lib/crunchrun/crunchrun.go

   1 // Copyright (C) The Arvados Authors. All rights reserved.
   2 //
   3 // SPDX-License-Identifier: AGPL-3.0
   4
   5 package crunchrun
   6
   7 import (
   8         "bytes"
   9         "context"
  10         "encoding/json"
  11         "errors"
  12         "flag"
  13         "fmt"
  14         "io"
  15         "io/ioutil"
  16         "log"
  17         "net"
  18         "net/http"
  19         "os"
  20         "os/exec"
  21         "os/signal"
  22         "os/user"
  23         "path"
  24         "path/filepath"
  25         "regexp"
  26         "runtime"
  27         "runtime/pprof"
  28         "sort"
  29         "strings"
  30         "sync"
  31         "syscall"
  32         "time"
  33
  34         "git.arvados.org/arvados.git/lib/cmd"
  35         "git.arvados.org/arvados.git/lib/config"
  36         "git.arvados.org/arvados.git/lib/crunchstat"
  37         "git.arvados.org/arvados.git/sdk/go/arvados"
  38         "git.arvados.org/arvados.git/sdk/go/arvadosclient"
  39         "git.arvados.org/arvados.git/sdk/go/ctxlog"
  40         "git.arvados.org/arvados.git/sdk/go/keepclient"
  41         "git.arvados.org/arvados.git/sdk/go/manifest"
  42         "golang.org/x/sys/unix"
  43 )
  44
  45 type command struct{}
  46
  47 var Command = command{}
  48
  49 // ConfigData contains environment variables and (when needed) cluster
  50 // configuration, passed from dispatchcloud to crunch-run on stdin.
  51 type ConfigData struct {
  52         Env         map[string]string
  53         KeepBuffers int
  54         Cluster     *arvados.Cluster
  55 }
  56
  57 // IArvadosClient is the minimal Arvados API methods used by crunch-run.
  58 type IArvadosClient interface {
  59         Create(resourceType string, parameters arvadosclient.Dict, output interface{}) error
  60         Get(resourceType string, uuid string, parameters arvadosclient.Dict, output interface{}) error
  61         Update(resourceType string, uuid string, parameters arvadosclient.Dict, output interface{}) error
  62         Call(method, resourceType, uuid, action string, parameters arvadosclient.Dict, output interface{}) error
  63         CallRaw(method string, resourceType string, uuid string, action string, parameters arvadosclient.Dict) (reader io.ReadCloser, err error)
  64         Discovery(key string) (interface{}, error)
  65 }
  66
  67 // ErrCancelled is the error returned when the container is cancelled.
  68 var ErrCancelled = errors.New("Cancelled")
  69
  70 // IKeepClient is the minimal Keep API methods used by crunch-run.
  71 type IKeepClient interface {
  72         BlockWrite(context.Context, arvados.BlockWriteOptions) (arvados.BlockWriteResponse, error)
  73         ReadAt(locator string, p []byte, off int) (int, error)
  74         ManifestFileReader(m manifest.Manifest, filename string) (arvados.File, error)
  75         LocalLocator(locator string) (string, error)
  76         ClearBlockCache()
  77         SetStorageClasses(sc []string)
  78 }
  79
  80 // NewLogWriter is a factory function to create a new log writer.
  81 type NewLogWriter func(name string) (io.WriteCloser, error)
  82
  83 type RunArvMount func(cmdline []string, tok string) (*exec.Cmd, error)
  84
  85 type MkTempDir func(string, string) (string, error)
  86
  87 type PsProcess interface {
  88         CmdlineSlice() ([]string, error)
  89 }
  90
  91 // ContainerRunner is the main stateful struct used for a single execution of a
  92 // container.
  93 type ContainerRunner struct {
  94         executor       containerExecutor
  95         executorStdin  io.Closer
  96         executorStdout io.Closer
  97         executorStderr io.Closer
  98
  99         // Dispatcher client is initialized with the Dispatcher token.
 100         // This is a privileged token used to manage container status
 101         // and logs.
 102         //
 103         // We have both dispatcherClient and DispatcherArvClient
 104         // because there are two different incompatible Arvados Go
 105         // SDKs and we have to use both (hopefully this gets fixed in
 106         // #14467)
 107         dispatcherClient     *arvados.Client
 108         DispatcherArvClient  IArvadosClient
 109         DispatcherKeepClient IKeepClient
 110
 111         // Container client is initialized with the Container token
 112         // This token controls the permissions of the container, and
 113         // must be used for operations such as reading collections.
 114         //
 115         // Same comment as above applies to
 116         // containerClient/ContainerArvClient.
 117         containerClient     *arvados.Client
 118         ContainerArvClient  IArvadosClient
 119         ContainerKeepClient IKeepClient
 120
 121         Container     arvados.Container
 122         token         string
 123         ExitCode      *int
 124         NewLogWriter  NewLogWriter
 125         CrunchLog     *ThrottledLogger
 126         logUUID       string
 127         logMtx        sync.Mutex
 128         LogCollection arvados.CollectionFileSystem
 129         LogsPDH       *string
 130         RunArvMount   RunArvMount
 131         MkTempDir     MkTempDir
 132         ArvMount      *exec.Cmd
 133         ArvMountPoint string
 134         HostOutputDir string
 135         Volumes       map[string]struct{}
 136         OutputPDH     *string
 137         SigChan       chan os.Signal
 138         ArvMountExit  chan error
 139         SecretMounts  map[string]arvados.Mount
 140         MkArvClient   func(token string) (IArvadosClient, IKeepClient, *arvados.Client, error)
 141         finalState    string
 142         parentTemp    string
 143         costStartTime time.Time
 144
 145         keepstore        *exec.Cmd
 146         keepstoreLogger  io.WriteCloser
 147         keepstoreLogbuf  *bufThenWrite
 148         statLogger       io.WriteCloser
 149         statReporter     *crunchstat.Reporter
 150         hoststatLogger   io.WriteCloser
 151         hoststatReporter *crunchstat.Reporter
 152         statInterval     time.Duration
 153         cgroupRoot       string
 154         // What we expect the container's cgroup parent to be.
 155         expectCgroupParent string
 156         // What we tell docker to use as the container's cgroup
 157         // parent. Note: Ideally we would use the same field for both
 158         // expectCgroupParent and setCgroupParent, and just make it
 159         // default to "docker". However, when using docker < 1.10 with
 160         // systemd, specifying a non-empty cgroup parent (even the
 161         // default value "docker") hits a docker bug
 162         // (https://github.com/docker/docker/issues/17126). Using two
 163         // separate fields makes it possible to use the "expect cgroup
 164         // parent to be X" feature even on sites where the "specify
 165         // cgroup parent" feature breaks.
 166         setCgroupParent string
 167
 168         cStateLock sync.Mutex
 169         cCancelled bool // StopContainer() invoked
 170
 171         enableMemoryLimit bool
 172         enableNetwork     string // one of "default" or "always"
 173         networkMode       string // "none", "host", or "" -- passed through to executor
 174         brokenNodeHook    string // script to run if node appears to be broken
 175         arvMountLog       *ThrottledLogger
 176
 177         containerWatchdogInterval time.Duration
 178
 179         gateway Gateway
 180 }
 181
 182 // setupSignals sets up signal handling to gracefully terminate the
 183 // underlying container and update state when receiving a TERM, INT or
 184 // QUIT signal.
 185 func (runner *ContainerRunner) setupSignals() {
 186         runner.SigChan = make(chan os.Signal, 1)
 187         signal.Notify(runner.SigChan, syscall.SIGTERM)
 188         signal.Notify(runner.SigChan, syscall.SIGINT)
 189         signal.Notify(runner.SigChan, syscall.SIGQUIT)
 190
 191         go func(sig chan os.Signal) {
 192                 for s := range sig {
 193                         runner.stop(s)
 194                 }
 195         }(runner.SigChan)
 196 }
 197
 198 // stop the underlying container.
 199 func (runner *ContainerRunner) stop(sig os.Signal) {
 200         runner.cStateLock.Lock()
 201         defer runner.cStateLock.Unlock()
 202         if sig != nil {
 203                 runner.CrunchLog.Printf("caught signal: %v", sig)
 204         }
 205         runner.cCancelled = true
 206         runner.CrunchLog.Printf("stopping container")
 207         err := runner.executor.Stop()
 208         if err != nil {
 209                 runner.CrunchLog.Printf("error stopping container: %s", err)
 210         }
 211 }
 212
 213 var errorBlacklist = []string{
 214         "(?ms).*[Cc]annot connect to the Docker daemon.*",
 215         "(?ms).*oci runtime error.*starting container process.*container init.*mounting.*to rootfs.*no such file or directory.*",
 216         "(?ms).*grpc: the connection is unavailable.*",
 217 }
 218
 219 func (runner *ContainerRunner) runBrokenNodeHook() {
 220         if runner.brokenNodeHook == "" {
 221                 path := filepath.Join(lockdir, brokenfile)
 222                 runner.CrunchLog.Printf("Writing %s to mark node as broken", path)
 223                 f, err := os.OpenFile(path, os.O_CREATE|os.O_RDWR, 0700)
 224                 if err != nil {
 225                         runner.CrunchLog.Printf("Error writing %s: %s", path, err)
 226                         return
 227                 }
 228                 f.Close()
 229         } else {
 230                 runner.CrunchLog.Printf("Running broken node hook %q", runner.brokenNodeHook)
 231                 // run killme script
 232                 c := exec.Command(runner.brokenNodeHook)
 233                 c.Stdout = runner.CrunchLog
 234                 c.Stderr = runner.CrunchLog
 235                 err := c.Run()
 236                 if err != nil {
 237                         runner.CrunchLog.Printf("Error running broken node hook: %v", err)
 238                 }
 239         }
 240 }
 241
 242 func (runner *ContainerRunner) checkBrokenNode(goterr error) bool {
 243         for _, d := range errorBlacklist {
 244                 if m, e := regexp.MatchString(d, goterr.Error()); m && e == nil {
 245                         runner.CrunchLog.Printf("Error suggests node is unable to run containers: %v", goterr)
 246                         runner.runBrokenNodeHook()
 247                         return true
 248                 }
 249         }
 250         return false
 251 }
 252
 253 // LoadImage determines the docker image id from the container record and
 254 // checks if it is available in the local Docker image store.  If not, it loads
 255 // the image from Keep.
 256 func (runner *ContainerRunner) LoadImage() (string, error) {
 257         runner.CrunchLog.Printf("Fetching Docker image from collection '%s'", runner.Container.ContainerImage)
 258
 259         d, err := os.Open(runner.ArvMountPoint + "/by_id/" + runner.Container.ContainerImage)
 260         if err != nil {
 261                 return "", err
 262         }
 263         defer d.Close()
 264         allfiles, err := d.Readdirnames(-1)
 265         if err != nil {
 266                 return "", err
 267         }
 268         var tarfiles []string
 269         for _, fnm := range allfiles {
 270                 if strings.HasSuffix(fnm, ".tar") {
 271                         tarfiles = append(tarfiles, fnm)
 272                 }
 273         }
 274         if len(tarfiles) == 0 {
 275                 return "", fmt.Errorf("image collection does not include a .tar image file")
 276         }
 277         if len(tarfiles) > 1 {
 278                 return "", fmt.Errorf("cannot choose from multiple tar files in image collection: %v", tarfiles)
 279         }
 280         imageID := tarfiles[0][:len(tarfiles[0])-4]
 281         imageTarballPath := runner.ArvMountPoint + "/by_id/" + runner.Container.ContainerImage + "/" + imageID + ".tar"
 282         runner.CrunchLog.Printf("Using Docker image id %q", imageID)
 283
 284         runner.CrunchLog.Print("Loading Docker image from keep")
 285         err = runner.executor.LoadImage(imageID, imageTarballPath, runner.Container, runner.ArvMountPoint,
 286                 runner.containerClient)
 287         if err != nil {
 288                 return "", err
 289         }
 290
 291         return imageID, nil
 292 }
 293
 294 func (runner *ContainerRunner) ArvMountCmd(cmdline []string, token string) (c *exec.Cmd, err error) {
 295         c = exec.Command(cmdline[0], cmdline[1:]...)
