services/crunch-run/crunchrun.go

   1 // Copyright (C) The Arvados Authors. All rights reserved.
   2 //
   3 // SPDX-License-Identifier: AGPL-3.0
   4
   5 package main
   6
   7 import (
   8         "bytes"
   9         "context"
  10         "encoding/json"
  11         "errors"
  12         "flag"
  13         "fmt"
  14         "io"
  15         "io/ioutil"
  16         "log"
  17         "os"
  18         "os/exec"
  19         "os/signal"
  20         "path"
  21         "path/filepath"
  22         "runtime"
  23         "runtime/pprof"
  24         "sort"
  25         "strings"
  26         "sync"
  27         "syscall"
  28         "time"
  29
  30         "git.curoverse.com/arvados.git/lib/crunchstat"
  31         "git.curoverse.com/arvados.git/sdk/go/arvados"
  32         "git.curoverse.com/arvados.git/sdk/go/arvadosclient"
  33         "git.curoverse.com/arvados.git/sdk/go/keepclient"
  34         "git.curoverse.com/arvados.git/sdk/go/manifest"
  35
  36         dockertypes "github.com/docker/docker/api/types"
  37         dockercontainer "github.com/docker/docker/api/types/container"
  38         dockernetwork "github.com/docker/docker/api/types/network"
  39         dockerclient "github.com/docker/docker/client"
  40 )
  41
  42 var version = "dev"
  43
  44 // IArvadosClient is the minimal Arvados API methods used by crunch-run.
  45 type IArvadosClient interface {
  46         Create(resourceType string, parameters arvadosclient.Dict, output interface{}) error
  47         Get(resourceType string, uuid string, parameters arvadosclient.Dict, output interface{}) error
  48         Update(resourceType string, uuid string, parameters arvadosclient.Dict, output interface{}) error
  49         Call(method, resourceType, uuid, action string, parameters arvadosclient.Dict, output interface{}) error
  50         CallRaw(method string, resourceType string, uuid string, action string, parameters arvadosclient.Dict) (reader io.ReadCloser, err error)
  51         Discovery(key string) (interface{}, error)
  52 }
  53
  54 // ErrCancelled is the error returned when the container is cancelled.
  55 var ErrCancelled = errors.New("Cancelled")
  56
  57 // IKeepClient is the minimal Keep API methods used by crunch-run.
  58 type IKeepClient interface {
  59         PutHB(hash string, buf []byte) (string, int, error)
  60         ManifestFileReader(m manifest.Manifest, filename string) (arvados.File, error)
  61         ClearBlockCache()
  62 }
  63
  64 // NewLogWriter is a factory function to create a new log writer.
  65 type NewLogWriter func(name string) io.WriteCloser
  66
  67 type RunArvMount func(args []string, tok string) (*exec.Cmd, error)
  68
  69 type MkTempDir func(string, string) (string, error)
  70
  71 // ThinDockerClient is the minimal Docker client interface used by crunch-run.
  72 type ThinDockerClient interface {
  73         ContainerAttach(ctx context.Context, container string, options dockertypes.ContainerAttachOptions) (dockertypes.HijackedResponse, error)
  74         ContainerCreate(ctx context.Context, config *dockercontainer.Config, hostConfig *dockercontainer.HostConfig,
  75                 networkingConfig *dockernetwork.NetworkingConfig, containerName string) (dockercontainer.ContainerCreateCreatedBody, error)
  76         ContainerStart(ctx context.Context, container string, options dockertypes.ContainerStartOptions) error
  77         ContainerStop(ctx context.Context, container string, timeout *time.Duration) error
  78         ContainerWait(ctx context.Context, container string, condition dockercontainer.WaitCondition) (<-chan dockercontainer.ContainerWaitOKBody, <-chan error)
  79         ImageInspectWithRaw(ctx context.Context, image string) (dockertypes.ImageInspect, []byte, error)
  80         ImageLoad(ctx context.Context, input io.Reader, quiet bool) (dockertypes.ImageLoadResponse, error)
  81         ImageRemove(ctx context.Context, image string, options dockertypes.ImageRemoveOptions) ([]dockertypes.ImageDeleteResponseItem, error)
  82 }
  83
  84 // ThinDockerClientProxy is a proxy implementation of ThinDockerClient
  85 // that executes the docker requests on dockerclient.Client
  86 type ThinDockerClientProxy struct {
  87         Docker *dockerclient.Client
  88 }
  89
  90 // ContainerAttach invokes dockerclient.Client.ContainerAttach
  91 func (proxy ThinDockerClientProxy) ContainerAttach(ctx context.Context, container string, options dockertypes.ContainerAttachOptions) (dockertypes.HijackedResponse, error) {
  92         return proxy.Docker.ContainerAttach(ctx, container, options)
  93 }
  94
  95 // ContainerCreate invokes dockerclient.Client.ContainerCreate
  96 func (proxy ThinDockerClientProxy) ContainerCreate(ctx context.Context, config *dockercontainer.Config, hostConfig *dockercontainer.HostConfig,
  97         networkingConfig *dockernetwork.NetworkingConfig, containerName string) (dockercontainer.ContainerCreateCreatedBody, error) {
  98         return proxy.Docker.ContainerCreate(ctx, config, hostConfig, networkingConfig, containerName)
  99 }
 100
 101 // ContainerStart invokes dockerclient.Client.ContainerStart
 102 func (proxy ThinDockerClientProxy) ContainerStart(ctx context.Context, container string, options dockertypes.ContainerStartOptions) error {
 103         return proxy.Docker.ContainerStart(ctx, container, options)
 104 }
 105
 106 // ContainerStop invokes dockerclient.Client.ContainerStop
 107 func (proxy ThinDockerClientProxy) ContainerStop(ctx context.Context, container string, timeout *time.Duration) error {
 108         return proxy.Docker.ContainerStop(ctx, container, timeout)
 109 }
 110
 111 // ContainerWait invokes dockerclient.Client.ContainerWait
 112 func (proxy ThinDockerClientProxy) ContainerWait(ctx context.Context, container string, condition dockercontainer.WaitCondition) (<-chan dockercontainer.ContainerWaitOKBody, <-chan error) {
 113         return proxy.Docker.ContainerWait(ctx, container, condition)
 114 }
 115
 116 // ImageInspectWithRaw invokes dockerclient.Client.ImageInspectWithRaw
 117 func (proxy ThinDockerClientProxy) ImageInspectWithRaw(ctx context.Context, image string) (dockertypes.ImageInspect, []byte, error) {
 118         return proxy.Docker.ImageInspectWithRaw(ctx, image)
 119 }
 120
 121 // ImageLoad invokes dockerclient.Client.ImageLoad
 122 func (proxy ThinDockerClientProxy) ImageLoad(ctx context.Context, input io.Reader, quiet bool) (dockertypes.ImageLoadResponse, error) {
 123         return proxy.Docker.ImageLoad(ctx, input, quiet)
 124 }
 125
 126 // ImageRemove invokes dockerclient.Client.ImageRemove
 127 func (proxy ThinDockerClientProxy) ImageRemove(ctx context.Context, image string, options dockertypes.ImageRemoveOptions) ([]dockertypes.ImageDeleteResponseItem, error) {
 128         return proxy.Docker.ImageRemove(ctx, image, options)
 129 }
 130
 131 // ContainerRunner is the main stateful struct used for a single execution of a
 132 // container.
 133 type ContainerRunner struct {
 134         Docker    ThinDockerClient
 135         ArvClient IArvadosClient
 136         Kc        IKeepClient
 137         arvados.Container
 138         ContainerConfig dockercontainer.Config
 139         dockercontainer.HostConfig
 140         token       string
 141         ContainerID string
 142         ExitCode    *int
 143         NewLogWriter
 144         loggingDone   chan bool
 145         CrunchLog     *ThrottledLogger
 146         Stdout        io.WriteCloser
 147         Stderr        io.WriteCloser
 148         LogCollection *CollectionWriter
 149         LogsPDH       *string
 150         RunArvMount
 151         MkTempDir
 152         ArvMount       *exec.Cmd
 153         ArvMountPoint  string
 154         HostOutputDir  string
 155         CleanupTempDir []string
 156         Binds          []string
 157         Volumes        map[string]struct{}
 158         OutputPDH      *string
 159         SigChan        chan os.Signal
 160         ArvMountExit   chan error
 161         finalState     string
 162
 163         statLogger   io.WriteCloser
 164         statReporter *crunchstat.Reporter
 165         statInterval time.Duration
 166         cgroupRoot   string
 167         // What we expect the container's cgroup parent to be.
