1 // Copyright (C) The Arvados Authors. All rights reserved.
3 // SPDX-License-Identifier: AGPL-3.0
30 "git.arvados.org/arvados.git/lib/cmd"
31 "git.arvados.org/arvados.git/lib/crunchstat"
32 "git.arvados.org/arvados.git/sdk/go/arvados"
33 "git.arvados.org/arvados.git/sdk/go/arvadosclient"
34 "git.arvados.org/arvados.git/sdk/go/keepclient"
35 "git.arvados.org/arvados.git/sdk/go/manifest"
36 "golang.org/x/net/context"
41 var Command = command{}
43 // IArvadosClient is the minimal Arvados API methods used by crunch-run.
44 type IArvadosClient interface {
45 Create(resourceType string, parameters arvadosclient.Dict, output interface{}) error
46 Get(resourceType string, uuid string, parameters arvadosclient.Dict, output interface{}) error
47 Update(resourceType string, uuid string, parameters arvadosclient.Dict, output interface{}) error
48 Call(method, resourceType, uuid, action string, parameters arvadosclient.Dict, output interface{}) error
49 CallRaw(method string, resourceType string, uuid string, action string, parameters arvadosclient.Dict) (reader io.ReadCloser, err error)
50 Discovery(key string) (interface{}, error)
53 // ErrCancelled is the error returned when the container is cancelled.
54 var ErrCancelled = errors.New("Cancelled")
56 // IKeepClient is the minimal Keep API methods used by crunch-run.
57 type IKeepClient interface {
58 BlockWrite(context.Context, arvados.BlockWriteOptions) (arvados.BlockWriteResponse, error)
59 ReadAt(locator string, p []byte, off int) (int, error)
60 ManifestFileReader(m manifest.Manifest, filename string) (arvados.File, error)
61 LocalLocator(locator string) (string, error)
63 SetStorageClasses(sc []string)
66 // NewLogWriter is a factory function to create a new log writer.
67 type NewLogWriter func(name string) (io.WriteCloser, error)
69 type RunArvMount func(cmdline []string, tok string) (*exec.Cmd, error)
71 type MkTempDir func(string, string) (string, error)
73 type PsProcess interface {
74 CmdlineSlice() ([]string, error)
77 // ContainerRunner is the main stateful struct used for a single execution of a
79 type ContainerRunner struct {
80 executor containerExecutor
81 executorStdin io.Closer
82 executorStdout io.Closer
83 executorStderr io.Closer
85 // Dispatcher client is initialized with the Dispatcher token.
86 // This is a privileged token used to manage container status
89 // We have both dispatcherClient and DispatcherArvClient
90 // because there are two different incompatible Arvados Go
91 // SDKs and we have to use both (hopefully this gets fixed in
93 dispatcherClient *arvados.Client
94 DispatcherArvClient IArvadosClient
95 DispatcherKeepClient IKeepClient
97 // Container client is initialized with the Container token
98 // This token controls the permissions of the container, and
99 // must be used for operations such as reading collections.
101 // Same comment as above applies to
102 // containerClient/ContainerArvClient.
103 containerClient *arvados.Client
104 ContainerArvClient IArvadosClient
105 ContainerKeepClient IKeepClient
107 Container arvados.Container
110 NewLogWriter NewLogWriter
111 CrunchLog *ThrottledLogger
114 LogCollection arvados.CollectionFileSystem
116 RunArvMount RunArvMount
121 Volumes map[string]struct{}
123 SigChan chan os.Signal
124 ArvMountExit chan error
125 SecretMounts map[string]arvados.Mount
126 MkArvClient func(token string) (IArvadosClient, IKeepClient, *arvados.Client, error)
130 statLogger io.WriteCloser
131 statReporter *crunchstat.Reporter
132 hoststatLogger io.WriteCloser
133 hoststatReporter *crunchstat.Reporter
134 statInterval time.Duration
136 // What we expect the container's cgroup parent to be.
137 expectCgroupParent string
138 // What we tell docker to use as the container's cgroup
139 // parent. Note: Ideally we would use the same field for both
140 // expectCgroupParent and setCgroupParent, and just make it
141 // default to "docker". However, when using docker < 1.10 with
142 // systemd, specifying a non-empty cgroup parent (even the
143 // default value "docker") hits a docker bug
144 // (https://github.com/docker/docker/issues/17126). Using two
145 // separate fields makes it possible to use the "expect cgroup
146 // parent to be X" feature even on sites where the "specify
147 // cgroup parent" feature breaks.
148 setCgroupParent string
150 cStateLock sync.Mutex
151 cCancelled bool // StopContainer() invoked
153 enableMemoryLimit bool
154 enableNetwork string // one of "default" or "always"
155 networkMode string // "none", "host", or "" -- passed through to executor
156 arvMountLog *ThrottledLogger
158 containerWatchdogInterval time.Duration
163 // setupSignals sets up signal handling to gracefully terminate the
164 // underlying container and update state when receiving a TERM, INT or
166 func (runner *ContainerRunner) setupSignals() {
167 runner.SigChan = make(chan os.Signal, 1)
168 signal.Notify(runner.SigChan, syscall.SIGTERM)
169 signal.Notify(runner.SigChan, syscall.SIGINT)
170 signal.Notify(runner.SigChan, syscall.SIGQUIT)
172 go func(sig chan os.Signal) {
179 // stop the underlying container.
180 func (runner *ContainerRunner) stop(sig os.Signal) {
181 runner.cStateLock.Lock()
182 defer runner.cStateLock.Unlock()
184 runner.CrunchLog.Printf("caught signal: %v", sig)
186 runner.cCancelled = true
187 runner.CrunchLog.Printf("stopping container")
188 err := runner.executor.Stop()
190 runner.CrunchLog.Printf("error stopping container: %s", err)
194 var errorBlacklist = []string{
195 "(?ms).*[Cc]annot connect to the Docker daemon.*",
196 "(?ms).*oci runtime error.*starting container process.*container init.*mounting.*to rootfs.*no such file or directory.*",
197 "(?ms).*grpc: the connection is unavailable.*",
199 var brokenNodeHook *string = flag.String("broken-node-hook", "", "Script to run if node is detected to be broken (for example, Docker daemon is not running)")
201 func (runner *ContainerRunner) runBrokenNodeHook() {
202 if *brokenNodeHook == "" {
203 path := filepath.Join(lockdir, brokenfile)
204 runner.CrunchLog.Printf("Writing %s to mark node as broken", path)
205 f, err := os.OpenFile(path, os.O_CREATE|os.O_RDWR, 0700)
207 runner.CrunchLog.Printf("Error writing %s: %s", path, err)
212 runner.CrunchLog.Printf("Running broken node hook %q", *brokenNodeHook)
214 c := exec.Command(*brokenNodeHook)
215 c.Stdout = runner.CrunchLog
216 c.Stderr = runner.CrunchLog
219 runner.CrunchLog.Printf("Error running broken node hook: %v", err)
224 func (runner *ContainerRunner) checkBrokenNode(goterr error) bool {
225 for _, d := range errorBlacklist {
226 if m, e := regexp.MatchString(d, goterr.Error()); m && e == nil {
227 runner.CrunchLog.Printf("Error suggests node is unable to run containers: %v", goterr)
228 runner.runBrokenNodeHook()
235 // LoadImage determines the docker image id from the container record and
236 // checks if it is available in the local Docker image store. If not, it loads
237 // the image from Keep.
