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         "encoding/json"
  10         "errors"
  11         "flag"
  12         "fmt"
  13         "io"
  14         "io/ioutil"
  15         "log"
  16         "os"
  17         "os/exec"
  18         "os/signal"
  19         "path"
  20         "path/filepath"
  21         "regexp"
  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         "github.com/shirou/gopsutil/process"
  36         "golang.org/x/net/context"
  37
  38         dockertypes "github.com/docker/docker/api/types"
  39         dockercontainer "github.com/docker/docker/api/types/container"
  40         dockernetwork "github.com/docker/docker/api/types/network"
  41         dockerclient "github.com/docker/docker/client"
  42 )
  43
  44 var version = "dev"
  45
  46 // IArvadosClient is the minimal Arvados API methods used by crunch-run.
  47 type IArvadosClient interface {
  48         Create(resourceType string, parameters arvadosclient.Dict, output interface{}) error
  49         Get(resourceType string, uuid string, parameters arvadosclient.Dict, output interface{}) error
  50         Update(resourceType string, uuid string, parameters arvadosclient.Dict, output interface{}) error
  51         Call(method, resourceType, uuid, action string, parameters arvadosclient.Dict, output interface{}) error
  52         CallRaw(method string, resourceType string, uuid string, action string, parameters arvadosclient.Dict) (reader io.ReadCloser, err error)
  53         Discovery(key string) (interface{}, error)
  54 }
  55
  56 // ErrCancelled is the error returned when the container is cancelled.
  57 var ErrCancelled = errors.New("Cancelled")
  58
  59 // IKeepClient is the minimal Keep API methods used by crunch-run.
  60 type IKeepClient interface {
  61         PutB(buf []byte) (string, int, error)
  62         ReadAt(locator string, p []byte, off int) (int, error)
  63         ManifestFileReader(m manifest.Manifest, filename string) (arvados.File, error)
  64         ClearBlockCache()
  65 }
  66
  67 // NewLogWriter is a factory function to create a new log writer.
  68 type NewLogWriter func(name string) (io.WriteCloser, error)
  69
  70 type RunArvMount func(args []string, tok string) (*exec.Cmd, error)
  71
  72 type MkTempDir func(string, string) (string, error)
  73
  74 // ThinDockerClient is the minimal Docker client interface used by crunch-run.
  75 type ThinDockerClient interface {
  76         ContainerAttach(ctx context.Context, container string, options dockertypes.ContainerAttachOptions) (dockertypes.HijackedResponse, error)
  77         ContainerCreate(ctx context.Context, config *dockercontainer.Config, hostConfig *dockercontainer.HostConfig,
  78                 networkingConfig *dockernetwork.NetworkingConfig, containerName string) (dockercontainer.ContainerCreateCreatedBody, error)
  79         ContainerStart(ctx context.Context, container string, options dockertypes.ContainerStartOptions) error
  80         ContainerRemove(ctx context.Context, container string, options dockertypes.ContainerRemoveOptions) error
  81         ContainerWait(ctx context.Context, container string, condition dockercontainer.WaitCondition) (<-chan dockercontainer.ContainerWaitOKBody, <-chan error)
  82         ContainerInspect(ctx context.Context, id string) (dockertypes.ContainerJSON, error)
  83         ImageInspectWithRaw(ctx context.Context, image string) (dockertypes.ImageInspect, []byte, error)
  84         ImageLoad(ctx context.Context, input io.Reader, quiet bool) (dockertypes.ImageLoadResponse, error)
  85         ImageRemove(ctx context.Context, image string, options dockertypes.ImageRemoveOptions) ([]dockertypes.ImageDeleteResponseItem, error)
  86 }
  87
  88 type PsProcess interface {
  89         CmdlineSlice() ([]string, error)
  90 }
  91
  92 // ContainerRunner is the main stateful struct used for a single execution of a
  93 // container.
  94 type ContainerRunner struct {
  95         Docker          ThinDockerClient
  96         client          *arvados.Client
  97         ArvClient       IArvadosClient
  98         Kc              IKeepClient
  99         Container       arvados.Container
 100         ContainerConfig dockercontainer.Config
 101         HostConfig      dockercontainer.HostConfig
 102         token           string
 103         ContainerID     string
 104         ExitCode        *int
 105         NewLogWriter    NewLogWriter
 106         loggingDone     chan bool
 107         CrunchLog       *ThrottledLogger
 108         Stdout          io.WriteCloser
 109         Stderr          io.WriteCloser
 110         logUUID         string
 111         logMtx          sync.Mutex
 112         LogCollection   arvados.CollectionFileSystem
 113         LogsPDH         *string
 114         RunArvMount     RunArvMount
 115         MkTempDir       MkTempDir
 116         ArvMount        *exec.Cmd
 117         ArvMountPoint   string
 118         HostOutputDir   string
 119         Binds           []string
 120         Volumes         map[string]struct{}
 121         OutputPDH       *string
 122         SigChan         chan os.Signal
 123         ArvMountExit    chan error
 124         SecretMounts    map[string]arvados.Mount
 125         MkArvClient     func(token string) (IArvadosClient, error)
 126         finalState      string
 127         parentTemp      string
 128
 129         ListProcesses func() ([]PsProcess, error)
 130
 131         statLogger       io.WriteCloser
 132         statReporter     *crunchstat.Reporter
 133         hoststatLogger   io.WriteCloser
 134         hoststatReporter *crunchstat.Reporter
 135         statInterval     time.Duration
 136         cgroupRoot       string
 137         // What we expect the container's cgroup parent to be.
 138         expectCgroupParent string
 139         // What we tell docker to use as the container's cgroup
 140         // parent. Note: Ideally we would use the same field for both
 141         // expectCgroupParent and setCgroupParent, and just make it
 142         // default to "docker". However, when using docker < 1.10 with
 143         // systemd, specifying a non-empty cgroup parent (even the
 144         // default value "docker") hits a docker bug
 145         // (https://github.com/docker/docker/issues/17126). Using two
 146         // separate fields makes it possible to use the "expect cgroup
 147         // parent to be X" feature even on sites where the "specify
 148         // cgroup parent" feature breaks.
 149         setCgroupParent string
 150
 151         cStateLock sync.Mutex
 152         cCancelled bool // StopContainer() invoked
 153         cRemoved   bool // docker confirmed the container no longer exists
 154
 155         enableNetwork   string // one of "default" or "always"
 156         networkMode     string // passed through to HostConfig.NetworkMode
 157         arvMountLog     *ThrottledLogger
 158         checkContainerd time.Duration
 159
 160         containerWatchdogInterval time.Duration
 161 }
 162
 163 // setupSignals sets up signal handling to gracefully terminate the underlying
 164 // Docker container and update state when receiving a TERM, INT or QUIT signal.
 165 func (runner *ContainerRunner) setupSignals() {
 166         runner.SigChan = make(chan os.Signal, 1)
 167         signal.Notify(runner.SigChan, syscall.SIGTERM)
 168         signal.Notify(runner.SigChan, syscall.SIGINT)
 169         signal.Notify(runner.SigChan, syscall.SIGQUIT)
 170
 171         go func(sig chan os.Signal) {
 172                 for s := range sig {
 173                         runner.stop(s)
 174                 }
 175         }(runner.SigChan)
 176 }
 177
 178 // stop the underlying Docker container.
 179 func (runner *ContainerRunner) stop(sig os.Signal) {
 180         runner.cStateLock.Lock()
 181         defer runner.cStateLock.Unlock()
 182         if sig != nil {
 183                 runner.CrunchLog.Printf("caught signal: %v", sig)
 184         }
 185         if runner.ContainerID == "" {
 186                 return
 187         }
 188         runner.cCancelled = true
 189         runner.CrunchLog.Printf("removing container")
 190         err := runner.Docker.ContainerRemove(context.TODO(), runner.ContainerID, dockertypes.ContainerRemoveOptions{Force: true})
 191         if err != nil {
 192                 runner.CrunchLog.Printf("error removing container: %s", err)
 193         }
 194         if err == nil || strings.Contains(err.Error(), "No such container: "+runner.ContainerID) {
 195                 runner.cRemoved = true
 196         }
 197 }
 198
 199 var errorBlacklist = []string{
 200         "(?ms).*[Cc]annot connect to the Docker daemon.*",
 201         "(?ms).*oci runtime error.*starting container process.*container init.*mounting.*to rootfs.*no such file or directory.*",
 202         "(?ms).*grpc: the connection is unavailable.*",
 203 }
 204 var brokenNodeHook *string = flag.String("broken-node-hook", "", "Script to run if node is detected to be broken (for example, Docker daemon is not running)")
 205
 206 func (runner *ContainerRunner) runBrokenNodeHook() {
 207         if *brokenNodeHook == "" {
 208                 runner.CrunchLog.Printf("No broken node hook provided, cannot mark node as broken.")