 168         expectCgroupParent string
 169         // What we tell docker to use as the container's cgroup
 170         // parent. Note: Ideally we would use the same field for both
 171         // expectCgroupParent and setCgroupParent, and just make it
 172         // default to "docker". However, when using docker < 1.10 with
 173         // systemd, specifying a non-empty cgroup parent (even the
 174         // default value "docker") hits a docker bug
 175         // (https://github.com/docker/docker/issues/17126). Using two
 176         // separate fields makes it possible to use the "expect cgroup
 177         // parent to be X" feature even on sites where the "specify
 178         // cgroup parent" feature breaks.
 179         setCgroupParent string
 180
 181         cStateLock sync.Mutex
 182         cStarted   bool // StartContainer() succeeded
 183         cCancelled bool // StopContainer() invoked
 184
 185         enableNetwork string // one of "default" or "always"
 186         networkMode   string // passed through to HostConfig.NetworkMode
 187         arvMountLog   *ThrottledLogger
 188 }
 189
 190 // setupSignals sets up signal handling to gracefully terminate the underlying
 191 // Docker container and update state when receiving a TERM, INT or QUIT signal.
 192 func (runner *ContainerRunner) setupSignals() {
 193         runner.SigChan = make(chan os.Signal, 1)
 194         signal.Notify(runner.SigChan, syscall.SIGTERM)
 195         signal.Notify(runner.SigChan, syscall.SIGINT)
 196         signal.Notify(runner.SigChan, syscall.SIGQUIT)
 197
 198         go func(sig chan os.Signal) {
 199                 s := <-sig
 200                 if s != nil {
 201                         runner.CrunchLog.Printf("Caught signal %v", s)
 202                 }
 203                 runner.stop()
 204         }(runner.SigChan)
 205 }
 206
 207 // stop the underlying Docker container.
 208 func (runner *ContainerRunner) stop() {
 209         runner.cStateLock.Lock()
 210         defer runner.cStateLock.Unlock()
 211         if runner.cCancelled {
 212                 return
 213         }
 214         runner.cCancelled = true
 215         if runner.cStarted {
 216                 timeout := time.Duration(10)
 217                 err := runner.Docker.ContainerStop(context.TODO(), runner.ContainerID, &(timeout))
 218                 if err != nil {
 219                         runner.CrunchLog.Printf("StopContainer failed: %s", err)
 220                 }
 221                 // Suppress multiple calls to stop()
 222                 runner.cStarted = false
 223         }
 224 }
 225
 226 func (runner *ContainerRunner) stopSignals() {
 227         if runner.SigChan != nil {
 228                 signal.Stop(runner.SigChan)
 229                 close(runner.SigChan)
 230         }
 231 }
 232
 233 // LoadImage determines the docker image id from the container record and
 234 // checks if it is available in the local Docker image store.  If not, it loads
 235 // the image from Keep.
 236 func (runner *ContainerRunner) LoadImage() (err error) {
 237
 238         runner.CrunchLog.Printf("Fetching Docker image from collection '%s'", runner.Container.ContainerImage)
 239
 240         var collection arvados.Collection
 241         err = runner.ArvClient.Get("collections", runner.Container.ContainerImage, nil, &collection)
 242         if err != nil {
 243                 return fmt.Errorf("While getting container image collection: %v", err)
 244         }
 245         manifest := manifest.Manifest{Text: collection.ManifestText}
 246         var img, imageID string
 247         for ms := range manifest.StreamIter() {
 248                 img = ms.FileStreamSegments[0].Name
 249                 if !strings.HasSuffix(img, ".tar") {
 250                         return fmt.Errorf("First file in the container image collection does not end in .tar")
 251                 }
 252                 imageID = img[:len(img)-4]
 253         }
 254
 255         runner.CrunchLog.Printf("Using Docker image id '%s'", imageID)
 256
 257         _, _, err = runner.Docker.ImageInspectWithRaw(context.TODO(), imageID)
 258         if err != nil {
 259                 runner.CrunchLog.Print("Loading Docker image from keep")
 260
 261                 var readCloser io.ReadCloser
 262                 readCloser, err = runner.Kc.ManifestFileReader(manifest, img)
 263                 if err != nil {
 264                         return fmt.Errorf("While creating ManifestFileReader for container image: %v", err)
 265                 }
 266
 267                 response, err := runner.Docker.ImageLoad(context.TODO(), readCloser, true)
 268                 if err != nil {
 269                         return fmt.Errorf("While loading container image into Docker: %v", err)
 270                 }
 271
 272                 defer response.Body.Close()
 273                 rbody, err := ioutil.ReadAll(response.Body)
 274                 if err != nil {
 275                         return fmt.Errorf("Reading response to image load: %v", err)
 276                 }
 277                 runner.CrunchLog.Printf("Docker response: %s", rbody)
 278         } else {
 279                 runner.CrunchLog.Print("Docker image is available")
 280         }
 281
 282         runner.ContainerConfig.Image = imageID
 283
 284         runner.Kc.ClearBlockCache()
 285
 286         return nil
 287 }
 288
 289 func (runner *ContainerRunner) ArvMountCmd(arvMountCmd []string, token string) (c *exec.Cmd, err error) {
 290         c = exec.Command("arv-mount", arvMountCmd...)
 291
 292         // Copy our environment, but override ARVADOS_API_TOKEN with
 293         // the container auth token.
 294         c.Env = nil
 295         for _, s := range os.Environ() {
 296                 if !strings.HasPrefix(s, "ARVADOS_API_TOKEN=") {
 297                         c.Env = append(c.Env, s)
 298                 }
 299         }
 300         c.Env = append(c.Env, "ARVADOS_API_TOKEN="+token)
 301
 302         runner.arvMountLog = NewThrottledLogger(runner.NewLogWriter("arv-mount"))
 303         c.Stdout = runner.arvMountLog
 304         c.Stderr = runner.arvMountLog
 305
 306         runner.CrunchLog.Printf("Running %v", c.Args)
 307
 308         err = c.Start()
 309         if err != nil {
 310                 return nil, err
 311         }
 312
 313         statReadme := make(chan bool)
 314         runner.ArvMountExit = make(chan error)
 315
 316         keepStatting := true
 317         go func() {
 318                 for keepStatting {
 319                         time.Sleep(100 * time.Millisecond)
 320                         _, err = os.Stat(fmt.Sprintf("%s/by_id/README", runner.ArvMountPoint))
 321                         if err == nil {
 322                                 keepStatting = false
 323                                 statReadme <- true
 324                         }
 325                 }
 326                 close(statReadme)
 327         }()
 328
 329         go func() {
 330                 mnterr := c.Wait()
 331                 if mnterr != nil {
 332                         runner.CrunchLog.Printf("Arv-mount exit error: %v", mnterr)
 333                 }
 334                 runner.ArvMountExit <- mnterr
 335                 close(runner.ArvMountExit)
 336         }()
 337
 338         select {
 339         case <-statReadme:
 340                 break
 341         case err := <-runner.ArvMountExit:
 342                 runner.ArvMount = nil
 343                 keepStatting = false
 344                 return nil, err
 345         }
 346
 347         return c, nil
 348 }
 349
 350 func (runner *ContainerRunner) SetupArvMountPoint(prefix string) (err error) {
 351         if runner.ArvMountPoint == "" {
 352                 runner.ArvMountPoint, err = runner.MkTempDir("", prefix)
 353         }
 354         return
 355 }
 356
 357 func (runner *ContainerRunner) SetupMounts() (err error) {
 358         err = runner.SetupArvMountPoint("keep")
 359         if err != nil {
 360                 return fmt.Errorf("While creating keep mount temp dir: %v", err)
 361         }
 362
 363         pdhOnly := true
 364         tmpcount := 0
 365         arvMountCmd := []string{
 366                 "--foreground",
 367                 "--allow-other",
 368                 "--read-write",
 369                 fmt.Sprintf("--crunchstat-interval=%v", runner.statInterval.Seconds())}
 370
 371         if runner.Container.RuntimeConstraints.KeepCacheRAM > 0 {
 372                 arvMountCmd = append(arvMountCmd, "--file-cache", fmt.Sprintf("%d", runner.Container.RuntimeConstraints.KeepCacheRAM))
 373         }
 374
 375         collectionPaths := []string{}
 376         runner.Binds = nil
 377         runner.Volumes = make(map[string]struct{})
 378         needCertMount := true
 379
 380         var binds []string
 381         for bind := range runner.Container.Mounts {
 382                 binds = append(binds, bind)
 383         }
 384         sort.Strings(binds)
 385
 386         for _, bind := range binds {
 387                 mnt := runner.Container.Mounts[bind]
 388                 if bind == "stdout" || bind == "stderr" {
 389                         // Is it a "file" mount kind?
 390                         if mnt.Kind != "file" {
 391                                 return fmt.Errorf("Unsupported mount kind '%s' for %s. Only 'file' is supported.", mnt.Kind, bind)
 392                         }
 393
 394                         // Does path start with OutputPath?