238 func (runner *ContainerRunner) LoadImage() (string, error) {
239 runner.CrunchLog.Printf("Fetching Docker image from collection '%s'", runner.Container.ContainerImage)
241 d, err := os.Open(runner.ArvMountPoint + "/by_id/" + runner.Container.ContainerImage)
246 allfiles, err := d.Readdirnames(-1)
250 var tarfiles []string
251 for _, fnm := range allfiles {
252 if strings.HasSuffix(fnm, ".tar") {
253 tarfiles = append(tarfiles, fnm)
256 if len(tarfiles) == 0 {
257 return "", fmt.Errorf("image collection does not include a .tar image file")
259 if len(tarfiles) > 1 {
260 return "", fmt.Errorf("cannot choose from multiple tar files in image collection: %v", tarfiles)
262 imageID := tarfiles[0][:len(tarfiles[0])-4]
263 imageTarballPath := runner.ArvMountPoint + "/by_id/" + runner.Container.ContainerImage + "/" + imageID + ".tar"
264 runner.CrunchLog.Printf("Using Docker image id %q", imageID)
266 runner.CrunchLog.Print("Loading Docker image from keep")
267 err = runner.executor.LoadImage(imageID, imageTarballPath, runner.Container, runner.ArvMountPoint,
268 runner.containerClient)
276 func (runner *ContainerRunner) ArvMountCmd(cmdline []string, token string) (c *exec.Cmd, err error) {
277 c = exec.Command(cmdline[0], cmdline[1:]...)
279 // Copy our environment, but override ARVADOS_API_TOKEN with
280 // the container auth token.
282 for _, s := range os.Environ() {
283 if !strings.HasPrefix(s, "ARVADOS_API_TOKEN=") {
284 c.Env = append(c.Env, s)
287 c.Env = append(c.Env, "ARVADOS_API_TOKEN="+token)
289 w, err := runner.NewLogWriter("arv-mount")
293 runner.arvMountLog = NewThrottledLogger(w)
294 scanner := logScanner{
297 "Block not found error",
298 "Unhandled exception during FUSE operation",
300 ReportFunc: runner.reportArvMountWarning,
302 c.Stdout = runner.arvMountLog
303 c.Stderr = io.MultiWriter(runner.arvMountLog, os.Stderr, &scanner)
305 runner.CrunchLog.Printf("Running %v", c.Args)
312 statReadme := make(chan bool)
313 runner.ArvMountExit = make(chan error)
318 time.Sleep(100 * time.Millisecond)
319 _, err = os.Stat(fmt.Sprintf("%s/by_id/README", runner.ArvMountPoint))
331 runner.CrunchLog.Printf("Arv-mount exit error: %v", mnterr)
333 runner.ArvMountExit <- mnterr
334 close(runner.ArvMountExit)
340 case err := <-runner.ArvMountExit:
341 runner.ArvMount = nil
349 func (runner *ContainerRunner) SetupArvMountPoint(prefix string) (err error) {
350 if runner.ArvMountPoint == "" {
351 runner.ArvMountPoint, err = runner.MkTempDir(runner.parentTemp, prefix)
356 func copyfile(src string, dst string) (err error) {
357 srcfile, err := os.Open(src)
362 os.MkdirAll(path.Dir(dst), 0777)
364 dstfile, err := os.Create(dst)
368 _, err = io.Copy(dstfile, srcfile)
373 err = srcfile.Close()
374 err2 := dstfile.Close()
387 func (runner *ContainerRunner) SetupMounts() (map[string]bindmount, error) {
388 bindmounts := map[string]bindmount{}
389 err := runner.SetupArvMountPoint("keep")
391 return nil, fmt.Errorf("While creating keep mount temp dir: %v", err)
394 token, err := runner.ContainerToken()
396 return nil, fmt.Errorf("could not get container token: %s", err)
398 runner.CrunchLog.Printf("container token %q", token)
402 arvMountCmd := []string{
406 "--storage-classes", strings.Join(runner.Container.OutputStorageClasses, ","),
407 fmt.Sprintf("--crunchstat-interval=%v", runner.statInterval.Seconds())}
409 if runner.executor.Runtime() == "docker" {
410 arvMountCmd = append(arvMountCmd, "--allow-other")
413 if runner.Container.RuntimeConstraints.KeepCacheRAM > 0 {
414 arvMountCmd = append(arvMountCmd, "--file-cache", fmt.Sprintf("%d", runner.Container.RuntimeConstraints.KeepCacheRAM))
417 collectionPaths := []string{}
418 needCertMount := true
419 type copyFile struct {
423 var copyFiles []copyFile
426 for bind := range runner.Container.Mounts {
427 binds = append(binds, bind)
429 for bind := range runner.SecretMounts {
430 if _, ok := runner.Container.Mounts[bind]; ok {
431 return nil, fmt.Errorf("secret mount %q conflicts with regular mount", bind)
433 if runner.SecretMounts[bind].Kind != "json" &&
434 runner.SecretMounts[bind].Kind != "text" {
435 return nil, fmt.Errorf("secret mount %q type is %q but only 'json' and 'text' are permitted",
436 bind, runner.SecretMounts[bind].Kind)
438 binds = append(binds, bind)
442 for _, bind := range binds {
443 mnt, notSecret := runner.Container.Mounts[bind]
445 mnt = runner.SecretMounts[bind]
447 if bind == "stdout" || bind == "stderr" {
448 // Is it a "file" mount kind?
449 if mnt.Kind != "file" {
450 return nil, fmt.Errorf("unsupported mount kind '%s' for %s: only 'file' is supported", mnt.Kind, bind)
453 // Does path start with OutputPath?
454 prefix := runner.Container.OutputPath
455 if !strings.HasSuffix(prefix, "/") {
458 if !strings.HasPrefix(mnt.Path, prefix) {
459 return nil, fmt.Errorf("%s path does not start with OutputPath: %s, %s", strings.Title(bind), mnt.Path, prefix)
464 // Is it a "collection" mount kind?
