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, ok := 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")
514 arvMountCmd = append(arvMountCmd, fmt.Sprintf("tmp%d", tmpcount))
518 if bind == runner.Container.OutputPath {
519 runner.HostOutputDir = src
520 bindmounts[bind] = bindmount{HostPath: src}
521 } else if strings.HasPrefix(bind, runner.Container.OutputPath+"/") {
522 copyFiles = append(copyFiles, copyFile{src, runner.HostOutputDir + bind[len(runner.Container.OutputPath):]})
524 bindmounts[bind] = bindmount{HostPath: src}
527 bindmounts[bind] = bindmount{HostPath: src, ReadOnly: true}
529 collectionPaths = append(collectionPaths, src)
531 case mnt.Kind == "tmp":
533 tmpdir, err = runner.MkTempDir(runner.parentTemp, "tmp")
535 return nil, fmt.Errorf("while creating mount temp dir: %v", err)
537 st, staterr := os.Stat(tmpdir)
539 return nil, fmt.Errorf("while Stat on temp dir: %v", staterr)
541 err = os.Chmod(tmpdir, st.Mode()|os.ModeSetgid|0777)
543 return nil, fmt.Errorf("while Chmod temp dir: %v", err)
545 bindmounts[bind] = bindmount{HostPath: tmpdir}
546 if bind == runner.Container.OutputPath {
547 runner.HostOutputDir = tmpdir
550 case mnt.Kind == "json" || mnt.Kind == "text":
552 if mnt.Kind == "json" {
553 filedata, err = json.Marshal(mnt.Content)
555 return nil, fmt.Errorf("encoding json data: %v", err)
558 text, ok := mnt.Content.(string)
560 return nil, fmt.Errorf("content for mount %q must be a string", bind)
562 filedata = []byte(text)
565 tmpdir, err := runner.MkTempDir(runner.parentTemp, mnt.Kind)
567 return nil, fmt.Errorf("creating temp dir: %v", err)
569 tmpfn := filepath.Join(tmpdir, "mountdata."+mnt.Kind)
570 err = ioutil.WriteFile(tmpfn, filedata, 0444)
572 return nil, fmt.Errorf("writing temp file: %v", err)
574 if strings.HasPrefix(bind, runner.Container.OutputPath+"/") {
575 copyFiles = append(copyFiles, copyFile{tmpfn, runner.HostOutputDir + bind[len(runner.Container.OutputPath):]})
577 bindmounts[bind] = bindmount{HostPath: tmpfn, ReadOnly: true}
580 case mnt.Kind == "git_tree":
581 tmpdir, err := runner.MkTempDir(runner.parentTemp, "git_tree")
583 return nil, fmt.Errorf("creating temp dir: %v", err)
585 err = gitMount(mnt).extractTree(runner.ContainerArvClient, tmpdir, token)
589 bindmounts[bind] = bindmount{HostPath: tmpdir, ReadOnly: true}
593 if runner.HostOutputDir == "" {
594 return nil, fmt.Errorf("output path does not correspond to a writable mount point")
597 if needCertMount && runner.Container.RuntimeConstraints.API {
598 for _, certfile := range arvadosclient.CertFiles {
599 _, err := os.Stat(certfile)
601 bindmounts["/etc/arvados/ca-certificates.crt"] = bindmount{HostPath: certfile, ReadOnly: true}
608 // If we are only mounting collections by pdh, make
609 // sure we don't subscribe to websocket events to
610 // avoid putting undesired load on the API server
611 arvMountCmd = append(arvMountCmd, "--mount-by-pdh", "by_id", "--disable-event-listening")
613 arvMountCmd = append(arvMountCmd, "--mount-by-id", "by_id")
615 // the by_uuid mount point is used by singularity when writing
616 // out docker images converted to SIF
617 arvMountCmd = append(arvMountCmd, "--mount-by-id", "by_uuid")
618 arvMountCmd = append(arvMountCmd, runner.ArvMountPoint)
620 runner.ArvMount, err = runner.RunArvMount(arvMountCmd, token)
622 return nil, fmt.Errorf("while trying to start arv-mount: %v", err)
625 for _, p := range collectionPaths {
628 return nil, fmt.Errorf("while checking that input files exist: %v", err)
632 for _, cp := range copyFiles {
633 st, err := os.Stat(cp.src)
635 return nil, fmt.Errorf("while staging writable file from %q to %q: %v", cp.src, cp.bind, err)
638 err = filepath.Walk(cp.src, func(walkpath string, walkinfo os.FileInfo, walkerr error) error {
642 target := path.Join(cp.bind, walkpath[len(cp.src):])
643 if walkinfo.Mode().IsRegular() {
644 copyerr := copyfile(walkpath, target)
648 return os.Chmod(target, walkinfo.Mode()|0777)
649 } else if walkinfo.Mode().IsDir() {
650 mkerr := os.MkdirAll(target, 0777)