 209         } else {
 210                 runner.CrunchLog.Printf("Running broken node hook %q", *brokenNodeHook)
 211                 // run killme script
 212                 c := exec.Command(*brokenNodeHook)
 213                 c.Stdout = runner.CrunchLog
 214                 c.Stderr = runner.CrunchLog
 215                 err := c.Run()
 216                 if err != nil {
 217                         runner.CrunchLog.Printf("Error running broken node hook: %v", err)
 218                 }
 219         }
 220 }
 221
 222 func (runner *ContainerRunner) checkBrokenNode(goterr error) bool {
 223         for _, d := range errorBlacklist {
 224                 if m, e := regexp.MatchString(d, goterr.Error()); m && e == nil {
 225                         runner.CrunchLog.Printf("Error suggests node is unable to run containers: %v", goterr)
 226                         runner.runBrokenNodeHook()
 227                         return true
 228                 }
 229         }
 230         return false
 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         w, err := runner.NewLogWriter("arv-mount")
 303         if err != nil {
 304                 return nil, err
 305         }
 306         runner.arvMountLog = NewThrottledLogger(w)
 307         c.Stdout = runner.arvMountLog
 308         c.Stderr = runner.arvMountLog
 309
 310         runner.CrunchLog.Printf("Running %v", c.Args)
 311
 312         err = c.Start()
 313         if err != nil {
 314                 return nil, err
 315         }
 316
 317         statReadme := make(chan bool)
 318         runner.ArvMountExit = make(chan error)
 319
 320         keepStatting := true
 321         go func() {
 322                 for keepStatting {
 323                         time.Sleep(100 * time.Millisecond)
 324                         _, err = os.Stat(fmt.Sprintf("%s/by_id/README", runner.ArvMountPoint))
 325                         if err == nil {
 326                                 keepStatting = false
 327                                 statReadme <- true
 328                         }
 329                 }
 330                 close(statReadme)
 331         }()
 332
 333         go func() {
 334                 mnterr := c.Wait()
 335                 if mnterr != nil {
 336                         runner.CrunchLog.Printf("Arv-mount exit error: %v", mnterr)
 337                 }
 338                 runner.ArvMountExit <- mnterr
 339                 close(runner.ArvMountExit)
 340         }()
 341
 342         select {
 343         case <-statReadme:
 344                 break
 345         case err := <-runner.ArvMountExit:
 346                 runner.ArvMount = nil
 347                 keepStatting = false
 348                 return nil, err
 349         }
 350
 351         return c, nil
 352 }
 353
 354 func (runner *ContainerRunner) SetupArvMountPoint(prefix string) (err error) {
 355         if runner.ArvMountPoint == "" {
 356                 runner.ArvMountPoint, err = runner.MkTempDir(runner.parentTemp, prefix)
 357         }
 358         return
 359 }
 360
 361 func copyfile(src string, dst string) (err error) {
 362         srcfile, err := os.Open(src)
 363         if err != nil {
 364                 return
 365         }
 366
 367         os.MkdirAll(path.Dir(dst), 0777)
 368
 369         dstfile, err := os.Create(dst)
 370         if err != nil {
 371                 return
 372         }
 373         _, err = io.Copy(dstfile, srcfile)
 374         if err != nil {
 375                 return
 376         }
 377
 378         err = srcfile.Close()
 379         err2 := dstfile.Close()
 380
 381         if err != nil {
 382                 return
 383         }
 384
 385         if err2 != nil {
 386                 return err2
 387         }
 388
 389         return nil
 390 }
 391
 392 func (runner *ContainerRunner) SetupMounts() (err error) {
 393         err = runner.SetupArvMountPoint("keep")
 394         if err != nil {
 395                 return fmt.Errorf("While creating keep mount temp dir: %v", err)
 396         }
 397
 398         token, err := runner.ContainerToken()
 399         if err != nil {
 400                 return fmt.Errorf("could not get container token: %s", err)
 401         }
 402
 403         pdhOnly := true
 404         tmpcount := 0
 405         arvMountCmd := []string{
 406                 "--foreground",
 407                 "--allow-other",
 408                 "--read-write",
 409                 fmt.Sprintf("--crunchstat-interval=%v", runner.statInterval.Seconds())}
 410
 411         if runner.Container.RuntimeConstraints.KeepCacheRAM > 0 {
 412                 arvMountCmd = append(arvMountCmd, "--file-cache", fmt.Sprintf("%d", runner.Container.RuntimeConstraints.KeepCacheRAM))
 413         }
 414
 415         collectionPaths := []string{}
 416         runner.Binds = nil
 417         runner.Volumes = make(map[string]struct{})
 418         needCertMount := true
 419         type copyFile struct {
 420                 src  string
 421                 bind string
 422         }
 423         var copyFiles []copyFile
 424
 425         var binds []string
 426         for bind := range runner.Container.Mounts {
 427                 binds = append(binds, bind)
 428         }
 429         for bind := range runner.SecretMounts {
 430                 if _, ok := runner.Container.Mounts[bind]; ok {
 431                         return fmt.Errorf("Secret mount %q conflicts with regular mount", bind)
 432                 }
 433                 if runner.SecretMounts[bind].Kind != "json" &&
 434                         runner.SecretMounts[bind].Kind != "text" {
 435                         return fmt.Errorf("Secret mount %q type is %q but only 'json' and 'text' are permitted.",
 436                                 bind, runner.SecretMounts[bind].Kind)
 437                 }
 438                 binds = append(binds, bind)
 439         }
 440         sort.Strings(binds)
 441
 442         for _, bind := range binds {
 443                 mnt, ok := runner.Container.Mounts[bind]
 444                 if !ok {
 445                         mnt = runner.SecretMounts[bind]
 446                 }
 447                 if bind == "stdout" || bind == "stderr" {
 448                         // Is it a "file" mount kind?
 449                         if mnt.Kind != "file" {
 450                                 return fmt.Errorf("Unsupported mount kind '%s' for %s. Only 'file' is supported.", mnt.Kind, bind)
 451                         }
 452
 453                         // Does path start with OutputPath?
 454                         prefix := runner.Container.OutputPath
 455                         if !strings.HasSuffix(prefix, "/") {
 456                                 prefix += "/"
 457                         }
 458                         if !strings.HasPrefix(mnt.Path, prefix) {
 459                                 return fmt.Errorf("%s path does not start with OutputPath: %s, %s", strings.Title(bind), mnt.Path, prefix)
 460                         }
 461                 }
 462
 463                 if bind == "stdin" {
 464                         // Is it a "collection" mount kind?
 465                         if mnt.Kind != "collection" && mnt.Kind != "json" {
 466                                 return fmt.Errorf("Unsupported mount kind '%s' for stdin. Only 'collection' or 'json' are supported.", mnt.Kind)
 467                         }
 468                 }
 469
 470                 if bind == "/etc/arvados/ca-certificates.crt" {
 471                         needCertMount = false
 472                 }
 473
 474                 if strings.HasPrefix(bind, runner.Container.OutputPath+"/") && bind != runner.Container.OutputPath+"/" {
 475                         if mnt.Kind != "collection" && mnt.Kind != "text" && mnt.Kind != "json" {
 476                                 return fmt.Errorf("Only mount points of kind 'collection', 'text' or 'json' are supported underneath the output_path for %q, was %q", bind, mnt.Kind)
 477                         }
 478                 }
 479
 480                 switch {
 481                 case mnt.Kind == "collection" && bind != "stdin":
 482                         var src string
 483                         if mnt.UUID != "" && mnt.PortableDataHash != "" {
 484                                 return fmt.Errorf("Cannot specify both 'uuid' and 'portable_data_hash' for a collection mount")
 485                         }
 486                         if mnt.UUID != "" {
 487                                 if mnt.Writable {
 488                                         return fmt.Errorf("Writing to existing collections currently not permitted.")
