig, containerName string) (dockercontainer.ContainerCreateCreatedBody, error) {
  89         return proxy.Docker.ContainerCreate(ctx, config, hostConfig, networkingConfig, containerName)
  90 }
  91
  92 // ContainerStart invokes dockerclient.Client.ContainerStart
  93 func (proxy ThinDockerClientProxy) ContainerStart(ctx context.Context, container string, options dockertypes.ContainerStartOptions) error {
  94         return proxy.Docker.ContainerStart(ctx, container, options)
  95 }
  96
  97 // ContainerStop invokes dockerclient.Client.ContainerStop
  98 func (proxy ThinDockerClientProxy) ContainerStop(ctx context.Context, container string, timeout *time.Duration) error {
  99         return proxy.Docker.ContainerStop(ctx, container, timeout)
 100 }
 101
 102 // ContainerWait invokes dockerclient.Client.ContainerWait
 103 func (proxy ThinDockerClientProxy) ContainerWait(ctx context.Context, container string, condition dockercontainer.WaitCondition) (<-chan dockercontainer.ContainerWaitOKBody, <-chan error) {
 104         return proxy.Docker.ContainerWait(ctx, container, condition)
 105 }
 106
 107 // ImageInspectWithRaw invokes dockerclient.Client.ImageInspectWithRaw
 108 func (proxy ThinDockerClientProxy) ImageInspectWithRaw(ctx context.Context, image string) (dockertypes.ImageInspect, []byte, error) {
 109         return proxy.Docker.ImageInspectWithRaw(ctx, image)
 110 }
 111
 112 // ImageLoad invokes dockerclient.Client.ImageLoad
 113 func (proxy ThinDockerClientProxy) ImageLoad(ctx context.Context, input io.Reader, quiet bool) (dockertypes.ImageLoadResponse, error) {
 114         return proxy.Docker.ImageLoad(ctx, input, quiet)
 115 }
 116
 117 // ImageRemove invokes dockerclient.Client.ImageRemove
 118 func (proxy ThinDockerClientProxy) ImageRemove(ctx context.Context, image string, options dockertypes.ImageRemoveOptions) ([]dockertypes.ImageDeleteResponseItem, error) {
 119         return proxy.Docker.ImageRemove(ctx, image, options)
 120 }
 121
 122 // ContainerRunner is the main stateful struct used for a single execution of a
 123 // container.
 124 type ContainerRunner struct {
 125         Docker    ThinDockerClient
 126         ArvClient IArvadosClient
 127         Kc        IKeepClient
 128         arvados.Container
 129         ContainerConfig dockercontainer.Config
 130         dockercontainer.HostConfig
 131         token       string
 132         ContainerID string
 133         ExitCode    *int
 134         NewLogWriter
 135         loggingDone   chan bool
 136         CrunchLog     *ThrottledLogger
 137         Stdout        io.WriteCloser
 138         Stderr        io.WriteCloser
 139         LogCollection *CollectionWriter
 140         LogsPDH       *string
 141         RunArvMount
 142         MkTempDir
 143         ArvMount       *exec.Cmd
 144         ArvMountPoint  string
 145         HostOutputDir  string
 146         CleanupTempDir []string
 147         Binds          []string
 148         Volumes        map[string]struct{}
 149         OutputPDH      *string
 150         SigChan        chan os.Signal
 151         ArvMountExit   chan error
 152         finalState     string
 153
 154         statLogger   io.WriteCloser
 155         statReporter *crunchstat.Reporter
 156         statInterval time.Duration
 157         cgroupRoot   string
 158         // What we expect the container's cgroup parent to be.
 159         expectCgroupParent string
 160         // What we tell docker to use as the container's cgroup
 161         // parent. Note: Ideally we would use the same field for both
 162         // expectCgroupParent and setCgroupParent, and just make it
 163         // default to "docker". However, when using docker < 1.10 with
 164         // systemd, specifying a non-empty cgroup parent (even the
 165         // default value "docker") hits a docker bug
 166         // (https://github.com/docker/docker/issues/17126). Using two
 167         // separate fields makes it possible to use the "expect cgroup
 168         // parent to be X" feature even on sites where the "specify
 169         // cgroup parent" feature breaks.
 170         setCgroupParent string
 171
 172         cStateLock sync.Mutex
 173         cStarted   bool // StartContainer() succeeded
 174         cCancelled bool // StopContainer() invoked
 175
 176         enableNetwork string // one of "default" or "always"
 177         networkMode   string // passed through to HostConfig.NetworkMode
 178 }
 179
 180 // SetupSignals sets up signal handling to gracefully terminate the underlying
 181 // Docker container and update state when receiving a TERM, INT or QUIT signal.
 182 func (runner *ContainerRunner) SetupSignals() {
 183         runner.SigChan = make(chan os.Signal, 1)
 184         signal.Notify(runner.SigChan, syscall.SIGTERM)
 185         signal.Notify(runner.SigChan, syscall.SIGINT)
 186         signal.Notify(runner.SigChan, syscall.SIGQUIT)
 187
 188         go func(sig chan os.Signal) {
 189                 <-sig
 190                 runner.stop()
 191                 signal.Stop(sig)
 192         }(runner.SigChan)
 193 }
 194
 195 // stop the underlying Docker container.
 196 func (runner *ContainerRunner) stop() {
 197         runner.cStateLock.Lock()
 198         defer runner.cStateLock.Unlock()
 199         if runner.cCancelled {
 200                 return
 201         }
 202         runner.cCancelled = true
 203         if runner.cStarted {
 204                 timeout := time.Duration(10)
 205                 err := runner.Docker.ContainerStop(context.TODO(), runner.ContainerID, &(timeout))
 206                 if err != nil {
 207                         log.Printf("StopContainer failed: %s", err)
 208                 }
 209         }
 210 }
 211
 212 // LoadImage determines the docker image id from the container record and
 213 // checks if it is available in the local Docker image store.  If not, it loads
 214 // the image from Keep.
 215 func (runner *ContainerRunner) LoadImage() (err error) {
 216
 217         runner.CrunchLog.Printf("Fetching Docker image from collection '%s'", runner.Container.ContainerImage)
 218
 219         var collection arvados.Collection
 220         err = runner.ArvClient.Get("collections", runner.Container.ContainerImage, nil, &collection)
 221         if err != nil {
 222                 return fmt.Errorf("While getting container image collection: %v", err)
 223         }
 224         manifest := manifest.Manifest{Text: collection.ManifestText}
 225         var img, imageID string
 226         for ms := range manifest.StreamIter() {
 227                 img = ms.FileStreamSegments[0].Name
 228                 if !strings.HasSuffix(img, ".tar") {
 229                         return fmt.Errorf("First file in the container image collection does not end in .tar")
 230                 }
 231                 imageID = img[:len(img)-4]
 232         }
 233
 234         runner.CrunchLog.Printf("Using Docker image id '%s'", imageID)
 235
 236         _, _, err = runner.Docker.ImageInspectWithRaw(context.TODO(), imageID)
 237         if err != nil {
 238                 runner.CrunchLog.Print("Loading Docker image from keep")
 239
 240                 var readCloser io.ReadCloser
 241                 readCloser, err = runner.Kc.ManifestFileReader(manifest, img)
 242                 if err != nil {
 243                         return fmt.Errorf("While creating ManifestFileReader for container image: %v", err)
 244                 }
 245
 246                 response, err := runner.Docker.ImageLoad(context.TODO(), readCloser, false)
 247                 if err != nil {
 248                         return fmt.Errorf("While loading container image into Docker: %v", err)
 249                 }
 250                 response.Body.Close()
 251         } else {
 252                 runner.CrunchLog.Print("Docker image is available")
 253         }
 254
 255         runner.ContainerConfig.Image = imageID
 256
 257         return nil
 258 }
 259
 260 func (runner *ContainerRunner) ArvMountCmd(arvMountCmd []string, token string) (c *exec.Cmd, err error) {
 261         c = exec.Command("arv-mount", arvMountCmd...)