654 return os.Chmod(target, walkinfo.Mode()|os.ModeSetgid|0777)
656 return fmt.Errorf("source %q is not a regular file or directory", cp.src)
659 } else if st.Mode().IsRegular() {
660 err = copyfile(cp.src, cp.bind)
662 err = os.Chmod(cp.bind, st.Mode()|0777)
666 return nil, fmt.Errorf("while staging writable file from %q to %q: %v", cp.src, cp.bind, err)
670 return bindmounts, nil
673 func (runner *ContainerRunner) stopHoststat() error {
674 if runner.hoststatReporter == nil {
677 runner.hoststatReporter.Stop()
678 err := runner.hoststatLogger.Close()
680 return fmt.Errorf("error closing hoststat logs: %v", err)
685 func (runner *ContainerRunner) startHoststat() error {
686 w, err := runner.NewLogWriter("hoststat")
690 runner.hoststatLogger = NewThrottledLogger(w)
691 runner.hoststatReporter = &crunchstat.Reporter{
692 Logger: log.New(runner.hoststatLogger, "", 0),
693 CgroupRoot: runner.cgroupRoot,
694 PollPeriod: runner.statInterval,
696 runner.hoststatReporter.Start()
700 func (runner *ContainerRunner) startCrunchstat() error {
701 w, err := runner.NewLogWriter("crunchstat")
705 runner.statLogger = NewThrottledLogger(w)
706 runner.statReporter = &crunchstat.Reporter{
707 CID: runner.executor.CgroupID(),
708 Logger: log.New(runner.statLogger, "", 0),
709 CgroupParent: runner.expectCgroupParent,
710 CgroupRoot: runner.cgroupRoot,
711 PollPeriod: runner.statInterval,
712 TempDir: runner.parentTemp,
714 runner.statReporter.Start()
718 type infoCommand struct {
723 // LogHostInfo logs info about the current host, for debugging and
724 // accounting purposes. Although it's logged as "node-info", this is
725 // about the environment where crunch-run is actually running, which
726 // might differ from what's described in the node record (see
728 func (runner *ContainerRunner) LogHostInfo() (err error) {
729 w, err := runner.NewLogWriter("node-info")
734 commands := []infoCommand{
736 label: "Host Information",
737 cmd: []string{"uname", "-a"},
740 label: "CPU Information",
741 cmd: []string{"cat", "/proc/cpuinfo"},
744 label: "Memory Information",
745 cmd: []string{"cat", "/proc/meminfo"},
749 cmd: []string{"df", "-m", "/", os.TempDir()},
752 label: "Disk INodes",
753 cmd: []string{"df", "-i", "/", os.TempDir()},
757 // Run commands with informational output to be logged.
758 for _, command := range commands {
759 fmt.Fprintln(w, command.label)
760 cmd := exec.Command(command.cmd[0], command.cmd[1:]...)
763 if err := cmd.Run(); err != nil {
764 err = fmt.Errorf("While running command %q: %v", command.cmd, err)
773 return fmt.Errorf("While closing node-info logs: %v", err)
778 // LogContainerRecord gets and saves the raw JSON container record from the API server
779 func (runner *ContainerRunner) LogContainerRecord() error {
780 logged, err := runner.logAPIResponse("container", "containers", map[string]interface{}{"filters": [][]string{{"uuid", "=", runner.Container.UUID}}}, nil)
781 if !logged && err == nil {
782 err = fmt.Errorf("error: no container record found for %s", runner.Container.UUID)
787 // LogNodeRecord logs the current host's InstanceType config entry (or
788 // the arvados#node record, if running via crunch-dispatch-slurm).
789 func (runner *ContainerRunner) LogNodeRecord() error {
790 if it := os.Getenv("InstanceType"); it != "" {
791 // Dispatched via arvados-dispatch-cloud. Save
792 // InstanceType config fragment received from
793 // dispatcher on stdin.
794 w, err := runner.LogCollection.OpenFile("node.json", os.O_CREATE|os.O_WRONLY, 0666)
799 _, err = io.WriteString(w, it)
805 // Dispatched via crunch-dispatch-slurm. Look up
806 // apiserver's node record corresponding to
808 hostname := os.Getenv("SLURMD_NODENAME")
810 hostname, _ = os.Hostname()
812 _, err := runner.logAPIResponse("node", "nodes", map[string]interface{}{"filters": [][]string{{"hostname", "=", hostname}}}, func(resp interface{}) {
813 // The "info" field has admin-only info when
814 // obtained with a privileged token, and
815 // should not be logged.