 489                                 }
 490                                 pdhOnly = false
 491                                 src = fmt.Sprintf("%s/by_id/%s", runner.ArvMountPoint, mnt.UUID)
 492                         } else if mnt.PortableDataHash != "" {
 493                                 if mnt.Writable && !strings.HasPrefix(bind, runner.Container.OutputPath+"/") {
 494                                         return fmt.Errorf("Can never write to a collection specified by portable data hash")
 495                                 }
 496                                 idx := strings.Index(mnt.PortableDataHash, "/")
 497                                 if idx > 0 {
 498                                         mnt.Path = path.Clean(mnt.PortableDataHash[idx:])
 499                                         mnt.PortableDataHash = mnt.PortableDataHash[0:idx]
 500                                         runner.Container.Mounts[bind] = mnt
 501                                 }
 502                                 src = fmt.Sprintf("%s/by_id/%s", runner.ArvMountPoint, mnt.PortableDataHash)
 503                                 if mnt.Path != "" && mnt.Path != "." {
 504                                         if strings.HasPrefix(mnt.Path, "./") {
 505                                                 mnt.Path = mnt.Path[2:]
 506                                         } else if strings.HasPrefix(mnt.Path, "/") {
 507                                                 mnt.Path = mnt.Path[1:]
 508                                         }
 509                                         src += "/" + mnt.Path
 510                                 }
 511                         } else {
 512                                 src = fmt.Sprintf("%s/tmp%d", runner.ArvMountPoint, tmpcount)
 513                                 arvMountCmd = append(arvMountCmd, "--mount-tmp")
 514                                 arvMountCmd = append(arvMountCmd, fmt.Sprintf("tmp%d", tmpcount))
 515                                 tmpcount += 1
 516                         }
 517                         if mnt.Writable {
 518                                 if bind == runner.Container.OutputPath {
 519                                         runner.HostOutputDir = src
 520                                         runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s", src, bind))
 521                                 } else if strings.HasPrefix(bind, runner.Container.OutputPath+"/") {
 522                                         copyFiles = append(copyFiles, copyFile{src, runner.HostOutputDir + bind[len(runner.Container.OutputPath):]})
 523                                 } else {
 524                                         runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s", src, bind))
 525                                 }
 526                         } else {
 527                                 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s:ro", src, bind))
 528                         }
 529                         collectionPaths = append(collectionPaths, src)
 530
 531                 case mnt.Kind == "tmp":
 532                         var tmpdir string
 533                         tmpdir, err = runner.MkTempDir(runner.parentTemp, "tmp")
 534                         if err != nil {
 535                                 return fmt.Errorf("While creating mount temp dir: %v", err)
 536                         }
 537                         st, staterr := os.Stat(tmpdir)
 538                         if staterr != nil {
 539                                 return fmt.Errorf("While Stat on temp dir: %v", staterr)
 540                         }
 541                         err = os.Chmod(tmpdir, st.Mode()|os.ModeSetgid|0777)
 542                         if staterr != nil {
 543                                 return fmt.Errorf("While Chmod temp dir: %v", err)
 544                         }
 545                         runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s", tmpdir, bind))
 546                         if bind == runner.Container.OutputPath {
 547                                 runner.HostOutputDir = tmpdir
 548                         }
 549
 550                 case mnt.Kind == "json" || mnt.Kind == "text":
 551                         var filedata []byte
 552                         if mnt.Kind == "json" {
 553                                 filedata, err = json.Marshal(mnt.Content)
 554                                 if err != nil {
 555                                         return fmt.Errorf("encoding json data: %v", err)
 556                                 }
 557                         } else {
 558                                 text, ok := mnt.Content.(string)
 559                                 if !ok {
 560                                         return fmt.Errorf("content for mount %q must be a string", bind)
 561                                 }
 562                                 filedata = []byte(text)
 563                         }
 564
 565                         tmpdir, err := runner.MkTempDir(runner.parentTemp, mnt.Kind)
 566                         if err != nil {
 567                                 return fmt.Errorf("creating temp dir: %v", err)
 568                         }
 569                         tmpfn := filepath.Join(tmpdir, "mountdata."+mnt.Kind)
 570                         err = ioutil.WriteFile(tmpfn, filedata, 0444)
 571                         if err != nil {
 572                                 return fmt.Errorf("writing temp file: %v", err)
 573                         }
 574                         if strings.HasPrefix(bind, runner.Container.OutputPath+"/") {
 575                                 copyFiles = append(copyFiles, copyFile{tmpfn, runner.HostOutputDir + bind[len(runner.Container.OutputPath):]})
 576                         } else {
 577                                 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s:ro", tmpfn, bind))
 578                         }
 579
 580                 case mnt.Kind == "git_tree":
 581                         tmpdir, err := runner.MkTempDir(runner.parentTemp, "git_tree")
 582                         if err != nil {
 583                                 return fmt.Errorf("creating temp dir: %v", err)
 584                         }
 585                         err = gitMount(mnt).extractTree(runner.ArvClient, tmpdir, token)
 586                         if err != nil {
 587                                 return err
 588                         }
 589                         runner.Binds = append(runner.Binds, tmpdir+":"+bind+":ro")
 590                 }
 591         }
 592
 593         if runner.HostOutputDir == "" {
 594                 return fmt.Errorf("Output path does not correspond to a writable mount point")
 595         }
 596
 597         if wantAPI := runner.Container.RuntimeConstraints.API; needCertMount && wantAPI != nil && *wantAPI {
 598                 for _, certfile := range arvadosclient.CertFiles {
 599                         _, err := os.Stat(certfile)
 600                         if err == nil {
 601                                 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:/etc/arvados/ca-certificates.crt:ro", certfile))
 602                                 break
 603                         }
 604                 }
 605         }
 606
 607         if pdhOnly {
 608                 arvMountCmd = append(arvMountCmd, "--mount-by-pdh", "by_id")
 609         } else {
 610                 arvMountCmd = append(arvMountCmd, "--mount-by-id", "by_id")
 611         }
 612         arvMountCmd = append(arvMountCmd, runner.ArvMountPoint)
 613
 614         runner.ArvMount, err = runner.RunArvMount(arvMountCmd, token)
 615         if err != nil {
 616                 return fmt.Errorf("While trying to start arv-mount: %v", err)
 617         }
 618
 619         for _, p := range collectionPaths {
 620                 _, err = os.Stat(p)
 621                 if err != nil {
 622                         return fmt.Errorf("While checking that input files exist: %v", err)
 623                 }
 624         }
 625
 626         for _, cp := range copyFiles {
 627                 st, err := os.Stat(cp.src)
 628                 if err != nil {
 629                         return fmt.Errorf("While staging writable file from %q to %q: %v", cp.src, cp.bind, err)
 630                 }
 631                 if st.IsDir() {
 632                         err = filepath.Walk(cp.src, func(walkpath string, walkinfo os.FileInfo, walkerr error) error {
 633                                 if walkerr != nil {
 634                                         return walkerr
 635                                 }
 636                                 target := path.Join(cp.bind, walkpath[len(cp.src):])
 637                                 if walkinfo.Mode().IsRegular() {
 638                                         copyerr := copyfile(walkpath, target)
 639                                         if copyerr != nil {
 640                                                 return copyerr
 641                                         }
 642                                         return os.Chmod(target, walkinfo.Mode()|0777)
 643                                 } else if walkinfo.Mode().IsDir() {
 644                                         mkerr := os.MkdirAll(target, 0777)
 645                                         if mkerr != nil {
 646                                                 return mkerr
 647                                         }
 648                                         return os.Chmod(target, walkinfo.Mode()|os.ModeSetgid|0777)
 649                                 } else {
 650                                         return fmt.Errorf("Source %q is not a regular file or directory", cp.src)
 651                                 }
 652                         })
 653                 } else if st.Mode().IsRegular() {
 654                         err = copyfile(cp.src, cp.bind)
 655                         if err == nil {
 656                                 err = os.Chmod(cp.bind, st.Mode()|0777)
 657                         }
 658                 }
 659                 if err != nil {
 660                         return fmt.Errorf("While staging writable file from %q to %q: %v", cp.src, cp.bind, err)
 661                 }
 662         }
 663
 664         return nil
 665 }
 666
 667 func (runner *ContainerRunner) ProcessDockerAttach(containerReader io.Reader) {
 668         // Handle docker log protocol
 669         // https://docs.docker.com/engine/reference/api/docker_remote_api_v1.15/#attach-to-a-container
 670         defer close(runner.loggingDone)
 671
 672         header := make([]byte, 8)
 673         var err error
 674         for err == nil {
 675                 _, err = io.ReadAtLeast(containerReader, header, 8)
 676                 if err != nil {
 677                         if err == io.EOF {
 678                                 err = nil
 679                         }
 680                         break
 681                 }
 682                 readsize := int64(header[7]) | (int64(header[6]) << 8) | (int64(header[5]) << 16) | (int64(header[4]) << 24)
 683                 if header[0] == 1 {
 684                         // stdout
 685                         _, err = io.CopyN(runner.Stdout, containerReader, readsize)
 686                 } else {
 687                         // stderr
 688                         _, err = io.CopyN(runner.Stderr, containerReader, readsize)
 689                 }
 690         }
 691
 692         if err != nil {
 693                 runner.CrunchLog.Printf("error reading docker logs: %v", err)
 694         }
 695
 696         err = runner.Stdout.Close()
 697         if err != nil {
 698                 runner.CrunchLog.Printf("error closing stdout logs: %v", err)
 699         }
 700
 701         err = runner.Stderr.Close()
 702         if err != nil {
 703                 runner.CrunchLog.Printf("error closing stderr logs: %v", err)
 704         }
 705
 706         if runner.statReporter != nil {
 707                 runner.statReporter.Stop()
 708                 err = runner.statLogger.Close()
 709                 if err != nil {
 710                         runner.CrunchLog.Printf("error closing crunchstat logs: %v", err)
 711                 }
 712         }
 713 }
 714
 715 func (runner *ContainerRunner) stopHoststat() error {
 716         if runner.hoststatReporter == nil {
 717                 return nil
 718         }
 719         runner.hoststatReporter.Stop()
 720         err := runner.hoststatLogger.Close()
 721         if err != nil {
 722                 return fmt.Errorf("error closing hoststat logs: %v", err)
 723         }
 724         return nil
 725 }
 726
 727 func (runner *ContainerRunner) startHoststat() error {
 728         w, err := runner.NewLogWriter("hoststat")
 729         if err != nil {
 730                 return err
 731         }
 732         runner.hoststatLogger = NewThrottledLogger(w)
 733         runner.hoststatReporter = &crunchstat.Reporter{
 734                 Logger:     log.New(runner.hoststatLogger, "", 0),
 735                 CgroupRoot: runner.cgroupRoot,
 736                 PollPeriod: runner.statInterval,
 737         }
 738         runner.hoststatReporter.Start()
 739         return nil
 740 }
 741
 742 func (runner *ContainerRunner) startCrunchstat() error {
 743         w, err := runner.NewLogWriter("crunchstat")
 744         if err != nil {
 745                 return err
 746         }
 747         runner.statLogger = NewThrottledLogger(w)
 748         runner.statReporter = &crunchstat.Reporter{
 749                 CID:          runner.ContainerID,
 750                 Logger:       log.New(runner.statLogger, "", 0),
 751                 CgroupParent: runner.expectCgroupParent,
 752                 CgroupRoot:   runner.cgroupRoot,
 753                 PollPeriod:   runner.statInterval,
 754                 TempDir:      runner.parentTemp,
 755         }
 756         runner.statReporter.Start()
 757         return nil
 758 }
 759
 760 type infoCommand struct {
 761         label string
 762         cmd   []string
 763 }
 764
 765 // LogHostInfo logs info about the current host, for debugging and
 766 // accounting purposes. Although it's logged as "node-info", this is
 767 // about the environment where crunch-run is actually running, which
 768 // might differ from what's described in the node record (see
 769 // LogNodeRecord).