 296
 297         // Copy our environment, but override ARVADOS_API_TOKEN with
 298         // the container auth token.
 299         c.Env = nil
 300         for _, s := range os.Environ() {
 301                 if !strings.HasPrefix(s, "ARVADOS_API_TOKEN=") {
 302                         c.Env = append(c.Env, s)
 303                 }
 304         }
 305         c.Env = append(c.Env, "ARVADOS_API_TOKEN="+token)
 306
 307         w, err := runner.NewLogWriter("arv-mount")
 308         if err != nil {
 309                 return nil, err
 310         }
 311         runner.arvMountLog = NewThrottledLogger(w)
 312         scanner := logScanner{
 313                 Patterns: []string{
 314                         "Keep write error",
 315                         "Block not found error",
 316                         "Unhandled exception during FUSE operation",
 317                 },
 318                 ReportFunc: runner.reportArvMountWarning,
 319         }
 320         c.Stdout = runner.arvMountLog
 321         c.Stderr = io.MultiWriter(runner.arvMountLog, os.Stderr, &scanner)
 322
 323         runner.CrunchLog.Printf("Running %v", c.Args)
 324
 325         err = c.Start()
 326         if err != nil {
 327                 return nil, err
 328         }
 329
 330         statReadme := make(chan bool)
 331         runner.ArvMountExit = make(chan error)
 332
 333         keepStatting := true
 334         go func() {
 335                 for keepStatting {
 336                         time.Sleep(100 * time.Millisecond)
 337                         _, err = os.Stat(fmt.Sprintf("%s/by_id/README", runner.ArvMountPoint))
 338                         if err == nil {
 339                                 keepStatting = false
 340                                 statReadme <- true
 341                         }
 342                 }
 343                 close(statReadme)
 344         }()
 345
 346         go func() {
 347                 mnterr := c.Wait()
 348                 if mnterr != nil {
 349                         runner.CrunchLog.Printf("Arv-mount exit error: %v", mnterr)
 350                 }
 351                 runner.ArvMountExit <- mnterr
 352                 close(runner.ArvMountExit)
 353         }()
 354
 355         select {
 356         case <-statReadme:
 357                 break
 358         case err := <-runner.ArvMountExit:
 359                 runner.ArvMount = nil
 360                 keepStatting = false
 361                 return nil, err
 362         }
 363
 364         return c, nil
 365 }
 366
 367 func (runner *ContainerRunner) SetupArvMountPoint(prefix string) (err error) {
 368         if runner.ArvMountPoint == "" {
 369                 runner.ArvMountPoint, err = runner.MkTempDir(runner.parentTemp, prefix)
 370         }
 371         return
 372 }
 373
 374 func copyfile(src string, dst string) (err error) {
 375         srcfile, err := os.Open(src)
 376         if err != nil {
 377                 return
 378         }
 379
 380         os.MkdirAll(path.Dir(dst), 0777)
 381
 382         dstfile, err := os.Create(dst)
 383         if err != nil {
 384                 return
 385         }
 386         _, err = io.Copy(dstfile, srcfile)
 387         if err != nil {
 388                 return
 389         }
 390
 391         err = srcfile.Close()
 392         err2 := dstfile.Close()
 393
 394         if err != nil {
 395                 return
 396         }
 397
 398         if err2 != nil {
 399                 return err2
 400         }
 401
 402         return nil
 403 }
 404
 405 func (runner *ContainerRunner) SetupMounts() (map[string]bindmount, error) {
 406         bindmounts := map[string]bindmount{}
 407         err := runner.SetupArvMountPoint("keep")
 408         if err != nil {
 409                 return nil, fmt.Errorf("While creating keep mount temp dir: %v", err)
 410         }
 411
 412         token, err := runner.ContainerToken()
 413         if err != nil {
 414                 return nil, fmt.Errorf("could not get container token: %s", err)
 415         }
 416         runner.CrunchLog.Printf("container token %q", token)
 417
 418         pdhOnly := true
 419         tmpcount := 0
 420         arvMountCmd := []string{
 421                 "arv-mount",
 422                 "--foreground",
 423                 "--read-write",
 424                 "--storage-classes", strings.Join(runner.Container.OutputStorageClasses, ","),
 425                 fmt.Sprintf("--crunchstat-interval=%v", runner.statInterval.Seconds())}
 426
 427         if _, isdocker := runner.executor.(*dockerExecutor); isdocker {
 428                 arvMountCmd = append(arvMountCmd, "--allow-other")
 429         }
 430
 431         if runner.Container.RuntimeConstraints.KeepCacheDisk > 0 {
 432                 keepcachedir, err := runner.MkTempDir(runner.parentTemp, "keepcache")
 433                 if err != nil {
 434                         return nil, fmt.Errorf("while creating keep cache temp dir: %v", err)
 435                 }
 436                 arvMountCmd = append(arvMountCmd, "--disk-cache", "--disk-cache-dir", keepcachedir, "--file-cache", fmt.Sprintf("%d", runner.Container.RuntimeConstraints.KeepCacheDisk))
 437         } else if runner.Container.RuntimeConstraints.KeepCacheRAM > 0 {
 438                 arvMountCmd = append(arvMountCmd, "--ram-cache", "--file-cache", fmt.Sprintf("%d", runner.Container.RuntimeConstraints.KeepCacheRAM))
 439         }
 440
 441         collectionPaths := []string{}
 442         needCertMount := true
 443         type copyFile struct {
 444                 src  string
 445                 bind string
 446         }
 447         var copyFiles []copyFile
 448
 449         var binds []string
 450         for bind := range runner.Container.Mounts {
 451                 binds = append(binds, bind)
 452         }
 453         for bind := range runner.SecretMounts {
 454                 if _, ok := runner.Container.Mounts[bind]; ok {
 455                         return nil, fmt.Errorf("secret mount %q conflicts with regular mount", bind)
 456                 }
 457                 if runner.SecretMounts[bind].Kind != "json" &&
 458                         runner.SecretMounts[bind].Kind != "text" {
 459                         return nil, fmt.Errorf("secret mount %q type is %q but only 'json' and 'text' are permitted",
 460                                 bind, runner.SecretMounts[bind].Kind)
 461                 }
 462                 binds = append(binds, bind)
 463         }
 464         sort.Strings(binds)
 465
 466         for _, bind := range binds {
 467                 mnt, notSecret := runner.Container.Mounts[bind]
 468                 if !notSecret {
 469                         mnt = runner.SecretMounts[bind]
 470                 }
 471                 if bind == "stdout" || bind == "stderr" {
 472                         // Is it a "file" mount kind?
 473                         if mnt.Kind != "file" {
 474                                 return nil, fmt.Errorf("unsupported mount kind '%s' for %s: only 'file' is supported", mnt.Kind, bind)
 475                         }
 476
 477                         // Does path start with OutputPath?
 478                         prefix := runner.Container.OutputPath
 479                         if !strings.HasSuffix(prefix, "/") {
 480                                 prefix += "/"
 481                         }
 482                         if !strings.HasPrefix(mnt.Path, prefix) {
 483                                 return nil, fmt.Errorf("%s path does not start with OutputPath: %s, %s", strings.Title(bind), mnt.Path, prefix)
 484                         }
 485                 }
 486
 487                 if bind == "stdin" {
 488                         // Is it a "collection" mount kind?
 489                         if mnt.Kind != "collection" && mnt.Kind != "json" {
 490                                 return nil, fmt.Errorf("unsupported mount kind '%s' for stdin: only 'collection' and 'json' are supported", mnt.Kind)
 491                         }
 492                 }
 493
 494                 if bind == "/etc/arvados/ca-certificates.crt" {
 495                         needCertMount = false
 496                 }
 497
 498                 if strings.HasPrefix(bind, runner.Container.OutputPath+"/") && bind != runner.Container.OutputPath+"/" {
 499                         if mnt.Kind != "collection" && mnt.Kind != "text" && mnt.Kind != "json" {
 500                                 return nil, fmt.Errorf("only mount points of kind 'collection', 'text' or 'json' are supported underneath the output_path for %q, was %q", bind, mnt.Kind)
 501                         }
 502                 }
 503
 504                 switch {
 505                 case mnt.Kind == "collection" && bind != "stdin":
 506                         var src string
 507                         if mnt.UUID != "" && mnt.PortableDataHash != "" {
 508                                 return nil, fmt.Errorf("cannot specify both 'uuid' and 'portable_data_hash' for a collection mount")
 509                         }
 510                         if mnt.UUID != "" {
 511                                 if mnt.Writable {
 512                                         return nil, fmt.Errorf("writing to existing collections currently not permitted")
 513                                 }
 514                                 pdhOnly = false
 515                                 src = fmt.Sprintf("%s/by_id/%s", runner.ArvMountPoint, mnt.UUID)
 516                         } else if mnt.PortableDataHash != "" {
 517                                 if mnt.Writable && !strings.HasPrefix(bind, runner.Container.OutputPath+"/") {
 518                                         return nil, fmt.Errorf("can never write to a collection specified by portable data hash")
 519                                 }
 520                                 idx := strings.Index(mnt.PortableDataHash, "/")
 521                                 if idx > 0 {
 522                                         mnt.Path = path.Clean(mnt.PortableDataHash[idx:])
 523                                         mnt.PortableDataHash = mnt.PortableDataHash[0:idx]
 524                                         runner.Container.Mounts[bind] = mnt
 525                                 }
 526                                 src = fmt.Sprintf("%s/by_id/%s", runner.ArvMountPoint, mnt.PortableDataHash)
 527                                 if mnt.Path != "" && mnt.Path != "." {
 528                                         if strings.HasPrefix(mnt.Path, "./") {
 529                                                 mnt.Path = mnt.Path[2:]
 530                                         } else if strings.HasPrefix(mnt.Path, "/") {
 531                                                 mnt.Path = mnt.Path[1:]
 532                                         }
 533                                         src += "/" + mnt.Path
 534                                 }
 535                         } else {
 536                                 src = fmt.Sprintf("%s/tmp%d", runner.ArvMountPoint, tmpcount)
 537                                 arvMountCmd = append(arvMountCmd, "--mount-tmp", fmt.Sprintf("tmp%d", tmpcount))
 538                                 tmpcount++
 539                         }
 540                         if mnt.Writable {
 541                                 if bind == runner.Container.OutputPath {
 542                                         runner.HostOutputDir = src
 543                                         bindmounts[bind] = bindmount{HostPath: src}
 544                                 } else if strings.HasPrefix(bind, runner.Container.OutputPath+"/") {
 545                                         copyFiles = append(copyFiles, copyFile{src, runner.HostOutputDir + bind[len(runner.Container.OutputPath):]})
 546                                 } else {
 547                                         bindmounts[bind] = bindmount{HostPath: src}
 548                                 }
 549                         } else {
 550                                 bindmounts[bind] = bindmount{HostPath: src, ReadOnly: true}
 551                         }
 552                         collectionPaths = append(collectionPaths, src)
 553
 554                 case mnt.Kind == "tmp":
 555                         var tmpdir string
 556                         tmpdir, err = runner.MkTempDir(runner.parentTemp, "tmp")
 557                         if err != nil {
 558                                 return nil, fmt.Errorf("while creating mount temp dir: %v", err)
 559                         }
 560                         st, staterr := os.Stat(tmpdir)
 561                         if staterr != nil {
 562                                 return nil, fmt.Errorf("while Stat on temp dir: %v", staterr)
 563                         }
 564                         err = os.Chmod(tmpdir, st.Mode()|os.ModeSetgid|0777)
 565                         if staterr != nil {
 566                                 return nil, fmt.Errorf("while Chmod temp dir: %v", err)
 567                         }
 568                         bindmounts[bind] = bindmount{HostPath: tmpdir}
 569                         if bind == runner.Container.OutputPath {
 570                                 runner.HostOutputDir = tmpdir
 571                         }
 572
 573                 case mnt.Kind == "json" || mnt.Kind == "text":
 574                         var filedata []byte
 575                         if mnt.Kind == "json" {
 576                                 filedata, err = json.Marshal(mnt.Content)
 577                                 if err != nil {
 578                                         return nil, fmt.Errorf("encoding json data: %v", err)
 579                                 }
 580                         } else {
 581                                 text, ok := mnt.Content.(string)
 582                                 if !ok {
 583                                         return nil, fmt.Errorf("content for mount %q must be a string", bind)
 584                                 }
 585                                 filedata = []byte(text)
 586                         }
 587
 588                         tmpdir, err := runner.MkTempDir(runner.parentTemp, mnt.Kind)
 589                         if err != nil {
 590                                 return nil, fmt.Errorf("creating temp dir: %v", err)
 591                         }
 592                         tmpfn := filepath.Join(tmpdir, "mountdata."+mnt.Kind)
 593                         err = ioutil.WriteFile(tmpfn, filedata, 0444)
 594                         if err != nil {
 595                                 return nil, fmt.Errorf("writing temp file: %v", err)
 596                         }
 597                         if strings.HasPrefix(bind, runner.Container.OutputPath+"/") && (notSecret || runner.Container.Mounts[runner.Container.OutputPath].Kind != "collection") {
 598                                 // In most cases, if the container
 599                                 // specifies a literal file inside the
 600                                 // output path, we copy it into the
 601                                 // output directory (either a mounted
 602                                 // collection or a staging area on the
 603                                 // host fs). If it's a secret, it will
 604                                 // be skipped when copying output from
 605                                 // staging to Keep later.