465 if mnt.Kind != "collection" && mnt.Kind != "json" {
466 return nil, fmt.Errorf("unsupported mount kind '%s' for stdin: only 'collection' and 'json' are supported", mnt.Kind)
470 if bind == "/etc/arvados/ca-certificates.crt" {
471 needCertMount = false
474 if strings.HasPrefix(bind, runner.Container.OutputPath+"/") && bind != runner.Container.OutputPath+"/" {
475 if mnt.Kind != "collection" && mnt.Kind != "text" && mnt.Kind != "json" {
476 return 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)
481 case mnt.Kind == "collection" && bind != "stdin":
483 if mnt.UUID != "" && mnt.PortableDataHash != "" {
484 return nil, fmt.Errorf("cannot specify both 'uuid' and 'portable_data_hash' for a collection mount")
488 return nil, fmt.Errorf("writing to existing collections currently not permitted")
491 src = fmt.Sprintf("%s/by_id/%s", runner.ArvMountPoint, mnt.UUID)
492 } else if mnt.PortableDataHash != "" {
493 if mnt.Writable && !strings.HasPrefix(bind, runner.Container.OutputPath+"/") {
494 return nil, fmt.Errorf("can never write to a collection specified by portable data hash")
496 idx := strings.Index(mnt.PortableDataHash, "/")
498 mnt.Path = path.Clean(mnt.PortableDataHash[idx:])
499 mnt.PortableDataHash = mnt.PortableDataHash[0:idx]
500 runner.Container.Mounts[bind] = mnt
502 src = fmt.Sprintf("%s/by_id/%s", runner.ArvMountPoint, mnt.PortableDataHash)
503 if mnt.Path != "" && mnt.Path != "." {
504 if strings.HasPrefix(mnt.Path, "./") {
505 mnt.Path = mnt.Path[2:]
506 } else if strings.HasPrefix(mnt.Path, "/") {
507 mnt.Path = mnt.Path[1:]
509 src += "/" + mnt.Path
512 src = fmt.Sprintf("%s/tmp%d", runner.ArvMountPoint, tmpcount)
513 arvMountCmd = append(arvMountCmd, "--mount-tmp", fmt.Sprintf("tmp%d", tmpcount))
517 if bind == runner.Container.OutputPath {
518 runner.HostOutputDir = src
519 bindmounts[bind] = bindmount{HostPath: src}
520 } else if strings.HasPrefix(bind, runner.Container.OutputPath+"/") {
521 copyFiles = append(copyFiles, copyFile{src, runner.HostOutputDir + bind[len(runner.Container.OutputPath):]})
523 bindmounts[bind] = bindmount{HostPath: src}
526 bindmounts[bind] = bindmount{HostPath: src, ReadOnly: true}
528 collectionPaths = append(collectionPaths, src)
530 case mnt.Kind == "tmp":
532 tmpdir, err = runner.MkTempDir(runner.parentTemp, "tmp")
534 return nil, fmt.Errorf("while creating mount temp dir: %v", err)
536 st, staterr := os.Stat(tmpdir)
538 return nil, fmt.Errorf("while Stat on temp dir: %v", staterr)
540 err = os.Chmod(tmpdir, st.Mode()|os.ModeSetgid|0777)
542 return nil, fmt.Errorf("while Chmod temp dir: %v", err)
544 bindmounts[bind] = bindmount{HostPath: tmpdir}
545 if bind == runner.Container.OutputPath {
546 runner.HostOutputDir = tmpdir
549 case mnt.Kind == "json" || mnt.Kind == "text":
551 if mnt.Kind == "json" {
552 filedata, err = json.Marshal(mnt.Content)
554 return nil, fmt.Errorf("encoding json data: %v", err)
557 text, ok := mnt.Content.(string)
559 return nil, fmt.Errorf("content for mount %q must be a string", bind)
561 filedata = []byte(text)
564 tmpdir, err := runner.MkTempDir(runner.parentTemp, mnt.Kind)
566 return nil, fmt.Errorf("creating temp dir: %v", err)
568 tmpfn := filepath.Join(tmpdir, "mountdata."+mnt.Kind)
569 err = ioutil.WriteFile(tmpfn, filedata, 0444)
571 return nil, fmt.Errorf("writing temp file: %v", err)
573 if strings.HasPrefix(bind, runner.Container.OutputPath+"/") && (notSecret || runner.Container.Mounts[runner.Container.OutputPath].Kind != "collection") {
574 // In most cases, if the container
575 // specifies a literal file inside the
576 // output path, we copy it into the
577 // output directory (either a mounted
578 // collection or a staging area on the
579 // host fs). If it's a secret, it will
580 // be skipped when copying output from
581 // staging to Keep later.
582 copyFiles = append(copyFiles, copyFile{tmpfn, runner.HostOutputDir + bind[len(runner.Container.OutputPath):]})
584 // If a secret is outside OutputPath,
585 // we bind mount the secret file
586 // directly just like other mounts. We
587 // also use this strategy when a
588 // secret is inside OutputPath but
589 // OutputPath is a live collection, to
590 // avoid writing the secret to
591 // Keep. Attempting to remove a
592 // bind-mounted secret file from
593 // inside the container will return a
594 // "Device or resource busy" error
595 // that might not be handled well by
596 // the container, which is why we
597 // don't use this strategy when
598 // OutputPath is a staging directory.
599 bindmounts[bind] = bindmount{HostPath: tmpfn, ReadOnly: true}
602 case mnt.Kind == "git_tree":
603 tmpdir, err := runner.MkTempDir(runner.parentTemp, "git_tree")
605 return nil, fmt.Errorf("creating temp dir: %v", err)
607 err = gitMount(mnt).extractTree(runner.ContainerArvClient, tmpdir, token)
611 bindmounts[bind] = bindmount{HostPath: tmpdir, ReadOnly: true}
615 if runner.HostOutputDir == "" {
616 return nil, fmt.Errorf("output path does not correspond to a writable mount point")
619 if needCertMount && runner.Container.RuntimeConstraints.API {
620 for _, certfile := range arvadosclient.CertFiles {
621 _, err := os.Stat(certfile)
623 bindmounts["/etc/arvados/ca-certificates.crt"] = bindmount{HostPath: certfile, ReadOnly: true}
630 // If we are only mounting collections by pdh, make
631 // sure we don't subscribe to websocket events to
632 // avoid putting undesired load on the API server
633 arvMountCmd = append(arvMountCmd, "--mount-by-pdh", "by_id", "--disable-event-listening")
635 arvMountCmd = append(arvMountCmd, "--mount-by-id", "by_id")
637 // the by_uuid mount point is used by singularity when writing
638 // out docker images converted to SIF
639 arvMountCmd = append(arvMountCmd, "--mount-by-id", "by_uuid")
640 arvMountCmd = append(arvMountCmd, runner.ArvMountPoint)
642 runner.ArvMount, err = runner.RunArvMount(arvMountCmd, token)
644 return nil, fmt.Errorf("while trying to start arv-mount: %v", err)
647 for _, p := range collectionPaths {
650 return nil, fmt.Errorf("while checking that input files exist: %v", err)
654 for _, cp := range copyFiles {
655 st, err := os.Stat(cp.src)
657 return nil, fmt.Errorf("while staging writable file from %q to %q: %v", cp.src, cp.bind, err)
660 err = filepath.Walk(cp.src, func(walkpath string, walkinfo os.FileInfo, walkerr error) error {
664 target := path.Join(cp.bind, walkpath[len(cp.src):])
665 if walkinfo.Mode().IsRegular() {
666 copyerr := copyfile(walkpath, target)
670 return os.Chmod(target, walkinfo.Mode()|0777)
671 } else if walkinfo.Mode().IsDir() {
672 mkerr := os.MkdirAll(target, 0777)
676 return os.Chmod(target, walkinfo.Mode()|os.ModeSetgid|0777)
678 return fmt.Errorf("source %q is not a regular file or directory", cp.src)
681 } else if st.Mode().IsRegular() {
682 err = copyfile(cp.src, cp.bind)
684 err = os.Chmod(cp.bind, st.Mode()|0777)
688 return nil, fmt.Errorf("while staging writable file from %q to %q: %v", cp.src, cp.bind, err)
692 return bindmounts, nil
695 func (runner *ContainerRunner) stopHoststat() error {
696 if runner.hoststatReporter == nil {
699 runner.hoststatReporter.Stop()
700 err := runner.hoststatLogger.Close()
702 return fmt.Errorf("error closing hoststat logs: %v", err)
707 func (runner *ContainerRunner) startHoststat() error {
708 w, err := runner.NewLogWriter("hoststat")
712 runner.hoststatLogger = NewThrottledLogger(w)
713 runner.hoststatReporter = &crunchstat.Reporter{
714 Logger: log.New(runner.hoststatLogger, "", 0),
715 CgroupRoot: runner.cgroupRoot,
716 PollPeriod: runner.statInterval,
718 runner.hoststatReporter.Start()
722 func (runner *ContainerRunner) startCrunchstat() error {
723 w, err := runner.NewLogWriter("crunchstat")
727 runner.statLogger = NewThrottledLogger(w)
728 runner.statReporter = &crunchstat.Reporter{
729 CID: runner.executor.CgroupID(),
730 Logger: log.New(runner.statLogger, "", 0),
731 CgroupParent: runner.expectCgroupParent,
732 CgroupRoot: runner.cgroupRoot,
733 PollPeriod: runner.statInterval,
734 TempDir: runner.parentTemp,
736 runner.statReporter.Start()
740 type infoCommand struct {
745 // LogHostInfo logs info about the current host, for debugging and
746 // accounting purposes. Although it's logged as "node-info", this is
747 // about the environment where crunch-run is actually running, which
748 // might differ from what's described in the node record (see
750 func (runner *ContainerRunner) LogHostInfo() (err error) {
751 w, err := runner.NewLogWriter("node-info")
756 commands := []infoCommand{
758 label: "Host Information",
759 cmd: []string{"uname", "-a"},
762 label: "CPU Information",
763 cmd: []string{"cat", "/proc/cpuinfo"},
766 label: "Memory Information",
767 cmd: []string{"cat", "/proc/meminfo"},
771 cmd: []string{"df", "-m", "/", os.TempDir()},
774 label: "Disk INodes",
775 cmd: []string{"df", "-i", "/", os.TempDir()},
779 // Run commands with informational output to be logged.