 395                         prefix := runner.Container.OutputPath
 396                         if !strings.HasSuffix(prefix, "/") {
 397                                 prefix += "/"
 398                         }
 399                         if !strings.HasPrefix(mnt.Path, prefix) {
 400                                 return fmt.Errorf("%s path does not start with OutputPath: %s, %s", strings.Title(bind), mnt.Path, prefix)
 401                         }
 402                 }
 403
 404                 if bind == "stdin" {
 405                         // Is it a "collection" mount kind?
 406                         if mnt.Kind != "collection" && mnt.Kind != "json" {
 407                                 return fmt.Errorf("Unsupported mount kind '%s' for stdin. Only 'collection' or 'json' are supported.", mnt.Kind)
 408                         }
 409                 }
 410
 411                 if bind == "/etc/arvados/ca-certificates.crt" {
 412                         needCertMount = false
 413                 }
 414
 415                 if strings.HasPrefix(bind, runner.Container.OutputPath+"/") && bind != runner.Container.OutputPath+"/" {
 416                         if mnt.Kind != "collection" {
 417                                 return fmt.Errorf("Only mount points of kind 'collection' are supported underneath the output_path: %v", bind)
 418                         }
 419                 }
 420
 421                 switch {
 422                 case mnt.Kind == "collection" && bind != "stdin":
 423                         var src string
 424                         if mnt.UUID != "" && mnt.PortableDataHash != "" {
 425                                 return fmt.Errorf("Cannot specify both 'uuid' and 'portable_data_hash' for a collection mount")
 426                         }
 427                         if mnt.UUID != "" {
 428                                 if mnt.Writable {
 429                                         return fmt.Errorf("Writing to existing collections currently not permitted.")
 430                                 }
 431                                 pdhOnly = false
 432                                 src = fmt.Sprintf("%s/by_id/%s", runner.ArvMountPoint, mnt.UUID)
 433                         } else if mnt.PortableDataHash != "" {
 434                                 if mnt.Writable {
 435                                         return fmt.Errorf("Can never write to a collection specified by portable data hash")
 436                                 }
 437                                 idx := strings.Index(mnt.PortableDataHash, "/")
 438                                 if idx > 0 {
 439                                         mnt.Path = path.Clean(mnt.PortableDataHash[idx:])
 440                                         mnt.PortableDataHash = mnt.PortableDataHash[0:idx]
 441                                         runner.Container.Mounts[bind] = mnt
 442                                 }
 443                                 src = fmt.Sprintf("%s/by_id/%s", runner.ArvMountPoint, mnt.PortableDataHash)
 444                                 if mnt.Path != "" && mnt.Path != "." {
 445                                         if strings.HasPrefix(mnt.Path, "./") {
 446                                                 mnt.Path = mnt.Path[2:]
 447                                         } else if strings.HasPrefix(mnt.Path, "/") {
 448                                                 mnt.Path = mnt.Path[1:]
 449                                         }
 450                                         src += "/" + mnt.Path
 451                                 }
 452                         } else {
 453                                 src = fmt.Sprintf("%s/tmp%d", runner.ArvMountPoint, tmpcount)
 454                                 arvMountCmd = append(arvMountCmd, "--mount-tmp")
 455                                 arvMountCmd = append(arvMountCmd, fmt.Sprintf("tmp%d", tmpcount))
 456                                 tmpcount += 1
 457                         }
 458                         if mnt.Writable {
 459                                 if bind == runner.Container.OutputPath {
 460                                         runner.HostOutputDir = src
 461                                 } else if strings.HasPrefix(bind, runner.Container.OutputPath+"/") {
 462                                         return fmt.Errorf("Writable mount points are not permitted underneath the output_path: %v", bind)
 463                                 }
 464                                 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s", src, bind))
 465                         } else {
 466                                 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s:ro", src, bind))
 467                         }
 468                         collectionPaths = append(collectionPaths, src)
 469
 470                 case mnt.Kind == "tmp":
 471                         var tmpdir string
 472                         tmpdir, err = runner.MkTempDir("", "")
 473                         if err != nil {
 474                                 return fmt.Errorf("While creating mount temp dir: %v", err)
 475                         }
 476                         st, staterr := os.Stat(tmpdir)
 477                         if staterr != nil {
 478                                 return fmt.Errorf("While Stat on temp dir: %v", staterr)
 479                         }
 480                         err = os.Chmod(tmpdir, st.Mode()|os.ModeSetgid|0777)
 481                         if staterr != nil {
 482                                 return fmt.Errorf("While Chmod temp dir: %v", err)
 483                         }
 484                         runner.CleanupTempDir = append(runner.CleanupTempDir, tmpdir)
 485                         runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s", tmpdir, bind))
 486                         if bind == runner.Container.OutputPath {
 487                                 runner.HostOutputDir = tmpdir
 488                         }
 489
 490                 case mnt.Kind == "json":
 491                         jsondata, err := json.Marshal(mnt.Content)
 492                         if err != nil {
 493                                 return fmt.Errorf("encoding json data: %v", err)
 494                         }
 495                         // Create a tempdir with a single file
 496                         // (instead of just a tempfile): this way we
 497                         // can ensure the file is world-readable
 498                         // inside the container, without having to
 499                         // make it world-readable on the docker host.
 500                         tmpdir, err := runner.MkTempDir("", "")
 501                         if err != nil {
 502                                 return fmt.Errorf("creating temp dir: %v", err)
 503                         }
 504                         runner.CleanupTempDir = append(runner.CleanupTempDir, tmpdir)
 505                         tmpfn := filepath.Join(tmpdir, "mountdata.json")
 506                         err = ioutil.WriteFile(tmpfn, jsondata, 0644)
 507                         if err != nil {
 508                                 return fmt.Errorf("writing temp file: %v", err)
 509                         }
 510                         runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s:ro", tmpfn, bind))
 511                 }
 512         }
 513
 514         if runner.HostOutputDir == "" {
 515                 return fmt.Errorf("Output path does not correspond to a writable mount point")
 516         }
 517
 518         if wantAPI := runner.Container.RuntimeConstraints.API; needCertMount && wantAPI != nil && *wantAPI {
 519                 for _, certfile := range arvadosclient.CertFiles {
 520                         _, err := os.Stat(certfile)
 521                         if err == nil {
 522                                 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:/etc/arvados/ca-certificates.crt:ro", certfile))
 523                                 break
 524                         }
 525                 }
 526         }
 527
 528         if pdhOnly {
 529                 arvMountCmd = append(arvMountCmd, "--mount-by-pdh", "by_id")
 530         } else {
 531                 arvMountCmd = append(arvMountCmd, "--mount-by-id", "by_id")
 532         }
 533         arvMountCmd = append(arvMountCmd, runner.ArvMountPoint)
 534
 535         token, err := runner.ContainerToken()
 536         if err != nil {
 537                 return fmt.Errorf("could not get container token: %s", err)
 538         }
 539
 540         runner.ArvMount, err = runner.RunArvMount(arvMountCmd, token)
 541         if err != nil {
 542                 return fmt.Errorf("While trying to start arv-mount: %v", err)
 543         }
 544
 545         for _, p := range collectionPaths {
 546                 _, err = os.Stat(p)
 547                 if err != nil {
 548                         return fmt.Errorf("While checking that input files exist: %v", err)
 549                 }
 550         }
 551
 552         return nil
 553 }
 554
 555 func (runner *ContainerRunner) ProcessDockerAttach(containerReader io.Reader) {
 556         // Handle docker log protocol
 557         // https://docs.docker.com/engine/reference/api/docker_remote_api_v1.15/#attach-to-a-container
 558
 559         header := make([]byte, 8)
 560         for {
 561                 _, readerr := io.ReadAtLeast(containerReader, header, 8)
 562
 563                 if readerr == nil {
 564                         readsize := int64(header[7]) | (int64(header[6]) << 8) | (int64(header[5]) << 16) | (int64(header[4]) << 24)
 565                         if header[0] == 1 {
 566                                 // stdout
 567                                 _, readerr = io.CopyN(runner.Stdout, containerReader, readsize)
 568                         } else {
 569                                 // stderr
 570                                 _, readerr = io.CopyN(runner.Stderr, containerReader, readsize)
 571                         }
 572                 }
 573
 574                 if readerr != nil {
 575                         if readerr != io.EOF {
 576                                 runner.CrunchLog.Printf("While reading docker logs: %v", readerr)
 577                         }
 578
 579                         closeerr := runner.Stdout.Close()
 580                         if closeerr != nil {
 581                                 runner.CrunchLog.Printf("While closing stdout logs: %v", closeerr)
 582                         }
 583
 584                         closeerr = runner.Stderr.Close()
 585                         if closeerr != nil {
 586                                 runner.CrunchLog.Printf("While closing stderr logs: %v", closeerr)
 587                         }
 588
 589                         if runner.statReporter != nil {
 590                                 runner.statReporter.Stop()
 591                                 closeerr = runner.statLogger.Close()
 592                                 if closeerr != nil {
 593                                         runner.CrunchLog.Printf("While closing crunchstat logs: %v", closeerr)
 594                                 }
 595                         }
 596
 597                         runner.loggingDone <- true
 598                         close(runner.loggingDone)
 599                         return
 600                 }
 601         }
 602 }
 603
 604 func (runner *ContainerRunner) StartCrunchstat() {
 605         runner.statLogger = NewThrottledLogger(runner.NewLogWriter("crunchstat"))
 606         runner.statReporter = &crunchstat.Reporter{
 607                 CID:          runner.ContainerID,
 608                 Logger:       log.New(runner.statLogger, "", 0),
 609                 CgroupParent: runner.expectCgroupParent,
 610                 CgroupRoot:   runner.cgroupRoot,
 611                 PollPeriod:   runner.statInterval,
 612         }
 613         runner.statReporter.Start()
 614 }
 615
 616 type infoCommand struct {
 617         label string
 618         cmd   []string
 619 }
 620
 621 // LogNodeInfo gathers node information and store it on the log for debugging
 622 // purposes.