 262
 263         // Copy our environment, but override ARVADOS_API_TOKEN with
 264         // the container auth token.
 265         c.Env = nil
 266         for _, s := range os.Environ() {
 267                 if !strings.HasPrefix(s, "ARVADOS_API_TOKEN=") {
 268                         c.Env = append(c.Env, s)
 269                 }
 270         }
 271         c.Env = append(c.Env, "ARVADOS_API_TOKEN="+token)
 272
 273         nt := NewThrottledLogger(runner.NewLogWriter("arv-mount"))
 274         c.Stdout = nt
 275         c.Stderr = nt
 276
 277         err = c.Start()
 278         if err != nil {
 279                 return nil, err
 280         }
 281
 282         statReadme := make(chan bool)
 283         runner.ArvMountExit = make(chan error)
 284
 285         keepStatting := true
 286         go func() {
 287                 for keepStatting {
 288                         time.Sleep(100 * time.Millisecond)
 289                         _, err = os.Stat(fmt.Sprintf("%s/by_id/README", runner.ArvMountPoint))
 290                         if err == nil {
 291                                 keepStatting = false
 292                                 statReadme <- true
 293                         }
 294                 }
 295                 close(statReadme)
 296         }()
 297
 298         go func() {
 299                 runner.ArvMountExit <- c.Wait()
 300                 close(runner.ArvMountExit)
 301         }()
 302
 303         select {
 304         case <-statReadme:
 305                 break
 306         case err := <-runner.ArvMountExit:
 307                 runner.ArvMount = nil
 308                 keepStatting = false
 309                 return nil, err
 310         }
 311
 312         return c, nil
 313 }
 314
 315 func (runner *ContainerRunner) SetupArvMountPoint(prefix string) (err error) {
 316         if runner.ArvMountPoint == "" {
 317                 runner.ArvMountPoint, err = runner.MkTempDir("", prefix)
 318         }
 319         return
 320 }
 321
 322 func (runner *ContainerRunner) SetupMounts() (err error) {
 323         err = runner.SetupArvMountPoint("keep")
 324         if err != nil {
 325                 return fmt.Errorf("While creating keep mount temp dir: %v", err)
 326         }
 327
 328         runner.CleanupTempDir = append(runner.CleanupTempDir, runner.ArvMountPoint)
 329
 330         pdhOnly := true
 331         tmpcount := 0
 332         arvMountCmd := []string{"--foreground", "--allow-other", "--read-write"}
 333
 334         if runner.Container.RuntimeConstraints.KeepCacheRAM > 0 {
 335                 arvMountCmd = append(arvMountCmd, "--file-cache", fmt.Sprintf("%d", runner.Container.RuntimeConstraints.KeepCacheRAM))
 336         }
 337
 338         collectionPaths := []string{}
 339         runner.Binds = nil
 340         runner.Volumes = make(map[string]struct{})
 341         needCertMount := true
 342
 343         var binds []string
 344         for bind, _ := range runner.Container.Mounts {
 345                 binds = append(binds, bind)
 346         }
 347         sort.Strings(binds)
 348
 349         for _, bind := range binds {
 350                 mnt := runner.Container.Mounts[bind]
 351                 if bind == "stdout" || bind == "stderr" {
 352                         // Is it a "file" mount kind?
 353                         if mnt.Kind != "file" {
 354                                 return fmt.Errorf("Unsupported mount kind '%s' for %s. Only 'file' is supported.", mnt.Kind, bind)
 355                         }
 356
 357                         // Does path start with OutputPath?
 358                         prefix := runner.Container.OutputPath
 359                         if !strings.HasSuffix(prefix, "/") {
 360                                 prefix += "/"
 361                         }
 362                         if !strings.HasPrefix(mnt.Path, prefix) {
 363                                 return fmt.Errorf("%s path does not start with OutputPath: %s, %s", strings.Title(bind), mnt.Path, prefix)
 364                         }
 365                 }
 366
 367                 if bind == "stdin" {
 368                         // Is it a "collection" mount kind?
 369                         if mnt.Kind != "collection" && mnt.Kind != "json" {
 370                                 return fmt.Errorf("Unsupported mount kind '%s' for stdin. Only 'collection' or 'json' are supported.", mnt.Kind)
 371                         }
 372                 }
 373
 374                 if bind == "/etc/arvados/ca-certificates.crt" {
 375                         needCertMount = false
 376                 }
 377
 378                 if strings.HasPrefix(bind, runner.Container.OutputPath+"/") && bind != runner.Container.OutputPath+"/" {
 379                         if mnt.Kind != "collection" {
 380                                 return fmt.Errorf("Only mount points of kind 'collection' are supported underneath the output_path: %v", bind)
 381                         }
 382                 }
 383
 384                 switch {
 385                 case mnt.Kind == "collection" && bind != "stdin":
 386                         var src string
 387                         if mnt.UUID != "" && mnt.PortableDataHash != "" {
 388                                 return fmt.Errorf("Cannot specify both 'uuid' and 'portable_data_hash' for a collection mount")
 389                         }
 390                         if mnt.UUID != "" {
 391                                 if mnt.Writable {
 392                                         return fmt.Errorf("Writing to existing collections currently not permitted.")