 606                                 copyFiles = append(copyFiles, copyFile{tmpfn, runner.HostOutputDir + bind[len(runner.Container.OutputPath):]})
 607                         } else {
 608                                 // If a secret is outside OutputPath,
 609                                 // we bind mount the secret file
 610                                 // directly just like other mounts. We
 611                                 // also use this strategy when a
 612                                 // secret is inside OutputPath but
 613                                 // OutputPath is a live collection, to
 614                                 // avoid writing the secret to
 615                                 // Keep. Attempting to remove a
 616                                 // bind-mounted secret file from
 617                                 // inside the container will return a
 618                                 // "Device or resource busy" error
 619                                 // that might not be handled well by
 620                                 // the container, which is why we
 621                                 // don't use this strategy when
 622                                 // OutputPath is a staging directory.
 623                                 bindmounts[bind] = bindmount{HostPath: tmpfn, ReadOnly: true}
 624                         }
 625
 626                 case mnt.Kind == "git_tree":
 627                         tmpdir, err := runner.MkTempDir(runner.parentTemp, "git_tree")
 628                         if err != nil {
 629                                 return nil, fmt.Errorf("creating temp dir: %v", err)
 630                         }
 631                         err = gitMount(mnt).extractTree(runner.ContainerArvClient, tmpdir, token)
 632                         if err != nil {
 633                                 return nil, err
 634                         }
 635                         bindmounts[bind] = bindmount{HostPath: tmpdir, ReadOnly: true}
 636                 }
 637         }
 638
 639         if runner.HostOutputDir == "" {
 640                 return nil, fmt.Errorf("output path does not correspond to a writable mount point")
 641         }
 642
 643         if needCertMount && runner.Container.RuntimeConstraints.API {
 644                 for _, certfile := range arvadosclient.CertFiles {
 645                         _, err := os.Stat(certfile)
 646                         if err == nil {
 647                                 bindmounts["/etc/arvados/ca-certificates.crt"] = bindmount{HostPath: certfile, ReadOnly: true}
 648                                 break
 649                         }
 650                 }
 651         }
 652
 653         if pdhOnly {
 654                 // If we are only mounting collections by pdh, make
 655                 // sure we don't subscribe to websocket events to
 656                 // avoid putting undesired load on the API server
 657                 arvMountCmd = append(arvMountCmd, "--mount-by-pdh", "by_id", "--disable-event-listening")
 658         } else {
 659                 arvMountCmd = append(arvMountCmd, "--mount-by-id", "by_id")
 660         }
 661         // the by_uuid mount point is used by singularity when writing
 662         // out docker images converted to SIF
 663         arvMountCmd = append(arvMountCmd, "--mount-by-id", "by_uuid")
 664         arvMountCmd = append(arvMountCmd, runner.ArvMountPoint)
 665
 666         runner.ArvMount, err = runner.RunArvMount(arvMountCmd, token)
 667         if err != nil {
 668                 return nil, fmt.Errorf("while trying to start arv-mount: %v", err)
 669         }
 670         if runner.hoststatReporter != nil && runner.ArvMount != nil {
 671                 runner.hoststatReporter.ReportPID("arv-mount", runner.ArvMount.Process.Pid)
 672         }
 673
 674         for _, p := range collectionPaths {
 675                 _, err = os.Stat(p)
 676                 if err != nil {
 677                         return nil, fmt.Errorf("while checking that input files exist: %v", err)
 678                 }
 679         }
 680
 681         for _, cp := range copyFiles {
 682                 st, err := os.Stat(cp.src)
 683                 if err != nil {
 684                         return nil, fmt.Errorf("while staging writable file from %q to %q: %v", cp.src, cp.bind, err)
 685                 }
 686                 if st.IsDir() {
 687                         err = filepath.Walk(cp.src, func(walkpath string, walkinfo os.FileInfo, walkerr error) error {
 688                                 if walkerr != nil {
 689                                         return walkerr
 690                                 }
 691                                 target := path.Join(cp.bind, walkpath[len(cp.src):])
 692                                 if walkinfo.Mode().IsRegular() {
 693                                         copyerr := copyfile(walkpath, target)
 694                                         if copyerr != nil {
 695                                                 return copyerr
 696                                         }
 697                                         return os.Chmod(target, walkinfo.Mode()|0777)
 698                                 } else if walkinfo.Mode().IsDir() {
 699                                         mkerr := os.MkdirAll(target, 0777)
 700                                         if mkerr != nil {
 701                                                 return mkerr
 702                                         }
 703                                         return os.Chmod(target, walkinfo.Mode()|os.ModeSetgid|0777)
 704                                 } else {
 705                                         return fmt.Errorf("source %q is not a regular file or directory", cp.src)
 706                                 }
 707                         })
 708                 } else if st.Mode().IsRegular() {
 709                         err = copyfile(cp.src, cp.bind)
 710                         if err == nil {
 711                                 err = os.Chmod(cp.bind, st.Mode()|0777)
 712                         }
 713                 }
 714                 if err != nil {
 715                         return nil, fmt.Errorf("while staging writable file from %q to %q: %v", cp.src, cp.bind, err)
 716                 }
 717         }
 718
 719         return bindmounts, nil
 720 }
 721
 722 func (runner *ContainerRunner) stopHoststat() error {
 723         if runner.hoststatReporter == nil {
 724                 return nil
 725         }
 726         runner.hoststatReporter.Stop()
 727         err := runner.hoststatLogger.Close()
 728         if err != nil {
 729                 return fmt.Errorf("error closing hoststat logs: %v", err)
 730         }
 731         return nil
 732 }
 733
 734 func (runner *ContainerRunner) startHoststat() error {
 735         w, err := runner.NewLogWriter("hoststat")
 736         if err != nil {
 737                 return err
 738         }
 739         runner.hoststatLogger = NewThrottledLogger(w)
 740         runner.hoststatReporter = &crunchstat.Reporter{
 741                 Logger:     log.New(runner.hoststatLogger, "", 0),
 742                 CgroupRoot: runner.cgroupRoot,
 743                 PollPeriod: runner.statInterval,
 744         }
 745         runner.hoststatReporter.Start()
 746         runner.hoststatReporter.ReportPID("crunch-run", os.Getpid())
 747         return nil
 748 }
 749
 750 func (runner *ContainerRunner) startCrunchstat() error {
 751         w, err := runner.NewLogWriter("crunchstat")
 752         if err != nil {
 753                 return err
 754         }
 755         runner.statLogger = NewThrottledLogger(w)
 756         runner.statReporter = &crunchstat.Reporter{
 757                 CID:          runner.executor.CgroupID(),
 758                 Logger:       log.New(runner.statLogger, "", 0),
 759                 CgroupParent: runner.expectCgroupParent,
 760                 CgroupRoot:   runner.cgroupRoot,
 761                 PollPeriod:   runner.statInterval,
 762                 TempDir:      runner.parentTemp,
 763         }
 764         runner.statReporter.Start()
 765         return nil
 766 }
 767
 768 type infoCommand struct {
 769         label string
 770         cmd   []string
 771 }
 772
 773 // LogHostInfo logs info about the current host, for debugging and
 774 // accounting purposes. Although it's logged as "node-info", this is
 775 // about the environment where crunch-run is actually running, which
 776 // might differ from what's described in the node record (see
 777 // LogNodeRecord).
 778 func (runner *ContainerRunner) LogHostInfo() (err error) {
 779         w, err := runner.NewLogWriter("node-info")
 780         if err != nil {
 781                 return
 782         }
 783
 784         commands := []infoCommand{
 785                 {
 786                         label: "Host Information",
 787                         cmd:   []string{"uname", "-a"},
 788                 },
 789                 {
 790                         label: "CPU Information",
 791                         cmd:   []string{"cat", "/proc/cpuinfo"},
 792                 },
 793                 {
 794                         label: "Memory Information",
 795                         cmd:   []string{"cat", "/proc/meminfo"},
 796                 },
 797                 {
 798                         label: "Disk Space",
 799                         cmd:   []string{"df", "-m", "/", os.TempDir()},
 800                 },
 801                 {
 802                         label: "Disk INodes",
 803                         cmd:   []string{"df", "-i", "/", os.TempDir()},
 804                 },
 805         }
 806
 807         // Run commands with informational output to be logged.
 808         for _, command := range commands {
 809                 fmt.Fprintln(w, command.label)
 810                 cmd := exec.Command(command.cmd[0], command.cmd[1:]...)
 811                 cmd.Stdout = w
 812                 cmd.Stderr = w
 813                 if err := cmd.Run(); err != nil {
 814                         err = fmt.Errorf("While running command %q: %v", command.cmd, err)
 815                         fmt.Fprintln(w, err)
 816                         return err
 817                 }
 818                 fmt.Fprintln(w, "")
 819         }
 820
 821         err = w.Close()
 822         if err != nil {
 823                 return fmt.Errorf("While closing node-info logs: %v", err)
 824         }
 825         return nil
 826 }
 827
 828 // LogContainerRecord gets and saves the raw JSON container record from the API server
 829 func (runner *ContainerRunner) LogContainerRecord() error {
 830         logged, err := runner.logAPIResponse("container", "containers", map[string]interface{}{"filters": [][]string{{"uuid", "=", runner.Container.UUID}}}, nil)
 831         if !logged && err == nil {
 832                 err = fmt.Errorf("error: no container record found for %s", runner.Container.UUID)
 833         }
 834         return err
 835 }
 836
 837 // LogNodeRecord logs the current host's InstanceType config entry (or
 838 // the arvados#node record, if running via crunch-dispatch-slurm).