 623 func (runner *ContainerRunner) LogNodeInfo() (err error) {
 624         w := runner.NewLogWriter("node-info")
 625         logger := log.New(w, "node-info", 0)
 626
 627         commands := []infoCommand{
 628                 {
 629                         label: "Host Information",
 630                         cmd:   []string{"uname", "-a"},
 631                 },
 632                 {
 633                         label: "CPU Information",
 634                         cmd:   []string{"cat", "/proc/cpuinfo"},
 635                 },
 636                 {
 637                         label: "Memory Information",
 638                         cmd:   []string{"cat", "/proc/meminfo"},
 639                 },
 640                 {
 641                         label: "Disk Space",
 642                         cmd:   []string{"df", "-m", "/", os.TempDir()},
 643                 },
 644                 {
 645                         label: "Disk INodes",
 646                         cmd:   []string{"df", "-i", "/", os.TempDir()},
 647                 },
 648         }
 649
 650         // Run commands with informational output to be logged.
 651         var out []byte
 652         for _, command := range commands {
 653                 out, err = exec.Command(command.cmd[0], command.cmd[1:]...).CombinedOutput()
 654                 if err != nil {
 655                         return fmt.Errorf("While running command %q: %v",
 656                                 command.cmd, err)
 657                 }
 658                 logger.Println(command.label)
 659                 for _, line := range strings.Split(string(out), "\n") {
 660                         logger.Println(" ", line)
 661                 }
 662         }
 663
 664         err = w.Close()
 665         if err != nil {
 666                 return fmt.Errorf("While closing node-info logs: %v", err)
 667         }
 668         return nil
 669 }
 670
 671 // LogContainerRecord gets and saves the raw JSON container record from the API server
 672 func (runner *ContainerRunner) LogContainerRecord() (err error) {
 673         w := &ArvLogWriter{
 674                 ArvClient:     runner.ArvClient,
 675                 UUID:          runner.Container.UUID,
 676                 loggingStream: "container",
 677                 writeCloser:   runner.LogCollection.Open("container.json"),
 678         }
 679
 680         // Get Container record JSON from the API Server
 681         reader, err := runner.ArvClient.CallRaw("GET", "containers", runner.Container.UUID, "", nil)
 682         if err != nil {
 683                 return fmt.Errorf("While retrieving container record from the API server: %v", err)
 684         }
 685         defer reader.Close()
 686
 687         dec := json.NewDecoder(reader)
 688         dec.UseNumber()
 689         var cr map[string]interface{}
 690         if err = dec.Decode(&cr); err != nil {
 691                 return fmt.Errorf("While decoding the container record JSON response: %v", err)
 692         }
 693         // Re-encode it using indentation to improve readability
 694         enc := json.NewEncoder(w)
 695         enc.SetIndent("", "    ")
 696         if err = enc.Encode(cr); err != nil {
 697                 return fmt.Errorf("While logging the JSON container record: %v", err)
 698         }
 699         err = w.Close()
 700         if err != nil {
 701                 return fmt.Errorf("While closing container.json log: %v", err)
 702         }
 703         return nil
 704 }
 705
 706 // AttachStreams connects the docker container stdin, stdout and stderr logs
 707 // to the Arvados logger which logs to Keep and the API server logs table.
 708 func (runner *ContainerRunner) AttachStreams() (err error) {
 709
 710         runner.CrunchLog.Print("Attaching container streams")
 711
 712         // If stdin mount is provided, attach it to the docker container
 713         var stdinRdr arvados.File
 714         var stdinJson []byte
 715         if stdinMnt, ok := runner.Container.Mounts["stdin"]; ok {
 716                 if stdinMnt.Kind == "collection" {
 717                         var stdinColl arvados.Collection
 718                         collId := stdinMnt.UUID
 719                         if collId == "" {
 720                                 collId = stdinMnt.PortableDataHash
 721                         }
 722                         err = runner.ArvClient.Get("collections", collId, nil, &stdinColl)
 723                         if err != nil {
 724                                 return fmt.Errorf("While getting stding collection: %v", err)
 725                         }
 726
 727                         stdinRdr, err = runner.Kc.ManifestFileReader(manifest.Manifest{Text: stdinColl.ManifestText}, stdinMnt.Path)
 728                         if os.IsNotExist(err) {
 729                                 return fmt.Errorf("stdin collection path not found: %v", stdinMnt.Path)
 730                         } else if err != nil {
 731                                 return fmt.Errorf("While getting stdin collection path %v: %v", stdinMnt.Path, err)
 732                         }
 733                 } else if stdinMnt.Kind == "json" {
 734                         stdinJson, err = json.Marshal(stdinMnt.Content)
 735                         if err != nil {
 736                                 return fmt.Errorf("While encoding stdin json data: %v", err)
 737                         }
 738                 }
 739         }
 740
 741         stdinUsed := stdinRdr != nil || len(stdinJson) != 0
 742         response, err := runner.Docker.ContainerAttach(context.TODO(), runner.ContainerID,
 743                 dockertypes.ContainerAttachOptions{Stream: true, Stdin: stdinUsed, Stdout: true, Stderr: true})
 744         if err != nil {
 745                 return fmt.Errorf("While attaching container stdout/stderr streams: %v", err)
 746         }
 747
 748         runner.loggingDone = make(chan bool)
 749
 750         if stdoutMnt, ok := runner.Container.Mounts["stdout"]; ok {
 751                 stdoutFile, err := runner.getStdoutFile(stdoutMnt.Path)
 752                 if err != nil {
 753                         return err
 754                 }
 755                 runner.Stdout = stdoutFile
 756         } else {
 757                 runner.Stdout = NewThrottledLogger(runner.NewLogWriter("stdout"))
 758         }
 759
 760         if stderrMnt, ok := runner.Container.Mounts["stderr"]; ok {
 761                 stderrFile, err := runner.getStdoutFile(stderrMnt.Path)
 762                 if err != nil {
 763                         return err
 764                 }
 765                 runner.Stderr = stderrFile
 766         } else {
 767                 runner.Stderr = NewThrottledLogger(runner.NewLogWriter("stderr"))
 768         }
 769
 770         if stdinRdr != nil {
 771                 go func() {
 772                         _, err := io.Copy(response.Conn, stdinRdr)
 773                         if err != nil {
 774                                 runner.CrunchLog.Print("While writing stdin collection to docker container %q", err)
 775                                 runner.stop()
 776                         }
 777                         stdinRdr.Close()
 778                         response.CloseWrite()
 779                 }()
 780         } else if len(stdinJson) != 0 {
 781                 go func() {
 782                         _, err := io.Copy(response.Conn, bytes.NewReader(stdinJson))
 783                         if err != nil {
 784                                 runner.CrunchLog.Print("While writing stdin json to docker container %q", err)
 785                                 runner.stop()
 786                         }
 787                         response.CloseWrite()
 788                 }()
 789         }
 790
 791         go runner.ProcessDockerAttach(response.Reader)
 792
 793         return nil
 794 }
 795
 796 func (runner *ContainerRunner) getStdoutFile(mntPath string) (*os.File, error) {
 797         stdoutPath := mntPath[len(runner.Container.OutputPath):]
 798         index := strings.LastIndex(stdoutPath, "/")
 799         if index > 0 {
 800                 subdirs := stdoutPath[:index]
 801                 if subdirs != "" {
 802                         st, err := os.Stat(runner.HostOutputDir)
 803                         if err != nil {
 804                                 return nil, fmt.Errorf("While Stat on temp dir: %v", err)
 805                         }
 806                         stdoutPath := filepath.Join(runner.HostOutputDir, subdirs)
 807                         err = os.MkdirAll(stdoutPath, st.Mode()|os.ModeSetgid|0777)
 808                         if err != nil {
 809                                 return nil, fmt.Errorf("While MkdirAll %q: %v", stdoutPath, err)
 810                         }
 811                 }
 812         }
 813         stdoutFile, err := os.Create(filepath.Join(runner.HostOutputDir, stdoutPath))
 814         if err != nil {
 815                 return nil, fmt.Errorf("While creating file %q: %v", stdoutPath, err)
 816         }
 817
 818         return stdoutFile, nil
 819 }
 820
 821 // CreateContainer creates the docker container.