816 node, ok := resp.(map[string]interface{})
824 func (runner *ContainerRunner) logAPIResponse(label, path string, params map[string]interface{}, munge func(interface{})) (logged bool, err error) {
825 writer, err := runner.LogCollection.OpenFile(label+".json", os.O_CREATE|os.O_WRONLY, 0666)
830 ArvClient: runner.DispatcherArvClient,
831 UUID: runner.Container.UUID,
832 loggingStream: label,
836 reader, err := runner.DispatcherArvClient.CallRaw("GET", path, "", "", arvadosclient.Dict(params))
838 return false, fmt.Errorf("error getting %s record: %v", label, err)
842 dec := json.NewDecoder(reader)
844 var resp map[string]interface{}
845 if err = dec.Decode(&resp); err != nil {
846 return false, fmt.Errorf("error decoding %s list response: %v", label, err)
848 items, ok := resp["items"].([]interface{})
850 return false, fmt.Errorf("error decoding %s list response: no \"items\" key in API list response", label)
851 } else if len(items) < 1 {
857 // Re-encode it using indentation to improve readability
858 enc := json.NewEncoder(w)
859 enc.SetIndent("", " ")
860 if err = enc.Encode(items[0]); err != nil {
861 return false, fmt.Errorf("error logging %s record: %v", label, err)
865 return false, fmt.Errorf("error closing %s.json in log collection: %v", label, err)
870 func (runner *ContainerRunner) getStdoutFile(mntPath string) (*os.File, error) {
871 stdoutPath := mntPath[len(runner.Container.OutputPath):]
872 index := strings.LastIndex(stdoutPath, "/")
874 subdirs := stdoutPath[:index]
876 st, err := os.Stat(runner.HostOutputDir)
878 return nil, fmt.Errorf("While Stat on temp dir: %v", err)
880 stdoutPath := filepath.Join(runner.HostOutputDir, subdirs)
881 err = os.MkdirAll(stdoutPath, st.Mode()|os.ModeSetgid|0777)
883 return nil, fmt.Errorf("While MkdirAll %q: %v", stdoutPath, err)
887 stdoutFile, err := os.Create(filepath.Join(runner.HostOutputDir, stdoutPath))
889 return nil, fmt.Errorf("While creating file %q: %v", stdoutPath, err)
892 return stdoutFile, nil
895 // CreateContainer creates the docker container.
896 func (runner *ContainerRunner) CreateContainer(imageID string, bindmounts map[string]bindmount) error {
897 var stdin io.ReadCloser = ioutil.NopCloser(bytes.NewReader(nil))
898 if mnt, ok := runner.Container.Mounts["stdin"]; ok {
905 collID = mnt.PortableDataHash
907 path := runner.ArvMountPoint + "/by_id/" + collID + "/" + mnt.Path
908 f, err := os.Open(path)
914 j, err := json.Marshal(mnt.Content)
916 return fmt.Errorf("error encoding stdin json data: %v", err)
918 stdin = ioutil.NopCloser(bytes.NewReader(j))
920 return fmt.Errorf("stdin mount has unsupported kind %q", mnt.Kind)
924 var stdout, stderr io.WriteCloser
925 if mnt, ok := runner.Container.Mounts["stdout"]; ok {
926 f, err := runner.getStdoutFile(mnt.Path)
931 } else if w, err := runner.NewLogWriter("stdout"); err != nil {
934 stdout = NewThrottledLogger(w)
937 if mnt, ok := runner.Container.Mounts["stderr"]; ok {
938 f, err := runner.getStdoutFile(mnt.Path)
943 } else if w, err := runner.NewLogWriter("stderr"); err != nil {
946 stderr = NewThrottledLogger(w)
949 env := runner.Container.Environment
950 enableNetwork := runner.enableNetwork == "always"
951 if runner.Container.RuntimeConstraints.API {
953 tok, err := runner.ContainerToken()
957 env = map[string]string{}
958 for k, v := range runner.Container.Environment {
961 env["ARVADOS_API_TOKEN"] = tok
962 env["ARVADOS_API_HOST"] = os.Getenv("ARVADOS_API_HOST")
963 env["ARVADOS_API_HOST_INSECURE"] = os.Getenv("ARVADOS_API_HOST_INSECURE")
965 workdir := runner.Container.Cwd
967 // both "" and "." mean default
970 ram := runner.Container.RuntimeConstraints.RAM
971 if !runner.enableMemoryLimit {
974 runner.executorStdin = stdin
975 runner.executorStdout = stdout
976 runner.executorStderr = stderr
977 return runner.executor.Create(containerSpec{
979 VCPUs: runner.Container.RuntimeConstraints.VCPUs,
983 BindMounts: bindmounts,
984 Command: runner.Container.Command,
985 EnableNetwork: enableNetwork,
986 NetworkMode: runner.networkMode,
987 CgroupParent: runner.setCgroupParent,
994 // StartContainer starts the docker container created by CreateContainer.
995 func (runner *ContainerRunner) StartContainer() error {
996 runner.CrunchLog.Printf("Starting container")
997 runner.cStateLock.Lock()
998 defer runner.cStateLock.Unlock()
999 if runner.cCancelled {
1002 err := runner.executor.Start()
1005 if m, e := regexp.MatchString("(?ms).*(exec|System error).*(no such file or directory|file not found).*", err.Error()); m && e == nil {
1006 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])
1008 return fmt.Errorf("could not start container: %v%s", err, advice)
1013 // WaitFinish waits for the container to terminate, capture the exit code, and
1014 // close the stdout/stderr logging.
1015 func (runner *ContainerRunner) WaitFinish() error {
1016 runner.CrunchLog.Print("Waiting for container to finish")
1017 var timeout <-chan time.Time
1018 if s := runner.Container.SchedulingParameters.MaxRunTime; s > 0 {
1019 timeout = time.After(time.Duration(s) * time.Second)
1021 ctx, cancel := context.WithCancel(context.Background())
1026 runner.CrunchLog.Printf("maximum run time exceeded. Stopping container.")
1028 case <-runner.ArvMountExit:
1029 runner.CrunchLog.Printf("arv-mount exited while container is still running. Stopping container.")