 393                                 }
 394                                 pdhOnly = false
 395                                 src = fmt.Sprintf("%s/by_id/%s", runner.ArvMountPoint, mnt.UUID)
 396                         } else if mnt.PortableDataHash != "" {
 397                                 if mnt.Writable {
 398                                         return fmt.Errorf("Can never write to a collection specified by portable data hash")
 399                                 }
 400                                 idx := strings.Index(mnt.PortableDataHash, "/")
 401                                 if idx > 0 {
 402                                         mnt.Path = path.Clean(mnt.PortableDataHash[idx:])
 403                                         mnt.PortableDataHash = mnt.PortableDataHash[0:idx]
 404                                         runner.Container.Mounts[bind] = mnt
 405                                 }
 406                                 src = fmt.Sprintf("%s/by_id/%s", runner.ArvMountPoint, mnt.PortableDataHash)
 407                                 if mnt.Path != "" && mnt.Path != "." {
 408                                         if strings.HasPrefix(mnt.Path, "./") {
 409                                                 mnt.Path = mnt.Path[2:]
 410                                         } else if strings.HasPrefix(mnt.Path, "/") {
 411                                                 mnt.Path = mnt.Path[1:]
 412                                         }
 413                                         src += "/" + mnt.Path
 414                                 }
 415                         } else {
 416                                 src = fmt.Sprintf("%s/tmp%d", runner.ArvMountPoint, tmpcount)
 417                                 arvMountCmd = append(arvMountCmd, "--mount-tmp")
 418                                 arvMountCmd = append(arvMountCmd, fmt.Sprintf("tmp%d", tmpcount))
 419                                 tmpcount += 1
 420                         }
 421                         if mnt.Writable {
 422                                 if bind == runner.Container.OutputPath {
 423                                         runner.HostOutputDir = src
 424                                 } else if strings.HasPrefix(bind, runner.Container.OutputPath+"/") {
 425                                         return fmt.Errorf("Writable mount points are not permitted underneath the output_path: %v", bind)
 426                                 }
 427                                 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s", src, bind))
 428                         } else {
 429                                 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s:ro", src, bind))
 430                         }
 431                         collectionPaths = append(collectionPaths, src)
 432
 433                 case mnt.Kind == "tmp":
 434                         var tmpdir string
 435                         tmpdir, err = runner.MkTempDir("", "")
 436                         if err != nil {
 437                                 return fmt.Errorf("While creating mount temp dir: %v", err)
 438                         }
 439                         st, staterr := os.Stat(tmpdir)
 440                         if staterr != nil {
 441                                 return fmt.Errorf("While Stat on temp dir: %v", staterr)
 442                         }
 443                         err = os.Chmod(tmpdir, st.Mode()|os.ModeSetgid|0777)
 444                         if staterr != nil {
 445                                 return fmt.Errorf("While Chmod temp dir: %v", err)
 446                         }
 447                         runner.CleanupTempDir = append(runner.CleanupTempDir, tmpdir)
 448                         runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s", tmpdir, bind))
 449                         if bind == runner.Container.OutputPath {
 450                                 runner.HostOutputDir = tmpdir
 451                         }
 452
 453                 case mnt.Kind == "json":
 454                         jsondata, err := json.Marshal(mnt.Content)
 455                         if err != nil {
 456                                 return fmt.Errorf("encoding json data: %v", err)
 457                         }
 458                         // Create a tempdir with a single file
 459                         // (instead of just a tempfile): this way we
 460                         // can ensure the file is world-readable
 461                         // inside the container, without having to
 462                         // make it world-readable on the docker host.
 463                         tmpdir, err := runner.MkTempDir("", "")
 464                         if err != nil {
 465                                 return fmt.Errorf("creating temp dir: %v", err)
 466                         }
 467                         runner.CleanupTempDir = append(runner.CleanupTempDir, tmpdir)
 468                         tmpfn := filepath.Join(tmpdir, "mountdata.json")
 469                         err = ioutil.WriteFile(tmpfn, jsondata, 0644)
 470                         if err != nil {
 471                                 return fmt.Errorf("writing temp file: %v", err)
 472                         }
 473                         runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s:ro", tmpfn, bind))
 474                 }
 475         }
 476
 477         if runner.HostOutputDir == "" {
 478                 return fmt.Errorf("Output path does not correspond to a writable mount point")
 479         }
 480
 481         if wantAPI := runner.Container.RuntimeConstraints.API; needCertMount && wantAPI != nil && *wantAPI {
 482                 for _, certfile := range arvadosclient.CertFiles {
 483                         _, err := os.Stat(certfile)
 484                         if err == nil {
 485                                 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:/etc/arvados/ca-certificates.crt:ro", certfile))
 486                                 break
 487                         }
 488                 }
 489         }
 490
 491         if pdhOnly {
 492                 arvMountCmd = append(arvMountCmd, "--mount-by-pdh", "by_id")
 493         } else {
 494                 arvMountCmd = append(arvMountCmd, "--mount-by-id", "by_id")
 495         }
 496         arvMountCmd = append(arvMountCmd, runner.ArvMountPoint)
 497
 498         token, err := runner.ContainerToken()
 499         if err != nil {
 500                 return fmt.Errorf("could not get container token: %s", err)
 501         }
 502
 503         runner.ArvMount, err = runner.RunArvMount(arvMountCmd, token)
 504         if err != nil {
 505                 return fmt.Errorf("While trying to start arv-mount: %v", err)
 506         }
 507
 508         for _, p := range collectionPaths {
 509                 _, err = os.Stat(p)
 510                 if err != nil {
 511                         return fmt.Errorf("While checking that input files exist: %v", err)
 512                 }
 513         }
 514
 515         return nil
 516 }
 517
 518 func (runner *ContainerRunner) ProcessDockerAttach(containerReader io.Reader) {
 519         // Handle docker log protocol
 520         // https://docs.docker.com/engine/reference/api/docker_remote_api_v1.15/#attach-to-a-container
 521
 522         header := make([]byte, 8)
 523         for {
 524                 _, readerr := io.ReadAtLeast(containerReader, header, 8)
 525
 526                 if readerr == nil {
 527                         readsize := int64(header[7]) | (int64(header[6]) << 8) | (int64(header[5]) << 16) | (int64(header[4]) << 24)
 528                         if header[0] == 1 {
 529                                 // stdout
 530                                 _, readerr = io.CopyN(runner.Stdout, containerReader, readsize)
 531                         } else {
 532                                 // stderr
 533                                 _, readerr = io.CopyN(runner.Stderr, containerReader, readsize)
 534                         }
 535                 }
 536
 537                 if readerr != nil {
 538                         if readerr != io.EOF {
 539                                 runner.CrunchLog.Printf("While reading docker logs: %v", readerr)
 540                         }
 541
 542                         closeerr := runner.Stdout.Close()
 543                         if closeerr != nil {
 544                                 runner.CrunchLog.Printf("While closing stdout logs: %v", closeerr)
 545                         }
 546
 547                         closeerr = runner.Stderr.Close()
 548                         if closeerr != nil {
 549                                 runner.CrunchLog.Printf("While closing stderr logs: %v", closeerr)
 550                         }
 551
 552                         if runner.statReporter != nil {
 553                                 runner.statReporter.Stop()
 554                                 closeerr = runner.statLogger.Close()
 555                                 if closeerr != nil {
 556                                         runner.CrunchLog.Printf("While closing crunchstat logs: %v", closeerr)
 557                                 }
 558                         }
 559
 560                         runner.loggingDone <- true
 561                         close(runner.loggingDone)
 562                         return
 563                 }
 564         }
 565 }
 566
 567 func (runner *ContainerRunner) StartCrunchstat() {
 568         runner.statLogger = NewThrottledLogger(runner.NewLogWriter("crunchstat"))
 569         runner.statReporter = &crunchstat.Reporter{
 570                 CID:          runner.ContainerID,
 571                 Logger:       log.New(runner.statLogger, "", 0),
 572                 CgroupParent: runner.expectCgroupParent,
 573                 CgroupRoot:   runner.cgroupRoot,
 574                 PollPeriod:   runner.statInterval,
 575         }
 576         runner.statReporter.Start()
 577 }
 578
 579 type infoCommand struct {
 580         label string
 581         cmd   []string
 582 }
 583
 584 // Gather node information and store it on the log for debugging
 585 // purposes.