 839 func (runner *ContainerRunner) LogNodeRecord() error {
 840         if it := os.Getenv("InstanceType"); it != "" {
 841                 // Dispatched via arvados-dispatch-cloud. Save
 842                 // InstanceType config fragment received from
 843                 // dispatcher on stdin.
 844                 w, err := runner.LogCollection.OpenFile("node.json", os.O_CREATE|os.O_WRONLY, 0666)
 845                 if err != nil {
 846                         return err
 847                 }
 848                 defer w.Close()
 849                 _, err = io.WriteString(w, it)
 850                 if err != nil {
 851                         return err
 852                 }
 853                 return w.Close()
 854         }
 855         // Dispatched via crunch-dispatch-slurm. Look up
 856         // apiserver's node record corresponding to
 857         // $SLURMD_NODENAME.
 858         hostname := os.Getenv("SLURMD_NODENAME")
 859         if hostname == "" {
 860                 hostname, _ = os.Hostname()
 861         }
 862         _, err := runner.logAPIResponse("node", "nodes", map[string]interface{}{"filters": [][]string{{"hostname", "=", hostname}}}, func(resp interface{}) {
 863                 // The "info" field has admin-only info when
 864                 // obtained with a privileged token, and
 865                 // should not be logged.
 866                 node, ok := resp.(map[string]interface{})
 867                 if ok {
 868                         delete(node, "info")
 869                 }
 870         })
 871         return err
 872 }
 873
 874 func (runner *ContainerRunner) logAPIResponse(label, path string, params map[string]interface{}, munge func(interface{})) (logged bool, err error) {
 875         writer, err := runner.LogCollection.OpenFile(label+".json", os.O_CREATE|os.O_WRONLY, 0666)
 876         if err != nil {
 877                 return false, err
 878         }
 879         w := &ArvLogWriter{
 880                 ArvClient:     runner.DispatcherArvClient,
 881                 UUID:          runner.Container.UUID,
 882                 loggingStream: label,
 883                 writeCloser:   writer,
 884         }
 885
 886         reader, err := runner.DispatcherArvClient.CallRaw("GET", path, "", "", arvadosclient.Dict(params))
 887         if err != nil {
 888                 return false, fmt.Errorf("error getting %s record: %v", label, err)
 889         }
 890         defer reader.Close()
 891
 892         dec := json.NewDecoder(reader)
 893         dec.UseNumber()
 894         var resp map[string]interface{}
 895         if err = dec.Decode(&resp); err != nil {
 896                 return false, fmt.Errorf("error decoding %s list response: %v", label, err)
 897         }
 898         items, ok := resp["items"].([]interface{})
 899         if !ok {
 900                 return false, fmt.Errorf("error decoding %s list response: no \"items\" key in API list response", label)
 901         } else if len(items) < 1 {
 902                 return false, nil
 903         }
 904         if munge != nil {
 905                 munge(items[0])
 906         }
 907         // Re-encode it using indentation to improve readability
 908         enc := json.NewEncoder(w)
 909         enc.SetIndent("", "    ")
 910         if err = enc.Encode(items[0]); err != nil {
 911                 return false, fmt.Errorf("error logging %s record: %v", label, err)
 912         }
 913         err = w.Close()
 914         if err != nil {
 915                 return false, fmt.Errorf("error closing %s.json in log collection: %v", label, err)
 916         }
 917         return true, nil
 918 }
 919
 920 func (runner *ContainerRunner) getStdoutFile(mntPath string) (*os.File, error) {
 921         stdoutPath := mntPath[len(runner.Container.OutputPath):]
 922         index := strings.LastIndex(stdoutPath, "/")
 923         if index > 0 {
 924                 subdirs := stdoutPath[:index]
 925                 if subdirs != "" {
 926                         st, err := os.Stat(runner.HostOutputDir)
 927                         if err != nil {
 928                                 return nil, fmt.Errorf("While Stat on temp dir: %v", err)
 929                         }
 930                         stdoutPath := filepath.Join(runner.HostOutputDir, subdirs)
 931                         err = os.MkdirAll(stdoutPath, st.Mode()|os.ModeSetgid|0777)
 932                         if err != nil {
 933                                 return nil, fmt.Errorf("While MkdirAll %q: %v", stdoutPath, err)
 934                         }
 935                 }
 936         }
 937         stdoutFile, err := os.Create(filepath.Join(runner.HostOutputDir, stdoutPath))
 938         if err != nil {
 939                 return nil, fmt.Errorf("While creating file %q: %v", stdoutPath, err)
 940         }
 941
 942         return stdoutFile, nil
 943 }
 944
 945 // CreateContainer creates the docker container.
 946 func (runner *ContainerRunner) CreateContainer(imageID string, bindmounts map[string]bindmount) error {
 947         var stdin io.ReadCloser = ioutil.NopCloser(bytes.NewReader(nil))
 948         if mnt, ok := runner.Container.Mounts["stdin"]; ok {
 949                 switch mnt.Kind {
 950                 case "collection":
 951                         var collID string
 952                         if mnt.UUID != "" {
 953                                 collID = mnt.UUID
 954                         } else {
 955                                 collID = mnt.PortableDataHash
 956                         }
 957                         path := runner.ArvMountPoint + "/by_id/" + collID + "/" + mnt.Path
 958                         f, err := os.Open(path)
 959                         if err != nil {
 960                                 return err
 961                         }
 962                         stdin = f
 963                 case "json":
 964                         j, err := json.Marshal(mnt.Content)
 965                         if err != nil {
 966                                 return fmt.Errorf("error encoding stdin json data: %v", err)
 967                         }
 968                         stdin = ioutil.NopCloser(bytes.NewReader(j))
 969                 default:
 970                         return fmt.Errorf("stdin mount has unsupported kind %q", mnt.Kind)
 971                 }
 972         }
 973
 974         var stdout, stderr io.WriteCloser
 975         if mnt, ok := runner.Container.Mounts["stdout"]; ok {
 976                 f, err := runner.getStdoutFile(mnt.Path)
 977                 if err != nil {
 978                         return err
 979                 }
 980                 stdout = f
 981         } else if w, err := runner.NewLogWriter("stdout"); err != nil {
 982                 return err
 983         } else {
 984                 stdout = NewThrottledLogger(w)
 985         }
 986
 987         if mnt, ok := runner.Container.Mounts["stderr"]; ok {
 988                 f, err := runner.getStdoutFile(mnt.Path)
 989                 if err != nil {
 990                         return err
 991                 }
 992                 stderr = f
 993         } else if w, err := runner.NewLogWriter("stderr"); err != nil {
 994                 return err
 995         } else {
 996                 stderr = NewThrottledLogger(w)
 997         }
 998
 999         env := runner.Container.Environment
1000         enableNetwork := runner.enableNetwork == "always"
1001         if runner.Container.RuntimeConstraints.API {
1002                 enableNetwork = true
1003                 tok, err := runner.ContainerToken()
1004                 if err != nil {
1005                         return err
1006                 }
1007                 env = map[string]string{}
1008                 for k, v := range runner.Container.Environment {
1009                         env[k] = v
1010                 }
1011                 env["ARVADOS_API_TOKEN"] = tok
1012                 env["ARVADOS_API_HOST"] = os.Getenv("ARVADOS_API_HOST")
1013                 env["ARVADOS_API_HOST_INSECURE"] = os.Getenv("ARVADOS_API_HOST_INSECURE")
1014                 env["ARVADOS_KEEP_SERVICES"] = os.Getenv("ARVADOS_KEEP_SERVICES")
1015         }
1016         workdir := runner.Container.Cwd
1017         if workdir == "." {
1018                 // both "" and "." mean default
1019                 workdir = ""
1020         }
1021         ram := runner.Container.RuntimeConstraints.RAM
1022         if !runner.enableMemoryLimit {
1023                 ram = 0
1024         }
1025         runner.executorStdin = stdin
1026         runner.executorStdout = stdout
1027         runner.executorStderr = stderr
1028
1029         if runner.Container.RuntimeConstraints.CUDA.DeviceCount > 0 {
1030                 nvidiaModprobe(runner.CrunchLog)
1031         }
1032
1033         return runner.executor.Create(containerSpec{
1034                 Image:           imageID,
1035                 VCPUs:           runner.Container.RuntimeConstraints.VCPUs,
1036                 RAM:             ram,
1037                 WorkingDir:      workdir,
1038                 Env:             env,
1039                 BindMounts:      bindmounts,
1040                 Command:         runner.Container.Command,
1041                 EnableNetwork:   enableNetwork,
1042                 CUDADeviceCount: runner.Container.RuntimeConstraints.CUDA.DeviceCount,
1043                 NetworkMode:     runner.networkMode,
1044                 CgroupParent:    runner.setCgroupParent,
1045                 Stdin:           stdin,
1046                 Stdout:          stdout,
1047                 Stderr:          stderr,
1048         })
1049 }
1050
1051 // StartContainer starts the docker container created by CreateContainer.
1052 func (runner *ContainerRunner) StartContainer() error {
1053         runner.CrunchLog.Printf("Starting container")
1054         runner.cStateLock.Lock()
1055         defer runner.cStateLock.Unlock()
1056         if runner.cCancelled {
1057                 return ErrCancelled
1058         }
1059         err := runner.executor.Start()
1060         if err != nil {
1061                 var advice string
1062                 if m, e := regexp.MatchString("(?ms).*(exec|System error).*(no such file or directory|file not found).*", err.Error()); m && e == nil {
1063                         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])
1064                 }
1065                 return fmt.Errorf("could not start container: %v%s", err, advice)
1066         }
1067         return nil
1068 }
1069
1070 // WaitFinish waits for the container to terminate, capture the exit code, and
1071 // close the stdout/stderr logging.
1072 func (runner *ContainerRunner) WaitFinish() error {
1073         runner.CrunchLog.Print("Waiting for container to finish")
1074         var timeout <-chan time.Time
1075         if s := runner.Container.SchedulingParameters.MaxRunTime; s > 0 {
1076                 timeout = time.After(time.Duration(s) * time.Second)
1077         }
1078         ctx, cancel := context.WithCancel(context.Background())
1079         defer cancel()
1080         go func() {
1081                 select {
1082                 case <-timeout:
1083                         runner.CrunchLog.Printf("maximum run time exceeded. Stopping container.")
1084                         runner.stop(nil)
1085                 case <-runner.ArvMountExit:
1086                         runner.CrunchLog.Printf("arv-mount exited while container is still running. Stopping container.")