 770 func (runner *ContainerRunner) LogHostInfo() (err error) {
 771         w, err := runner.NewLogWriter("node-info")
 772         if err != nil {
 773                 return
 774         }
 775
 776         commands := []infoCommand{
 777                 {
 778                         label: "Host Information",
 779                         cmd:   []string{"uname", "-a"},
 780                 },
 781                 {
 782                         label: "CPU Information",
 783                         cmd:   []string{"cat", "/proc/cpuinfo"},
 784                 },
 785                 {
 786                         label: "Memory Information",
 787                         cmd:   []string{"cat", "/proc/meminfo"},
 788                 },
 789                 {
 790                         label: "Disk Space",
 791                         cmd:   []string{"df", "-m", "/", os.TempDir()},
 792                 },
 793                 {
 794                         label: "Disk INodes",
 795                         cmd:   []string{"df", "-i", "/", os.TempDir()},
 796                 },
 797         }
 798
 799         // Run commands with informational output to be logged.
 800         for _, command := range commands {
 801                 fmt.Fprintln(w, command.label)
 802                 cmd := exec.Command(command.cmd[0], command.cmd[1:]...)
 803                 cmd.Stdout = w
 804                 cmd.Stderr = w
 805                 if err := cmd.Run(); err != nil {
 806                         err = fmt.Errorf("While running command %q: %v", command.cmd, err)
 807                         fmt.Fprintln(w, err)
 808                         return err
 809                 }
 810                 fmt.Fprintln(w, "")
 811         }
 812
 813         err = w.Close()
 814         if err != nil {
 815                 return fmt.Errorf("While closing node-info logs: %v", err)
 816         }
 817         return nil
 818 }
 819
 820 // LogContainerRecord gets and saves the raw JSON container record from the API server
 821 func (runner *ContainerRunner) LogContainerRecord() error {
 822         logged, err := runner.logAPIResponse("container", "containers", map[string]interface{}{"filters": [][]string{{"uuid", "=", runner.Container.UUID}}}, nil)
 823         if !logged && err == nil {
 824                 err = fmt.Errorf("error: no container record found for %s", runner.Container.UUID)
 825         }
 826         return err
 827 }
 828
 829 // LogNodeRecord logs arvados#node record corresponding to the current host.
 830 func (runner *ContainerRunner) LogNodeRecord() error {
 831         hostname := os.Getenv("SLURMD_NODENAME")
 832         if hostname == "" {
 833                 hostname, _ = os.Hostname()
 834         }
 835         _, err := runner.logAPIResponse("node", "nodes", map[string]interface{}{"filters": [][]string{{"hostname", "=", hostname}}}, func(resp interface{}) {
 836                 // The "info" field has admin-only info when obtained
 837                 // with a privileged token, and should not be logged.
 838                 node, ok := resp.(map[string]interface{})
 839                 if ok {
 840                         delete(node, "info")
 841                 }
 842         })
 843         return err
 844 }
 845
 846 func (runner *ContainerRunner) logAPIResponse(label, path string, params map[string]interface{}, munge func(interface{})) (logged bool, err error) {
 847         writer, err := runner.LogCollection.OpenFile(label+".json", os.O_CREATE|os.O_WRONLY, 0666)
 848         if err != nil {
 849                 return false, err
 850         }
 851         w := &ArvLogWriter{
 852                 ArvClient:     runner.ArvClient,
 853                 UUID:          runner.Container.UUID,
 854                 loggingStream: label,
 855                 writeCloser:   writer,
 856         }
 857
 858         reader, err := runner.ArvClient.CallRaw("GET", path, "", "", arvadosclient.Dict(params))
 859         if err != nil {
 860                 return false, fmt.Errorf("error getting %s record: %v", label, err)
 861         }
 862         defer reader.Close()
 863
 864         dec := json.NewDecoder(reader)
 865         dec.UseNumber()
 866         var resp map[string]interface{}
 867         if err = dec.Decode(&resp); err != nil {
 868                 return false, fmt.Errorf("error decoding %s list response: %v", label, err)
 869         }
 870         items, ok := resp["items"].([]interface{})
 871         if !ok {
 872                 return false, fmt.Errorf("error decoding %s list response: no \"items\" key in API list response", label)
 873         } else if len(items) < 1 {
 874                 return false, nil
 875         }
 876         if munge != nil {
 877                 munge(items[0])
 878         }
 879         // Re-encode it using indentation to improve readability
 880         enc := json.NewEncoder(w)
 881         enc.SetIndent("", "    ")
 882         if err = enc.Encode(items[0]); err != nil {
 883                 return false, fmt.Errorf("error logging %s record: %v", label, err)
 884         }
 885         err = w.Close()
 886         if err != nil {
 887                 return false, fmt.Errorf("error closing %s.json in log collection: %v", label, err)
 888         }
 889         return true, nil
 890 }
 891
 892 // AttachStreams connects the docker container stdin, stdout and stderr logs
 893 // to the Arvados logger which logs to Keep and the API server logs table.