 586 func (runner *ContainerRunner) LogNodeInfo() (err error) {
 587         w := runner.NewLogWriter("node-info")
 588         logger := log.New(w, "node-info", 0)
 589
 590         commands := []infoCommand{
 591                 infoCommand{
 592                         label: "Host Information",
 593                         cmd:   []string{"uname", "-a"},
 594                 },
 595                 infoCommand{
 596                         label: "CPU Information",
 597                         cmd:   []string{"cat", "/proc/cpuinfo"},
 598                 },
 599                 infoCommand{
 600                         label: "Memory Information",
 601                         cmd:   []string{"cat", "/proc/meminfo"},
 602                 },
 603                 infoCommand{
 604                         label: "Disk Space",
 605                         cmd:   []string{"df", "-m", "/", os.TempDir()},
 606                 },
 607                 infoCommand{
 608                         label: "Disk INodes",
 609                         cmd:   []string{"df", "-i", "/", os.TempDir()},
 610                 },
 611         }
 612
 613         // Run commands with informational output to be logged.
 614         var out []byte
 615         for _, command := range commands {
 616                 out, err = exec.Command(command.cmd[0], command.cmd[1:]...).CombinedOutput()
 617                 if err != nil {
 618                         return fmt.Errorf("While running command %q: %v",
 619                                 command.cmd, err)
 620                 }
 621                 logger.Println(command.label)
 622                 for _, line := range strings.Split(string(out), "\n") {
 623                         logger.Println(" ", line)
 624                 }
 625         }
 626
 627         err = w.Close()
 628         if err != nil {
 629                 return fmt.Errorf("While closing node-info logs: %v", err)
 630         }
 631         return nil
 632 }
 633
 634 // Get and save the raw JSON container record from the API server
 635 func (runner *ContainerRunner) LogContainerRecord() (err error) {
 636         w := &ArvLogWriter{
 637                 ArvClient:     runner.ArvClient,
 638                 UUID:          runner.Container.UUID,
 639                 loggingStream: "container",
 640                 writeCloser:   runner.LogCollection.Open("container.json"),
 641         }
 642
 643         // Get Container record JSON from the API Server
 644         reader, err := runner.ArvClient.CallRaw("GET", "containers", runner.Container.UUID, "", nil)
 645         if err != nil {
 646                 return fmt.Errorf("While retrieving container record from the API server: %v", err)
 647         }
 648         defer reader.Close()
 649         // Read the API server response as []byte
 650         json_bytes, err := ioutil.ReadAll(reader)
 651         if err != nil {
 652                 return fmt.Errorf("While reading container record API server response: %v", err)
 653         }
 654         // Decode the JSON []byte
 655         var cr map[string]interface{}
 656         if err = json.Unmarshal(json_bytes, &cr); err != nil {
 657                 return fmt.Errorf("While decoding the container record JSON response: %v", err)
 658         }
 659         // Re-encode it using indentation to improve readability
 660         enc := json.NewEncoder(w)
 661         enc.SetIndent("", "    ")
 662         if err = enc.Encode(cr); err != nil {
 663                 return fmt.Errorf("While logging the JSON container record: %v", err)
 664         }
 665         err = w.Close()
 666         if err != nil {
 667                 return fmt.Errorf("While closing container.json log: %v", err)
 668         }
 669         return nil
 670 }
 671
 672 // AttachStreams connects the docker container stdin, stdout and stderr logs
 673 // to the Arvados logger which logs to Keep and the API server logs table.
 674 func (runner *ContainerRunner) AttachStreams() (err error) {
 675
 676         runner.CrunchLog.Print("Attaching container streams")
 677
 678         // If stdin mount is provided, attach it to the docker container
 679         var stdinRdr arvados.File
 680         var stdinJson []byte
 681         if stdinMnt, ok := runner.Container.Mounts["stdin"]; ok {
 682                 if stdinMnt.Kind == "collection" {
 683                         var stdinColl arvados.Collection
 684                         collId := stdinMnt.UUID
 685                         if collId == "" {
 686                                 collId = stdinMnt.PortableDataHash
 687                         }
 688                         err = runner.ArvClient.Get("collections", collId, nil, &stdinColl)
 689                         if err != nil {
 690                                 return fmt.Errorf("While getting stding collection: %v", err)
 691                         }
 692
 693                         stdinRdr, err = runner.Kc.ManifestFileReader(manifest.Manifest{Text: stdinColl.ManifestText}, stdinMnt.Path)
 694                         if os.IsNotExist(err) {
 695                                 return fmt.Errorf("stdin collection path not found: %v", stdinMnt.Path)
 696                         } else if err != nil {
 697                                 return fmt.Errorf("While getting stdin collection path %v: %v", stdinMnt.Path, err)
 698                         }
 699                 } else if stdinMnt.Kind == "json" {
 700                         stdinJson, err = json.Marshal(stdinMnt.Content)
 701                         if err != nil {
 702                                 return fmt.Errorf("While encoding stdin json data: %v", err)
 703                         }
 704                 }
 705         }
 706
 707         stdinUsed := stdinRdr != nil || len(stdinJson) != 0
 708         response, err := runner.Docker.ContainerAttach(context.TODO(), runner.ContainerID,
 709                 dockertypes.ContainerAttachOptions{Stream: true, Stdin: stdinUsed, Stdout: true, Stderr: true})
 710         if err != nil {
 711                 return fmt.Errorf("While attaching container stdout/stderr streams: %v", err)
 712         }
 713
 714         runner.loggingDone = make(chan bool)
 715
 716         if stdoutMnt, ok := runner.Container.Mounts["stdout"]; ok {
 717                 stdoutFile, err := runner.getStdoutFile(stdoutMnt.Path)
 718                 if err != nil {
 719                         return err
 720                 }
 721                 runner.Stdout = stdoutFile
 722         } else {
 723                 runner.Stdout = NewThrottledLogger(runner.NewLogWriter("stdout"))
 724         }
 725
 726         if stderrMnt, ok := runner.Container.Mounts["stderr"]; ok {
 727                 stderrFile, err := runner.getStdoutFile(stderrMnt.Path)
 728                 if err != nil {
 729                         return err
 730                 }
 731                 runner.Stderr = stderrFile
 732         } else {
 733                 runner.Stderr = NewThrottledLogger(runner.NewLogWriter("stderr"))
 734         }
 735
 736         if stdinRdr != nil {
 737                 go func() {
 738                         _, err := io.Copy(response.Conn, stdinRdr)
 739                         if err != nil {
 740                                 runner.CrunchLog.Print("While writing stdin collection to docker container %q", err)
 741                                 runner.stop()
 742                         }
 743                         stdinRdr.Close()
 744                         response.CloseWrite()
 745                 }()
 746         } else if len(stdinJson) != 0 {
 747                 go func() {
 748                         _, err := io.Copy(response.Conn, bytes.NewReader(stdinJson))
 749                         if err != nil {
 750                                 runner.CrunchLog.Print("While writing stdin json to docker container %q", err)
 751                                 runner.stop()
 752                         }
 753                         response.CloseWrite()
 754                 }()
 755         }
 756
 757         go runner.ProcessDockerAttach(response.Reader)
 758
 759         return nil
 760 }
 761
 762 func (runner *ContainerRunner) getStdoutFile(mntPath string) (*os.File, error) {
 763         stdoutPath := mntPath[len(runner.Container.OutputPath):]
 764         index := strings.LastIndex(stdoutPath, "/")
 765         if index > 0 {
 766                 subdirs := stdoutPath[:index]
 767                 if subdirs != "" {
 768                         st, err := os.Stat(runner.HostOutputDir)
 769                         if err != nil {
 770                                 return nil, fmt.Errorf("While Stat on temp dir: %v", err)
 771                         }
 772                         stdoutPath := filepath.Join(runner.HostOutputDir, subdirs)
 773                         err = os.MkdirAll(stdoutPath, st.Mode()|os.ModeSetgid|0777)
 774                         if err != nil {
 775                                 return nil, fmt.Errorf("While MkdirAll %q: %v", stdoutPath, err)
 776                         }
 777                 }
 778         }
 779         stdoutFile, err := os.Create(filepath.Join(runner.HostOutputDir, stdoutPath))
 780         if err != nil {
 781                 return nil, fmt.Errorf("While creating file %q: %v", stdoutPath, err)
 782         }
 783
 784         return stdoutFile, nil
 785 }
 786
 787 // CreateContainer creates the docker container.