1034 exitcode, err := runner.executor.Wait(ctx)
1036 runner.checkBrokenNode(err)
1039 runner.ExitCode = &exitcode
1042 if err = runner.executorStdin.Close(); err != nil {
1043 err = fmt.Errorf("error closing container stdin: %s", err)
1044 runner.CrunchLog.Printf("%s", err)
1047 if err = runner.executorStdout.Close(); err != nil {
1048 err = fmt.Errorf("error closing container stdout: %s", err)
1049 runner.CrunchLog.Printf("%s", err)
1050 if returnErr == nil {
1054 if err = runner.executorStderr.Close(); err != nil {
1055 err = fmt.Errorf("error closing container stderr: %s", err)
1056 runner.CrunchLog.Printf("%s", err)
1057 if returnErr == nil {
1062 if runner.statReporter != nil {
1063 runner.statReporter.Stop()
1064 err = runner.statLogger.Close()
1066 runner.CrunchLog.Printf("error closing crunchstat logs: %v", err)
1072 func (runner *ContainerRunner) updateLogs() {
1073 ticker := time.NewTicker(crunchLogUpdatePeriod / 360)
1076 sigusr1 := make(chan os.Signal, 1)
1077 signal.Notify(sigusr1, syscall.SIGUSR1)
1078 defer signal.Stop(sigusr1)
1080 saveAtTime := time.Now().Add(crunchLogUpdatePeriod)
1081 saveAtSize := crunchLogUpdateSize
1087 saveAtTime = time.Now()
1089 runner.logMtx.Lock()
1090 done := runner.LogsPDH != nil
1091 runner.logMtx.Unlock()
1095 size := runner.LogCollection.Size()
1096 if size == savedSize || (time.Now().Before(saveAtTime) && size < saveAtSize) {
1099 saveAtTime = time.Now().Add(crunchLogUpdatePeriod)
1100 saveAtSize = runner.LogCollection.Size() + crunchLogUpdateSize
1101 saved, err := runner.saveLogCollection(false)
1103 runner.CrunchLog.Printf("error updating log collection: %s", err)
1107 var updated arvados.Container
1108 err = runner.DispatcherArvClient.Update("containers", runner.Container.UUID, arvadosclient.Dict{
1109 "container": arvadosclient.Dict{"log": saved.PortableDataHash},
1112 runner.CrunchLog.Printf("error updating container log to %s: %s", saved.PortableDataHash, err)
1120 func (runner *ContainerRunner) reportArvMountWarning(pattern, text string) {
1121 var updated arvados.Container
1122 err := runner.DispatcherArvClient.Update("containers", runner.Container.UUID, arvadosclient.Dict{
1123 "container": arvadosclient.Dict{
1124 "runtime_status": arvadosclient.Dict{
1125 "warning": "arv-mount: " + pattern,
1126 "warningDetail": text,
1131 runner.CrunchLog.Printf("error updating container runtime_status: %s", err)
1135 // CaptureOutput saves data from the container's output directory if
1136 // needed, and updates the container output accordingly.
1137 func (runner *ContainerRunner) CaptureOutput(bindmounts map[string]bindmount) error {
1138 if runner.Container.RuntimeConstraints.API {
1139 // Output may have been set directly by the container, so
1140 // refresh the container record to check.
1141 err := runner.DispatcherArvClient.Get("containers", runner.Container.UUID,
1142 nil, &runner.Container)
1146 if runner.Container.Output != "" {
1147 // Container output is already set.
1148 runner.OutputPDH = &runner.Container.Output
1153 txt, err := (&copier{
1154 client: runner.containerClient,
1155 arvClient: runner.ContainerArvClient,
1156 keepClient: runner.ContainerKeepClient,
1157 hostOutputDir: runner.HostOutputDir,
1158 ctrOutputDir: runner.Container.OutputPath,
1159 bindmounts: bindmounts,
1160 mounts: runner.Container.Mounts,
1161 secretMounts: runner.SecretMounts,
1162 logger: runner.CrunchLog,
1167 if n := len(regexp.MustCompile(` [0-9a-f]+\+\S*\+R`).FindAllStringIndex(txt, -1)); n > 0 {
1168 runner.CrunchLog.Printf("Copying %d data blocks from remote input collections...", n)
1169 fs, err := (&arvados.Collection{ManifestText: txt}).FileSystem(runner.containerClient, runner.ContainerKeepClient)
1173 txt, err = fs.MarshalManifest(".")
1178 var resp arvados.Collection
1179 err = runner.ContainerArvClient.Create("collections", arvadosclient.Dict{
1180 "ensure_unique_name": true,
1181 "collection": arvadosclient.Dict{
1183 "name": "output for " + runner.Container.UUID,
1184 "manifest_text": txt,
1188 return fmt.Errorf("error creating output collection: %v", err)
1190 runner.OutputPDH = &resp.PortableDataHash
1194 func (runner *ContainerRunner) CleanupDirs() {
1195 if runner.ArvMount != nil {
1197 umount := exec.Command("arv-mount", fmt.Sprintf("--unmount-timeout=%d", delay), "--unmount", runner.ArvMountPoint)
1198 umount.Stdout = runner.CrunchLog
1199 umount.Stderr = runner.CrunchLog
1200 runner.CrunchLog.Printf("Running %v", umount.Args)
1201 umnterr := umount.Start()
1204 runner.CrunchLog.Printf("Error unmounting: %v", umnterr)
1205 runner.ArvMount.Process.Kill()
1207 // If arv-mount --unmount gets stuck for any reason, we
1208 // don't want to wait for it forever. Do Wait() in a goroutine
1209 // so it doesn't block crunch-run.