1087                         runner.stop(nil)
1088                 case <-ctx.Done():
1089                 }
1090         }()
1091         exitcode, err := runner.executor.Wait(ctx)
1092         if err != nil {
1093                 runner.checkBrokenNode(err)
1094                 return err
1095         }
1096         runner.ExitCode = &exitcode
1097
1098         extra := ""
1099         if exitcode&0x80 != 0 {
1100                 // Convert raw exit status (0x80 + signal number) to a
1101                 // string to log after the code, like " (signal 101)"
1102                 // or " (signal 9, killed)"
1103                 sig := syscall.WaitStatus(exitcode).Signal()
1104                 if name := unix.SignalName(sig); name != "" {
1105                         extra = fmt.Sprintf(" (signal %d, %s)", sig, name)
1106                 } else {
1107                         extra = fmt.Sprintf(" (signal %d)", sig)
1108                 }
1109         }
1110         runner.CrunchLog.Printf("Container exited with status code %d%s", exitcode, extra)
1111         err = runner.DispatcherArvClient.Update("containers", runner.Container.UUID, arvadosclient.Dict{
1112                 "container": arvadosclient.Dict{"exit_code": exitcode},
1113         }, nil)
1114         if err != nil {
1115                 runner.CrunchLog.Printf("ignoring error updating exit_code: %s", err)
1116         }
1117
1118         var returnErr error
1119         if err = runner.executorStdin.Close(); err != nil {
1120                 err = fmt.Errorf("error closing container stdin: %s", err)
1121                 runner.CrunchLog.Printf("%s", err)
1122                 returnErr = err
1123         }
1124         if err = runner.executorStdout.Close(); err != nil {
1125                 err = fmt.Errorf("error closing container stdout: %s", err)
1126                 runner.CrunchLog.Printf("%s", err)
1127                 if returnErr == nil {
1128                         returnErr = err
1129                 }
1130         }
1131         if err = runner.executorStderr.Close(); err != nil {
1132                 err = fmt.Errorf("error closing container stderr: %s", err)
1133                 runner.CrunchLog.Printf("%s", err)
1134                 if returnErr == nil {
1135                         returnErr = err
1136                 }
1137         }
1138
1139         if runner.statReporter != nil {
1140                 runner.statReporter.Stop()
1141                 err = runner.statLogger.Close()
1142                 if err != nil {
1143                         runner.CrunchLog.Printf("error closing crunchstat logs: %v", err)
1144                 }
1145         }
1146         return returnErr
1147 }
1148
1149 func (runner *ContainerRunner) updateLogs() {
1150         ticker := time.NewTicker(crunchLogUpdatePeriod / 360)
1151         defer ticker.Stop()
1152
1153         sigusr1 := make(chan os.Signal, 1)
1154         signal.Notify(sigusr1, syscall.SIGUSR1)
1155         defer signal.Stop(sigusr1)
1156
1157         saveAtTime := time.Now().Add(crunchLogUpdatePeriod)
1158         saveAtSize := crunchLogUpdateSize
1159         var savedSize int64
1160         for {
1161                 select {
1162                 case <-ticker.C:
1163                 case <-sigusr1:
1164                         saveAtTime = time.Now()
1165                 }
1166                 runner.logMtx.Lock()
1167                 done := runner.LogsPDH != nil
1168                 runner.logMtx.Unlock()
1169                 if done {
1170                         return
1171                 }
1172                 size := runner.LogCollection.Size()
1173                 if size == savedSize || (time.Now().Before(saveAtTime) && size < saveAtSize) {
1174                         continue
1175                 }
1176                 saveAtTime = time.Now().Add(crunchLogUpdatePeriod)
1177                 saveAtSize = runner.LogCollection.Size() + crunchLogUpdateSize
1178                 saved, err := runner.saveLogCollection(false)
1179                 if err != nil {
1180                         runner.CrunchLog.Printf("error updating log collection: %s", err)
1181                         continue
1182                 }
1183
1184                 err = runner.DispatcherArvClient.Update("containers", runner.Container.UUID, arvadosclient.Dict{
1185                         "container": arvadosclient.Dict{"log": saved.PortableDataHash},
1186                 }, nil)
1187                 if err != nil {
1188                         runner.CrunchLog.Printf("error updating container log to %s: %s", saved.PortableDataHash, err)
1189                         continue
1190                 }
1191
1192                 savedSize = size
1193         }
1194 }
1195
1196 func (runner *ContainerRunner) reportArvMountWarning(pattern, text string) {
1197         var updated arvados.Container
1198         err := runner.DispatcherArvClient.Update("containers", runner.Container.UUID, arvadosclient.Dict{
1199                 "container": arvadosclient.Dict{
1200                         "runtime_status": arvadosclient.Dict{
1201                                 "warning":       "arv-mount: " + pattern,
1202                                 "warningDetail": text,
1203                         },
1204                 },
1205         }, &updated)
1206         if err != nil {
1207                 runner.CrunchLog.Printf("error updating container runtime_status: %s", err)
1208         }
1209 }
1210
1211 // CaptureOutput saves data from the container's output directory if
1212 // needed, and updates the container output accordingly.
1213 func (runner *ContainerRunner) CaptureOutput(bindmounts map[string]bindmount) error {
1214         if runner.Container.RuntimeConstraints.API {
1215                 // Output may have been set directly by the container, so
1216                 // refresh the container record to check.
1217                 err := runner.DispatcherArvClient.Get("containers", runner.Container.UUID,
1218                         nil, &runner.Container)
1219                 if err != nil {
1220                         return err
1221                 }
1222                 if runner.Container.Output != "" {
1223                         // Container output is already set.
1224                         runner.OutputPDH = &runner.Container.Output
1225                         return nil
1226                 }
1227         }
1228
1229         txt, err := (&copier{
1230                 client:        runner.containerClient,
1231                 arvClient:     runner.ContainerArvClient,
1232                 keepClient:    runner.ContainerKeepClient,
1233                 hostOutputDir: runner.HostOutputDir,
1234                 ctrOutputDir:  runner.Container.OutputPath,
1235                 bindmounts:    bindmounts,
1236                 mounts:        runner.Container.Mounts,
1237                 secretMounts:  runner.SecretMounts,
1238                 logger:        runner.CrunchLog,
1239         }).Copy()
1240         if err != nil {
1241                 return err
1242         }
1243         if n := len(regexp.MustCompile(` [0-9a-f]+\+\S*\+R`).FindAllStringIndex(txt, -1)); n > 0 {
1244                 runner.CrunchLog.Printf("Copying %d data blocks from remote input collections...", n)
1245                 fs, err := (&arvados.Collection{ManifestText: txt}).FileSystem(runner.containerClient, runner.ContainerKeepClient)
1246                 if err != nil {
1247                         return err
1248                 }
1249                 txt, err = fs.MarshalManifest(".")
1250                 if err != nil {
1251                         return err
1252                 }
1253         }
1254         var resp arvados.Collection
1255         err = runner.ContainerArvClient.Create("collections", arvadosclient.Dict{
1256                 "ensure_unique_name": true,
1257                 "collection": arvadosclient.Dict{
1258                         "is_trashed":    true,
1259                         "name":          "output for " + runner.Container.UUID,
1260                         "manifest_text": txt,
1261                 },
1262         }, &resp)
1263         if err != nil {
1264                 return fmt.Errorf("error creating output collection: %v", err)
1265         }
1266         runner.OutputPDH = &resp.PortableDataHash
1267         return nil
1268 }
1269
1270 func (runner *ContainerRunner) CleanupDirs() {
1271         if runner.ArvMount != nil {
1272                 var delay int64 = 8
1273                 umount := exec.Command("arv-mount", fmt.Sprintf("--unmount-timeout=%d", delay), "--unmount", runner.ArvMountPoint)
1274                 umount.Stdout = runner.CrunchLog
1275                 umount.Stderr = runner.CrunchLog
1276                 runner.CrunchLog.Printf("Running %v", umount.Args)
1277                 umnterr := umount.Start()
1278
1279                 if umnterr != nil {
1280                         runner.CrunchLog.Printf("Error unmounting: %v", umnterr)
1281                         runner.ArvMount.Process.Kill()
1282                 } else {
1283                         // If arv-mount --unmount gets stuck for any reason, we
1284                         // don't want to wait for it forever.  Do Wait() in a goroutine
1285                         // so it doesn't block crunch-run.
1286                         umountExit := make(chan error)
1287                         go func() {
1288                                 mnterr := umount.Wait()
1289                                 if mnterr != nil {
1290                                         runner.CrunchLog.Printf("Error unmounting: %v", mnterr)
1291                                 }
1292                                 umountExit <- mnterr
1293                         }()
1294
1295                         for again := true; again; {
1296                                 again = false
1297                                 select {
1298                                 case <-umountExit:
1299                                         umount = nil
1300                                         again = true
1301                                 case <-runner.ArvMountExit:
1302                                         break
1303                                 case <-time.After(time.Duration((delay + 1) * int64(time.Second))):
1304                                         runner.CrunchLog.Printf("Timed out waiting for unmount")
1305                                         if umount != nil {
1306                                                 umount.Process.Kill()
1307                                         }
1308                                         runner.ArvMount.Process.Kill()
1309                                 }
1310                         }
1311                 }
1312                 runner.ArvMount = nil
1313         }
1314
1315         if runner.ArvMountPoint != "" {
1316                 if rmerr := os.Remove(runner.ArvMountPoint); rmerr != nil {
1317                         runner.CrunchLog.Printf("While cleaning up arv-mount directory %s: %v", runner.ArvMountPoint, rmerr)
1318                 }
1319                 runner.ArvMountPoint = ""
1320         }
1321
1322         if rmerr := os.RemoveAll(runner.parentTemp); rmerr != nil {
1323                 runner.CrunchLog.Printf("While cleaning up temporary directory %s: %v", runner.parentTemp, rmerr)
1324         }
1325 }
1326
1327 // CommitLogs posts the collection containing the final container logs.
1328 func (runner *ContainerRunner) CommitLogs() error {
1329         func() {
1330                 // Hold cStateLock to prevent races on CrunchLog (e.g., stop()).
1331                 runner.cStateLock.Lock()
1332                 defer runner.cStateLock.Unlock()
1333
1334                 runner.CrunchLog.Print(runner.finalState)
1335
1336                 if runner.arvMountLog != nil {
1337                         runner.arvMountLog.Close()
1338                 }
1339                 runner.CrunchLog.Close()
1340
1341                 // Closing CrunchLog above allows them to be committed to Keep at this
1342                 // point, but re-open crunch log with ArvClient in case there are any
1343                 // other further errors (such as failing to write the log to Keep!)
1344                 // while shutting down
1345                 runner.CrunchLog = NewThrottledLogger(&ArvLogWriter{
1346                         ArvClient:     runner.DispatcherArvClient,
1347                         UUID:          runner.Container.UUID,
1348                         loggingStream: "crunch-run",
1349                         writeCloser:   nil,
1350                 })
1351                 runner.CrunchLog.Immediate = log.New(os.Stderr, runner.Container.UUID+" ", 0)
1352         }()
1353
1354         if runner.keepstoreLogger != nil {
1355                 // Flush any buffered logs from our local keepstore
1356                 // process.  Discard anything logged after this point
1357                 // -- it won't end up in the log collection, so
1358                 // there's no point writing it to the collectionfs.
1359                 runner.keepstoreLogbuf.SetWriter(io.Discard)
1360                 runner.keepstoreLogger.Close()
1361                 runner.keepstoreLogger = nil
1362         }
1363
1364         if runner.LogsPDH != nil {
1365                 // If we have already assigned something to LogsPDH,
1366                 // we must be closing the re-opened log, which won't
1367                 // end up getting attached to the container record and
1368                 // therefore doesn't need to be saved as a collection
1369                 // -- it exists only to send logs to other channels.
1370                 return nil
1371         }
1372
1373         saved, err := runner.saveLogCollection(true)
1374         if err != nil {
1375                 return fmt.Errorf("error saving log collection: %s", err)
1376         }
1377         runner.logMtx.Lock()
1378         defer runner.logMtx.Unlock()
1379         runner.LogsPDH = &saved.PortableDataHash
1380         return nil
1381 }
1382
1383 func (runner *ContainerRunner) saveLogCollection(final bool) (response arvados.Collection, err error) {
1384         runner.logMtx.Lock()
1385         defer runner.logMtx.Unlock()
1386         if runner.LogsPDH != nil {
1387                 // Already finalized.
1388                 return
1389         }
1390         updates := arvadosclient.Dict{
1391                 "name": "logs for " + runner.Container.UUID,
1392         }
1393         mt, err1 := runner.LogCollection.MarshalManifest(".")
1394         if err1 == nil {
1395                 // Only send updated manifest text if there was no
1396                 // error.
1397                 updates["manifest_text"] = mt
1398         }
1399
1400         // Even if flushing the manifest had an error, we still want
1401         // to update the log record, if possible, to push the trash_at
1402         // and delete_at times into the future.  Details on bug
1403         // #17293.