 788 func (runner *ContainerRunner) CreateContainer() error {
 789         runner.CrunchLog.Print("Creating Docker container")
 790
 791         runner.ContainerConfig.Cmd = runner.Container.Command
 792         if runner.Container.Cwd != "." {
 793                 runner.ContainerConfig.WorkingDir = runner.Container.Cwd
 794         }
 795
 796         for k, v := range runner.Container.Environment {
 797                 runner.ContainerConfig.Env = append(runner.ContainerConfig.Env, k+"="+v)
 798         }
 799
 800         runner.ContainerConfig.Volumes = runner.Volumes
 801
 802         runner.HostConfig = dockercontainer.HostConfig{
 803                 Binds: runner.Binds,
 804                 LogConfig: dockercontainer.LogConfig{
 805                         Type: "none",
 806                 },
 807                 Resources: dockercontainer.Resources{
 808                         CgroupParent: runner.setCgroupParent,
 809                 },
 810         }
 811
 812         if wantAPI := runner.Container.RuntimeConstraints.API; wantAPI != nil && *wantAPI {
 813                 tok, err := runner.ContainerToken()
 814                 if err != nil {
 815                         return err
 816                 }
 817                 runner.ContainerConfig.Env = append(runner.ContainerConfig.Env,
 818                         "ARVADOS_API_TOKEN="+tok,
 819                         "ARVADOS_API_HOST="+os.Getenv("ARVADOS_API_HOST"),
 820                         "ARVADOS_API_HOST_INSECURE="+os.Getenv("ARVADOS_API_HOST_INSECURE"),
 821                 )
 822                 runner.HostConfig.NetworkMode = dockercontainer.NetworkMode(runner.networkMode)
 823         } else {
 824                 if runner.enableNetwork == "always" {
 825                         runner.HostConfig.NetworkMode = dockercontainer.NetworkMode(runner.networkMode)
 826                 } else {
 827                         runner.HostConfig.NetworkMode = dockercontainer.NetworkMode("none")
 828                 }
 829         }
 830
 831         _, stdinUsed := runner.Container.Mounts["stdin"]
 832         runner.ContainerConfig.OpenStdin = stdinUsed
 833         runner.ContainerConfig.StdinOnce = stdinUsed
 834         runner.ContainerConfig.AttachStdin = stdinUsed
 835         runner.ContainerConfig.AttachStdout = true
 836         runner.ContainerConfig.AttachStderr = true
 837
 838         createdBody, err := runner.Docker.ContainerCreate(context.TODO(), &runner.ContainerConfig, &runner.HostConfig, nil, runner.Container.UUID)
 839         if err != nil {
 840                 return fmt.Errorf("While creating container: %v", err)
 841         }
 842
 843         runner.ContainerID = createdBody.ID
 844
 845         return runner.AttachStreams()
 846 }
 847
 848 // StartContainer starts the docker container created by CreateContainer.
 849 func (runner *ContainerRunner) StartContainer() error {
 850         runner.CrunchLog.Printf("Starting Docker container id '%s'", runner.ContainerID)
 851         runner.cStateLock.Lock()
 852         defer runner.cStateLock.Unlock()
 853         if runner.cCancelled {
 854                 return ErrCancelled
 855         }
 856         err := runner.Docker.ContainerStart(context.TODO(), runner.ContainerID,
 857                 dockertypes.ContainerStartOptions{})
 858         if err != nil {
 859                 return fmt.Errorf("could not start container: %v", err)
 860         }
 861         runner.cStarted = true
 862         return nil
 863 }
 864
 865 // WaitFinish waits for the container to terminate, capture the exit code, and
 866 // close the stdout/stderr logging.
 867 func (runner *ContainerRunner) WaitFinish() (err error) {
 868         runner.CrunchLog.Print("Waiting for container to finish")
 869
 870         waitOk, waitErr := runner.Docker.ContainerWait(context.TODO(), runner.ContainerID, "not-running")
 871
 872         var waitBody dockercontainer.ContainerWaitOKBody
 873         select {
 874         case waitBody = <-waitOk:
 875         case err = <-waitErr:
 876         }
 877
 878         if err != nil {
 879                 return fmt.Errorf("container wait: %v", err)
 880         }
 881
 882         runner.CrunchLog.Printf("Container exited with code: %v", waitBody.StatusCode)
 883         code := int(waitBody.StatusCode)
 884         runner.ExitCode = &code
 885
 886         waitMount := runner.ArvMountExit
 887         select {
 888         case err = <-waitMount:
 889                 runner.CrunchLog.Printf("arv-mount exited before container finished: %v", err)
 890                 waitMount = nil
 891                 runner.stop()
 892         default:
 893         }
 894
 895         // wait for stdout/stderr to complete
 896         <-runner.loggingDone
 897
 898         return nil
 899 }
 900
 901 // HandleOutput sets the output, unmounts the FUSE mount, and deletes temporary directories
 902 func (runner *ContainerRunner) CaptureOutput() error {
 903         if runner.finalState != "Complete" {
 904                 return nil
 905         }
 906
 907         if wantAPI := runner.Container.RuntimeConstraints.API; wantAPI != nil && *wantAPI {
 908                 // Output may have been set directly by the container, so
 909                 // refresh the container record to check.