 822 func (runner *ContainerRunner) CreateContainer() error {
 823         runner.CrunchLog.Print("Creating Docker container")
 824
 825         runner.ContainerConfig.Cmd = runner.Container.Command
 826         if runner.Container.Cwd != "." {
 827                 runner.ContainerConfig.WorkingDir = runner.Container.Cwd
 828         }
 829
 830         for k, v := range runner.Container.Environment {
 831                 runner.ContainerConfig.Env = append(runner.ContainerConfig.Env, k+"="+v)
 832         }
 833
 834         runner.ContainerConfig.Volumes = runner.Volumes
 835
 836         runner.HostConfig = dockercontainer.HostConfig{
 837                 Binds: runner.Binds,
 838                 LogConfig: dockercontainer.LogConfig{
 839                         Type: "none",
 840                 },
 841                 Resources: dockercontainer.Resources{
 842                         CgroupParent: runner.setCgroupParent,
 843                 },
 844         }
 845
 846         if wantAPI := runner.Container.RuntimeConstraints.API; wantAPI != nil && *wantAPI {
 847                 tok, err := runner.ContainerToken()
 848                 if err != nil {
 849                         return err
 850                 }
 851                 runner.ContainerConfig.Env = append(runner.ContainerConfig.Env,
 852                         "ARVADOS_API_TOKEN="+tok,
 853                         "ARVADOS_API_HOST="+os.Getenv("ARVADOS_API_HOST"),
 854                         "ARVADOS_API_HOST_INSECURE="+os.Getenv("ARVADOS_API_HOST_INSECURE"),
 855                 )
 856                 runner.HostConfig.NetworkMode = dockercontainer.NetworkMode(runner.networkMode)
 857         } else {
 858                 if runner.enableNetwork == "always" {
 859                         runner.HostConfig.NetworkMode = dockercontainer.NetworkMode(runner.networkMode)
 860                 } else {
 861                         runner.HostConfig.NetworkMode = dockercontainer.NetworkMode("none")
 862                 }
 863         }
 864
 865         _, stdinUsed := runner.Container.Mounts["stdin"]
 866         runner.ContainerConfig.OpenStdin = stdinUsed
 867         runner.ContainerConfig.StdinOnce = stdinUsed
 868         runner.ContainerConfig.AttachStdin = stdinUsed
 869         runner.ContainerConfig.AttachStdout = true
 870         runner.ContainerConfig.AttachStderr = true
 871
 872         createdBody, err := runner.Docker.ContainerCreate(context.TODO(), &runner.ContainerConfig, &runner.HostConfig, nil, runner.Container.UUID)
 873         if err != nil {
 874                 return fmt.Errorf("While creating container: %v", err)
 875         }
 876
 877         runner.ContainerID = createdBody.ID
 878
 879         return runner.AttachStreams()
 880 }
 881
 882 // StartContainer starts the docker container created by CreateContainer.
 883 func (runner *ContainerRunner) StartContainer() error {
 884         runner.CrunchLog.Printf("Starting Docker container id '%s'", runner.ContainerID)
 885         runner.cStateLock.Lock()
 886         defer runner.cStateLock.Unlock()
 887         if runner.cCancelled {
 888                 return ErrCancelled
 889         }
 890         err := runner.Docker.ContainerStart(context.TODO(), runner.ContainerID,
 891                 dockertypes.ContainerStartOptions{})
 892         if err != nil {
 893                 var advice string
 894                 if strings.Contains(err.Error(), "no such file or directory") {
 895                         advice = fmt.Sprintf("\nPossible causes: command %q is missing, the interpreter given in #! is missing, or script has Windows line endings.", runner.Container.Command[0])
 896                 }
 897                 return fmt.Errorf("could not start container: %v%s", err, advice)
 898         }
 899         runner.cStarted = true
 900         return nil
 901 }
 902
 903 // WaitFinish waits for the container to terminate, capture the exit code, and
 904 // close the stdout/stderr logging.
 905 func (runner *ContainerRunner) WaitFinish() (err error) {
 906         runner.CrunchLog.Print("Waiting for container to finish")
 907
 908         waitOk, waitErr := runner.Docker.ContainerWait(context.TODO(), runner.ContainerID, "not-running")
 909
 910         go func() {
 911                 <-runner.ArvMountExit
 912                 if runner.cStarted {
 913                         runner.CrunchLog.Printf("arv-mount exited while container is still running.  Stopping container.")
 914                         runner.stop()
 915                 }
 916         }()
 917
 918         var waitBody dockercontainer.ContainerWaitOKBody
 919         select {
 920         case waitBody = <-waitOk:
 921         case err = <-waitErr:
 922         }
 923
 924         // Container isn't running any more
 925         runner.cStarted = false
 926
 927         if err != nil {
 928                 return fmt.Errorf("container wait: %v", err)
 929         }
 930
 931         runner.CrunchLog.Printf("Container exited with code: %v", waitBody.StatusCode)
 932         code := int(waitBody.StatusCode)
 933         runner.ExitCode = &code
 934
 935         // wait for stdout/stderr to complete
 936         <-runner.loggingDone
 937
 938         return nil
 939 }
 940
 941 var ErrNotInOutputDir = fmt.Errorf("Must point to path within the output directory")
 942
 943 func (runner *ContainerRunner) derefOutputSymlink(path string, startinfo os.FileInfo) (tgt string, readlinktgt string, info os.FileInfo, err error) {
 944         // Follow symlinks if necessary
 945         info = startinfo
 946         tgt = path
 947         readlinktgt = ""
 948         nextlink := path
 949         for followed := 0; info.Mode()&os.ModeSymlink != 0; followed++ {
 950                 if followed >= limitFollowSymlinks {
 951                         // Got stuck in a loop or just a pathological number of links, give up.
 952                         err = fmt.Errorf("Followed more than %v symlinks from path %q", limitFollowSymlinks, path)
 953                         return
 954                 }
 955
 956                 readlinktgt, err = os.Readlink(nextlink)
 957                 if err != nil {
 958                         return
 959                 }
 960
 961                 tgt = readlinktgt
 962                 if !strings.HasPrefix(tgt, "/") {
 963                         // Relative symlink, resolve it to host path
 964                         tgt = filepath.Join(filepath.Dir(path), tgt)
 965                 }
 966                 if strings.HasPrefix(tgt, runner.Container.OutputPath+"/") && !strings.HasPrefix(tgt, runner.HostOutputDir+"/") {
 967                         // Absolute symlink to container output path, adjust it to host output path.
 968                         tgt = filepath.Join(runner.HostOutputDir, tgt[len(runner.Container.OutputPath):])
 969                 }
 970                 if !strings.HasPrefix(tgt, runner.HostOutputDir+"/") {
 971                         // After dereferencing, symlink target must either be
 972                         // within output directory, or must point to a
 973                         // collection mount.
 974                         err = ErrNotInOutputDir
 975                         return
 976                 }
 977
 978                 info, err = os.Lstat(tgt)
 979                 if err != nil {
 980                         // tgt
 981                         err = fmt.Errorf("Symlink in output %q points to invalid location %q: %v",
 982                                 path[len(runner.HostOutputDir):], readlinktgt, err)
 983                         return
 984                 }
 985
 986                 nextlink = tgt
 987         }
 988
 989         return
 990 }
 991
 992 var limitFollowSymlinks = 10
 993
 994 // UploadFile uploads files within the output directory, with special handling
 995 // for symlinks. If the symlink leads to a keep mount, copy the manifest text
 996 // from the keep mount into the output manifestText.  Ensure that whether
 997 // symlinks are relative or absolute, every symlink target (even targets that
 998 // are symlinks themselves) must point to a path in either the output directory
 999 // or a collection mount.
1000 //
1001 // Assumes initial value of "path" is absolute, and located within runner.HostOutputDir.
