services/crunch-run/crunchrun.go

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