 910                 err := runner.ArvClient.Get("containers", runner.Container.UUID,
 911                         nil, &runner.Container)
 912                 if err != nil {
 913                         return err
 914                 }
 915                 if runner.Container.Output != "" {
 916                         // Container output is already set.
 917                         runner.OutputPDH = &runner.Container.Output
 918                         return nil
 919                 }
 920         }
 921
 922         if runner.HostOutputDir == "" {
 923                 return nil
 924         }
 925
 926         _, err := os.Stat(runner.HostOutputDir)
 927         if err != nil {
 928                 return fmt.Errorf("While checking host output path: %v", err)
 929         }
 930
 931         // Pre-populate output from the configured mount points
 932         var binds []string
 933         for bind, mnt := range runner.Container.Mounts {
 934                 if mnt.Kind == "collection" {
 935                         binds = append(binds, bind)
 936                 }
 937         }
 938         sort.Strings(binds)
 939
 940         var manifestText string
 941
 942         collectionMetafile := fmt.Sprintf("%s/.arvados#collection", runner.HostOutputDir)
 943         _, err = os.Stat(collectionMetafile)
 944         if err != nil {
 945                 // Regular directory
 946
 947                 // Find symlinks to arv-mounted files & dirs.
 948                 err = filepath.Walk(runner.HostOutputDir, func(path string, info os.FileInfo, err error) error {
 949                         if err != nil {
 950                                 return err
 951                         }
 952                         if info.Mode()&os.ModeSymlink == 0 {
 953                                 return nil
 954                         }
 955                         // read link to get container internal path
 956                         // only support 1 level of symlinking here.
 957                         var tgt string
 958                         tgt, err = os.Readlink(path)
 959                         if err != nil {
 960                                 return err
 961                         }
 962
 963                         // get path relative to output dir
 964                         outputSuffix := path[len(runner.HostOutputDir):]
 965
 966                         if strings.HasPrefix(tgt, "/") {
 967                                 // go through mounts and try reverse map to collection reference
 968                                 for _, bind := range binds {
 969                                         mnt := runner.Container.Mounts[bind]
 970                                         if tgt == bind || strings.HasPrefix(tgt, bind+"/") {
 971                                                 // get path relative to bind
 972                                                 targetSuffix := tgt[len(bind):]
 973
 974                                                 // Copy mount and adjust the path to add path relative to the bind
 975                                                 adjustedMount := mnt
 976                                                 adjustedMount.Path = filepath.Join(adjustedMount.Path, targetSuffix)
 977
 978                                                 // get manifest text
 979                                                 var m string
 980                                                 m, err = runner.getCollectionManifestForPath(adjustedMount, outputSuffix)
 981                                                 if err != nil {
 982                                                         return err
 983                                                 }
 984                                                 manifestText = manifestText + m
 985                                                 // delete symlink so WriteTree won't try to to dereference it.
 986                                                 os.Remove(path)
 987                                                 return nil
 988                                         }
 989                                 }
 990                         }
 991
 992                         // Not a link to a mount.  Must be dereferencible and
 993                         // point into the output directory.
 994                         tgt, err = filepath.EvalSymlinks(path)
 995                         if err != nil {
 996                                 os.Remove(path)
 997                                 return err
 998                         }
 999
1000                         // Symlink target must be within the output directory otherwise it's an error.
1001                         if !strings.HasPrefix(tgt, runner.HostOutputDir+"/") {
1002                                 os.Remove(path)
1003                                 return fmt.Errorf("Output directory symlink %q points to invalid location %q, must point to mount or output directory.",
1004                                         outputSuffix, tgt)
1005                         }
1006                         return nil
1007                 })
1008                 if err != nil {
1009                         return fmt.Errorf("While checking output symlinks: %v", err)
1010                 }
1011
1012                 cw := CollectionWriter{0, runner.Kc, nil, nil, sync.Mutex{}}
1013                 var m string
1014                 m, err = cw.WriteTree(runner.HostOutputDir, runner.CrunchLog.Logger)
1015                 manifestText = manifestText + m
1016                 if err != nil {
1017                         return fmt.Errorf("While uploading output files: %v", err)
1018                 }
1019         } else {
1020                 // FUSE mount directory
1021                 file, openerr := os.Open(collectionMetafile)
1022                 if openerr != nil {
1023                         return fmt.Errorf("While opening FUSE metafile: %v", err)
1024                 }
1025                 defer file.Close()
1026
1027                 var rec arvados.Collection
1028                 err = json.NewDecoder(file).Decode(&rec)
1029                 if err != nil {
1030                         return fmt.Errorf("While reading FUSE metafile: %v", err)
1031                 }
1032                 manifestText = rec.ManifestText
1033         }
1034
1035         for _, bind := range binds {
1036                 mnt := runner.Container.Mounts[bind]
1037
1038                 bindSuffix := strings.TrimPrefix(bind, runner.Container.OutputPath)
1039
1040                 if bindSuffix == bind || len(bindSuffix) <= 0 {
1041                         // either does not start with OutputPath or is OutputPath itself
1042                         continue
1043                 }
1044
1045                 if mnt.ExcludeFromOutput == true {
1046                         continue
1047                 }
1048
1049                 // append to manifest_text
1050                 m, err := runner.getCollectionManifestForPath(mnt, bindSuffix)
1051                 if err != nil {
1052                         return err
1053                 }
1054
1055                 manifestText = manifestText + m
1056         }
1057
1058         // Save output
1059         var response arvados.Collection
1060         manifest := manifest.Manifest{Text: manifestText}
1061         manifestText = manifest.Extract(".", ".").Text
1062         err = runner.ArvClient.Create("collections",
1063                 arvadosclient.Dict{
1064                         "ensure_unique_name": true,
1065                         "collection": arvadosclient.Dict{
1066                                 "is_trashed":    true,
1067                                 "name":          "output for " + runner.Container.UUID,
1068                                 "manifest_text": manifestText}},
1069                 &response)
1070         if err != nil {
1071                 return fmt.Errorf("While creating output collection: %v", err)
1072         }
1073         runner.OutputPDH = &response.PortableDataHash
1074         return nil
1075 }
1076
1077 var outputCollections = make(map[string]arvados.Collection)
1078
1079 // Fetch the collection for the mnt.PortableDataHash
1080 // Return the manifest_text fragment corresponding to the specified mnt.Path
1081 //  after making any required updates.
