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(args []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 runner.CrunchLog.Printf("Using Docker image id %q", imageID)
265 runner.CrunchLog.Print("Loading Docker image from keep")
266 err = runner.executor.LoadImage(imageID, runner.Container, runner.ArvMountPoint,
267 runner.containerClient, runner.ContainerKeepClient)
275 func (runner *ContainerRunner) ArvMountCmd(arvMountCmd []string, token string) (c *exec.Cmd, err error) {
276 c = exec.Command("arv-mount", arvMountCmd...)
278 // Copy our environment, but override ARVADOS_API_TOKEN with
279 // the container auth token.
281 for _, s := range os.Environ() {
282 if !strings.HasPrefix(s, "ARVADOS_API_TOKEN=") {
283 c.Env = append(c.Env, s)
286 c.Env = append(c.Env, "ARVADOS_API_TOKEN="+token)
288 w, err := runner.NewLogWriter("arv-mount")
292 runner.arvMountLog = NewThrottledLogger(w)
293 c.Stdout = runner.arvMountLog
294 c.Stderr = io.MultiWriter(runner.arvMountLog, os.Stderr)
296 runner.CrunchLog.Printf("Running %v", c.Args)
303 statReadme := make(chan bool)
304 runner.ArvMountExit = make(chan error)
309 time.Sleep(100 * time.Millisecond)
310 _, err = os.Stat(fmt.Sprintf("%s/by_id/README", runner.ArvMountPoint))
322 runner.CrunchLog.Printf("Arv-mount exit error: %v", mnterr)
324 runner.ArvMountExit <- mnterr
325 close(runner.ArvMountExit)
331 case err := <-runner.ArvMountExit:
332 runner.ArvMount = nil
340 func (runner *ContainerRunner) SetupArvMountPoint(prefix string) (err error) {
341 if runner.ArvMountPoint == "" {
342 runner.ArvMountPoint, err = runner.MkTempDir(runner.parentTemp, prefix)
347 func copyfile(src string, dst string) (err error) {
348 srcfile, err := os.Open(src)
353 os.MkdirAll(path.Dir(dst), 0777)
355 dstfile, err := os.Create(dst)
359 _, err = io.Copy(dstfile, srcfile)
364 err = srcfile.Close()
365 err2 := dstfile.Close()
378 func (runner *ContainerRunner) SetupMounts() (map[string]bindmount, error) {
379 bindmounts := map[string]bindmount{}
380 err := runner.SetupArvMountPoint("keep")
382 return nil, fmt.Errorf("While creating keep mount temp dir: %v", err)
385 token, err := runner.ContainerToken()
387 return nil, fmt.Errorf("could not get container token: %s", err)
389 runner.CrunchLog.Printf("container token %q", token)
393 arvMountCmd := []string{
397 "--storage-classes", strings.Join(runner.Container.OutputStorageClasses, ","),
398 fmt.Sprintf("--crunchstat-interval=%v", runner.statInterval.Seconds())}
400 if runner.Container.RuntimeConstraints.KeepCacheRAM > 0 {
401 arvMountCmd = append(arvMountCmd, "--file-cache", fmt.Sprintf("%d", runner.Container.RuntimeConstraints.KeepCacheRAM))
404 collectionPaths := []string{}
405 needCertMount := true
406 type copyFile struct {
410 var copyFiles []copyFile
413 for bind := range runner.Container.Mounts {
414 binds = append(binds, bind)
416 for bind := range runner.SecretMounts {
417 if _, ok := runner.Container.Mounts[bind]; ok {
418 return nil, fmt.Errorf("secret mount %q conflicts with regular mount", bind)
420 if runner.SecretMounts[bind].Kind != "json" &&
421 runner.SecretMounts[bind].Kind != "text" {
422 return nil, fmt.Errorf("secret mount %q type is %q but only 'json' and 'text' are permitted",
423 bind, runner.SecretMounts[bind].Kind)
425 binds = append(binds, bind)
429 for _, bind := range binds {
430 mnt, ok := runner.Container.Mounts[bind]
432 mnt = runner.SecretMounts[bind]
434 if bind == "stdout" || bind == "stderr" {
435 // Is it a "file" mount kind?
436 if mnt.Kind != "file" {
437 return nil, fmt.Errorf("unsupported mount kind '%s' for %s: only 'file' is supported", mnt.Kind, bind)
440 // Does path start with OutputPath?
441 prefix := runner.Container.OutputPath
442 if !strings.HasSuffix(prefix, "/") {
445 if !strings.HasPrefix(mnt.Path, prefix) {
446 return nil, fmt.Errorf("%s path does not start with OutputPath: %s, %s", strings.Title(bind), mnt.Path, prefix)
451 // Is it a "collection" mount kind?