780 for _, command := range commands {
781 fmt.Fprintln(w, command.label)
782 cmd := exec.Command(command.cmd[0], command.cmd[1:]...)
785 if err := cmd.Run(); err != nil {
786 err = fmt.Errorf("While running command %q: %v", command.cmd, err)
795 return fmt.Errorf("While closing node-info logs: %v", err)
800 // LogContainerRecord gets and saves the raw JSON container record from the API server
801 func (runner *ContainerRunner) LogContainerRecord() error {
802 logged, err := runner.logAPIResponse("container", "containers", map[string]interface{}{"filters": [][]string{{"uuid", "=", runner.Container.UUID}}}, nil)
803 if !logged && err == nil {
804 err = fmt.Errorf("error: no container record found for %s", runner.Container.UUID)
809 // LogNodeRecord logs the current host's InstanceType config entry (or
810 // the arvados#node record, if running via crunch-dispatch-slurm).
811 func (runner *ContainerRunner) LogNodeRecord() error {
812 if it := os.Getenv("InstanceType"); it != "" {
813 // Dispatched via arvados-dispatch-cloud. Save
814 // InstanceType config fragment received from
815 // dispatcher on stdin.
816 w, err := runner.LogCollection.OpenFile("node.json", os.O_CREATE|os.O_WRONLY, 0666)
821 _, err = io.WriteString(w, it)
827 // Dispatched via crunch-dispatch-slurm. Look up
828 // apiserver's node record corresponding to
830 hostname := os.Getenv("SLURMD_NODENAME")
832 hostname, _ = os.Hostname()
834 _, err := runner.logAPIResponse("node", "nodes", map[string]interface{}{"filters": [][]string{{"hostname", "=", hostname}}}, func(resp interface{}) {
835 // The "info" field has admin-only info when
836 // obtained with a privileged token, and
837 // should not be logged.
838 node, ok := resp.(map[string]interface{})
846 func (runner *ContainerRunner) logAPIResponse(label, path string, params map[string]interface{}, munge func(interface{})) (logged bool, err error) {
847 writer, err := runner.LogCollection.OpenFile(label+".json", os.O_CREATE|os.O_WRONLY, 0666)
852 ArvClient: runner.DispatcherArvClient,
853 UUID: runner.Container.UUID,
854 loggingStream: label,
858 reader, err := runner.DispatcherArvClient.CallRaw("GET", path, "", "", arvadosclient.Dict(params))
860 return false, fmt.Errorf("error getting %s record: %v", label, err)
864 dec := json.NewDecoder(reader)
866 var resp map[string]interface{}
867 if err = dec.Decode(&resp); err != nil {
868 return false, fmt.Errorf("error decoding %s list response: %v", label, err)
870 items, ok := resp["items"].([]interface{})
872 return false, fmt.Errorf("error decoding %s list response: no \"items\" key in API list response", label)
873 } else if len(items) < 1 {
879 // Re-encode it using indentation to improve readability
880 enc := json.NewEncoder(w)
881 enc.SetIndent("", " ")
882 if err = enc.Encode(items[0]); err != nil {
883 return false, fmt.Errorf("error logging %s record: %v", label, err)
887 return false, fmt.Errorf("error closing %s.json in log collection: %v", label, err)
892 func (runner *ContainerRunner) getStdoutFile(mntPath string) (*os.File, error) {
893 stdoutPath := mntPath[len(runner.Container.OutputPath):]
894 index := strings.LastIndex(stdoutPath, "/")
896 subdirs := stdoutPath[:index]
898 st, err := os.Stat(runner.HostOutputDir)
900 return nil, fmt.Errorf("While Stat on temp dir: %v", err)
902 stdoutPath := filepath.Join(runner.HostOutputDir, subdirs)
903 err = os.MkdirAll(stdoutPath, st.Mode()|os.ModeSetgid|0777)
905 return nil, fmt.Errorf("While MkdirAll %q: %v", stdoutPath, err)
909 stdoutFile, err := os.Create(filepath.Join(runner.HostOutputDir, stdoutPath))
911 return nil, fmt.Errorf("While creating file %q: %v", stdoutPath, err)
914 return stdoutFile, nil
917 // CreateContainer creates the docker container.