1082 //  Ex:
1083 //    If mnt.Path is not specified,
1084 //      return the entire manifest_text after replacing any "." with bindSuffix
1085 //    If mnt.Path corresponds to one stream,
1086 //      return the manifest_text for that stream after replacing that stream name with bindSuffix
1087 //    Otherwise, check if a filename in any one stream is being sought. Return the manifest_text
1088 //      for that stream after replacing stream name with bindSuffix minus the last word
1089 //      and the file name with last word of the bindSuffix
1090 //  Allowed path examples:
1091 //    "path":"/"
1092 //    "path":"/subdir1"
1093 //    "path":"/subdir1/subdir2"
1094 //    "path":"/subdir/filename" etc
1095 func (runner *ContainerRunner) getCollectionManifestForPath(mnt arvados.Mount, bindSuffix string) (string, error) {
1096         collection := outputCollections[mnt.PortableDataHash]
1097         if collection.PortableDataHash == "" {
1098                 err := runner.ArvClient.Get("collections", mnt.PortableDataHash, nil, &collection)
1099                 if err != nil {
1100                         return "", fmt.Errorf("While getting collection for %v: %v", mnt.PortableDataHash, err)
1101                 }
1102                 outputCollections[mnt.PortableDataHash] = collection
1103         }
1104
1105         if collection.ManifestText == "" {
1106                 runner.CrunchLog.Printf("No manifest text for collection %v", collection.PortableDataHash)
1107                 return "", nil
1108         }
1109
1110         mft := manifest.Manifest{Text: collection.ManifestText}
1111         extracted := mft.Extract(mnt.Path, bindSuffix)
1112         if extracted.Err != nil {
1113                 return "", fmt.Errorf("Error parsing manifest for %v: %v", mnt.PortableDataHash, extracted.Err.Error())
1114         }
1115         return extracted.Text, nil
1116 }
1117
1118 func (runner *ContainerRunner) CleanupDirs() {
1119         if runner.ArvMount != nil {
1120                 umount := exec.Command("fusermount", "-z", "-u", runner.ArvMountPoint)
1121                 umnterr := umount.Run()
1122                 if umnterr != nil {
1123                         runner.CrunchLog.Printf("While running fusermount: %v", umnterr)
1124                 }
1125
1126                 mnterr := <-runner.ArvMountExit
1127                 if mnterr != nil {
1128                         runner.CrunchLog.Printf("Arv-mount exit error: %v", mnterr)
1129                 }
1130         }
1131
1132         for _, tmpdir := range runner.CleanupTempDir {
1133                 rmerr := os.RemoveAll(tmpdir)
1134                 if rmerr != nil {
1135                         runner.CrunchLog.Printf("While cleaning up temporary directory %s: %v", tmpdir, rmerr)
1136                 }
1137         }
1138 }
1139
1140 // CommitLogs posts the collection containing the final container logs.
1141 func (runner *ContainerRunner) CommitLogs() error {
1142         runner.CrunchLog.Print(runner.finalState)
1143         runner.CrunchLog.Close()
1144
1145         // Closing CrunchLog above allows it to be committed to Keep at this
1146         // point, but re-open crunch log with ArvClient in case there are any
1147         // other further (such as failing to write the log to Keep!) while
1148         // shutting down
1149         runner.CrunchLog = NewThrottledLogger(&ArvLogWriter{ArvClient: runner.ArvClient,
1150                 UUID: runner.Container.UUID, loggingStream: "crunch-run", writeCloser: nil})
1151
1152         if runner.LogsPDH != nil {
1153                 // If we have already assigned something to LogsPDH,
1154                 // we must be closing the re-opened log, which won't
1155                 // end up getting attached to the container record and
1156                 // therefore doesn't need to be saved as a collection
1157                 // -- it exists only to send logs to other channels.
1158                 return nil
1159         }
1160
1161         mt, err := runner.LogCollection.ManifestText()
1162         if err != nil {
1163                 return fmt.Errorf("While creating log manifest: %v", err)
1164         }
1165
1166         var response arvados.Collection
1167         err = runner.ArvClient.Create("collections",
1168                 arvadosclient.Dict{
1169                         "ensure_unique_name": true,
1170                         "collection": arvadosclient.Dict{
1171                                 "is_trashed":    true,
1172                                 "name":          "logs for " + runner.Container.UUID,
1173                                 "manifest_text": mt}},
1174                 &response)
1175         if err != nil {
1176                 return fmt.Errorf("While creating log collection: %v", err)
1177         }
1178         runner.LogsPDH = &response.PortableDataHash
1179         return nil
1180 }
1181
1182 // UpdateContainerRunning updates the container state to "Running"
1183 func (runner *ContainerRunner) UpdateContainerRunning() error {
1184         runner.cStateLock.Lock()
1185         defer runner.cStateLock.Unlock()
1186         if runner.cCancelled {
1187                 return ErrCancelled
1188         }
1189         return runner.ArvClient.Update("containers", runner.Container.UUID,
1190                 arvadosclient.Dict{"container": arvadosclient.Dict{"state": "Running"}}, nil)
1191 }
1192
1193 // ContainerToken returns the api_token the container (and any
1194 // arv-mount processes) are allowed to use.
1195 func (runner *ContainerRunner) ContainerToken() (string, error) {
1196         if runner.token != "" {
1197                 return runner.token, nil
1198         }
1199
1200         var auth arvados.APIClientAuthorization
1201         err := runner.ArvClient.Call("GET", "containers", runner.Container.UUID, "auth", nil, &auth)
1202         if err != nil {
1203                 return "", err
1204         }
1205         runner.token = auth.APIToken
1206         return runner.token, nil
1207 }
1208
1209 // UpdateContainerComplete updates the container record state on API
1210 // server to "Complete" or "Cancelled"
1211 func (runner *ContainerRunner) UpdateContainerFinal() error {
1212         update := arvadosclient.Dict{}
1213         update["state"] = runner.finalState
1214         if runner.LogsPDH != nil {
1215                 update["log"] = *runner.LogsPDH
1216         }
1217         if runner.finalState == "Complete" {
1218                 if runner.ExitCode != nil {
1219                         update["exit_code"] = *runner.ExitCode
1220                 }
1221                 if runner.OutputPDH != nil {
1222                         update["output"] = *runner.OutputPDH
1223                 }
1224         }
1225         return runner.ArvClient.Update("containers", runner.Container.UUID, arvadosclient.Dict{"container": update}, nil)
1226 }
1227
1228 // IsCancelled returns the value of Cancelled, with goroutine safety.
1229 func (runner *ContainerRunner) IsCancelled() bool {
1230         runner.cStateLock.Lock()
1231         defer runner.cStateLock.Unlock()
1232         return runner.cCancelled
1233 }
1234
1235 // NewArvLogWriter creates an ArvLogWriter
1236 func (runner *ContainerRunner) NewArvLogWriter(name string) io.WriteCloser {
1237         return &ArvLogWriter{ArvClient: runner.ArvClient, UUID: runner.Container.UUID, loggingStream: name,
1238                 writeCloser: runner.LogCollection.Open(name + ".txt")}
1239 }
1240
1241 // Run the full container lifecycle.
1242 func (runner *ContainerRunner) Run() (err error) {
1243         runner.CrunchLog.Printf("Executing container '%s'", runner.Container.UUID)
1244
1245         hostname, hosterr := os.Hostname()
1246         if hosterr != nil {
1247                 runner.CrunchLog.Printf("Error getting hostname '%v'", hosterr)
1248         } else {
1249                 runner.CrunchLog.Printf("Executing on host '%s'", hostname)
1250         }
1251
1252         // Clean up temporary directories _after_ finalizing
1253         // everything (if we've made any by then)
1254         defer runner.CleanupDirs()
1255
1256         runner.finalState = "Queued"
1257
1258         defer func() {
1259                 // checkErr prints e (unless it's nil) and sets err to
1260                 // e (unless err is already non-nil). Thus, if err
1261                 // hasn't already been assigned when Run() returns,
1262                 // this cleanup func will cause Run() to return the
1263                 // first non-nil error that is passed to checkErr().