1210 umountExit := make(chan error)
1212 mnterr := umount.Wait()
1214 runner.CrunchLog.Printf("Error unmounting: %v", mnterr)
1216 umountExit <- mnterr
1219 for again := true; again; {
1225 case <-runner.ArvMountExit:
1227 case <-time.After(time.Duration((delay + 1) * int64(time.Second))):
1228 runner.CrunchLog.Printf("Timed out waiting for unmount")
1230 umount.Process.Kill()
1232 runner.ArvMount.Process.Kill()
1236 runner.ArvMount = nil
1239 if runner.ArvMountPoint != "" {
1240 if rmerr := os.Remove(runner.ArvMountPoint); rmerr != nil {
1241 runner.CrunchLog.Printf("While cleaning up arv-mount directory %s: %v", runner.ArvMountPoint, rmerr)
1243 runner.ArvMountPoint = ""
1246 if rmerr := os.RemoveAll(runner.parentTemp); rmerr != nil {
1247 runner.CrunchLog.Printf("While cleaning up temporary directory %s: %v", runner.parentTemp, rmerr)
1251 // CommitLogs posts the collection containing the final container logs.
1252 func (runner *ContainerRunner) CommitLogs() error {
1254 // Hold cStateLock to prevent races on CrunchLog (e.g., stop()).
1255 runner.cStateLock.Lock()
1256 defer runner.cStateLock.Unlock()
1258 runner.CrunchLog.Print(runner.finalState)
1260 if runner.arvMountLog != nil {
1261 runner.arvMountLog.Close()
1263 runner.CrunchLog.Close()
1265 // Closing CrunchLog above allows them to be committed to Keep at this
1266 // point, but re-open crunch log with ArvClient in case there are any
1267 // other further errors (such as failing to write the log to Keep!)
1268 // while shutting down
1269 runner.CrunchLog = NewThrottledLogger(&ArvLogWriter{
1270 ArvClient: runner.DispatcherArvClient,
1271 UUID: runner.Container.UUID,
1272 loggingStream: "crunch-run",
1275 runner.CrunchLog.Immediate = log.New(os.Stderr, runner.Container.UUID+" ", 0)
1278 if runner.LogsPDH != nil {
1279 // If we have already assigned something to LogsPDH,
1280 // we must be closing the re-opened log, which won't
1281 // end up getting attached to the container record and
1282 // therefore doesn't need to be saved as a collection
1283 // -- it exists only to send logs to other channels.
1286 saved, err := runner.saveLogCollection(true)
1288 return fmt.Errorf("error saving log collection: %s", err)
1290 runner.logMtx.Lock()
1291 defer runner.logMtx.Unlock()
1292 runner.LogsPDH = &saved.PortableDataHash
1296 func (runner *ContainerRunner) saveLogCollection(final bool) (response arvados.Collection, err error) {
1297 runner.logMtx.Lock()
1298 defer runner.logMtx.Unlock()
1299 if runner.LogsPDH != nil {
1300 // Already finalized.
1303 updates := arvadosclient.Dict{
1304 "name": "logs for " + runner.Container.UUID,
1306 mt, err1 := runner.LogCollection.MarshalManifest(".")
1308 // Only send updated manifest text if there was no
1310 updates["manifest_text"] = mt
1313 // Even if flushing the manifest had an error, we still want
1314 // to update the log record, if possible, to push the trash_at
1315 // and delete_at times into the future. Details on bug
1318 updates["is_trashed"] = true
1320 exp := time.Now().Add(crunchLogUpdatePeriod * 24)
1321 updates["trash_at"] = exp
1322 updates["delete_at"] = exp
1324 reqBody := arvadosclient.Dict{"collection": updates}
1326 if runner.logUUID == "" {
1327 reqBody["ensure_unique_name"] = true
1328 err2 = runner.DispatcherArvClient.Create("collections", reqBody, &response)
1330 err2 = runner.DispatcherArvClient.Update("collections", runner.logUUID, reqBody, &response)
1333 runner.logUUID = response.UUID
1336 if err1 != nil || err2 != nil {
1337 err = fmt.Errorf("error recording logs: %q, %q", err1, err2)
1342 // UpdateContainerRunning updates the container state to "Running"
1343 func (runner *ContainerRunner) UpdateContainerRunning() error {
1344 runner.cStateLock.Lock()
1345 defer runner.cStateLock.Unlock()
1346 if runner.cCancelled {
1349 return runner.DispatcherArvClient.Update("containers", runner.Container.UUID,
1350 arvadosclient.Dict{"container": arvadosclient.Dict{"state": "Running", "gateway_address": runner.gateway.Address}}, nil)
1353 // ContainerToken returns the api_token the container (and any
1354 // arv-mount processes) are allowed to use.