918 func (runner *ContainerRunner) CreateContainer(imageID string, bindmounts map[string]bindmount) error {
919 var stdin io.ReadCloser = ioutil.NopCloser(bytes.NewReader(nil))
920 if mnt, ok := runner.Container.Mounts["stdin"]; ok {
927 collID = mnt.PortableDataHash
929 path := runner.ArvMountPoint + "/by_id/" + collID + "/" + mnt.Path
930 f, err := os.Open(path)
936 j, err := json.Marshal(mnt.Content)
938 return fmt.Errorf("error encoding stdin json data: %v", err)
940 stdin = ioutil.NopCloser(bytes.NewReader(j))
942 return fmt.Errorf("stdin mount has unsupported kind %q", mnt.Kind)
946 var stdout, stderr io.WriteCloser
947 if mnt, ok := runner.Container.Mounts["stdout"]; ok {
948 f, err := runner.getStdoutFile(mnt.Path)
953 } else if w, err := runner.NewLogWriter("stdout"); err != nil {
956 stdout = NewThrottledLogger(w)
959 if mnt, ok := runner.Container.Mounts["stderr"]; ok {
960 f, err := runner.getStdoutFile(mnt.Path)
965 } else if w, err := runner.NewLogWriter("stderr"); err != nil {
968 stderr = NewThrottledLogger(w)
971 env := runner.Container.Environment
972 enableNetwork := runner.enableNetwork == "always"
973 if runner.Container.RuntimeConstraints.API {
975 tok, err := runner.ContainerToken()
979 env = map[string]string{}
980 for k, v := range runner.Container.Environment {
983 env["ARVADOS_API_TOKEN"] = tok
984 env["ARVADOS_API_HOST"] = os.Getenv("ARVADOS_API_HOST")
985 env["ARVADOS_API_HOST_INSECURE"] = os.Getenv("ARVADOS_API_HOST_INSECURE")
987 workdir := runner.Container.Cwd
989 // both "" and "." mean default
992 ram := runner.Container.RuntimeConstraints.RAM
993 if !runner.enableMemoryLimit {
996 runner.executorStdin = stdin
997 runner.executorStdout = stdout
998 runner.executorStderr = stderr
999 return runner.executor.Create(containerSpec{
1001 VCPUs: runner.Container.RuntimeConstraints.VCPUs,
1003 WorkingDir: workdir,
1005 BindMounts: bindmounts,
1006 Command: runner.Container.Command,
1007 EnableNetwork: enableNetwork,
1008 NetworkMode: runner.networkMode,
1009 CgroupParent: runner.setCgroupParent,
1016 // StartContainer starts the docker container created by CreateContainer.
1017 func (runner *ContainerRunner) StartContainer() error {
1018 runner.CrunchLog.Printf("Starting container")
1019 runner.cStateLock.Lock()
1020 defer runner.cStateLock.Unlock()
1021 if runner.cCancelled {
1024 err := runner.executor.Start()
1027 if m, e := regexp.MatchString("(?ms).*(exec|System error).*(no such file or directory|file not found).*", err.Error()); m && e == nil {
1028 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])
1030 return fmt.Errorf("could not start container: %v%s", err, advice)
1035 // WaitFinish waits for the container to terminate, capture the exit code, and
1036 // close the stdout/stderr logging.
1037 func (runner *ContainerRunner) WaitFinish() error {
1038 runner.CrunchLog.Print("Waiting for container to finish")
1039 var timeout <-chan time.Time
1040 if s := runner.Container.SchedulingParameters.MaxRunTime; s > 0 {
1041 timeout = time.After(time.Duration(s) * time.Second)
1043 ctx, cancel := context.WithCancel(context.Background())
1048 runner.CrunchLog.Printf("maximum run time exceeded. Stopping container.")
1050 case <-runner.ArvMountExit:
1051 runner.CrunchLog.Printf("arv-mount exited while container is still running. Stopping container.")
1056 exitcode, err := runner.executor.Wait(ctx)
1058 runner.checkBrokenNode(err)
1061 runner.ExitCode = &exitcode
1064 if err = runner.executorStdin.Close(); err != nil {
1065 err = fmt.Errorf("error closing container stdin: %s", err)
1066 runner.CrunchLog.Printf("%s", err)
1069 if err = runner.executorStdout.Close(); err != nil {
1070 err = fmt.Errorf("error closing container stdout: %s", err)
1071 runner.CrunchLog.Printf("%s", err)
1072 if returnErr == nil {
1076 if err = runner.executorStderr.Close(); err != nil {
1077 err = fmt.Errorf("error closing container stderr: %s", err)
1078 runner.CrunchLog.Printf("%s", err)
1079 if returnErr == nil {
1084 if runner.statReporter != nil {
1085 runner.statReporter.Stop()
1086 err = runner.statLogger.Close()
1088 runner.CrunchLog.Printf("error closing crunchstat logs: %v", err)
1094 func (runner *ContainerRunner) updateLogs() {
1095 ticker := time.NewTicker(crunchLogUpdatePeriod / 360)
1098 sigusr1 := make(chan os.Signal, 1)
1099 signal.Notify(sigusr1, syscall.SIGUSR1)
1100 defer signal.Stop(sigusr1)
1102 saveAtTime := time.Now().Add(crunchLogUpdatePeriod)
1103 saveAtSize := crunchLogUpdateSize
1109 saveAtTime = time.Now()
1111 runner.logMtx.Lock()
1112 done := runner.LogsPDH != nil
1113 runner.logMtx.Unlock()
1117 size := runner.LogCollection.Size()
1118 if size == savedSize || (time.Now().Before(saveAtTime) && size < saveAtSize) {
1121 saveAtTime = time.Now().Add(crunchLogUpdatePeriod)
1122 saveAtSize = runner.LogCollection.Size() + crunchLogUpdateSize
1123 saved, err := runner.saveLogCollection(false)
1125 runner.CrunchLog.Printf("error updating log collection: %s", err)
1129 var updated arvados.Container
1130 err = runner.DispatcherArvClient.Update("containers", runner.Container.UUID, arvadosclient.Dict{
1131 "container": arvadosclient.Dict{"log": saved.PortableDataHash},
1134 runner.CrunchLog.Printf("error updating container log to %s: %s", saved.PortableDataHash, err)
1142 func (runner *ContainerRunner) reportArvMountWarning(pattern, text string) {
1143 var updated arvados.Container
1144 err := runner.DispatcherArvClient.Update("containers", runner.Container.UUID, arvadosclient.Dict{
1145 "container": arvadosclient.Dict{
1146 "runtime_status": arvadosclient.Dict{
1147 "warning": "arv-mount: " + pattern,
1148 "warningDetail": text,
1153 runner.CrunchLog.Printf("error updating container runtime_status: %s", err)
1157 // CaptureOutput saves data from the container's output directory if
1158 // needed, and updates the container output accordingly.
1159 func (runner *ContainerRunner) CaptureOutput(bindmounts map[string]bindmount) error {
1160 if runner.Container.RuntimeConstraints.API {
1161 // Output may have been set directly by the container, so
1162 // refresh the container record to check.
1163 err := runner.DispatcherArvClient.Get("containers", runner.Container.UUID,
1164 nil, &runner.Container)
1168 if runner.Container.Output != "" {
1169 // Container output is already set.
1170 runner.OutputPDH = &runner.Container.Output
1175 txt, err := (&copier{
1176 client: runner.containerClient,
1177 arvClient: runner.ContainerArvClient,
1178 keepClient: runner.ContainerKeepClient,
1179 hostOutputDir: runner.HostOutputDir,
1180 ctrOutputDir: runner.Container.OutputPath,
1181 bindmounts: bindmounts,
1182 mounts: runner.Container.Mounts,
1183 secretMounts: runner.SecretMounts,
1184 logger: runner.CrunchLog,
1189 if n := len(regexp.MustCompile(` [0-9a-f]+\+\S*\+R`).FindAllStringIndex(txt, -1)); n > 0 {
1190 runner.CrunchLog.Printf("Copying %d data blocks from remote input collections...", n)
1191 fs, err := (&arvados.Collection{ManifestText: txt}).FileSystem(runner.containerClient, runner.ContainerKeepClient)
1195 txt, err = fs.MarshalManifest(".")