1264                 checkErr := func(e error) {
1265                         if e == nil {
1266                                 return
1267                         }
1268                         runner.CrunchLog.Print(e)
1269                         if err == nil {
1270                                 err = e
1271                         }
1272                         if runner.finalState == "Complete" {
1273                                 // There was an error in the finalization.
1274                                 runner.finalState = "Cancelled"
1275                         }
1276                 }
1277
1278                 // Log the error encountered in Run(), if any
1279                 checkErr(err)
1280
1281                 if runner.finalState == "Queued" {
1282                         runner.CrunchLog.Close()
1283                         runner.UpdateContainerFinal()
1284                         return
1285                 }
1286
1287                 if runner.IsCancelled() {
1288                         runner.finalState = "Cancelled"
1289                         // but don't return yet -- we still want to
1290                         // capture partial output and write logs
1291                 }
1292
1293                 checkErr(runner.CaptureOutput())
1294                 checkErr(runner.CommitLogs())
1295                 checkErr(runner.UpdateContainerFinal())
1296
1297                 // The real log is already closed, but then we opened
1298                 // a new one in case we needed to log anything while
1299                 // finalizing.
1300                 runner.CrunchLog.Close()
1301         }()
1302
1303         err = runner.ArvClient.Get("containers", runner.Container.UUID, nil, &runner.Container)
1304         if err != nil {
1305                 err = fmt.Errorf("While getting container record: %v", err)
1306                 return
1307         }
1308
1309         // setup signal handling
1310         runner.SetupSignals()
1311
1312         // check for and/or load image
1313         err = runner.LoadImage()
1314         if err != nil {
1315                 runner.finalState = "Cancelled"
1316                 err = fmt.Errorf("While loading container image: %v", err)
1317                 return
1318         }
1319
1320         // set up FUSE mount and binds
1321         err = runner.SetupMounts()
1322         if err != nil {
1323                 runner.finalState = "Cancelled"
1324                 err = fmt.Errorf("While setting up mounts: %v", err)
1325                 return
1326         }
1327
1328         err = runner.CreateContainer()
1329         if err != nil {
1330                 return
1331         }
1332
1333         // Gather and record node information
1334         err = runner.LogNodeInfo()
1335         if err != nil {
1336                 return
1337         }
1338         // Save container.json record on log collection
1339         err = runner.LogContainerRecord()
1340         if err != nil {
1341                 return
1342         }
1343
1344         runner.StartCrunchstat()
1345
1346         if runner.IsCancelled() {
1347                 return
1348         }
1349
1350         err = runner.UpdateContainerRunning()
1351         if err != nil {
1352                 return
1353         }
1354         runner.finalState = "Cancelled"
1355
1356         err = runner.StartContainer()
1357         if err != nil {
1358                 return
1359         }
1360
1361         err = runner.WaitFinish()
1362         if err == nil {
1363                 runner.finalState = "Complete"
1364         }
1365         return
1366 }
1367
1368 // NewContainerRunner creates a new container runner.
1369 func NewContainerRunner(api IArvadosClient,
1370         kc IKeepClient,
1371         docker ThinDockerClient,
1372         containerUUID string) *ContainerRunner {
1373
1374         cr := &ContainerRunner{ArvClient: api, Kc: kc, Docker: docker}
1375         cr.NewLogWriter = cr.NewArvLogWriter
1376         cr.RunArvMount = cr.ArvMountCmd
1377         cr.MkTempDir = ioutil.TempDir
1378         cr.LogCollection = &CollectionWriter{0, kc, nil, nil, sync.Mutex{}}
1379         cr.Container.UUID = containerUUID
1380         cr.CrunchLog = NewThrottledLogger(cr.NewLogWriter("crunch-run"))
1381         cr.CrunchLog.Immediate = log.New(os.Stderr, containerUUID+" ", 0)
1382
1383         loadLogThrottleParams(api)
1384
1385         return cr
1386 }
1387
1388 func main() {
1389         statInterval := flag.Duration("crunchstat-interval", 10*time.Second, "sampling period for periodic resource usage reporting")
1390         cgroupRoot := flag.String("cgroup-root", "/sys/fs/cgroup", "path to sysfs cgroup tree")
1391         cgroupParent := flag.String("cgroup-parent", "docker", "name of container's parent cgroup (ignored if -cgroup-parent-subsystem is used)")
1392         cgroupParentSubsystem := flag.String("cgroup-parent-subsystem", "", "use current cgroup for given subsystem as parent cgroup for container")
1393         caCertsPath := flag.String("ca-certs", "", "Path to TLS root certificates")
1394         enableNetwork := flag.String("container-enable-networking", "default",
1395                 `Specify if networking should be enabled for container.  One of 'default', 'always':
1396         default: only enable networking if container requests it.
1397         always:  containers always have networking enabled
1398         `)
1399         networkMode := flag.String("container-network-mode", "default",
1400                 `Set networking mode for container.  Corresponds to Docker network mode (--net).
1401         `)
1402         flag.Parse()
1403
1404         containerId := flag.Arg(0)
1405
1406         if *caCertsPath != "" {
1407                 arvadosclient.CertFiles = []string{*caCertsPath}
1408         }
1409
1410         api, err := arvadosclient.MakeArvadosClient()
1411         if err != nil {
1412                 log.Fatalf("%s: %v", containerId, err)
1413         }
1414         api.Retries = 8
1415
1416         var kc *keepclient.KeepClient
1417         kc, err = keepclient.MakeKeepClient(api)
1418         if err != nil {
1419                 log.Fatalf("%s: %v", containerId, err)
1420         }
1421         kc.Retries = 4
1422
1423         var docker *dockerclient.Client
1424         // API version 1.21 corresponds to Docker 1.9, which is currently the
1425         // minimum version we want to support.
1426         docker, err = dockerclient.NewClient(dockerclient.DefaultDockerHost, "1.21", nil, nil)
1427         if err != nil {
1428                 log.Fatalf("%s: %v", containerId, err)
1429         }
1430
1431         dockerClientProxy := ThinDockerClientProxy{Docker: docker}
1432
1433         cr := NewContainerRunner(api, kc, dockerClientProxy, containerId)
1434         cr.statInterval = *statInterval
1435         cr.cgroupRoot = *cgroupRoot
1436         cr.expectCgroupParent = *cgroupParent
1437         cr.enableNetwork = *enableNetwork
1438         cr.networkMode = *networkMode
1439         if *cgroupParentSubsystem != "" {
1440                 p := findCgroup(*cgroupParentSubsystem)
1441                 cr.setCgroupParent = p
1442                 cr.expectCgroupParent = p
1443         }
1444
1445         err = cr.Run()
1446         if err != nil {
1447                 log.Fatalf("%s: %v", containerId, err)
1448         }
1449
1450 }