 894 func (runner *ContainerRunner) AttachStreams() (err error) {
 895
 896         runner.CrunchLog.Print("Attaching container streams")
 897
 898         // If stdin mount is provided, attach it to the docker container
 899         var stdinRdr arvados.File
 900         var stdinJson []byte
 901         if stdinMnt, ok := runner.Container.Mounts["stdin"]; ok {
 902                 if stdinMnt.Kind == "collection" {
 903                         var stdinColl arvados.Collection
 904                         collId := stdinMnt.UUID
 905                         if collId == "" {
 906                                 collId = stdinMnt.PortableDataHash
 907                         }
 908                         err = runner.ArvClient.Get("collections", collId, nil, &stdinColl)
 909                         if err != nil {
 910                                 return fmt.Errorf("While getting stding collection: %v", err)
 911                         }
 912
 913                         stdinRdr, err = runner.Kc.ManifestFileReader(manifest.Manifest{Text: stdinColl.ManifestText}, stdinMnt.Path)
 914                         if os.IsNotExist(err) {
 915                                 return fmt.Errorf("stdin collection path not found: %v", stdinMnt.Path)
 916                         } else if err != nil {
 917                                 return fmt.Errorf("While getting stdin collection path %v: %v", stdinMnt.Path, err)
 918                         }
 919                 } else if stdinMnt.Kind == "json" {
 920                         stdinJson, err = json.Marshal(stdinMnt.Content)
 921                         if err != nil {
 922                                 return fmt.Errorf("While encoding stdin json data: %v", err)
 923                         }
 924                 }
 925         }
 926
 927         stdinUsed := stdinRdr != nil || len(stdinJson) != 0
 928         response, err := runner.Docker.ContainerAttach(context.TODO(), runner.ContainerID,
 929                 dockertypes.ContainerAttachOptions{Stream: true, Stdin: stdinUsed, Stdout: true, Stderr: true})
 930         if err != nil {
 931                 return fmt.Errorf("While attaching container stdout/stderr streams: %v", err)
 932         }
 933
 934         runner.loggingDone = make(chan bool)
 935
 936         if stdoutMnt, ok := runner.Container.Mounts["stdout"]; ok {
 937                 stdoutFile, err := runner.getStdoutFile(stdoutMnt.Path)
 938                 if err != nil {
 939                         return err
 940                 }
 941                 runner.Stdout = stdoutFile
 942         } else if w, err := runner.NewLogWriter("stdout"); err != nil {
 943                 return err
 944         } else {
 945                 runner.Stdout = NewThrottledLogger(w)
 946         }
 947
 948         if stderrMnt, ok := runner.Container.Mounts["stderr"]; ok {
 949                 stderrFile, err := runner.getStdoutFile(stderrMnt.Path)
 950                 if err != nil {
 951                         return err
 952                 }
 953                 runner.Stderr = stderrFile
 954         } else if w, err := runner.NewLogWriter("stderr"); err != nil {
 955                 return err
 956         } else {
 957                 runner.Stderr = NewThrottledLogger(w)
 958         }
 959
 960         if stdinRdr != nil {
 961                 go func() {
 962                         _, err := io.Copy(response.Conn, stdinRdr)
 963                         if err != nil {
 964                                 runner.CrunchLog.Print("While writing stdin collection to docker container %q", err)
 965                                 runner.stop(nil)
 966                         }
 967                         stdinRdr.Close()
 968                         response.CloseWrite()
 969                 }()
 970         } else if len(stdinJson) != 0 {
 971                 go func() {
 972                         _, err := io.Copy(response.Conn, bytes.NewReader(stdinJson))
 973                         if err != nil {
 974                                 runner.CrunchLog.Print("While writing stdin json to docker container %q", err)
 975                                 runner.stop(nil)
 976                         }
 977                         response.CloseWrite()
 978                 }()
 979         }
 980
 981         go runner.ProcessDockerAttach(response.Reader)
 982
 983         return nil
 984 }
 985
 986 func (runner *ContainerRunner) getStdoutFile(mntPath string) (*os.File, error) {
 987         stdoutPath := mntPath[len(runner.Container.OutputPath):]
 988         index := strings.LastIndex(stdoutPath, "/")
 989         if index > 0 {
 990                 subdirs := stdoutPath[:index]
 991                 if subdirs != "" {
 992                         st, err := os.Stat(runner.HostOutputDir)
 993                         if err != nil {
 994                                 return nil, fmt.Errorf("While Stat on temp dir: %v", err)
 995                         }
 996                         stdoutPath := filepath.Join(runner.HostOutputDir, subdirs)
 997                         err = os.MkdirAll(stdoutPath, st.Mode()|os.ModeSetgid|0777)
 998                         if err != nil {
 999                                 return nil, fmt.Errorf("While MkdirAll %q: %v", stdoutPath, err)
1000                         }
1001                 }
1002         }
1003         stdoutFile, err := os.Create(filepath.Join(runner.HostOutputDir, stdoutPath))
1004         if err != nil {
1005                 return nil, fmt.Errorf("While creating file %q: %v", stdoutPath, err)
1006         }
1007
1008         return stdoutFile, nil
1009 }
1010
1011 // CreateContainer creates the docker container.
1012 func (runner *ContainerRunner) CreateContainer() error {
1013         runner.CrunchLog.Print("Creating Docker container")
1014
1015         runner.ContainerConfig.Cmd = runner.Container.Command
1016         if runner.Container.Cwd != "." {
1017                 runner.ContainerConfig.WorkingDir = runner.Container.Cwd
1018         }
1019
1020         for k, v := range runner.Container.Environment {
1021                 runner.ContainerConfig.Env = append(runner.ContainerConfig.Env, k+"="+v)
1022         }
1023
1024         runner.ContainerConfig.Volumes = runner.Volumes
1025
1026         maxRAM := int64(runner.Container.RuntimeConstraints.RAM)
1027         if maxRAM < 4*1024*1024 {
1028                 // Docker daemon won't let you set a limit less than 4 MiB
1029                 maxRAM = 4 * 1024 * 1024
1030         }
1031         runner.HostConfig = dockercontainer.HostConfig{
1032                 Binds: runner.Binds,
1033                 LogConfig: dockercontainer.LogConfig{
1034                         Type: "none",
1035                 },
1036                 Resources: dockercontainer.Resources{
1037                         CgroupParent: runner.setCgroupParent,
1038                         NanoCPUs:     int64(runner.Container.RuntimeConstraints.VCPUs) * 1000000000,
1039                         Memory:       maxRAM, // RAM
1040                         MemorySwap:   maxRAM, // RAM+swap
1041                         KernelMemory: maxRAM, // kernel portion
1042                 },
1043         }
1044
1045         if wantAPI := runner.Container.RuntimeConstraints.API; wantAPI != nil && *wantAPI {
1046                 tok, err := runner.ContainerToken()
1047                 if err != nil {
1048                         return err
1049                 }
1050                 runner.ContainerConfig.Env = append(runner.ContainerConfig.Env,
1051                         "ARVADOS_API_TOKEN="+tok,
1052                         "ARVADOS_API_HOST="+os.Getenv("ARVADOS_API_HOST"),
1053                         "ARVADOS_API_HOST_INSECURE="+os.Getenv("ARVADOS_API_HOST_INSECURE"),
1054                 )
1055                 runner.HostConfig.NetworkMode = dockercontainer.NetworkMode(runner.networkMode)
1056         } else {
1057                 if runner.enableNetwork == "always" {
1058                         runner.HostConfig.NetworkMode = dockercontainer.NetworkMode(runner.networkMode)
1059                 } else {
1060                         runner.HostConfig.NetworkMode = dockercontainer.NetworkMode("none")
1061                 }
1062         }
1063
1064         _, stdinUsed := runner.Container.Mounts["stdin"]
1065         runner.ContainerConfig.OpenStdin = stdinUsed
1066         runner.ContainerConfig.StdinOnce = stdinUsed
1067         runner.ContainerConfig.AttachStdin = stdinUsed
1068         runner.ContainerConfig.AttachStdout = true
1069         runner.ContainerConfig.AttachStderr = true
1070
1071         createdBody, err := runner.Docker.ContainerCreate(context.TODO(), &runner.ContainerConfig, &runner.HostConfig, nil, runner.Container.UUID)
1072         if err != nil {
1073                 return fmt.Errorf("While creating container: %v", err)
1074         }
1075
1076         runner.ContainerID = createdBody.ID
1077
1078         return runner.AttachStreams()
1079 }
1080
1081 // StartContainer starts the docker container created by CreateContainer.
1082 func (runner *ContainerRunner) StartContainer() error {
1083         runner.CrunchLog.Printf("Starting Docker container id '%s'", runner.ContainerID)
1084         runner.cStateLock.Lock()
1085         defer runner.cStateLock.Unlock()
1086         if runner.cCancelled {
1087                 return ErrCancelled
1088         }
1089         err := runner.Docker.ContainerStart(context.TODO(), runner.ContainerID,
1090                 dockertypes.ContainerStartOptions{})
1091         if err != nil {
1092                 var advice string
1093                 if m, e := regexp.MatchString("(?ms).*(exec|System error).*(no such file or directory|file not found).*", err.Error()); m && e == nil {
1094                         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])
1095                 }
1096                 return fmt.Errorf("could not start container: %v%s", err, advice)
1097         }
1098         return nil
1099 }
1100
1101 // checkContainerd checks if "containerd" is present in the process list.
1102 func (runner *ContainerRunner) CheckContainerd() error {
1103         if runner.checkContainerd == 0 {
1104                 return nil
1105         }
1106         p, _ := runner.ListProcesses()
1107         for _, i := range p {
1108                 e, _ := i.CmdlineSlice()
1109                 if len(e) > 0 {
1110                         if strings.Index(e[0], "containerd") > -1 {
1111                                 return nil
1112                         }
1113                 }
1114         }
1115
1116         // Not found
1117         runner.runBrokenNodeHook()
1118         runner.stop(nil)
1119         return fmt.Errorf("'containerd' not found in process list.")
1120 }
1121
1122 // WaitFinish waits for the container to terminate, capture the exit code, and
1123 // close the stdout/stderr logging.