1200 var resp arvados.Collection
1201 err = runner.ContainerArvClient.Create("collections", arvadosclient.Dict{
1202 "ensure_unique_name": true,
1203 "collection": arvadosclient.Dict{
1205 "name": "output for " + runner.Container.UUID,
1206 "manifest_text": txt,
1210 return fmt.Errorf("error creating output collection: %v", err)
1212 runner.OutputPDH = &resp.PortableDataHash
1216 func (runner *ContainerRunner) CleanupDirs() {
1217 if runner.ArvMount != nil {
1219 umount := exec.Command("arv-mount", fmt.Sprintf("--unmount-timeout=%d", delay), "--unmount", runner.ArvMountPoint)
1220 umount.Stdout = runner.CrunchLog
1221 umount.Stderr = runner.CrunchLog
1222 runner.CrunchLog.Printf("Running %v", umount.Args)
1223 umnterr := umount.Start()
1226 runner.CrunchLog.Printf("Error unmounting: %v", umnterr)
1227 runner.ArvMount.Process.Kill()
1229 // If arv-mount --unmount gets stuck for any reason, we
1230 // don't want to wait for it forever. Do Wait() in a goroutine
1231 // so it doesn't block crunch-run.
1232 umountExit := make(chan error)
1234 mnterr := umount.Wait()
1236 runner.CrunchLog.Printf("Error unmounting: %v", mnterr)
1238 umountExit <- mnterr
1241 for again := true; again; {
1247 case <-runner.ArvMountExit:
1249 case <-time.After(time.Duration((delay + 1) * int64(time.Second))):
1250 runner.CrunchLog.Printf("Timed out waiting for unmount")
1252 umount.Process.Kill()
1254 runner.ArvMount.Process.Kill()
1258 runner.ArvMount = nil
1261 if runner.ArvMountPoint != "" {
1262 if rmerr := os.Remove(runner.ArvMountPoint); rmerr != nil {
1263 runner.CrunchLog.Printf("While cleaning up arv-mount directory %s: %v", runner.ArvMountPoint, rmerr)
1265 runner.ArvMountPoint = ""
1268 if rmerr := os.RemoveAll(runner.parentTemp); rmerr != nil {
1269 runner.CrunchLog.Printf("While cleaning up temporary directory %s: %v", runner.parentTemp, rmerr)
1273 // CommitLogs posts the collection containing the final container logs.
1274 func (runner *ContainerRunner) CommitLogs() error {
1276 // Hold cStateLock to prevent races on CrunchLog (e.g., stop()).
1277 runner.cStateLock.Lock()
1278 defer runner.cStateLock.Unlock()
1280 runner.CrunchLog.Print(runner.finalState)
1282 if runner.arvMountLog != nil {
1283 runner.arvMountLog.Close()
1285 runner.CrunchLog.Close()
1287 // Closing CrunchLog above allows them to be committed to Keep at this
1288 // point, but re-open crunch log with ArvClient in case there are any
1289 // other further errors (such as failing to write the log to Keep!)
1290 // while shutting down
1291 runner.CrunchLog = NewThrottledLogger(&ArvLogWriter{
1292 ArvClient: runner.DispatcherArvClient,
1293 UUID: runner.Container.UUID,
1294 loggingStream: "crunch-run",
1297 runner.CrunchLog.Immediate = log.New(os.Stderr, runner.Container.UUID+" ", 0)
1300 if runner.LogsPDH != nil {
1301 // If we have already assigned something to LogsPDH,
1302 // we must be closing the re-opened log, which won't
1303 // end up getting attached to the container record and
1304 // therefore doesn't need to be saved as a collection
1305 // -- it exists only to send logs to other channels.
1308 saved, err := runner.saveLogCollection(true)
1310 return fmt.Errorf("error saving log collection: %s", err)
1312 runner.logMtx.Lock()
1313 defer runner.logMtx.Unlock()
1314 runner.LogsPDH = &saved.PortableDataHash
1318 func (runner *ContainerRunner) saveLogCollection(final bool) (response arvados.Collection, err error) {
1319 runner.logMtx.Lock()
1320 defer runner.logMtx.Unlock()
1321 if runner.LogsPDH != nil {
1322 // Already finalized.
1325 updates := arvadosclient.Dict{
1326 "name": "logs for " + runner.Container.UUID,
1328 mt, err1 := runner.LogCollection.MarshalManifest(".")
1330 // Only send updated manifest text if there was no
1332 updates["manifest_text"] = mt
1335 // Even if flushing the manifest had an error, we still want
1336 // to update the log record, if possible, to push the trash_at
1337 // and delete_at times into the future. Details on bug
1340 updates["is_trashed"] = true
1342 exp := time.Now().Add(crunchLogUpdatePeriod * 24)
1343 updates["trash_at"] = exp
1344 updates["delete_at"] = exp
1346 reqBody := arvadosclient.Dict{"collection": updates}
1348 if runner.logUUID == "" {
1349 reqBody["ensure_unique_name"] = true
1350 err2 = runner.DispatcherArvClient.Create("collections", reqBody, &response)
1352 err2 = runner.DispatcherArvClient.Update("collections", runner.logUUID, reqBody, &response)
1355 runner.logUUID = response.UUID
1358 if err1 != nil || err2 != nil {
1359 err = fmt.Errorf("error recording logs: %q, %q", err1, err2)
1364 // UpdateContainerRunning updates the container state to "Running"
1365 func (runner *ContainerRunner) UpdateContainerRunning() error {
1366 runner.cStateLock.Lock()
1367 defer runner.cStateLock.Unlock()
1368 if runner.cCancelled {
1371 return runner.DispatcherArvClient.Update("containers", runner.Container.UUID,
1372 arvadosclient.Dict{"container": arvadosclient.Dict{"state": "Running", "gateway_address": runner.gateway.Address}}, nil)
1375 // ContainerToken returns the api_token the container (and any
1376 // arv-mount processes) are allowed to use.
1377 func (runner *ContainerRunner) ContainerToken() (string, error) {
1378 if runner.token != "" {
1379 return runner.token, nil
1382 var auth arvados.APIClientAuthorization
1383 err := runner.DispatcherArvClient.Call("GET", "containers", runner.Container.UUID, "auth", nil, &auth)
1387 runner.token = fmt.Sprintf("v2/%s/%s/%s", auth.UUID, auth.APIToken, runner.Container.UUID)
1388 return runner.token, nil
1391 // UpdateContainerFinal updates the container record state on API
1392 // server to "Complete" or "Cancelled"
1393 func (runner *ContainerRunner) UpdateContainerFinal() error {
1394 update := arvadosclient.Dict{}
1395 update["state"] = runner.finalState
1396 if runner.LogsPDH != nil {
1397 update["log"] = *runner.LogsPDH
1399 if runner.finalState == "Complete" {
1400 if runner.ExitCode != nil {
1401 update["exit_code"] = *runner.ExitCode
1403 if runner.OutputPDH != nil {
1404 update["output"] = *runner.OutputPDH
1407 return runner.DispatcherArvClient.Update("containers", runner.Container.UUID, arvadosclient.Dict{"container": update}, nil)
1410 // IsCancelled returns the value of Cancelled, with goroutine safety.