1404         if final {
1405                 updates["is_trashed"] = true
1406         } else {
1407                 exp := time.Now().Add(crunchLogUpdatePeriod * 24)
1408                 updates["trash_at"] = exp
1409                 updates["delete_at"] = exp
1410         }
1411         reqBody := arvadosclient.Dict{"collection": updates}
1412         var err2 error
1413         if runner.logUUID == "" {
1414                 reqBody["ensure_unique_name"] = true
1415                 err2 = runner.DispatcherArvClient.Create("collections", reqBody, &response)
1416         } else {
1417                 err2 = runner.DispatcherArvClient.Update("collections", runner.logUUID, reqBody, &response)
1418         }
1419         if err2 == nil {
1420                 runner.logUUID = response.UUID
1421         }
1422
1423         if err1 != nil || err2 != nil {
1424                 err = fmt.Errorf("error recording logs: %q, %q", err1, err2)
1425         }
1426         return
1427 }
1428
1429 // UpdateContainerRunning updates the container state to "Running"
1430 func (runner *ContainerRunner) UpdateContainerRunning() error {
1431         runner.cStateLock.Lock()
1432         defer runner.cStateLock.Unlock()
1433         if runner.cCancelled {
1434                 return ErrCancelled
1435         }
1436         return runner.DispatcherArvClient.Update("containers", runner.Container.UUID,
1437                 arvadosclient.Dict{"container": arvadosclient.Dict{"state": "Running", "gateway_address": runner.gateway.Address}}, nil)
1438 }
1439
1440 // ContainerToken returns the api_token the container (and any
1441 // arv-mount processes) are allowed to use.
1442 func (runner *ContainerRunner) ContainerToken() (string, error) {
1443         if runner.token != "" {
1444                 return runner.token, nil
1445         }
1446
1447         var auth arvados.APIClientAuthorization
1448         err := runner.DispatcherArvClient.Call("GET", "containers", runner.Container.UUID, "auth", nil, &auth)
1449         if err != nil {
1450                 return "", err
1451         }
1452         runner.token = fmt.Sprintf("v2/%s/%s/%s", auth.UUID, auth.APIToken, runner.Container.UUID)
1453         return runner.token, nil
1454 }
1455
1456 // UpdateContainerFinal updates the container record state on API
1457 // server to "Complete" or "Cancelled"
1458 func (runner *ContainerRunner) UpdateContainerFinal() error {
1459         update := arvadosclient.Dict{}
1460         update["state"] = runner.finalState
1461         if runner.LogsPDH != nil {
1462                 update["log"] = *runner.LogsPDH
1463         }
1464         if runner.ExitCode != nil {
1465                 update["exit_code"] = *runner.ExitCode
1466         } else {
1467                 update["exit_code"] = nil
1468         }
1469         if runner.finalState == "Complete" && runner.OutputPDH != nil {
1470                 update["output"] = *runner.OutputPDH
1471         }
1472         var it arvados.InstanceType
1473         if j := os.Getenv("InstanceType"); j != "" && json.Unmarshal([]byte(j), &it) == nil && it.Price > 0 {
1474                 update["cost"] = it.Price * time.Now().Sub(runner.costStartTime).Seconds() / time.Hour.Seconds()
1475         }
1476         return runner.DispatcherArvClient.Update("containers", runner.Container.UUID, arvadosclient.Dict{"container": update}, nil)
1477 }
1478
1479 // IsCancelled returns the value of Cancelled, with goroutine safety.
1480 func (runner *ContainerRunner) IsCancelled() bool {
1481         runner.cStateLock.Lock()
1482         defer runner.cStateLock.Unlock()
1483         return runner.cCancelled
1484 }
1485
1486 // NewArvLogWriter creates an ArvLogWriter
1487 func (runner *ContainerRunner) NewArvLogWriter(name string) (io.WriteCloser, error) {
1488         writer, err := runner.LogCollection.OpenFile(name+".txt", os.O_CREATE|os.O_WRONLY, 0666)
1489         if err != nil {
1490                 return nil, err
1491         }
1492         return &ArvLogWriter{
1493                 ArvClient:     runner.DispatcherArvClient,
1494                 UUID:          runner.Container.UUID,
1495                 loggingStream: name,
1496                 writeCloser:   writer,
1497         }, nil
1498 }
1499
1500 // Run the full container lifecycle.
1501 func (runner *ContainerRunner) Run() (err error) {
1502         runner.CrunchLog.Printf("crunch-run %s started", cmd.Version.String())
1503         runner.CrunchLog.Printf("%s", currentUserAndGroups())
1504         v, _ := exec.Command("arv-mount", "--version").CombinedOutput()
1505         runner.CrunchLog.Printf("Using FUSE mount: %s", v)
1506         runner.CrunchLog.Printf("Using container runtime: %s", runner.executor.Runtime())
1507         runner.CrunchLog.Printf("Executing container: %s", runner.Container.UUID)
1508         runner.costStartTime = time.Now()
1509
1510         hostname, hosterr := os.Hostname()
1511         if hosterr != nil {
1512                 runner.CrunchLog.Printf("Error getting hostname '%v'", hosterr)
1513         } else {
1514                 runner.CrunchLog.Printf("Executing on host '%s'", hostname)
1515         }
1516
1517         runner.finalState = "Queued"
1518
1519         defer func() {
1520                 runner.CleanupDirs()
1521
1522                 runner.CrunchLog.Printf("crunch-run finished")
1523                 runner.CrunchLog.Close()
1524         }()
1525
1526         err = runner.fetchContainerRecord()
1527         if err != nil {
1528                 return
1529         }
1530         if runner.Container.State != "Locked" {
1531                 return fmt.Errorf("dispatch error detected: container %q has state %q", runner.Container.UUID, runner.Container.State)
1532         }
1533
1534         var bindmounts map[string]bindmount
1535         defer func() {
1536                 // checkErr prints e (unless it's nil) and sets err to
1537                 // e (unless err is already non-nil). Thus, if err
1538                 // hasn't already been assigned when Run() returns,
1539                 // this cleanup func will cause Run() to return the
1540                 // first non-nil error that is passed to checkErr().
1541                 checkErr := func(errorIn string, e error) {
1542                         if e == nil {
1543                                 return
1544                         }
1545                         runner.CrunchLog.Printf("error in %s: %v", errorIn, e)
1546                         if err == nil {
1547                                 err = e
1548                         }
1549                         if runner.finalState == "Complete" {
1550                                 // There was an error in the finalization.
1551                                 runner.finalState = "Cancelled"
1552                         }
1553                 }
1554
1555                 // Log the error encountered in Run(), if any
1556                 checkErr("Run", err)
1557
1558                 if runner.finalState == "Queued" {
1559                         runner.UpdateContainerFinal()
1560                         return
1561                 }
1562
1563                 if runner.IsCancelled() {
1564                         runner.finalState = "Cancelled"
1565                         // but don't return yet -- we still want to
1566                         // capture partial output and write logs
1567                 }
1568
1569                 if bindmounts != nil {
1570                         checkErr("CaptureOutput", runner.CaptureOutput(bindmounts))
1571                 }
1572                 checkErr("stopHoststat", runner.stopHoststat())
1573                 checkErr("CommitLogs", runner.CommitLogs())
1574                 runner.CleanupDirs()
1575                 checkErr("UpdateContainerFinal", runner.UpdateContainerFinal())
1576         }()
1577
1578         runner.setupSignals()
1579         err = runner.startHoststat()
1580         if err != nil {
1581                 return
1582         }
1583         if runner.keepstore != nil {
1584                 runner.hoststatReporter.ReportPID("keepstore", runner.keepstore.Process.Pid)
1585         }
1586
1587         // set up FUSE mount and binds
1588         bindmounts, err = runner.SetupMounts()
1589         if err != nil {
1590                 runner.finalState = "Cancelled"
1591                 err = fmt.Errorf("While setting up mounts: %v", err)
1592                 return
1593         }
1594
1595         // check for and/or load image
1596         imageID, err := runner.LoadImage()
1597         if err != nil {
1598                 if !runner.checkBrokenNode(err) {
1599                         // Failed to load image but not due to a "broken node"
1600                         // condition, probably user error.
1601                         runner.finalState = "Cancelled"
1602                 }
1603                 err = fmt.Errorf("While loading container image: %v", err)
1604                 return
1605         }
1606
1607         err = runner.CreateContainer(imageID, bindmounts)
1608         if err != nil {
1609                 return
1610         }
1611         err = runner.LogHostInfo()
1612         if err != nil {
1613                 return
1614         }
1615         err = runner.LogNodeRecord()
1616         if err != nil {
1617                 return
1618         }
1619         err = runner.LogContainerRecord()
1620         if err != nil {
1621                 return
1622         }
1623
1624         if runner.IsCancelled() {
1625                 return
1626         }
1627
1628         err = runner.UpdateContainerRunning()
1629         if err != nil {
1630                 return
1631         }
1632         runner.finalState = "Cancelled"
1633
1634         err = runner.startCrunchstat()
1635         if err != nil {
1636                 return
1637         }
1638
1639         err = runner.StartContainer()
1640         if err != nil {
1641                 runner.checkBrokenNode(err)
1642                 return
1643         }
1644
1645         err = runner.WaitFinish()
1646         if err == nil && !runner.IsCancelled() {
1647                 runner.finalState = "Complete"
1648         }
1649         return
1650 }
1651
1652 // Fetch the current container record (uuid = runner.Container.UUID)
1653 // into runner.Container.
1654 func (runner *ContainerRunner) fetchContainerRecord() error {
1655         reader, err := runner.DispatcherArvClient.CallRaw("GET", "containers", runner.Container.UUID, "", nil)
1656         if err != nil {
1657                 return fmt.Errorf("error fetching container record: %v", err)
1658         }
1659         defer reader.Close()
1660
1661         dec := json.NewDecoder(reader)
1662         dec.UseNumber()
1663         err = dec.Decode(&runner.Container)
1664         if err != nil {
1665                 return fmt.Errorf("error decoding container record: %v", err)
1666         }
1667
1668         var sm struct {
1669                 SecretMounts map[string]arvados.Mount `json:"secret_mounts"`
1670         }
1671
1672         containerToken, err := runner.ContainerToken()
1673         if err != nil {
1674                 return fmt.Errorf("error getting container token: %v", err)
1675         }
1676
1677         runner.ContainerArvClient, runner.ContainerKeepClient,
1678                 runner.containerClient, err = runner.MkArvClient(containerToken)
1679         if err != nil {
1680                 return fmt.Errorf("error creating container API client: %v", err)
1681         }
1682
1683         runner.ContainerKeepClient.SetStorageClasses(runner.Container.OutputStorageClasses)
1684         runner.DispatcherKeepClient.SetStorageClasses(runner.Container.OutputStorageClasses)
1685
1686         err = runner.ContainerArvClient.Call("GET", "containers", runner.Container.UUID, "secret_mounts", nil, &sm)
1687         if err != nil {
1688                 if apierr, ok := err.(arvadosclient.APIServerError); !ok || apierr.HttpStatusCode != 404 {
1689                         return fmt.Errorf("error fetching secret_mounts: %v", err)
1690                 }
1691                 // ok && apierr.HttpStatusCode == 404, which means
1692                 // secret_mounts isn't supported by this API server.
1693         }
1694         runner.SecretMounts = sm.SecretMounts
1695
1696         return nil
1697 }
1698
1699 // NewContainerRunner creates a new container runner.