1124 func (runner *ContainerRunner) WaitFinish() error {
1125         var runTimeExceeded <-chan time.Time
1126         runner.CrunchLog.Print("Waiting for container to finish")
1127
1128         waitOk, waitErr := runner.Docker.ContainerWait(context.TODO(), runner.ContainerID, dockercontainer.WaitConditionNotRunning)
1129         arvMountExit := runner.ArvMountExit
1130         if timeout := runner.Container.SchedulingParameters.MaxRunTime; timeout > 0 {
1131                 runTimeExceeded = time.After(time.Duration(timeout) * time.Second)
1132         }
1133
1134         containerGone := make(chan struct{})
1135         go func() {
1136                 defer close(containerGone)
1137                 if runner.containerWatchdogInterval < 1 {
1138                         runner.containerWatchdogInterval = time.Minute
1139                 }
1140                 for range time.NewTicker(runner.containerWatchdogInterval).C {
1141                         ctx, cancel := context.WithDeadline(context.Background(), time.Now().Add(runner.containerWatchdogInterval))
1142                         ctr, err := runner.Docker.ContainerInspect(ctx, runner.ContainerID)
1143                         cancel()
1144                         runner.cStateLock.Lock()
1145                         done := runner.cRemoved || runner.ExitCode != nil
1146                         runner.cStateLock.Unlock()
1147                         if done {
1148                                 return
1149                         } else if err != nil {
1150                                 runner.CrunchLog.Printf("Error inspecting container: %s", err)
1151                                 runner.checkBrokenNode(err)
1152                                 return
1153                         } else if ctr.State == nil || !(ctr.State.Running || ctr.State.Status == "created") {
1154                                 runner.CrunchLog.Printf("Container is not running: State=%v", ctr.State)
1155                                 return
1156                         }
1157                 }
1158         }()
1159
1160         containerdGone := make(chan error)
1161         defer close(containerdGone)
1162         if runner.checkContainerd > 0 {
1163                 go func() {
1164                         ticker := time.NewTicker(time.Duration(runner.checkContainerd))
1165                         defer ticker.Stop()
1166                         for {
1167                                 select {
1168                                 case <-ticker.C:
1169                                         if ck := runner.CheckContainerd(); ck != nil {
1170                                                 containerdGone <- ck
1171                                                 return
1172                                         }
1173                                 case <-containerdGone:
1174                                         // Channel closed, quit goroutine
1175                                         return
1176                                 }
1177                         }
1178                 }()
1179         }
1180
1181         for {
1182                 select {
1183                 case waitBody := <-waitOk:
1184                         runner.CrunchLog.Printf("Container exited with code: %v", waitBody.StatusCode)
1185                         code := int(waitBody.StatusCode)
1186                         runner.ExitCode = &code
1187
1188                         // wait for stdout/stderr to complete
1189                         <-runner.loggingDone
1190                         return nil
1191
1192                 case err := <-waitErr:
1193                         return fmt.Errorf("container wait: %v", err)
1194
1195                 case <-arvMountExit:
1196                         runner.CrunchLog.Printf("arv-mount exited while container is still running.  Stopping container.")
1197                         runner.stop(nil)
1198                         // arvMountExit will always be ready now that
1199                         // it's closed, but that doesn't interest us.
1200                         arvMountExit = nil
1201
1202                 case <-runTimeExceeded:
1203                         runner.CrunchLog.Printf("maximum run time exceeded. Stopping container.")
1204                         runner.stop(nil)
1205                         runTimeExceeded = nil
1206
1207                 case <-containerGone:
1208                         return errors.New("docker client never returned status")
1209
1210                 case err := <-containerdGone:
1211                         return err
1212                 }
1213         }
1214 }
1215
1216 func (runner *ContainerRunner) updateLogs() {
1217         ticker := time.NewTicker(crunchLogUpdatePeriod / 360)
1218         defer ticker.Stop()
1219
1220         sigusr1 := make(chan os.Signal, 1)
1221         signal.Notify(sigusr1, syscall.SIGUSR1)
1222         defer signal.Stop(sigusr1)
1223
1224         saveAtTime := time.Now().Add(crunchLogUpdatePeriod)
1225         saveAtSize := crunchLogUpdateSize
1226         var savedSize int64
1227         for {
1228                 select {
1229                 case <-ticker.C:
1230                 case <-sigusr1:
1231                         saveAtTime = time.Now()
1232                 }
1233                 runner.logMtx.Lock()
1234                 done := runner.LogsPDH != nil
1235                 runner.logMtx.Unlock()
1236                 if done {
1237                         return
1238                 }
1239                 size := runner.LogCollection.Size()
1240                 if size == savedSize || (time.Now().Before(saveAtTime) && size < saveAtSize) {
1241                         continue
1242                 }
1243                 saveAtTime = time.Now().Add(crunchLogUpdatePeriod)
1244                 saveAtSize = runner.LogCollection.Size() + crunchLogUpdateSize
1245                 saved, err := runner.saveLogCollection(false)
1246                 if err != nil {
1247                         runner.CrunchLog.Printf("error updating log collection: %s", err)
1248                         continue
1249                 }
1250
1251                 var updated arvados.Container
1252                 err = runner.ArvClient.Update("containers", runner.Container.UUID, arvadosclient.Dict{
1253                         "container": arvadosclient.Dict{"log": saved.PortableDataHash},
1254                 }, &updated)
1255                 if err != nil {
1256                         runner.CrunchLog.Printf("error updating container log to %s: %s", saved.PortableDataHash, err)
1257                         continue
1258                 }
1259
1260                 savedSize = size
1261         }
1262 }
1263
1264 // CaptureOutput saves data from the container's output directory if
1265 // needed, and updates the container output accordingly.
1266 func (runner *ContainerRunner) CaptureOutput() error {
1267         if wantAPI := runner.Container.RuntimeConstraints.API; wantAPI != nil && *wantAPI {
1268                 // Output may have been set directly by the container, so
1269                 // refresh the container record to check.
1270                 err := runner.ArvClient.Get("containers", runner.Container.UUID,
1271                         nil, &runner.Container)
1272                 if err != nil {
1273                         return err
1274                 }
1275                 if runner.Container.Output != "" {
1276                         // Container output is already set.
1277                         runner.OutputPDH = &runner.Container.Output
1278                         return nil
1279                 }
1280         }
1281
1282         txt, err := (&copier{
1283                 client:        runner.client,
1284                 arvClient:     runner.ArvClient,
1285                 keepClient:    runner.Kc,
1286                 hostOutputDir: runner.HostOutputDir,
1287                 ctrOutputDir:  runner.Container.OutputPath,
1288                 binds:         runner.Binds,
1289                 mounts:        runner.Container.Mounts,
1290                 secretMounts:  runner.SecretMounts,
1291                 logger:        runner.CrunchLog,
1292         }).Copy()
1293         if err != nil {
1294                 return err
1295         }
1296         var resp arvados.Collection
1297         err = runner.ArvClient.Create("collections", arvadosclient.Dict{
1298                 "ensure_unique_name": true,
1299                 "collection": arvadosclient.Dict{
1300                         "is_trashed":    true,
1301                         "name":          "output for " + runner.Container.UUID,
1302                         "manifest_text": txt,
1303                 },
1304         }, &resp)
1305         if err != nil {
1306                 return fmt.Errorf("error creating output collection: %v", err)
1307         }
1308         runner.OutputPDH = &resp.PortableDataHash
1309         return nil
1310 }
1311
1312 func (runner *ContainerRunner) CleanupDirs() {
1313         if runner.ArvMount != nil {
1314                 var delay int64 = 8
1315                 umount := exec.Command("arv-mount", fmt.Sprintf("--unmount-timeout=%d", delay), "--unmount", runner.ArvMountPoint)
1316                 umount.Stdout = runner.CrunchLog
1317                 umount.Stderr = runner.CrunchLog
1318                 runner.CrunchLog.Printf("Running %v", umount.Args)
1319                 umnterr := umount.Start()
1320
1321                 if umnterr != nil {
1322                         runner.CrunchLog.Printf("Error unmounting: %v", umnterr)
1323                 } else {
1324                         // If arv-mount --unmount gets stuck for any reason, we
1325                         // don't want to wait for it forever.  Do Wait() in a goroutine
1326                         // so it doesn't block crunch-run.
1327                         umountExit := make(chan error)
1328                         go func() {
1329                                 mnterr := umount.Wait()
1330                                 if mnterr != nil {
1331                                         runner.CrunchLog.Printf("Error unmounting: %v", mnterr)
1332                                 }
1333                                 umountExit <- mnterr
1334                         }()
1335
1336                         for again := true; again; {
1337                                 again = false
1338                                 select {
1339                                 case <-umountExit:
1340                                         umount = nil
1341                                         again = true
1342                                 case <-runner.ArvMountExit:
1343                                         break
1344                                 case <-time.After(time.Duration((delay + 1) * int64(time.Second))):
1345                                         runner.CrunchLog.Printf("Timed out waiting for unmount")
1346                                         if umount != nil {
1347                                                 umount.Process.Kill()
1348                                         }
1349                                         runner.ArvMount.Process.Kill()
1350                                 }
1351                         }
1352                 }
1353         }
1354
1355         if runner.ArvMountPoint != "" {
1356                 if rmerr := os.Remove(runner.ArvMountPoint); rmerr != nil {
1357                         runner.CrunchLog.Printf("While cleaning up arv-mount directory %s: %v", runner.ArvMountPoint, rmerr)
1358                 }
1359         }
1360
1361         if rmerr := os.RemoveAll(runner.parentTemp); rmerr != nil {
1362                 runner.CrunchLog.Printf("While cleaning up temporary directory %s: %v", runner.parentTemp, rmerr)
1363         }
1364 }
1365
1366 // CommitLogs posts the collection containing the final container logs.
1367 func (runner *ContainerRunner) CommitLogs() error {
1368         func() {
1369                 // Hold cStateLock to prevent races on CrunchLog (e.g., stop()).
1370                 runner.cStateLock.Lock()
1371                 defer runner.cStateLock.Unlock()
1372
1373                 runner.CrunchLog.Print(runner.finalState)
1374
1375                 if runner.arvMountLog != nil {
1376                         runner.arvMountLog.Close()
1377                 }
1378                 runner.CrunchLog.Close()
1379
1380                 // Closing CrunchLog above allows them to be committed to Keep at this
1381                 // point, but re-open crunch log with ArvClient in case there are any
1382                 // other further errors (such as failing to write the log to Keep!)