1002 func (runner *ContainerRunner) UploadOutputFile(
1003         path string,
1004         info os.FileInfo,
1005         infoerr error,
1006         binds []string,
1007         walkUpload *WalkUpload,
1008         relocateFrom string,
1009         relocateTo string,
1010         followed int) (manifestText string, err error) {
1011
1012         if info.Mode().IsDir() {
1013                 return
1014         }
1015
1016         if infoerr != nil {
1017                 return "", infoerr
1018         }
1019
1020         if followed >= limitFollowSymlinks {
1021                 // Got stuck in a loop or just a pathological number of
1022                 // directory links, give up.
1023                 err = fmt.Errorf("Followed more than %v symlinks from path %q", limitFollowSymlinks, path)
1024                 return
1025         }
1026
1027         // When following symlinks, the source path may need to be logically
1028         // relocated to some other path within the output collection.  Remove
1029         // the relocateFrom prefix and replace it with relocateTo.
1030         relocated := relocateTo + path[len(relocateFrom):]
1031
1032         tgt, readlinktgt, info, derefErr := runner.derefOutputSymlink(path, info)
1033         if derefErr != nil && derefErr != ErrNotInOutputDir {
1034                 return "", derefErr
1035         }
1036
1037         // go through mounts and try reverse map to collection reference
1038         for _, bind := range binds {
1039                 mnt := runner.Container.Mounts[bind]
1040                 if tgt == bind || strings.HasPrefix(tgt, bind+"/") {
1041                         // get path relative to bind
1042                         targetSuffix := tgt[len(bind):]
1043
1044                         // Copy mount and adjust the path to add path relative to the bind
1045                         adjustedMount := mnt
1046                         adjustedMount.Path = filepath.Join(adjustedMount.Path, targetSuffix)
1047
1048                         // Terminates in this keep mount, so add the
1049                         // manifest text at appropriate location.
1050                         outputSuffix := path[len(runner.HostOutputDir):]
1051                         manifestText, err = runner.getCollectionManifestForPath(adjustedMount, outputSuffix)
1052                         return
1053                 }
1054         }
1055
1056         // If target is not a collection mount, it must be located within the
1057         // output directory, otherwise it is an error.
1058         if derefErr == ErrNotInOutputDir {
1059                 err = fmt.Errorf("Symlink in output %q points to invalid location %q, must point to path within the output directory.",
1060                         path[len(runner.HostOutputDir):], readlinktgt)
1061                 return
1062         }
1063
1064         if info.Mode().IsRegular() {
1065                 return "", walkUpload.UploadFile(relocated, tgt)
1066         }
1067
1068         if info.Mode().IsDir() {
1069                 // Symlink leads to directory.  Walk() doesn't follow
1070                 // directory symlinks, so we walk the target directory
1071                 // instead.  Within the walk, file paths are relocated
1072                 // so they appear under the original symlink path.
1073                 err = filepath.Walk(tgt, func(walkpath string, walkinfo os.FileInfo, walkerr error) error {
1074                         var m string
1075                         m, walkerr = runner.UploadOutputFile(walkpath, walkinfo, walkerr,
1076                                 binds, walkUpload, tgt, relocated, followed+1)
1077                         if walkerr == nil {
1078                                 manifestText = manifestText + m
1079                         }
1080                         return walkerr
1081                 })
1082                 return
1083         }
1084
1085         return
1086 }
1087
1088 // HandleOutput sets the output, unmounts the FUSE mount, and deletes temporary directories
1089 func (runner *ContainerRunner) CaptureOutput() error {
1090         if runner.finalState != "Complete" {
1091                 return nil
1092         }
1093
1094         if wantAPI := runner.Container.RuntimeConstraints.API; wantAPI != nil && *wantAPI {
1095                 // Output may have been set directly by the container, so
1096                 // refresh the container record to check.
1097                 err := runner.ArvClient.Get("containers", runner.Container.UUID,
1098                         nil, &runner.Container)
1099                 if err != nil {
1100                         return err
1101                 }
1102                 if runner.Container.Output != "" {
1103                         // Container output is already set.
1104                         runner.OutputPDH = &runner.Container.Output
1105                         return nil
1106                 }
1107         }
1108
1109         if runner.HostOutputDir == "" {
1110                 return nil
1111         }
1112
1113         _, err := os.Stat(runner.HostOutputDir)
1114         if err != nil {
1115                 return fmt.Errorf("While checking host output path: %v", err)
1116         }
1117
1118         // Pre-populate output from the configured mount points
1119         var binds []string
1120         for bind, mnt := range runner.Container.Mounts {
1121                 if mnt.Kind == "collection" {
1122                         binds = append(binds, bind)
1123                 }
1124         }
1125         sort.Strings(binds)
1126
1127         var manifestText string
1128
1129         collectionMetafile := fmt.Sprintf("%s/.arvados#collection", runner.HostOutputDir)
1130         _, err = os.Stat(collectionMetafile)
1131         if err != nil {
1132                 // Regular directory
1133
1134                 cw := CollectionWriter{0, runner.Kc, nil, nil, sync.Mutex{}}
1135                 walkUpload := cw.BeginUpload(runner.HostOutputDir, runner.CrunchLog.Logger)
1136
1137                 var m string
1138                 err = filepath.Walk(runner.HostOutputDir, func(path string, info os.FileInfo, err error) error {
1139                         m, err = runner.UploadOutputFile(path, info, err, binds, walkUpload, "", "", 0)
1140                         if err == nil {
1141                                 manifestText = manifestText + m
1142                         }
1143                         return err
1144                 })
1145
1146                 cw.EndUpload(walkUpload)
1147
1148                 if err != nil {
1149                         return fmt.Errorf("While uploading output files: %v", err)
1150                 }
1151
1152                 m, err = cw.ManifestText()
1153                 manifestText = manifestText + m
1154                 if err != nil {
1155                         return fmt.Errorf("While uploading output files: %v", err)
1156                 }
1157         } else {
1158                 // FUSE mount directory
1159                 file, openerr := os.Open(collectionMetafile)
1160                 if openerr != nil {
1161                         return fmt.Errorf("While opening FUSE metafile: %v", err)
1162                 }
1163                 defer file.Close()
1164
1165                 var rec arvados.Collection
1166                 err = json.NewDecoder(file).Decode(&rec)
1167                 if err != nil {
1168                         return fmt.Errorf("While reading FUSE metafile: %v", err)
1169                 }
1170                 manifestText = rec.ManifestText
1171         }
1172
1173         for _, bind := range binds {
1174                 mnt := runner.Container.Mounts[bind]
1175
1176                 bindSuffix := strings.TrimPrefix(bind, runner.Container.OutputPath)
1177
1178                 if bindSuffix == bind || len(bindSuffix) <= 0 {
1179                         // either does not start with OutputPath or is OutputPath itself
1180                         continue
1181                 }
1182
1183                 if mnt.ExcludeFromOutput == true {
1184                         continue
1185                 }
1186
1187                 // append to manifest_text
1188                 m, err := runner.getCollectionManifestForPath(mnt, bindSuffix)
1189                 if err != nil {
1190                         return err
1191                 }
1192
1193                 manifestText = manifestText + m
1194         }
1195
1196         // Save output
1197         var response arvados.Collection
1198         manifest := manifest.Manifest{Text: manifestText}
1199         manifestText = manifest.Extract(".", ".").Text
1200         err = runner.ArvClient.Create("collections",
1201                 arvadosclient.Dict{
1202                         "ensure_unique_name": true,
1203                         "collection": arvadosclient.Dict{
1204                                 "is_trashed":    true,
1205                                 "name":          "output for " + runner.Container.UUID,
1206                                 "manifest_text": manifestText}},
1207                 &response)
1208         if err != nil {
1209                 return fmt.Errorf("While creating output collection: %v", err)
1210         }
1211         runner.OutputPDH = &response.PortableDataHash
1212         return nil
1213 }
1214
1215 var outputCollections = make(map[string]arvados.Collection)
1216
1217 // Fetch the collection for the mnt.PortableDataHash
1218 // Return the manifest_text fragment corresponding to the specified mnt.Path
1219 //  after making any required updates.