1700 func NewContainerRunner(dispatcherClient *arvados.Client,
1701         dispatcherArvClient IArvadosClient,
1702         dispatcherKeepClient IKeepClient,
1703         containerUUID string) (*ContainerRunner, error) {
1704
1705         cr := &ContainerRunner{
1706                 dispatcherClient:     dispatcherClient,
1707                 DispatcherArvClient:  dispatcherArvClient,
1708                 DispatcherKeepClient: dispatcherKeepClient,
1709         }
1710         cr.NewLogWriter = cr.NewArvLogWriter
1711         cr.RunArvMount = cr.ArvMountCmd
1712         cr.MkTempDir = ioutil.TempDir
1713         cr.MkArvClient = func(token string) (IArvadosClient, IKeepClient, *arvados.Client, error) {
1714                 cl, err := arvadosclient.MakeArvadosClient()
1715                 if err != nil {
1716                         return nil, nil, nil, err
1717                 }
1718                 cl.ApiToken = token
1719                 kc, err := keepclient.MakeKeepClient(cl)
1720                 if err != nil {
1721                         return nil, nil, nil, err
1722                 }
1723                 c2 := arvados.NewClientFromEnv()
1724                 c2.AuthToken = token
1725                 return cl, kc, c2, nil
1726         }
1727         var err error
1728         cr.LogCollection, err = (&arvados.Collection{}).FileSystem(cr.dispatcherClient, cr.DispatcherKeepClient)
1729         if err != nil {
1730                 return nil, err
1731         }
1732         cr.Container.UUID = containerUUID
1733         w, err := cr.NewLogWriter("crunch-run")
1734         if err != nil {
1735                 return nil, err
1736         }
1737         cr.CrunchLog = NewThrottledLogger(w)
1738         cr.CrunchLog.Immediate = log.New(os.Stderr, containerUUID+" ", 0)
1739
1740         loadLogThrottleParams(dispatcherArvClient)
1741         go cr.updateLogs()
1742
1743         return cr, nil
1744 }
1745
1746 func (command) RunCommand(prog string, args []string, stdin io.Reader, stdout, stderr io.Writer) int {
1747         log := log.New(stderr, "", 0)
1748         flags := flag.NewFlagSet(prog, flag.ContinueOnError)
1749         statInterval := flags.Duration("crunchstat-interval", 10*time.Second, "sampling period for periodic resource usage reporting")
1750         cgroupRoot := flags.String("cgroup-root", "/sys/fs/cgroup", "path to sysfs cgroup tree")
1751         cgroupParent := flags.String("cgroup-parent", "docker", "name of container's parent cgroup (ignored if -cgroup-parent-subsystem is used)")
1752         cgroupParentSubsystem := flags.String("cgroup-parent-subsystem", "", "use current cgroup for given subsystem as parent cgroup for container")
1753         caCertsPath := flags.String("ca-certs", "", "Path to TLS root certificates")
1754         detach := flags.Bool("detach", false, "Detach from parent process and run in the background")
1755         stdinConfig := flags.Bool("stdin-config", false, "Load config and environment variables from JSON message on stdin")
1756         configFile := flags.String("config", arvados.DefaultConfigFile, "filename of cluster config file to try loading if -stdin-config=false (default is $ARVADOS_CONFIG)")
1757         sleep := flags.Duration("sleep", 0, "Delay before starting (testing use only)")
1758         kill := flags.Int("kill", -1, "Send signal to an existing crunch-run process for given UUID")
1759         list := flags.Bool("list", false, "List UUIDs of existing crunch-run processes")
1760         enableMemoryLimit := flags.Bool("enable-memory-limit", true, "tell container runtime to limit container's memory usage")
1761         enableNetwork := flags.String("container-enable-networking", "default", "enable networking \"always\" (for all containers) or \"default\" (for containers that request it)")
1762         networkMode := flags.String("container-network-mode", "default", `Docker network mode for container (use any argument valid for docker --net)`)
1763         memprofile := flags.String("memprofile", "", "write memory profile to `file` after running container")
1764         runtimeEngine := flags.String("runtime-engine", "docker", "container runtime: docker or singularity")
1765         brokenNodeHook := flags.String("broken-node-hook", "", "script to run if node is detected to be broken (for example, Docker daemon is not running)")
1766         flags.Duration("check-containerd", 0, "Ignored. Exists for compatibility with older versions.")
1767         version := flags.Bool("version", false, "Write version information to stdout and exit 0.")
1768
1769         ignoreDetachFlag := false
1770         if len(args) > 0 && args[0] == "-no-detach" {
1771                 // This process was invoked by a parent process, which
1772                 // has passed along its own arguments, including
1773                 // -detach, after the leading -no-detach flag.  Strip
1774                 // the leading -no-detach flag (it's not recognized by
1775                 // flags.Parse()) and ignore the -detach flag that
1776                 // comes later.
1777                 args = args[1:]
1778                 ignoreDetachFlag = true
1779         }
1780
1781         if ok, code := cmd.ParseFlags(flags, prog, args, "container-uuid", stderr); !ok {
1782                 return code
1783         } else if *version {
1784                 fmt.Fprintln(stdout, prog, cmd.Version.String())
1785                 return 0
1786         } else if !*list && flags.NArg() != 1 {
1787                 fmt.Fprintf(stderr, "missing required argument: container-uuid (try -help)\n")
1788                 return 2
1789         }
1790
1791         containerUUID := flags.Arg(0)
1792
1793         switch {
1794         case *detach && !ignoreDetachFlag:
1795                 return Detach(containerUUID, prog, args, os.Stdin, os.Stdout, os.Stderr)
1796         case *kill >= 0:
1797                 return KillProcess(containerUUID, syscall.Signal(*kill), os.Stdout, os.Stderr)
1798         case *list:
1799                 return ListProcesses(os.Stdout, os.Stderr)
1800         }
1801
1802         if len(containerUUID) != 27 {
1803                 log.Printf("usage: %s [options] UUID", prog)
1804                 return 1
1805         }
1806
1807         var keepstoreLogbuf bufThenWrite
1808         var conf ConfigData
1809         if *stdinConfig {
1810                 err := json.NewDecoder(stdin).Decode(&conf)
1811                 if err != nil {
1812                         log.Printf("decode stdin: %s", err)
1813                         return 1
1814                 }
1815                 for k, v := range conf.Env {
1816                         err = os.Setenv(k, v)
1817                         if err != nil {
1818                                 log.Printf("setenv(%q): %s", k, err)
1819                                 return 1
1820                         }
1821                 }
1822                 if conf.Cluster != nil {
1823                         // ClusterID is missing from the JSON
1824                         // representation, but we need it to generate
1825                         // a valid config file for keepstore, so we
1826                         // fill it using the container UUID prefix.
1827                         conf.Cluster.ClusterID = containerUUID[:5]
1828                 }
1829         } else {
1830                 conf = hpcConfData(containerUUID, *configFile, io.MultiWriter(&keepstoreLogbuf, stderr))
1831         }
1832
1833         log.Printf("crunch-run %s started", cmd.Version.String())
1834         time.Sleep(*sleep)
1835
1836         if *caCertsPath != "" {
1837                 arvadosclient.CertFiles = []string{*caCertsPath}
1838         }
1839
1840         keepstore, err := startLocalKeepstore(conf, io.MultiWriter(&keepstoreLogbuf, stderr))
1841         if err != nil {
1842                 log.Print(err)
1843                 return 1
1844         }
1845         if keepstore != nil {
1846                 defer keepstore.Process.Kill()
1847         }
1848
1849         api, err := arvadosclient.MakeArvadosClient()
1850         if err != nil {
1851                 log.Printf("%s: %v", containerUUID, err)
1852                 return 1
1853         }
1854         api.Retries = 8
1855
1856         kc, err := keepclient.MakeKeepClient(api)
1857         if err != nil {
1858                 log.Printf("%s: %v", containerUUID, err)
1859                 return 1
1860         }
1861         kc.BlockCache = &keepclient.BlockCache{MaxBlocks: 2}
1862         kc.Retries = 4
1863
1864         cr, err := NewContainerRunner(arvados.NewClientFromEnv(), api, kc, containerUUID)
1865         if err != nil {
1866                 log.Print(err)
1867                 return 1
1868         }
1869
1870         cr.keepstore = keepstore
1871         if keepstore == nil {
1872                 // Log explanation (if any) for why we're not running
1873                 // a local keepstore.
1874                 var buf bytes.Buffer
1875                 keepstoreLogbuf.SetWriter(&buf)
1876                 if buf.Len() > 0 {
1877                         cr.CrunchLog.Printf("%s", strings.TrimSpace(buf.String()))
1878                 }
1879         } else if logWhat := conf.Cluster.Containers.LocalKeepLogsToContainerLog; logWhat == "none" {
1880                 cr.CrunchLog.Printf("using local keepstore process (pid %d) at %s", keepstore.Process.Pid, os.Getenv("ARVADOS_KEEP_SERVICES"))
1881                 keepstoreLogbuf.SetWriter(io.Discard)
1882         } else {
1883                 cr.CrunchLog.Printf("using local keepstore process (pid %d) at %s, writing logs to keepstore.txt in log collection", keepstore.Process.Pid, os.Getenv("ARVADOS_KEEP_SERVICES"))
1884                 logwriter, err := cr.NewLogWriter("keepstore")
1885                 if err != nil {
1886                         log.Print(err)
1887                         return 1
1888                 }
1889                 cr.keepstoreLogger = NewThrottledLogger(logwriter)
1890
1891                 var writer io.WriteCloser = cr.keepstoreLogger
1892                 if logWhat == "errors" {
1893                         writer = &filterKeepstoreErrorsOnly{WriteCloser: writer}
1894                 } else if logWhat != "all" {
1895                         // should have been caught earlier by
1896                         // dispatcher's config loader
1897                         log.Printf("invalid value for Containers.LocalKeepLogsToContainerLog: %q", logWhat)
1898                         return 1
1899                 }
1900                 err = keepstoreLogbuf.SetWriter(writer)
1901                 if err != nil {
1902                         log.Print(err)
1903                         return 1
1904                 }
1905                 cr.keepstoreLogbuf = &keepstoreLogbuf
1906         }
1907
1908         switch *runtimeEngine {
1909         case "docker":
1910                 cr.executor, err = newDockerExecutor(containerUUID, cr.CrunchLog.Printf, cr.containerWatchdogInterval)
1911         case "singularity":
1912                 cr.executor, err = newSingularityExecutor(cr.CrunchLog.Printf)
1913         default:
1914                 cr.CrunchLog.Printf("%s: unsupported RuntimeEngine %q", containerUUID, *runtimeEngine)
1915                 cr.CrunchLog.Close()
1916                 return 1
1917         }
1918         if err != nil {
1919                 cr.CrunchLog.Printf("%s: %v", containerUUID, err)
1920                 cr.checkBrokenNode(err)
1921                 cr.CrunchLog.Close()
1922                 return 1
1923         }
1924         defer cr.executor.Close()
1925
1926         cr.brokenNodeHook = *brokenNodeHook
1927
1928         gwAuthSecret := os.Getenv("GatewayAuthSecret")
1929         os.Unsetenv("GatewayAuthSecret")
1930         if gwAuthSecret == "" {
1931                 // not safe to run a gateway service without an auth
1932                 // secret
1933                 cr.CrunchLog.Printf("Not starting a gateway server (GatewayAuthSecret was not provided by dispatcher)")
1934         } else {
1935                 gwListen := os.Getenv("GatewayAddress")
1936                 cr.gateway = Gateway{
1937                         Address:       gwListen,
1938                         AuthSecret:    gwAuthSecret,
1939                         ContainerUUID: containerUUID,
1940                         Target:        cr.executor,
1941                         Log:           cr.CrunchLog,
1942                 }
1943                 if gwListen == "" {
1944                         // Direct connection won't work, so we use the
1945                         // gateway_address field to indicate the
1946                         // internalURL of the controller process that
1947                         // has the current tunnel connection.