1355 func (runner *ContainerRunner) ContainerToken() (string, error) {
1356 if runner.token != "" {
1357 return runner.token, nil
1360 var auth arvados.APIClientAuthorization
1361 err := runner.DispatcherArvClient.Call("GET", "containers", runner.Container.UUID, "auth", nil, &auth)
1365 runner.token = fmt.Sprintf("v2/%s/%s/%s", auth.UUID, auth.APIToken, runner.Container.UUID)
1366 return runner.token, nil
1369 // UpdateContainerFinal updates the container record state on API
1370 // server to "Complete" or "Cancelled"
1371 func (runner *ContainerRunner) UpdateContainerFinal() error {
1372 update := arvadosclient.Dict{}
1373 update["state"] = runner.finalState
1374 if runner.LogsPDH != nil {
1375 update["log"] = *runner.LogsPDH
1377 if runner.finalState == "Complete" {
1378 if runner.ExitCode != nil {
1379 update["exit_code"] = *runner.ExitCode
1381 if runner.OutputPDH != nil {
1382 update["output"] = *runner.OutputPDH
1385 return runner.DispatcherArvClient.Update("containers", runner.Container.UUID, arvadosclient.Dict{"container": update}, nil)
1388 // IsCancelled returns the value of Cancelled, with goroutine safety.
1389 func (runner *ContainerRunner) IsCancelled() bool {
1390 runner.cStateLock.Lock()
1391 defer runner.cStateLock.Unlock()
1392 return runner.cCancelled
1395 // NewArvLogWriter creates an ArvLogWriter
1396 func (runner *ContainerRunner) NewArvLogWriter(name string) (io.WriteCloser, error) {
1397 writer, err := runner.LogCollection.OpenFile(name+".txt", os.O_CREATE|os.O_WRONLY, 0666)
1401 return &ArvLogWriter{
1402 ArvClient: runner.DispatcherArvClient,
1403 UUID: runner.Container.UUID,
1404 loggingStream: name,
1405 writeCloser: writer,
1409 // Run the full container lifecycle.
1410 func (runner *ContainerRunner) Run() (err error) {
1411 runner.CrunchLog.Printf("crunch-run %s started", cmd.Version.String())
1412 runner.CrunchLog.Printf("Executing container '%s' using %s runtime", runner.Container.UUID, runner.executor.Runtime())
1414 hostname, hosterr := os.Hostname()
1416 runner.CrunchLog.Printf("Error getting hostname '%v'", hosterr)
1418 runner.CrunchLog.Printf("Executing on host '%s'", hostname)
1421 runner.finalState = "Queued"
1424 runner.CleanupDirs()
1426 runner.CrunchLog.Printf("crunch-run finished")
1427 runner.CrunchLog.Close()
1430 err = runner.fetchContainerRecord()
1434 if runner.Container.State != "Locked" {
1435 return fmt.Errorf("dispatch error detected: container %q has state %q", runner.Container.UUID, runner.Container.State)
1438 var bindmounts map[string]bindmount
1440 // checkErr prints e (unless it's nil) and sets err to
1441 // e (unless err is already non-nil). Thus, if err
1442 // hasn't already been assigned when Run() returns,
1443 // this cleanup func will cause Run() to return the
1444 // first non-nil error that is passed to checkErr().
1445 checkErr := func(errorIn string, e error) {
1449 runner.CrunchLog.Printf("error in %s: %v", errorIn, e)
1453 if runner.finalState == "Complete" {
1454 // There was an error in the finalization.
1455 runner.finalState = "Cancelled"
1459 // Log the error encountered in Run(), if any
1460 checkErr("Run", err)
1462 if runner.finalState == "Queued" {
1463 runner.UpdateContainerFinal()
1467 if runner.IsCancelled() {
1468 runner.finalState = "Cancelled"
1469 // but don't return yet -- we still want to
1470 // capture partial output and write logs
1473 if bindmounts != nil {
1474 checkErr("CaptureOutput", runner.CaptureOutput(bindmounts))
1476 checkErr("stopHoststat", runner.stopHoststat())
1477 checkErr("CommitLogs", runner.CommitLogs())
1478 runner.CleanupDirs()
1479 checkErr("UpdateContainerFinal", runner.UpdateContainerFinal())
1482 runner.setupSignals()
1483 err = runner.startHoststat()
1488 // set up FUSE mount and binds
1489 bindmounts, err = runner.SetupMounts()
1491 runner.finalState = "Cancelled"
1492 err = fmt.Errorf("While setting up mounts: %v", err)
1496 // check for and/or load image
1497 imageID, err := runner.LoadImage()
1499 if !runner.checkBrokenNode(err) {
1500 // Failed to load image but not due to a "broken node"
1501 // condition, probably user error.
1502 runner.finalState = "Cancelled"
1504 err = fmt.Errorf("While loading container image: %v", err)
1508 err = runner.CreateContainer(imageID, bindmounts)
1512 err = runner.LogHostInfo()
1516 err = runner.LogNodeRecord()
1520 err = runner.LogContainerRecord()
1525 if runner.IsCancelled() {
1529 err = runner.UpdateContainerRunning()
1533 runner.finalState = "Cancelled"
1535 err = runner.startCrunchstat()
1540 err = runner.StartContainer()
1542 runner.checkBrokenNode(err)
1546 err = runner.WaitFinish()
1547 if err == nil && !runner.IsCancelled() {
1548 runner.finalState = "Complete"
1553 // Fetch the current container record (uuid = runner.Container.UUID)
1554 // into runner.Container.