1220 //  Ex:
1221 //    If mnt.Path is not specified,
1222 //      return the entire manifest_text after replacing any "." with bindSuffix
1223 //    If mnt.Path corresponds to one stream,
1224 //      return the manifest_text for that stream after replacing that stream name with bindSuffix
1225 //    Otherwise, check if a filename in any one stream is being sought. Return the manifest_text
1226 //      for that stream after replacing stream name with bindSuffix minus the last word
1227 //      and the file name with last word of the bindSuffix
1228 //  Allowed path examples:
1229 //    "path":"/"
1230 //    "path":"/subdir1"
1231 //    "path":"/subdir1/subdir2"
1232 //    "path":"/subdir/filename" etc
1233 func (runner *ContainerRunner) getCollectionManifestForPath(mnt arvados.Mount, bindSuffix string) (string, error) {
1234         collection := outputCollections[mnt.PortableDataHash]
1235         if collection.PortableDataHash == "" {
1236                 err := runner.ArvClient.Get("collections", mnt.PortableDataHash, nil, &collection)
1237                 if err != nil {
1238                         return "", fmt.Errorf("While getting collection for %v: %v", mnt.PortableDataHash, err)
1239                 }
1240                 outputCollections[mnt.PortableDataHash] = collection
1241         }
1242
1243         if collection.ManifestText == "" {
1244                 runner.CrunchLog.Printf("No manifest text for collection %v", collection.PortableDataHash)
1245                 return "", nil
1246         }
1247
1248         mft := manifest.Manifest{Text: collection.ManifestText}
1249         extracted := mft.Extract(mnt.Path, bindSuffix)
1250         if extracted.Err != nil {
1251                 return "", fmt.Errorf("Error parsing manifest for %v: %v", mnt.PortableDataHash, extracted.Err.Error())
1252         }
1253         return extracted.Text, nil
1254 }
1255
1256 func (runner *ContainerRunner) CleanupDirs() {
1257         if runner.ArvMount != nil {
1258                 var delay int64 = 8
1259                 umount := exec.Command("arv-mount", fmt.Sprintf("--unmount-timeout=%d", delay), "--unmount", runner.ArvMountPoint)
1260                 umount.Stdout = runner.CrunchLog
1261                 umount.Stderr = runner.CrunchLog
1262                 runner.CrunchLog.Printf("Running %v", umount.Args)
1263                 umnterr := umount.Start()
1264
1265                 if umnterr != nil {
1266                         runner.CrunchLog.Printf("Error unmounting: %v", umnterr)
1267                 } else {
1268                         // If arv-mount --unmount gets stuck for any reason, we
1269                         // don't want to wait for it forever.  Do Wait() in a goroutine
1270                         // so it doesn't block crunch-run.
1271                         umountExit := make(chan error)
1272                         go func() {
1273                                 mnterr := umount.Wait()
1274                                 if mnterr != nil {
1275                                         runner.CrunchLog.Printf("Error unmounting: %v", mnterr)
1276                                 }
1277                                 umountExit <- mnterr
1278                         }()
1279
1280                         for again := true; again; {
1281                                 again = false
1282                                 select {
1283                                 case <-umountExit:
1284                                         umount = nil
1285                                         again = true
1286                                 case <-runner.ArvMountExit:
1287                                         break
1288                                 case <-time.After(time.Duration((delay + 1) * int64(time.Second))):
1289                                         runner.CrunchLog.Printf("Timed out waiting for unmount")
1290                                         if umount != nil {
1291                                                 umount.Process.Kill()
1292                                         }
1293                                         runner.ArvMount.Process.Kill()
1294                                 }
1295                         }
1296                 }
1297         }
1298
1299         if runner.ArvMountPoint != "" {
1300                 if rmerr := os.Remove(runner.ArvMountPoint); rmerr != nil {
1301                         runner.CrunchLog.Printf("While cleaning up arv-mount directory %s: %v", runner.ArvMountPoint, rmerr)
1302                 }
1303         }
1304
1305         for _, tmpdir := range runner.CleanupTempDir {
1306                 if rmerr := os.RemoveAll(tmpdir); rmerr != nil {
1307                         runner.CrunchLog.Printf("While cleaning up temporary directory %s: %v", tmpdir, rmerr)
1308                 }
1309         }
1310 }
1311
1312 // CommitLogs posts the collection containing the final container logs.
1313 func (runner *ContainerRunner) CommitLogs() error {
1314         runner.CrunchLog.Print(runner.finalState)
1315
1316         if runner.arvMountLog != nil {
1317                 runner.arvMountLog.Close()
1318         }
1319         runner.CrunchLog.Close()
1320
1321         // Closing CrunchLog above allows them to be committed to Keep at this
1322         // point, but re-open crunch log with ArvClient in case there are any
1323         // other further errors (such as failing to write the log to Keep!)
1324         // while shutting down
1325         runner.CrunchLog = NewThrottledLogger(&ArvLogWriter{ArvClient: runner.ArvClient,
1326                 UUID: runner.Container.UUID, loggingStream: "crunch-run", writeCloser: nil})
1327         runner.CrunchLog.Immediate = log.New(os.Stderr, runner.Container.UUID+" ", 0)
1328
1329         if runner.LogsPDH != nil {
1330                 // If we have already assigned something to LogsPDH,
1331                 // we must be closing the re-opened log, which won't
1332                 // end up getting attached to the container record and
1333                 // therefore doesn't need to be saved as a collection
1334                 // -- it exists only to send logs to other channels.
1335                 return nil
1336         }
1337
1338         mt, err := runner.LogCollection.ManifestText()
1339         if err != nil {
1340                 return fmt.Errorf("While creating log manifest: %v", err)
1341         }
1342
1343         var response arvados.Collection
1344         err = runner.ArvClient.Create("collections",
1345                 arvadosclient.Dict{
1346                         "ensure_unique_name": true,
1347                         "collection": arvadosclient.Dict{
1348                                 "is_trashed":    true,
1349                                 "name":          "logs for " + runner.Container.UUID,
1350                                 "manifest_text": mt}},
1351                 &response)
1352         if err != nil {
1353                 return fmt.Errorf("While creating log collection: %v", err)
1354         }
1355         runner.LogsPDH = &response.PortableDataHash
1356         return nil
1357 }
1358
1359 // UpdateContainerRunning updates the container state to "Running"
1360 func (runner *ContainerRunner) UpdateContainerRunning() error {
1361         runner.cStateLock.Lock()
1362         defer runner.cStateLock.Unlock()
1363         if runner.cCancelled {
1364                 return ErrCancelled
1365         }
1366         return runner.ArvClient.Update("containers", runner.Container.UUID,
1367                 arvadosclient.Dict{"container": arvadosclient.Dict{"state": "Running"}}, nil)
1368 }
1369
1370 // ContainerToken returns the api_token the container (and any
1371 // arv-mount processes) are allowed to use.
1372 func (runner *ContainerRunner) ContainerToken() (string, error) {
1373         if runner.token != "" {
1374                 return runner.token, nil
1375         }
1376
1377         var auth arvados.APIClientAuthorization
1378         err := runner.ArvClient.Call("GET", "containers", runner.Container.UUID, "auth", nil, &auth)
1379         if err != nil {
1380                 return "", err
1381         }
1382         runner.token = auth.APIToken
1383         return runner.token, nil
1384 }
1385
1386 // UpdateContainerComplete updates the container record state on API
1387 // server to "Complete" or "Cancelled"
1388 func (runner *ContainerRunner) UpdateContainerFinal() error {
1389         update := arvadosclient.Dict{}
1390         update["state"] = runner.finalState
1391         if runner.LogsPDH != nil {
1392                 update["log"] = *runner.LogsPDH
1393         }
1394         if runner.finalState == "Complete" {
1395                 if runner.ExitCode != nil {
1396                         update["exit_code"] = *runner.ExitCode
1397                 }
1398                 if runner.OutputPDH != nil {
1399                         update["output"] = *runner.OutputPDH
1400                 }
1401         }
1402         return runner.ArvClient.Update("containers", runner.Container.UUID, arvadosclient.Dict{"container": update}, nil)
1403 }
1404
1405 // IsCancelled returns the value of Cancelled, with goroutine safety.
1406 func (runner *ContainerRunner) IsCancelled() bool {
1407         runner.cStateLock.Lock()
1408         defer runner.cStateLock.Unlock()
1409         return runner.cCancelled
1410 }
1411
1412 // NewArvLogWriter creates an ArvLogWriter
1413 func (runner *ContainerRunner) NewArvLogWriter(name string) io.WriteCloser {
1414         return &ArvLogWriter{
1415                 ArvClient:     runner.ArvClient,
1416                 UUID:          runner.Container.UUID,
1417                 loggingStream: name,
1418                 writeCloser:   runner.LogCollection.Open(name + ".txt")}
1419 }
1420
1421 // Run the full container lifecycle.