1948                         cr.gateway.ArvadosClient = cr.dispatcherClient
1949                         cr.gateway.UpdateTunnelURL = func(url string) {
1950                                 cr.gateway.Address = "tunnel " + url
1951                                 cr.DispatcherArvClient.Update("containers", containerUUID,
1952                                         arvadosclient.Dict{"container": arvadosclient.Dict{"gateway_address": cr.gateway.Address}}, nil)
1953                         }
1954                 }
1955                 err = cr.gateway.Start()
1956                 if err != nil {
1957                         log.Printf("error starting gateway server: %s", err)
1958                         return 1
1959                 }
1960         }
1961
1962         parentTemp, tmperr := cr.MkTempDir("", "crunch-run."+containerUUID+".")
1963         if tmperr != nil {
1964                 log.Printf("%s: %v", containerUUID, tmperr)
1965                 return 1
1966         }
1967
1968         cr.parentTemp = parentTemp
1969         cr.statInterval = *statInterval
1970         cr.cgroupRoot = *cgroupRoot
1971         cr.expectCgroupParent = *cgroupParent
1972         cr.enableMemoryLimit = *enableMemoryLimit
1973         cr.enableNetwork = *enableNetwork
1974         cr.networkMode = *networkMode
1975         if *cgroupParentSubsystem != "" {
1976                 p, err := findCgroup(*cgroupParentSubsystem)
1977                 if err != nil {
1978                         log.Printf("fatal: cgroup parent subsystem: %s", err)
1979                         return 1
1980                 }
1981                 cr.setCgroupParent = p
1982                 cr.expectCgroupParent = p
1983         }
1984
1985         runerr := cr.Run()
1986
1987         if *memprofile != "" {
1988                 f, err := os.Create(*memprofile)
1989                 if err != nil {
1990                         log.Printf("could not create memory profile: %s", err)
1991                 }
1992                 runtime.GC() // get up-to-date statistics
1993                 if err := pprof.WriteHeapProfile(f); err != nil {
1994                         log.Printf("could not write memory profile: %s", err)
1995                 }
1996                 closeerr := f.Close()
1997                 if closeerr != nil {
1998                         log.Printf("closing memprofile file: %s", err)
1999                 }
2000         }
2001
2002         if runerr != nil {
2003                 log.Printf("%s: %v", containerUUID, runerr)
2004                 return 1
2005         }
2006         return 0
2007 }
2008
2009 // Try to load ConfigData in hpc (slurm/lsf) environment. This means
2010 // loading the cluster config from the specified file and (if that
2011 // works) getting the runtime_constraints container field from
2012 // controller to determine # VCPUs so we can calculate KeepBuffers.
2013 func hpcConfData(uuid string, configFile string, stderr io.Writer) ConfigData {
2014         var conf ConfigData
2015         conf.Cluster = loadClusterConfigFile(configFile, stderr)
2016         if conf.Cluster == nil {
2017                 // skip loading the container record -- we won't be
2018                 // able to start local keepstore anyway.
2019                 return conf
2020         }
2021         arv, err := arvadosclient.MakeArvadosClient()
2022         if err != nil {
2023                 fmt.Fprintf(stderr, "error setting up arvadosclient: %s\n", err)
2024                 return conf
2025         }
2026         arv.Retries = 8
2027         var ctr arvados.Container
2028         err = arv.Call("GET", "containers", uuid, "", arvadosclient.Dict{"select": []string{"runtime_constraints"}}, &ctr)
2029         if err != nil {
2030                 fmt.Fprintf(stderr, "error getting container record: %s\n", err)
2031                 return conf
2032         }
2033         if ctr.RuntimeConstraints.VCPUs > 0 {
2034                 conf.KeepBuffers = ctr.RuntimeConstraints.VCPUs * conf.Cluster.Containers.LocalKeepBlobBuffersPerVCPU
2035         }
2036         return conf
2037 }
2038
2039 // Load cluster config file from given path. If an error occurs, log
2040 // the error to stderr and return nil.
2041 func loadClusterConfigFile(path string, stderr io.Writer) *arvados.Cluster {
2042         ldr := config.NewLoader(&bytes.Buffer{}, ctxlog.New(stderr, "plain", "info"))
2043         ldr.Path = path
2044         cfg, err := ldr.Load()
2045         if err != nil {
2046                 fmt.Fprintf(stderr, "could not load config file %s: %s\n", path, err)
2047                 return nil
2048         }
2049         cluster, err := cfg.GetCluster("")
2050         if err != nil {
2051                 fmt.Fprintf(stderr, "could not use config file %s: %s\n", path, err)
2052                 return nil
2053         }
2054         fmt.Fprintf(stderr, "loaded config file %s\n", path)
2055         return cluster
2056 }
2057
2058 func startLocalKeepstore(configData ConfigData, logbuf io.Writer) (*exec.Cmd, error) {
2059         if configData.KeepBuffers < 1 {
2060                 fmt.Fprintf(logbuf, "not starting a local keepstore process because KeepBuffers=%v in config\n", configData.KeepBuffers)
2061                 return nil, nil
2062         }
2063         if configData.Cluster == nil {
2064                 fmt.Fprint(logbuf, "not starting a local keepstore process because cluster config file was not loaded\n")
2065                 return nil, nil
2066         }
2067         for uuid, vol := range configData.Cluster.Volumes {
2068                 if len(vol.AccessViaHosts) > 0 {
2069                         fmt.Fprintf(logbuf, "not starting a local keepstore process because a volume (%s) uses AccessViaHosts\n", uuid)
2070                         return nil, nil
2071                 }
2072                 if !vol.ReadOnly && vol.Replication < configData.Cluster.Collections.DefaultReplication {
2073                         fmt.Fprintf(logbuf, "not starting a local keepstore process because a writable volume (%s) has replication less than Collections.DefaultReplication (%d < %d)\n", uuid, vol.Replication, configData.Cluster.Collections.DefaultReplication)
2074                         return nil, nil
2075                 }
2076         }
2077
2078         // Rather than have an alternate way to tell keepstore how
2079         // many buffers to use when starting it this way, we just
2080         // modify the cluster configuration that we feed it on stdin.
2081         configData.Cluster.API.MaxKeepBlobBuffers = configData.KeepBuffers
2082
2083         localaddr := localKeepstoreAddr()
2084         ln, err := net.Listen("tcp", net.JoinHostPort(localaddr, "0"))
2085         if err != nil {
2086                 return nil, err
2087         }
2088         _, port, err := net.SplitHostPort(ln.Addr().String())
2089         if err != nil {
2090                 ln.Close()
2091                 return nil, err
2092         }
2093         ln.Close()
2094         url := "http://" + net.JoinHostPort(localaddr, port)
2095
2096         fmt.Fprintf(logbuf, "starting keepstore on %s\n", url)
2097
2098         var confJSON bytes.Buffer
2099         err = json.NewEncoder(&confJSON).Encode(arvados.Config{
2100                 Clusters: map[string]arvados.Cluster{
2101                         configData.Cluster.ClusterID: *configData.Cluster,
2102                 },
2103         })
2104         if err != nil {
2105                 return nil, err
2106         }
2107         cmd := exec.Command("/proc/self/exe", "keepstore", "-config=-")
2108         if target, err := os.Readlink(cmd.Path); err == nil && strings.HasSuffix(target, ".test") {
2109                 // If we're a 'go test' process, running
2110                 // /proc/self/exe would start the test suite in a
2111                 // child process, which is not what we want.
2112                 cmd.Path, _ = exec.LookPath("go")
2113                 cmd.Args = append([]string{"go", "run", "../../cmd/arvados-server"}, cmd.Args[1:]...)
2114                 cmd.Env = os.Environ()
2115         }
2116         cmd.Stdin = &confJSON
2117         cmd.Stdout = logbuf
2118         cmd.Stderr = logbuf
2119         cmd.Env = append(cmd.Env,
2120                 "GOGC=10",
2121                 "ARVADOS_SERVICE_INTERNAL_URL="+url)
2122         err = cmd.Start()
2123         if err != nil {
2124                 return nil, fmt.Errorf("error starting keepstore process: %w", err)
2125         }
2126         cmdExited := false
2127         go func() {
2128                 cmd.Wait()
2129                 cmdExited = true
2130         }()
2131         ctx, cancel := context.WithDeadline(context.Background(), time.Now().Add(time.Second*10))
2132         defer cancel()
2133         poll := time.NewTicker(time.Second / 10)
2134         defer poll.Stop()
2135         client := http.Client{}
2136         for range poll.C {
2137                 testReq, err := http.NewRequestWithContext(ctx, "GET", url+"/_health/ping", nil)
2138                 testReq.Header.Set("Authorization", "Bearer "+configData.Cluster.ManagementToken)
2139                 if err != nil {
2140                         return nil, err
2141                 }
2142                 resp, err := client.Do(testReq)
2143                 if err == nil {
2144                         resp.Body.Close()
2145                         if resp.StatusCode == http.StatusOK {
2146                                 break
2147                         }
2148                 }
2149                 if cmdExited {
2150                         return nil, fmt.Errorf("keepstore child process exited")
2151                 }
2152                 if ctx.Err() != nil {
2153                         return nil, fmt.Errorf("timed out waiting for new keepstore process to report healthy")
2154                 }
2155         }
2156         os.Setenv("ARVADOS_KEEP_SERVICES", url)
2157         return cmd, nil
2158 }
2159
2160 // return current uid, gid, groups in a format suitable for logging:
2161 // "crunch-run process has uid=1234(arvados) gid=1234(arvados)
2162 // groups=1234(arvados),114(fuse)"
2163 func currentUserAndGroups() string {
2164         u, err := user.Current()
2165         if err != nil {
2166                 return fmt.Sprintf("error getting current user ID: %s", err)
2167         }
2168         s := fmt.Sprintf("crunch-run process has uid=%s(%s) gid=%s", u.Uid, u.Username, u.Gid)
2169         if g, err := user.LookupGroupId(u.Gid); err == nil {
2170                 s += fmt.Sprintf("(%s)", g.Name)
2171         }
2172         s += " groups="
2173         if gids, err := u.GroupIds(); err == nil {
2174                 for i, gid := range gids {
2175                         if i > 0 {
2176                                 s += ","
2177                         }
2178                         s += gid
2179                         if g, err := user.LookupGroupId(gid); err == nil {
2180                                 s += fmt.Sprintf("(%s)", g.Name)
2181                         }
2182                 }
2183         }
2184         return s
2185 }
2186
2187 // Return a suitable local interface address for a local keepstore
2188 // service. Currently this is the numerically lowest non-loopback ipv4
2189 // address assigned to a local interface that is not in any of the
2190 // link-local/vpn/loopback ranges 169.254/16, 100.64/10, or 127/8.
2191 func localKeepstoreAddr() string {
2192         var ips []net.IP
2193         // Ignore error (proceed with zero IPs)
2194         addrs, _ := processIPs(os.Getpid())
2195         for addr := range addrs {
2196                 ip := net.ParseIP(addr)
2197                 if ip == nil {
2198                         // invalid
2199                         continue
2200                 }
2201                 if ip.Mask(net.CIDRMask(8, 32)).Equal(net.IPv4(127, 0, 0, 0)) ||
2202                         ip.Mask(net.CIDRMask(10, 32)).Equal(net.IPv4(100, 64, 0, 0)) ||
2203                         ip.Mask(net.CIDRMask(16, 32)).Equal(net.IPv4(169, 254, 0, 0)) {
2204                         // unsuitable
2205                         continue
2206                 }
2207                 ips = append(ips, ip)
2208         }
2209         if len(ips) == 0 {
2210                 return "0.0.0.0"
2211         }
2212         sort.Slice(ips, func(ii, jj int) bool {
2213                 i, j := ips[ii], ips[jj]
2214                 if len(i) != len(j) {
2215                         return len(i) < len(j)
2216                 }
2217                 for x := range i {
2218                         if i[x] != j[x] {
2219                                 return i[x] < j[x]
2220                         }
2221                 }
2222                 return false
2223         })
2224         return ips[0].String()
2225 }