452 if mnt.Kind != "collection" && mnt.Kind != "json" {
453 return nil, fmt.Errorf("unsupported mount kind '%s' for stdin: only 'collection' and 'json' are supported", mnt.Kind)
457 if bind == "/etc/arvados/ca-certificates.crt" {
458 needCertMount = false
461 if strings.HasPrefix(bind, runner.Container.OutputPath+"/") && bind != runner.Container.OutputPath+"/" {
462 if mnt.Kind != "collection" && mnt.Kind != "text" && mnt.Kind != "json" {
463 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)
468 case mnt.Kind == "collection" && bind != "stdin":
470 if mnt.UUID != "" && mnt.PortableDataHash != "" {
471 return nil, fmt.Errorf("cannot specify both 'uuid' and 'portable_data_hash' for a collection mount")
475 return nil, fmt.Errorf("writing to existing collections currently not permitted")
478 src = fmt.Sprintf("%s/by_id/%s", runner.ArvMountPoint, mnt.UUID)
479 } else if mnt.PortableDataHash != "" {
480 if mnt.Writable && !strings.HasPrefix(bind, runner.Container.OutputPath+"/") {
481 return nil, fmt.Errorf("can never write to a collection specified by portable data hash")
483 idx := strings.Index(mnt.PortableDataHash, "/")
485 mnt.Path = path.Clean(mnt.PortableDataHash[idx:])
486 mnt.PortableDataHash = mnt.PortableDataHash[0:idx]
487 runner.Container.Mounts[bind] = mnt
489 src = fmt.Sprintf("%s/by_id/%s", runner.ArvMountPoint, mnt.PortableDataHash)
490 if mnt.Path != "" && mnt.Path != "." {
491 if strings.HasPrefix(mnt.Path, "./") {
492 mnt.Path = mnt.Path[2:]
493 } else if strings.HasPrefix(mnt.Path, "/") {
494 mnt.Path = mnt.Path[1:]
496 src += "/" + mnt.Path
499 src = fmt.Sprintf("%s/tmp%d", runner.ArvMountPoint, tmpcount)
500 arvMountCmd = append(arvMountCmd, "--mount-tmp")
501 arvMountCmd = append(arvMountCmd, fmt.Sprintf("tmp%d", tmpcount))
505 if bind == runner.Container.OutputPath {
506 runner.HostOutputDir = src
507 bindmounts[bind] = bindmount{HostPath: src}
508 } else if strings.HasPrefix(bind, runner.Container.OutputPath+"/") {
509 copyFiles = append(copyFiles, copyFile{src, runner.HostOutputDir + bind[len(runner.Container.OutputPath):]})
511 bindmounts[bind] = bindmount{HostPath: src}
514 bindmounts[bind] = bindmount{HostPath: src, ReadOnly: true}
516 collectionPaths = append(collectionPaths, src)
518 case mnt.Kind == "tmp":
520 tmpdir, err = runner.MkTempDir(runner.parentTemp, "tmp")
522 return nil, fmt.Errorf("while creating mount temp dir: %v", err)
524 st, staterr := os.Stat(tmpdir)
526 return nil, fmt.Errorf("while Stat on temp dir: %v", staterr)
528 err = os.Chmod(tmpdir, st.Mode()|os.ModeSetgid|0777)
530 return nil, fmt.Errorf("while Chmod temp dir: %v", err)
532 bindmounts[bind] = bindmount{HostPath: tmpdir}
533 if bind == runner.Container.OutputPath {
534 runner.HostOutputDir = tmpdir
537 case mnt.Kind == "json" || mnt.Kind == "text":
539 if mnt.Kind == "json" {
540 filedata, err = json.Marshal(mnt.Content)
542 return nil, fmt.Errorf("encoding json data: %v", err)
545 text, ok := mnt.Content.(string)
547 return nil, fmt.Errorf("content for mount %q must be a string", bind)
549 filedata = []byte(text)
552 tmpdir, err := runner.MkTempDir(runner.parentTemp, mnt.Kind)
554 return nil, fmt.Errorf("creating temp dir: %v", err)
556 tmpfn := filepath.Join(tmpdir, "mountdata."+mnt.Kind)
557 err = ioutil.WriteFile(tmpfn, filedata, 0444)
559 return nil, fmt.Errorf("writing temp file: %v", err)
561 if strings.HasPrefix(bind, runner.Container.OutputPath+"/") {
562 copyFiles = append(copyFiles, copyFile{tmpfn, runner.HostOutputDir + bind[len(runner.Container.OutputPath):]})
564 bindmounts[bind] = bindmount{HostPath: tmpfn, ReadOnly: true}
567 case mnt.Kind == "git_tree":
568 tmpdir, err := runner.MkTempDir(runner.parentTemp, "git_tree")
570 return nil, fmt.Errorf("creating temp dir: %v", err)
572 err = gitMount(mnt).extractTree(runner.ContainerArvClient, tmpdir, token)