1383                 // while shutting down
1384                 runner.CrunchLog = NewThrottledLogger(&ArvLogWriter{
1385                         ArvClient:     runner.ArvClient,
1386                         UUID:          runner.Container.UUID,
1387                         loggingStream: "crunch-run",
1388                         writeCloser:   nil,
1389                 })
1390                 runner.CrunchLog.Immediate = log.New(os.Stderr, runner.Container.UUID+" ", 0)
1391         }()
1392
1393         if runner.LogsPDH != nil {
1394                 // If we have already assigned something to LogsPDH,
1395                 // we must be closing the re-opened log, which won't
1396                 // end up getting attached to the container record and
1397                 // therefore doesn't need to be saved as a collection
1398                 // -- it exists only to send logs to other channels.
1399                 return nil
1400         }
1401         saved, err := runner.saveLogCollection(true)
1402         if err != nil {
1403                 return fmt.Errorf("error saving log collection: %s", err)
1404         }
1405         runner.logMtx.Lock()
1406         defer runner.logMtx.Unlock()
1407         runner.LogsPDH = &saved.PortableDataHash
1408         return nil
1409 }
1410
1411 func (runner *ContainerRunner) saveLogCollection(final bool) (response arvados.Collection, err error) {
1412         runner.logMtx.Lock()
1413         defer runner.logMtx.Unlock()
1414         if runner.LogsPDH != nil {
1415                 // Already finalized.
1416                 return
1417         }
1418         mt, err := runner.LogCollection.MarshalManifest(".")
1419         if err != nil {
1420                 err = fmt.Errorf("error creating log manifest: %v", err)
1421                 return
1422         }
1423         updates := arvadosclient.Dict{
1424                 "name":          "logs for " + runner.Container.UUID,
1425                 "manifest_text": mt,
1426         }
1427         if final {
1428                 updates["is_trashed"] = true
1429         } else {
1430                 exp := time.Now().Add(crunchLogUpdatePeriod * 24)
1431                 updates["trash_at"] = exp
1432                 updates["delete_at"] = exp
1433         }
1434         reqBody := arvadosclient.Dict{"collection": updates}
1435         if runner.logUUID == "" {
1436                 reqBody["ensure_unique_name"] = true
1437                 err = runner.ArvClient.Create("collections", reqBody, &response)
1438         } else {
1439                 err = runner.ArvClient.Update("collections", runner.logUUID, reqBody, &response)
1440         }
1441         if err != nil {
1442                 return
1443         }
1444         runner.logUUID = response.UUID
1445         return
1446 }
1447
1448 // UpdateContainerRunning updates the container state to "Running"
1449 func (runner *ContainerRunner) UpdateContainerRunning() error {
1450         runner.cStateLock.Lock()
1451         defer runner.cStateLock.Unlock()
1452         if runner.cCancelled {
1453                 return ErrCancelled
1454         }
1455         return runner.ArvClient.Update("containers", runner.Container.UUID,
1456                 arvadosclient.Dict{"container": arvadosclient.Dict{"state": "Running"}}, nil)
1457 }
1458
1459 // ContainerToken returns the api_token the container (and any
1460 // arv-mount processes) are allowed to use.
1461 func (runner *ContainerRunner) ContainerToken() (string, error) {
1462         if runner.token != "" {
1463                 return runner.token, nil
1464         }
1465
1466         var auth arvados.APIClientAuthorization
1467         err := runner.ArvClient.Call("GET", "containers", runner.Container.UUID, "auth", nil, &auth)
1468         if err != nil {
1469                 return "", err
1470         }
1471         runner.token = fmt.Sprintf("v2/%s/%s/%s", auth.UUID, auth.APIToken, runner.Container.UUID)
1472         return runner.token, nil
1473 }
1474
1475 // UpdateContainerComplete updates the container record state on API
1476 // server to "Complete" or "Cancelled"
1477 func (runner *ContainerRunner) UpdateContainerFinal() error {
1478         update := arvadosclient.Dict{}
1479         update["state"] = runner.finalState
1480         if runner.LogsPDH != nil {
1481                 update["log"] = *runner.LogsPDH
1482         }
1483         if runner.finalState == "Complete" {
1484                 if runner.ExitCode != nil {
1485                         update["exit_code"] = *runner.ExitCode
1486                 }
1487                 if runner.OutputPDH != nil {
1488                         update["output"] = *runner.OutputPDH
1489                 }
1490         }
1491         return runner.ArvClient.Update("containers", runner.Container.UUID, arvadosclient.Dict{"container": update}, nil)
1492 }
1493
1494 // IsCancelled returns the value of Cancelled, with goroutine safety.
1495 func (runner *ContainerRunner) IsCancelled() bool {
1496         runner.cStateLock.Lock()
1497         defer runner.cStateLock.Unlock()
1498         return runner.cCancelled
1499 }
1500
1501 // NewArvLogWriter creates an ArvLogWriter
1502 func (runner *ContainerRunner) NewArvLogWriter(name string) (io.WriteCloser, error) {
1503         writer, err := runner.LogCollection.OpenFile(name+".txt", os.O_CREATE|os.O_WRONLY, 0666)
1504         if err != nil {
1505                 return nil, err
1506         }
1507         return &ArvLogWriter{
1508                 ArvClient:     runner.ArvClient,
1509                 UUID:          runner.Container.UUID,
1510                 loggingStream: name,
1511                 writeCloser:   writer,
1512         }, nil
1513 }
1514
1515 // Run the full container lifecycle.
1516 func (runner *ContainerRunner) Run() (err error) {
1517         runner.CrunchLog.Printf("crunch-run %s started", version)
1518         runner.CrunchLog.Printf("Executing container '%s'", runner.Container.UUID)
1519
1520         hostname, hosterr := os.Hostname()
1521         if hosterr != nil {
1522                 runner.CrunchLog.Printf("Error getting hostname '%v'", hosterr)
1523         } else {
1524                 runner.CrunchLog.Printf("Executing on host '%s'", hostname)
1525         }
1526
1527         runner.finalState = "Queued"
1528
1529         defer func() {
1530                 runner.CleanupDirs()
1531
1532                 runner.CrunchLog.Printf("crunch-run finished")
1533                 runner.CrunchLog.Close()
1534         }()
1535
1536         defer func() {
1537                 // checkErr prints e (unless it's nil) and sets err to
1538                 // e (unless err is already non-nil). Thus, if err
1539                 // hasn't already been assigned when Run() returns,
1540                 // this cleanup func will cause Run() to return the
1541                 // first non-nil error that is passed to checkErr().
1542                 checkErr := func(errorIn string, e error) {
1543                         if e == nil {
1544                                 return
1545                         }
1546                         runner.CrunchLog.Printf("error in %s: %v", errorIn, e)
1547                         if err == nil {
1548                                 err = e
1549                         }
1550                         if runner.finalState == "Complete" {
1551                                 // There was an error in the finalization.
1552                                 runner.finalState = "Cancelled"
1553                         }
1554                 }
1555
1556                 // Log the error encountered in Run(), if any
1557                 checkErr("Run", err)
1558
1559                 if runner.finalState == "Queued" {
1560                         runner.UpdateContainerFinal()
1561                         return
1562                 }
1563
1564                 if runner.IsCancelled() {
1565                         runner.finalState = "Cancelled"
1566                         // but don't return yet -- we still want to
1567                         // capture partial output and write logs
1568                 }
1569
1570                 checkErr("CaptureOutput", runner.CaptureOutput())
1571                 checkErr("stopHoststat", runner.stopHoststat())
1572                 checkErr("CommitLogs", runner.CommitLogs())
1573                 checkErr("UpdateContainerFinal", runner.UpdateContainerFinal())
1574         }()
1575
1576         err = runner.fetchContainerRecord()
1577         if err != nil {
1578                 return
1579         }
1580         runner.setupSignals()
1581         err = runner.startHoststat()
1582         if err != nil {
1583                 return
1584         }
1585
1586         // Sanity check that containerd is running.
1587         err = runner.CheckContainerd()
1588         if err != nil {
1589                 return
1590         }
1591
1592         // check for and/or load image
1593         err = runner.LoadImage()
1594         if err != nil {
1595                 if !runner.checkBrokenNode(err) {
1596                         // Failed to load image but not due to a "broken node"
1597                         // condition, probably user error.