1411 func (runner *ContainerRunner) IsCancelled() bool {
1412 runner.cStateLock.Lock()
1413 defer runner.cStateLock.Unlock()
1414 return runner.cCancelled
1417 // NewArvLogWriter creates an ArvLogWriter
1418 func (runner *ContainerRunner) NewArvLogWriter(name string) (io.WriteCloser, error) {
1419 writer, err := runner.LogCollection.OpenFile(name+".txt", os.O_CREATE|os.O_WRONLY, 0666)
1423 return &ArvLogWriter{
1424 ArvClient: runner.DispatcherArvClient,
1425 UUID: runner.Container.UUID,
1426 loggingStream: name,
1427 writeCloser: writer,
1431 // Run the full container lifecycle.
1432 func (runner *ContainerRunner) Run() (err error) {
1433 runner.CrunchLog.Printf("crunch-run %s started", cmd.Version.String())
1434 runner.CrunchLog.Printf("Executing container '%s' using %s runtime", runner.Container.UUID, runner.executor.Runtime())
1436 hostname, hosterr := os.Hostname()
1438 runner.CrunchLog.Printf("Error getting hostname '%v'", hosterr)
1440 runner.CrunchLog.Printf("Executing on host '%s'", hostname)
1443 runner.finalState = "Queued"
1446 runner.CleanupDirs()
1448 runner.CrunchLog.Printf("crunch-run finished")
1449 runner.CrunchLog.Close()
1452 err = runner.fetchContainerRecord()
1456 if runner.Container.State != "Locked" {
1457 return fmt.Errorf("dispatch error detected: container %q has state %q", runner.Container.UUID, runner.Container.State)
1460 var bindmounts map[string]bindmount
1462 // checkErr prints e (unless it's nil) and sets err to
1463 // e (unless err is already non-nil). Thus, if err
1464 // hasn't already been assigned when Run() returns,
1465 // this cleanup func will cause Run() to return the
1466 // first non-nil error that is passed to checkErr().
1467 checkErr := func(errorIn string, e error) {
1471 runner.CrunchLog.Printf("error in %s: %v", errorIn, e)
1475 if runner.finalState == "Complete" {
1476 // There was an error in the finalization.
1477 runner.finalState = "Cancelled"
1481 // Log the error encountered in Run(), if any
1482 checkErr("Run", err)
1484 if runner.finalState == "Queued" {
1485 runner.UpdateContainerFinal()
1489 if runner.IsCancelled() {
1490 runner.finalState = "Cancelled"
1491 // but don't return yet -- we still want to
1492 // capture partial output and write logs
1495 if bindmounts != nil {
1496 checkErr("CaptureOutput", runner.CaptureOutput(bindmounts))
1498 checkErr("stopHoststat", runner.stopHoststat())
1499 checkErr("CommitLogs", runner.CommitLogs())
1500 runner.CleanupDirs()
1501 checkErr("UpdateContainerFinal", runner.UpdateContainerFinal())
1504 runner.setupSignals()
1505 err = runner.startHoststat()
1510 // set up FUSE mount and binds
1511 bindmounts, err = runner.SetupMounts()
1513 runner.finalState = "Cancelled"
1514 err = fmt.Errorf("While setting up mounts: %v", err)
1518 // check for and/or load image
1519 imageID, err := runner.LoadImage()
1521 if !runner.checkBrokenNode(err) {
1522 // Failed to load image but not due to a "broken node"
1523 // condition, probably user error.
1524 runner.finalState = "Cancelled"
1526 err = fmt.Errorf("While loading container image: %v", err)
1530 err = runner.CreateContainer(imageID, bindmounts)
1534 err = runner.LogHostInfo()
1538 err = runner.LogNodeRecord()
1542 err = runner.LogContainerRecord()
1547 if runner.IsCancelled() {
1551 err = runner.UpdateContainerRunning()
1555 runner.finalState = "Cancelled"
1557 err = runner.startCrunchstat()
1562 err = runner.StartContainer()
1564 runner.checkBrokenNode(err)
1568 err = runner.WaitFinish()
1569 if err == nil && !runner.IsCancelled() {
1570 runner.finalState = "Complete"
1575 // Fetch the current container record (uuid = runner.Container.UUID)
1576 // into runner.Container.
1577 func (runner *ContainerRunner) fetchContainerRecord() error {
1578 reader, err := runner.DispatcherArvClient.CallRaw("GET", "containers", runner.Container.UUID, "", nil)
1580 return fmt.Errorf("error fetching container record: %v", err)
1582 defer reader.Close()
1584 dec := json.NewDecoder(reader)
1586 err = dec.Decode(&runner.Container)
1588 return fmt.Errorf("error decoding container record: %v", err)
1592 SecretMounts map[string]arvados.Mount `json:"secret_mounts"`
1595 containerToken, err := runner.ContainerToken()
1597 return fmt.Errorf("error getting container token: %v", err)
1600 runner.ContainerArvClient, runner.ContainerKeepClient,
1601 runner.containerClient, err = runner.MkArvClient(containerToken)
1603 return fmt.Errorf("error creating container API client: %v", err)
1606 runner.ContainerKeepClient.SetStorageClasses(runner.Container.OutputStorageClasses)
1607 runner.DispatcherKeepClient.SetStorageClasses(runner.Container.OutputStorageClasses)
1609 err = runner.ContainerArvClient.Call("GET", "containers", runner.Container.UUID, "secret_mounts", nil, &sm)
1611 if apierr, ok := err.(arvadosclient.APIServerError); !ok || apierr.HttpStatusCode != 404 {
1612 return fmt.Errorf("error fetching secret_mounts: %v", err)
1614 // ok && apierr.HttpStatusCode == 404, which means
1615 // secret_mounts isn't supported by this API server.
1617 runner.SecretMounts = sm.SecretMounts
1622 // NewContainerRunner creates a new container runner.