1422 func (runner *ContainerRunner) Run() (err error) {
1423         runner.CrunchLog.Printf("crunch-run %q started", version)
1424         runner.CrunchLog.Printf("Executing container '%s'", runner.Container.UUID)
1425
1426         hostname, hosterr := os.Hostname()
1427         if hosterr != nil {
1428                 runner.CrunchLog.Printf("Error getting hostname '%v'", hosterr)
1429         } else {
1430                 runner.CrunchLog.Printf("Executing on host '%s'", hostname)
1431         }
1432
1433         runner.finalState = "Queued"
1434
1435         defer func() {
1436                 runner.stopSignals()
1437                 runner.CleanupDirs()
1438
1439                 runner.CrunchLog.Printf("crunch-run finished")
1440                 runner.CrunchLog.Close()
1441         }()
1442
1443         defer func() {
1444                 // checkErr prints e (unless it's nil) and sets err to
1445                 // e (unless err is already non-nil). Thus, if err
1446                 // hasn't already been assigned when Run() returns,
1447                 // this cleanup func will cause Run() to return the
1448                 // first non-nil error that is passed to checkErr().
1449                 checkErr := func(e error) {
1450                         if e == nil {
1451                                 return
1452                         }
1453                         runner.CrunchLog.Print(e)
1454                         if err == nil {
1455                                 err = e
1456                         }
1457                         if runner.finalState == "Complete" {
1458                                 // There was an error in the finalization.
1459                                 runner.finalState = "Cancelled"
1460                         }
1461                 }
1462
1463                 // Log the error encountered in Run(), if any
1464                 checkErr(err)
1465
1466                 if runner.finalState == "Queued" {
1467                         runner.UpdateContainerFinal()
1468                         return
1469                 }
1470
1471                 if runner.IsCancelled() {
1472                         runner.finalState = "Cancelled"
1473                         // but don't return yet -- we still want to
1474                         // capture partial output and write logs
1475                 }
1476
1477                 checkErr(runner.CaptureOutput())
1478                 checkErr(runner.CommitLogs())
1479                 checkErr(runner.UpdateContainerFinal())
1480         }()
1481
1482         err = runner.fetchContainerRecord()
1483         if err != nil {
1484                 return
1485         }
1486
1487         // setup signal handling
1488         runner.setupSignals()
1489
1490         // check for and/or load image
1491         err = runner.LoadImage()
1492         if err != nil {
1493                 runner.finalState = "Cancelled"
1494                 err = fmt.Errorf("While loading container image: %v", err)
1495                 return
1496         }
1497
1498         // set up FUSE mount and binds
1499         err = runner.SetupMounts()
1500         if err != nil {
1501                 runner.finalState = "Cancelled"
1502                 err = fmt.Errorf("While setting up mounts: %v", err)
1503                 return
1504         }
1505
1506         err = runner.CreateContainer()
1507         if err != nil {
1508                 return
1509         }
1510
1511         // Gather and record node information
1512         err = runner.LogNodeInfo()
1513         if err != nil {
1514                 return
1515         }
1516         // Save container.json record on log collection
1517         err = runner.LogContainerRecord()
1518         if err != nil {
1519                 return
1520         }
1521
1522         runner.StartCrunchstat()
1523
1524         if runner.IsCancelled() {
1525                 return
1526         }
1527
1528         err = runner.UpdateContainerRunning()
1529         if err != nil {
1530                 return
1531         }
1532         runner.finalState = "Cancelled"
1533
1534         err = runner.StartContainer()
1535         if err != nil {
1536                 return
1537         }
1538
1539         err = runner.WaitFinish()
1540         if err == nil {
1541                 runner.finalState = "Complete"
1542         }
1543         return
1544 }
1545
1546 // Fetch the current container record (uuid = runner.Container.UUID)
1547 // into runner.Container.
1548 func (runner *ContainerRunner) fetchContainerRecord() error {
1549         reader, err := runner.ArvClient.CallRaw("GET", "containers", runner.Container.UUID, "", nil)
1550         if err != nil {
1551                 return fmt.Errorf("error fetching container record: %v", err)
1552         }
1553         defer reader.Close()
1554
1555         dec := json.NewDecoder(reader)
1556         dec.UseNumber()
1557         err = dec.Decode(&runner.Container)
1558         if err != nil {
1559                 return fmt.Errorf("error decoding container record: %v", err)
1560         }
1561         return nil
1562 }
1563
1564 // NewContainerRunner creates a new container runner.
1565 func NewContainerRunner(api IArvadosClient,
1566         kc IKeepClient,
1567         docker ThinDockerClient,
1568         containerUUID string) *ContainerRunner {
1569
1570         cr := &ContainerRunner{ArvClient: api, Kc: kc, Docker: docker}
1571         cr.NewLogWriter = cr.NewArvLogWriter
1572         cr.RunArvMount = cr.ArvMountCmd
1573         cr.MkTempDir = ioutil.TempDir
1574         cr.LogCollection = &CollectionWriter{0, kc, nil, nil, sync.Mutex{}}
1575         cr.Container.UUID = containerUUID
1576         cr.CrunchLog = NewThrottledLogger(cr.NewLogWriter("crunch-run"))
1577         cr.CrunchLog.Immediate = log.New(os.Stderr, containerUUID+" ", 0)
1578
1579         loadLogThrottleParams(api)
1580
1581         return cr
1582 }
1583
1584 func main() {
1585         statInterval := flag.Duration("crunchstat-interval", 10*time.Second, "sampling period for periodic resource usage reporting")
1586         cgroupRoot := flag.String("cgroup-root", "/sys/fs/cgroup", "path to sysfs cgroup tree")
1587         cgroupParent := flag.String("cgroup-parent", "docker", "name of container's parent cgroup (ignored if -cgroup-parent-subsystem is used)")
1588         cgroupParentSubsystem := flag.String("cgroup-parent-subsystem", "", "use current cgroup for given subsystem as parent cgroup for container")
1589         caCertsPath := flag.String("ca-certs", "", "Path to TLS root certificates")
1590         enableNetwork := flag.String("container-enable-networking", "default",
1591                 `Specify if networking should be enabled for container.  One of 'default', 'always':
1592         default: only enable networking if container requests it.
1593         always:  containers always have networking enabled
1594         `)
1595         networkMode := flag.String("container-network-mode", "default",
1596                 `Set networking mode for container.  Corresponds to Docker network mode (--net).
1597         `)
1598         memprofile := flag.String("memprofile", "", "write memory profile to `file` after running container")
1599         getVersion := flag.Bool("version", false, "Print version information and exit.")
1600         flag.Parse()
1601
1602         // Print version information if requested
1603         if *getVersion {
1604                 fmt.Printf("Version: %s\n", version)
1605                 os.Exit(0)
1606         }
1607
1608         log.Printf("crunch-run %q started", version)
1609
1610         containerId := flag.Arg(0)
1611
1612         if *caCertsPath != "" {
1613                 arvadosclient.CertFiles = []string{*caCertsPath}
1614         }
1615
1616         api, err := arvadosclient.MakeArvadosClient()
1617         if err != nil {
1618                 log.Fatalf("%s: %v", containerId, err)
1619         }
1620         api.Retries = 8
1621
1622         var kc *keepclient.KeepClient
1623         kc, err = keepclient.MakeKeepClient(api)
1624         if err != nil {
1625                 log.Fatalf("%s: %v", containerId, err)
1626         }
1627         kc.BlockCache = &keepclient.BlockCache{MaxBlocks: 2}
1628         kc.Retries = 4
1629
1630         var docker *dockerclient.Client
1631         // API version 1.21 corresponds to Docker 1.9, which is currently the
1632         // minimum version we want to support.
1633         docker, err = dockerclient.NewClient(dockerclient.DefaultDockerHost, "1.21", nil, nil)
1634         if err != nil {
1635                 log.Fatalf("%s: %v", containerId, err)
1636         }
1637
1638         dockerClientProxy := ThinDockerClientProxy{Docker: docker}
1639
1640         cr := NewContainerRunner(api, kc, dockerClientProxy, containerId)
1641         cr.statInterval = *statInterval
1642         cr.cgroupRoot = *cgroupRoot
1643         cr.expectCgroupParent = *cgroupParent
1644         cr.enableNetwork = *enableNetwork
1645         cr.networkMode = *networkMode
1646         if *cgroupParentSubsystem != "" {
1647                 p := findCgroup(*cgroupParentSubsystem)
1648                 cr.setCgroupParent = p
1649                 cr.expectCgroupParent = p
1650         }
1651
1652         runerr := cr.Run()
1653
1654         if *memprofile != "" {
1655                 f, err := os.Create(*memprofile)
1656                 if err != nil {
1657                         log.Printf("could not create memory profile: ", err)
1658                 }
1659                 runtime.GC() // get up-to-date statistics
1660                 if err := pprof.WriteHeapProfile(f); err != nil {
1661                         log.Printf("could not write memory profile: ", err)
1662                 }
1663                 closeerr := f.Close()
1664                 if closeerr != nil {
1665                         log.Printf("closing memprofile file: ", err)
1666                 }
1667         }
1668
1669         if runerr != nil {
1670                 log.Fatalf("%s: %v", containerId, runerr)
1671         }
1672 }