1598                         runner.finalState = "Cancelled"
1599                 }
1600                 err = fmt.Errorf("While loading container image: %v", err)
1601                 return
1602         }
1603
1604         // set up FUSE mount and binds
1605         err = runner.SetupMounts()
1606         if err != nil {
1607                 runner.finalState = "Cancelled"
1608                 err = fmt.Errorf("While setting up mounts: %v", err)
1609                 return
1610         }
1611
1612         err = runner.CreateContainer()
1613         if err != nil {
1614                 return
1615         }
1616         err = runner.LogHostInfo()
1617         if err != nil {
1618                 return
1619         }
1620         err = runner.LogNodeRecord()
1621         if err != nil {
1622                 return
1623         }
1624         err = runner.LogContainerRecord()
1625         if err != nil {
1626                 return
1627         }
1628
1629         if runner.IsCancelled() {
1630                 return
1631         }
1632
1633         err = runner.UpdateContainerRunning()
1634         if err != nil {
1635                 return
1636         }
1637         runner.finalState = "Cancelled"
1638
1639         err = runner.startCrunchstat()
1640         if err != nil {
1641                 return
1642         }
1643
1644         err = runner.StartContainer()
1645         if err != nil {
1646                 runner.checkBrokenNode(err)
1647                 return
1648         }
1649
1650         err = runner.WaitFinish()
1651         if err == nil && !runner.IsCancelled() {
1652                 runner.finalState = "Complete"
1653         }
1654         return
1655 }
1656
1657 // Fetch the current container record (uuid = runner.Container.UUID)
1658 // into runner.Container.
1659 func (runner *ContainerRunner) fetchContainerRecord() error {
1660         reader, err := runner.ArvClient.CallRaw("GET", "containers", runner.Container.UUID, "", nil)
1661         if err != nil {
1662                 return fmt.Errorf("error fetching container record: %v", err)
1663         }
1664         defer reader.Close()
1665
1666         dec := json.NewDecoder(reader)
1667         dec.UseNumber()
1668         err = dec.Decode(&runner.Container)
1669         if err != nil {
1670                 return fmt.Errorf("error decoding container record: %v", err)
1671         }
1672
1673         var sm struct {
1674                 SecretMounts map[string]arvados.Mount `json:"secret_mounts"`
1675         }
1676
1677         containerToken, err := runner.ContainerToken()
1678         if err != nil {
1679                 return fmt.Errorf("error getting container token: %v", err)
1680         }
1681
1682         containerClient, err := runner.MkArvClient(containerToken)
1683         if err != nil {
1684                 return fmt.Errorf("error creating container API client: %v", err)
1685         }
1686
1687         err = containerClient.Call("GET", "containers", runner.Container.UUID, "secret_mounts", nil, &sm)
1688         if err != nil {
1689                 if apierr, ok := err.(arvadosclient.APIServerError); !ok || apierr.HttpStatusCode != 404 {
1690                         return fmt.Errorf("error fetching secret_mounts: %v", err)
1691                 }
1692                 // ok && apierr.HttpStatusCode == 404, which means
1693                 // secret_mounts isn't supported by this API server.
1694         }
1695         runner.SecretMounts = sm.SecretMounts
1696
1697         return nil
1698 }
1699
1700 // NewContainerRunner creates a new container runner.
1701 func NewContainerRunner(client *arvados.Client, api IArvadosClient, kc IKeepClient, docker ThinDockerClient, containerUUID string) (*ContainerRunner, error) {
1702         cr := &ContainerRunner{
1703                 client:    client,
1704                 ArvClient: api,
1705                 Kc:        kc,
1706                 Docker:    docker,
1707         }
1708         cr.NewLogWriter = cr.NewArvLogWriter
1709         cr.RunArvMount = cr.ArvMountCmd
1710         cr.MkTempDir = ioutil.TempDir
1711         cr.ListProcesses = func() ([]PsProcess, error) {
1712                 pr, err := process.Processes()
1713                 if err != nil {
1714                         return nil, err
1715                 }
1716                 ps := make([]PsProcess, len(pr))
1717                 for i, j := range pr {
1718                         ps[i] = j
1719                 }
1720                 return ps, nil
1721         }
1722         cr.MkArvClient = func(token string) (IArvadosClient, error) {
1723                 cl, err := arvadosclient.MakeArvadosClient()
1724                 if err != nil {
1725                         return nil, err
1726                 }
1727                 cl.ApiToken = token
1728                 return cl, nil
1729         }
1730         var err error
1731         cr.LogCollection, err = (&arvados.Collection{}).FileSystem(cr.client, cr.Kc)
1732         if err != nil {
1733                 return nil, err
1734         }
1735         cr.Container.UUID = containerUUID
1736         w, err := cr.NewLogWriter("crunch-run")
1737         if err != nil {
1738                 return nil, err
1739         }
1740         cr.CrunchLog = NewThrottledLogger(w)
1741         cr.CrunchLog.Immediate = log.New(os.Stderr, containerUUID+" ", 0)
1742
1743         loadLogThrottleParams(api)
1744         go cr.updateLogs()
1745
1746         return cr, nil
1747 }
1748
1749 func main() {
1750         statInterval := flag.Duration("crunchstat-interval", 10*time.Second, "sampling period for periodic resource usage reporting")
1751         cgroupRoot := flag.String("cgroup-root", "/sys/fs/cgroup", "path to sysfs cgroup tree")
1752         cgroupParent := flag.String("cgroup-parent", "docker", "name of container's parent cgroup (ignored if -cgroup-parent-subsystem is used)")
1753         cgroupParentSubsystem := flag.String("cgroup-parent-subsystem", "", "use current cgroup for given subsystem as parent cgroup for container")
1754         caCertsPath := flag.String("ca-certs", "", "Path to TLS root certificates")
1755         enableNetwork := flag.String("container-enable-networking", "default",
1756                 `Specify if networking should be enabled for container.  One of 'default', 'always':
1757         default: only enable networking if container requests it.
1758         always:  containers always have networking enabled
1759         `)
1760         networkMode := flag.String("container-network-mode", "default",
1761                 `Set networking mode for container.  Corresponds to Docker network mode (--net).
1762         `)
1763         memprofile := flag.String("memprofile", "", "write memory profile to `file` after running container")
1764         getVersion := flag.Bool("version", false, "Print version information and exit.")
1765         checkContainerd := flag.Duration("check-containerd", 60*time.Second, "Periodic check if (docker-)containerd is running (use 0s to disable).")
1766         flag.Parse()
1767
1768         // Print version information if requested
1769         if *getVersion {
1770                 fmt.Printf("crunch-run %s\n", version)
1771                 return
1772         }
1773
1774         log.Printf("crunch-run %s started", version)
1775
1776         containerId := flag.Arg(0)
1777
1778         if *caCertsPath != "" {
1779                 arvadosclient.CertFiles = []string{*caCertsPath}
1780         }
1781
1782         api, err := arvadosclient.MakeArvadosClient()
1783         if err != nil {
1784                 log.Fatalf("%s: %v", containerId, err)
1785         }
1786         api.Retries = 8
1787
1788         kc, kcerr := keepclient.MakeKeepClient(api)
1789         if kcerr != nil {
1790                 log.Fatalf("%s: %v", containerId, kcerr)
1791         }
1792         kc.BlockCache = &keepclient.BlockCache{MaxBlocks: 2}
1793         kc.Retries = 4
1794
1795         // API version 1.21 corresponds to Docker 1.9, which is currently the
1796         // minimum version we want to support.
1797         docker, dockererr := dockerclient.NewClient(dockerclient.DefaultDockerHost, "1.21", nil, nil)
1798
1799         cr, err := NewContainerRunner(arvados.NewClientFromEnv(), api, kc, docker, containerId)
1800         if err != nil {
1801                 log.Fatal(err)
1802         }
1803         if dockererr != nil {
1804                 cr.CrunchLog.Printf("%s: %v", containerId, dockererr)
1805                 cr.checkBrokenNode(dockererr)
1806                 cr.CrunchLog.Close()
1807                 os.Exit(1)
1808         }
1809
1810         parentTemp, tmperr := cr.MkTempDir("", "crunch-run."+containerId+".")
1811         if tmperr != nil {
1812                 log.Fatalf("%s: %v", containerId, tmperr)
1813         }
1814
1815         cr.parentTemp = parentTemp
1816         cr.statInterval = *statInterval
1817         cr.cgroupRoot = *cgroupRoot
1818         cr.expectCgroupParent = *cgroupParent
1819         cr.enableNetwork = *enableNetwork
1820         cr.networkMode = *networkMode
1821         cr.checkContainerd = *checkContainerd
1822         if *cgroupParentSubsystem != "" {
1823                 p := findCgroup(*cgroupParentSubsystem)
1824                 cr.setCgroupParent = p
1825                 cr.expectCgroupParent = p
1826         }
1827
1828         runerr := cr.Run()
1829
1830         if *memprofile != "" {
1831                 f, err := os.Create(*memprofile)
1832                 if err != nil {
1833                         log.Printf("could not create memory profile: %s", err)
1834                 }
1835                 runtime.GC() // get up-to-date statistics
1836                 if err := pprof.WriteHeapProfile(f); err != nil {
1837                         log.Printf("could not write memory profile: %s", err)
1838                 }
1839                 closeerr := f.Close()
1840                 if closeerr != nil {
1841                         log.Printf("closing memprofile file: %s", err)
1842                 }
1843         }
1844
1845         if runerr != nil {
1846                 log.Fatalf("%s: %v", containerId, runerr)
1847         }
1848 }