1555 func (runner *ContainerRunner) fetchContainerRecord() error {
1556 reader, err := runner.DispatcherArvClient.CallRaw("GET", "containers", runner.Container.UUID, "", nil)
1558 return fmt.Errorf("error fetching container record: %v", err)
1560 defer reader.Close()
1562 dec := json.NewDecoder(reader)
1564 err = dec.Decode(&runner.Container)
1566 return fmt.Errorf("error decoding container record: %v", err)
1570 SecretMounts map[string]arvados.Mount `json:"secret_mounts"`
1573 containerToken, err := runner.ContainerToken()
1575 return fmt.Errorf("error getting container token: %v", err)
1578 runner.ContainerArvClient, runner.ContainerKeepClient,
1579 runner.containerClient, err = runner.MkArvClient(containerToken)
1581 return fmt.Errorf("error creating container API client: %v", err)
1584 runner.ContainerKeepClient.SetStorageClasses(runner.Container.OutputStorageClasses)
1585 runner.DispatcherKeepClient.SetStorageClasses(runner.Container.OutputStorageClasses)
1587 err = runner.ContainerArvClient.Call("GET", "containers", runner.Container.UUID, "secret_mounts", nil, &sm)
1589 if apierr, ok := err.(arvadosclient.APIServerError); !ok || apierr.HttpStatusCode != 404 {
1590 return fmt.Errorf("error fetching secret_mounts: %v", err)
1592 // ok && apierr.HttpStatusCode == 404, which means
1593 // secret_mounts isn't supported by this API server.
1595 runner.SecretMounts = sm.SecretMounts
1600 // NewContainerRunner creates a new container runner.
1601 func NewContainerRunner(dispatcherClient *arvados.Client,
1602 dispatcherArvClient IArvadosClient,
1603 dispatcherKeepClient IKeepClient,
1604 containerUUID string) (*ContainerRunner, error) {
1606 cr := &ContainerRunner{
1607 dispatcherClient: dispatcherClient,
1608 DispatcherArvClient: dispatcherArvClient,
1609 DispatcherKeepClient: dispatcherKeepClient,
1611 cr.NewLogWriter = cr.NewArvLogWriter
1612 cr.RunArvMount = cr.ArvMountCmd
1613 cr.MkTempDir = ioutil.TempDir
1614 cr.MkArvClient = func(token string) (IArvadosClient, IKeepClient, *arvados.Client, error) {
1615 cl, err := arvadosclient.MakeArvadosClient()
1617 return nil, nil, nil, err
1620 kc, err := keepclient.MakeKeepClient(cl)
1622 return nil, nil, nil, err
1624 c2 := arvados.NewClientFromEnv()
1625 c2.AuthToken = token
1626 return cl, kc, c2, nil
1629 cr.LogCollection, err = (&arvados.Collection{}).FileSystem(cr.dispatcherClient, cr.DispatcherKeepClient)
1633 cr.Container.UUID = containerUUID
1634 w, err := cr.NewLogWriter("crunch-run")
1638 cr.CrunchLog = NewThrottledLogger(w)
1639 cr.CrunchLog.Immediate = log.New(os.Stderr, containerUUID+" ", 0)
1641 loadLogThrottleParams(dispatcherArvClient)
1647 func (command) RunCommand(prog string, args []string, stdin io.Reader, stdout, stderr io.Writer) int {
1648 flags := flag.NewFlagSet(prog, flag.ContinueOnError)
1649 statInterval := flags.Duration("crunchstat-interval", 10*time.Second, "sampling period for periodic resource usage reporting")
1650 cgroupRoot := flags.String("cgroup-root", "/sys/fs/cgroup", "path to sysfs cgroup tree")
1651 cgroupParent := flags.String("cgroup-parent", "docker", "name of container's parent cgroup (ignored if -cgroup-parent-subsystem is used)")
1652 cgroupParentSubsystem := flags.String("cgroup-parent-subsystem", "", "use current cgroup for given subsystem as parent cgroup for container")
1653 caCertsPath := flags.String("ca-certs", "", "Path to TLS root certificates")
1654 detach := flags.Bool("detach", false, "Detach from parent process and run in the background")
1655 stdinEnv := flags.Bool("stdin-env", false, "Load environment variables from JSON message on stdin")
1656 sleep := flags.Duration("sleep", 0, "Delay before starting (testing use only)")
1657 kill := flags.Int("kill", -1, "Send signal to an existing crunch-run process for given UUID")
1658 list := flags.Bool("list", false, "List UUIDs of existing crunch-run processes")
1659 enableMemoryLimit := flags.Bool("enable-memory-limit", true, "tell container runtime to limit container's memory usage")
1660 enableNetwork := flags.String("container-enable-networking", "default", "enable networking \"always\" (for all containers) or \"default\" (for containers that request it)")
1661 networkMode := flags.String("container-network-mode", "default", `Docker network mode for container (use any argument valid for docker --net)`)
1662 memprofile := flags.String("memprofile", "", "write memory profile to `file` after running container")
1663 runtimeEngine := flags.String("runtime-engine", "docker", "container runtime: docker or singularity")
1664 flags.Duration("check-containerd", 0, "Ignored. Exists for compatibility with older versions.")