1623 func NewContainerRunner(dispatcherClient *arvados.Client,
1624 dispatcherArvClient IArvadosClient,
1625 dispatcherKeepClient IKeepClient,
1626 containerUUID string) (*ContainerRunner, error) {
1628 cr := &ContainerRunner{
1629 dispatcherClient: dispatcherClient,
1630 DispatcherArvClient: dispatcherArvClient,
1631 DispatcherKeepClient: dispatcherKeepClient,
1633 cr.NewLogWriter = cr.NewArvLogWriter
1634 cr.RunArvMount = cr.ArvMountCmd
1635 cr.MkTempDir = ioutil.TempDir
1636 cr.MkArvClient = func(token string) (IArvadosClient, IKeepClient, *arvados.Client, error) {
1637 cl, err := arvadosclient.MakeArvadosClient()
1639 return nil, nil, nil, err
1642 kc, err := keepclient.MakeKeepClient(cl)
1644 return nil, nil, nil, err
1646 c2 := arvados.NewClientFromEnv()
1647 c2.AuthToken = token
1648 return cl, kc, c2, nil
1651 cr.LogCollection, err = (&arvados.Collection{}).FileSystem(cr.dispatcherClient, cr.DispatcherKeepClient)
1655 cr.Container.UUID = containerUUID
1656 w, err := cr.NewLogWriter("crunch-run")
1660 cr.CrunchLog = NewThrottledLogger(w)
1661 cr.CrunchLog.Immediate = log.New(os.Stderr, containerUUID+" ", 0)
1663 loadLogThrottleParams(dispatcherArvClient)
1669 func (command) RunCommand(prog string, args []string, stdin io.Reader, stdout, stderr io.Writer) int {
1670 flags := flag.NewFlagSet(prog, flag.ContinueOnError)
1671 statInterval := flags.Duration("crunchstat-interval", 10*time.Second, "sampling period for periodic resource usage reporting")
1672 cgroupRoot := flags.String("cgroup-root", "/sys/fs/cgroup", "path to sysfs cgroup tree")
1673 cgroupParent := flags.String("cgroup-parent", "docker", "name of container's parent cgroup (ignored if -cgroup-parent-subsystem is used)")
1674 cgroupParentSubsystem := flags.String("cgroup-parent-subsystem", "", "use current cgroup for given subsystem as parent cgroup for container")
1675 caCertsPath := flags.String("ca-certs", "", "Path to TLS root certificates")
1676 detach := flags.Bool("detach", false, "Detach from parent process and run in the background")
1677 stdinEnv := flags.Bool("stdin-env", false, "Load environment variables from JSON message on stdin")
1678 sleep := flags.Duration("sleep", 0, "Delay before starting (testing use only)")
1679 kill := flags.Int("kill", -1, "Send signal to an existing crunch-run process for given UUID")
1680 list := flags.Bool("list", false, "List UUIDs of existing crunch-run processes")
1681 enableMemoryLimit := flags.Bool("enable-memory-limit", true, "tell container runtime to limit container's memory usage")
1682 enableNetwork := flags.String("container-enable-networking", "default", "enable networking \"always\" (for all containers) or \"default\" (for containers that request it)")
1683 networkMode := flags.String("container-network-mode", "default", `Docker network mode for container (use any argument valid for docker --net)`)
1684 memprofile := flags.String("memprofile", "", "write memory profile to `file` after running container")
1685 runtimeEngine := flags.String("runtime-engine", "docker", "container runtime: docker or singularity")
1686 flags.Duration("check-containerd", 0, "Ignored. Exists for compatibility with older versions.")
1688 ignoreDetachFlag := false
1689 if len(args) > 0 && args[0] == "-no-detach" {
1690 // This process was invoked by a parent process, which
1691 // has passed along its own arguments, including
1692 // -detach, after the leading -no-detach flag. Strip
1693 // the leading -no-detach flag (it's not recognized by
1694 // flags.Parse()) and ignore the -detach flag that
1697 ignoreDetachFlag = true
1700 if err := flags.Parse(args); err == flag.ErrHelp {
1702 } else if err != nil {
1707 if *stdinEnv && !ignoreDetachFlag {
1708 // Load env vars on stdin if asked (but not in a
1709 // detached child process, in which case stdin is
1711 err := loadEnv(os.Stdin)
1718 containerUUID := flags.Arg(0)
1721 case *detach && !ignoreDetachFlag:
1722 return Detach(containerUUID, prog, args, os.Stdout, os.Stderr)
1724 return KillProcess(containerUUID, syscall.Signal(*kill), os.Stdout, os.Stderr)
1726 return ListProcesses(os.Stdout, os.Stderr)
1729 if containerUUID == "" {
1730 log.Printf("usage: %s [options] UUID", prog)
1734 log.Printf("crunch-run %s started", cmd.Version.String())
1737 if *caCertsPath != "" {
1738 arvadosclient.CertFiles = []string{*caCertsPath}
1741 api, err := arvadosclient.MakeArvadosClient()
1743 log.Printf("%s: %v", containerUUID, err)
1748 kc, kcerr := keepclient.MakeKeepClient(api)
1750 log.Printf("%s: %v", containerUUID, kcerr)
1753 kc.BlockCache = &keepclient.BlockCache{MaxBlocks: 2}
1756 cr, err := NewContainerRunner(arvados.NewClientFromEnv(), api, kc, containerUUID)
1762 switch *runtimeEngine {
1764 cr.executor, err = newDockerExecutor(containerUUID, cr.CrunchLog.Printf, cr.containerWatchdogInterval)
1766 cr.executor, err = newSingularityExecutor(cr.CrunchLog.Printf)
1768 cr.CrunchLog.Printf("%s: unsupported RuntimeEngine %q", containerUUID, *runtimeEngine)
1769 cr.CrunchLog.Close()
1773 cr.CrunchLog.Printf("%s: %v", containerUUID, err)
1774 cr.checkBrokenNode(err)
1775 cr.CrunchLog.Close()
1778 defer cr.executor.Close()
1780 gwAuthSecret := os.Getenv("GatewayAuthSecret")
1781 os.Unsetenv("GatewayAuthSecret")
1782 if gwAuthSecret == "" {
1783 // not safe to run a gateway service without an auth
1785 cr.CrunchLog.Printf("Not starting a gateway server (GatewayAuthSecret was not provided by dispatcher)")
1786 } else if gwListen := os.Getenv("GatewayAddress"); gwListen == "" {
1787 // dispatcher did not tell us which external IP
1788 // address to advertise --> no gateway service
1789 cr.CrunchLog.Printf("Not starting a gateway server (GatewayAddress was not provided by dispatcher)")
1790 } else if de, ok := cr.executor.(*dockerExecutor); ok {
1791 cr.gateway = Gateway{
1793 AuthSecret: gwAuthSecret,
1794 ContainerUUID: containerUUID,
1795 DockerContainerID: &de.containerID,
1797 ContainerIPAddress: dockerContainerIPAddress(&de.containerID),
1799 err = cr.gateway.Start()
1801 log.Printf("error starting gateway server: %s", err)
1806 parentTemp, tmperr := cr.MkTempDir("", "crunch-run."+containerUUID+".")
1808 log.Printf("%s: %v", containerUUID, tmperr)
1812 cr.parentTemp = parentTemp
1813 cr.statInterval = *statInterval
1814 cr.cgroupRoot = *cgroupRoot
1815 cr.expectCgroupParent = *cgroupParent
1816 cr.enableMemoryLimit = *enableMemoryLimit
1817 cr.enableNetwork = *enableNetwork
1818 cr.networkMode = *networkMode
1819 if *cgroupParentSubsystem != "" {
1820 p := findCgroup(*cgroupParentSubsystem)
1821 cr.setCgroupParent = p
1822 cr.expectCgroupParent = p
1827 if *memprofile != "" {
1828 f, err := os.Create(*memprofile)
1830 log.Printf("could not create memory profile: %s", err)
1832 runtime.GC() // get up-to-date statistics
1833 if err := pprof.WriteHeapProfile(f); err != nil {
1834 log.Printf("could not write memory profile: %s", err)
1836 closeerr := f.Close()
1837 if closeerr != nil {
1838 log.Printf("closing memprofile file: %s", err)
1843 log.Printf("%s: %v", containerUUID, runerr)
1849 func loadEnv(rdr io.Reader) error {
1850 buf, err := ioutil.ReadAll(rdr)
1852 return fmt.Errorf("read stdin: %s", err)
1854 var env map[string]string
1855 err = json.Unmarshal(buf, &env)
1857 return fmt.Errorf("decode stdin: %s", err)
1859 for k, v := range env {
1860 err = os.Setenv(k, v)
1862 return fmt.Errorf("setenv(%q): %s", k, err)