576 bindmounts[bind] = bindmount{HostPath: tmpdir, ReadOnly: true}
580 if runner.HostOutputDir == "" {
581 return nil, fmt.Errorf("output path does not correspond to a writable mount point")
584 if needCertMount && runner.Container.RuntimeConstraints.API {
585 for _, certfile := range arvadosclient.CertFiles {
586 _, err := os.Stat(certfile)
588 bindmounts["/etc/arvados/ca-certificates.crt"] = bindmount{HostPath: certfile, ReadOnly: true}
595 arvMountCmd = append(arvMountCmd, "--mount-by-pdh", "by_id")
597 arvMountCmd = append(arvMountCmd, "--mount-by-id", "by_id")
599 arvMountCmd = append(arvMountCmd, "--mount-by-id", "by_uuid")
600 arvMountCmd = append(arvMountCmd, runner.ArvMountPoint)
602 runner.ArvMount, err = runner.RunArvMount(arvMountCmd, token)
604 return nil, fmt.Errorf("while trying to start arv-mount: %v", err)
607 for _, p := range collectionPaths {
610 return nil, fmt.Errorf("while checking that input files exist: %v", err)
614 for _, cp := range copyFiles {
615 st, err := os.Stat(cp.src)
617 return nil, fmt.Errorf("while staging writable file from %q to %q: %v", cp.src, cp.bind, err)
620 err = filepath.Walk(cp.src, func(walkpath string, walkinfo os.FileInfo, walkerr error) error {
624 target := path.Join(cp.bind, walkpath[len(cp.src):])
625 if walkinfo.Mode().IsRegular() {
626 copyerr := copyfile(walkpath, target)
630 return os.Chmod(target, walkinfo.Mode()|0777)
631 } else if walkinfo.Mode().IsDir() {
632 mkerr := os.MkdirAll(target, 0777)
636 return os.Chmod(target, walkinfo.Mode()|os.ModeSetgid|0777)
638 return fmt.Errorf("source %q is not a regular file or directory", cp.src)
641 } else if st.Mode().IsRegular() {
642 err = copyfile(cp.src, cp.bind)
644 err = os.Chmod(cp.bind, st.Mode()|0777)
648 return nil, fmt.Errorf("while staging writable file from %q to %q: %v", cp.src, cp.bind, err)
652 return bindmounts, nil
655 func (runner *ContainerRunner) stopHoststat() error {
656 if runner.hoststatReporter == nil {
659 runner.hoststatReporter.Stop()
660 err := runner.hoststatLogger.Close()
662 return fmt.Errorf("error closing hoststat logs: %v", err)
667 func (runner *ContainerRunner) startHoststat() error {
668 w, err := runner.NewLogWriter("hoststat")
672 runner.hoststatLogger = NewThrottledLogger(w)
673 runner.hoststatReporter = &crunchstat.Reporter{
674 Logger: log.New(runner.hoststatLogger, "", 0),
675 CgroupRoot: runner.cgroupRoot,
676 PollPeriod: runner.statInterval,
678 runner.hoststatReporter.Start()
682 func (runner *ContainerRunner) startCrunchstat() error {
683 w, err := runner.NewLogWriter("crunchstat")
687 runner.statLogger = NewThrottledLogger(w)
688 runner.statReporter = &crunchstat.Reporter{
689 CID: runner.executor.CgroupID(),
690 Logger: log.New(runner.statLogger, "", 0),
691 CgroupParent: runner.expectCgroupParent,
692 CgroupRoot: runner.cgroupRoot,
693 PollPeriod: runner.statInterval,
694 TempDir: runner.parentTemp,
696 runner.statReporter.Start()
700 type infoCommand struct {
705 // LogHostInfo logs info about the current host, for debugging and
706 // accounting purposes. Although it's logged as "node-info", this is
707 // about the environment where crunch-run is actually running, which
708 // might differ from what's described in the node record (see
710 func (runner *ContainerRunner) LogHostInfo() (err error) {
711 w, err := runner.NewLogWriter("node-info")
716 commands := []infoCommand{
718 label: "Host Information",
719 cmd: []string{"uname", "-a"},
722 label: "CPU Information",
723 cmd: []string{"cat", "/proc/cpuinfo"},
726 label: "Memory Information",
727 cmd: []string{"cat", "/proc/meminfo"},
731 cmd: []string{"df", "-m", "/", os.TempDir()},
734 label: "Disk INodes",
735 cmd: []string{"df", "-i", "/", os.TempDir()},
739 // Run commands with informational output to be logged.