1666 ignoreDetachFlag := false
1667 if len(args) > 0 && args[0] == "-no-detach" {
1668 // This process was invoked by a parent process, which
1669 // has passed along its own arguments, including
1670 // -detach, after the leading -no-detach flag. Strip
1671 // the leading -no-detach flag (it's not recognized by
1672 // flags.Parse()) and ignore the -detach flag that
1675 ignoreDetachFlag = true
1678 if err := flags.Parse(args); err == flag.ErrHelp {
1680 } else if err != nil {
1685 if *stdinEnv && !ignoreDetachFlag {
1686 // Load env vars on stdin if asked (but not in a
1687 // detached child process, in which case stdin is
1689 err := loadEnv(os.Stdin)
1696 containerUUID := flags.Arg(0)
1699 case *detach && !ignoreDetachFlag:
1700 return Detach(containerUUID, prog, args, os.Stdout, os.Stderr)
1702 return KillProcess(containerUUID, syscall.Signal(*kill), os.Stdout, os.Stderr)
1704 return ListProcesses(os.Stdout, os.Stderr)
1707 if containerUUID == "" {
1708 log.Printf("usage: %s [options] UUID", prog)
1712 log.Printf("crunch-run %s started", cmd.Version.String())
1715 if *caCertsPath != "" {
1716 arvadosclient.CertFiles = []string{*caCertsPath}
1719 api, err := arvadosclient.MakeArvadosClient()
1721 log.Printf("%s: %v", containerUUID, err)
1726 kc, kcerr := keepclient.MakeKeepClient(api)
1728 log.Printf("%s: %v", containerUUID, kcerr)
1731 kc.BlockCache = &keepclient.BlockCache{MaxBlocks: 2}
1734 cr, err := NewContainerRunner(arvados.NewClientFromEnv(), api, kc, containerUUID)
1740 switch *runtimeEngine {
1742 cr.executor, err = newDockerExecutor(containerUUID, cr.CrunchLog.Printf, cr.containerWatchdogInterval)
1744 cr.executor, err = newSingularityExecutor(cr.CrunchLog.Printf)
1746 cr.CrunchLog.Printf("%s: unsupported RuntimeEngine %q", containerUUID, *runtimeEngine)
1747 cr.CrunchLog.Close()
1751 cr.CrunchLog.Printf("%s: %v", containerUUID, err)
1752 cr.checkBrokenNode(err)
1753 cr.CrunchLog.Close()
1756 defer cr.executor.Close()
1758 gwAuthSecret := os.Getenv("GatewayAuthSecret")
1759 os.Unsetenv("GatewayAuthSecret")
1760 if gwAuthSecret == "" {
1761 // not safe to run a gateway service without an auth
1763 cr.CrunchLog.Printf("Not starting a gateway server (GatewayAuthSecret was not provided by dispatcher)")
1764 } else if gwListen := os.Getenv("GatewayAddress"); gwListen == "" {
1765 // dispatcher did not tell us which external IP
1766 // address to advertise --> no gateway service
1767 cr.CrunchLog.Printf("Not starting a gateway server (GatewayAddress was not provided by dispatcher)")
1768 } else if de, ok := cr.executor.(*dockerExecutor); ok {
1769 cr.gateway = Gateway{
1771 AuthSecret: gwAuthSecret,
1772 ContainerUUID: containerUUID,
1773 DockerContainerID: &de.containerID,
1775 ContainerIPAddress: dockerContainerIPAddress(&de.containerID),
1777 err = cr.gateway.Start()
1779 log.Printf("error starting gateway server: %s", err)
1784 parentTemp, tmperr := cr.MkTempDir("", "crunch-run."+containerUUID+".")
1786 log.Printf("%s: %v", containerUUID, tmperr)
1790 cr.parentTemp = parentTemp
1791 cr.statInterval = *statInterval
1792 cr.cgroupRoot = *cgroupRoot
1793 cr.expectCgroupParent = *cgroupParent
1794 cr.enableMemoryLimit = *enableMemoryLimit
1795 cr.enableNetwork = *enableNetwork
1796 cr.networkMode = *networkMode
1797 if *cgroupParentSubsystem != "" {
1798 p := findCgroup(*cgroupParentSubsystem)
1799 cr.setCgroupParent = p
1800 cr.expectCgroupParent = p
1805 if *memprofile != "" {
1806 f, err := os.Create(*memprofile)
1808 log.Printf("could not create memory profile: %s", err)
1810 runtime.GC() // get up-to-date statistics
1811 if err := pprof.WriteHeapProfile(f); err != nil {
1812 log.Printf("could not write memory profile: %s", err)
1814 closeerr := f.Close()
1815 if closeerr != nil {
1816 log.Printf("closing memprofile file: %s", err)
1821 log.Printf("%s: %v", containerUUID, runerr)
1827 func loadEnv(rdr io.Reader) error {
1828 buf, err := ioutil.ReadAll(rdr)
1830 return fmt.Errorf("read stdin: %s", err)
1832 var env map[string]string
1833 err = json.Unmarshal(buf, &env)
1835 return fmt.Errorf("decode stdin: %s", err)
1837 for k, v := range env {
1838 err = os.Setenv(k, v)
1840 return fmt.Errorf("setenv(%q): %s", k, err)