740 for _, command := range commands {
741 fmt.Fprintln(w, command.label)
742 cmd := exec.Command(command.cmd[0], command.cmd[1:]...)
745 if err := cmd.Run(); err != nil {
746 err = fmt.Errorf("While running command %q: %v", command.cmd, err)
755 return fmt.Errorf("While closing node-info logs: %v", err)
760 // LogContainerRecord gets and saves the raw JSON container record from the API server
761 func (runner *ContainerRunner) LogContainerRecord() error {
762 logged, err := runner.logAPIResponse("container", "containers", map[string]interface{}{"filters": [][]string{{"uuid", "=", runner.Container.UUID}}}, nil)
763 if !logged && err == nil {
764 err = fmt.Errorf("error: no container record found for %s", runner.Container.UUID)
769 // LogNodeRecord logs the current host's InstanceType config entry (or
770 // the arvados#node record, if running via crunch-dispatch-slurm).
771 func (runner *ContainerRunner) LogNodeRecord() error {
772 if it := os.Getenv("InstanceType"); it != "" {
773 // Dispatched via arvados-dispatch-cloud. Save
774 // InstanceType config fragment received from
775 // dispatcher on stdin.
776 w, err := runner.LogCollection.OpenFile("node.json", os.O_CREATE|os.O_WRONLY, 0666)
781 _, err = io.WriteString(w, it)
787 // Dispatched via crunch-dispatch-slurm. Look up
788 // apiserver's node record corresponding to
790 hostname := os.Getenv("SLURMD_NODENAME")
792 hostname, _ = os.Hostname()
794 _, err := runner.logAPIResponse("node", "nodes", map[string]interface{}{"filters": [][]string{{"hostname", "=", hostname}}}, func(resp interface{}) {
795 // The "info" field has admin-only info when
796 // obtained with a privileged token, and
797 // should not be logged.
798 node, ok := resp.(map[string]interface{})
806 func (runner *ContainerRunner) logAPIResponse(label, path string, params map[string]interface{}, munge func(interface{})) (logged bool, err error) {
807 writer, err := runner.LogCollection.OpenFile(label+".json", os.O_CREATE|os.O_WRONLY, 0666)
812 ArvClient: runner.DispatcherArvClient,
813 UUID: runner.Container.UUID,
814 loggingStream: label,
818 reader, err := runner.DispatcherArvClient.CallRaw("GET", path, "", "", arvadosclient.Dict(params))
820 return false, fmt.Errorf("error getting %s record: %v", label, err)
824 dec := json.NewDecoder(reader)
826 var resp map[string]interface{}
827 if err = dec.Decode(&resp); err != nil {
828 return false, fmt.Errorf("error decoding %s list response: %v", label, err)
830 items, ok := resp["items"].([]interface{})
832 return false, fmt.Errorf("error decoding %s list response: no \"items\" key in API list response", label)
833 } else if len(items) < 1 {
839 // Re-encode it using indentation to improve readability
840 enc := json.NewEncoder(w)
841 enc.SetIndent("", " ")
842 if err = enc.Encode(items[0]); err != nil {
843 return false, fmt.Errorf("error logging %s record: %v", label, err)
847 return false, fmt.Errorf("error closing %s.json in log collection: %v", label, err)
852 func (runner *ContainerRunner) getStdoutFile(mntPath string) (*os.File, error) {
853 stdoutPath := mntPath[len(runner.Container.OutputPath):]
854 index := strings.LastIndex(stdoutPath, "/")
856 subdirs := stdoutPath[:index]
858 st, err := os.Stat(runner.HostOutputDir)
860 return nil, fmt.Errorf("While Stat on temp dir: %v", err)
862 stdoutPath := filepath.Join(runner.HostOutputDir, subdirs)
863 err = os.MkdirAll(stdoutPath, st.Mode()|os.ModeSetgid|0777)
865 return nil, fmt.Errorf("While MkdirAll %q: %v", stdoutPath, err)
869 stdoutFile, err := os.Create(filepath.Join(runner.HostOutputDir, stdoutPath))
871 return nil, fmt.Errorf("While creating file %q: %v", stdoutPath, err)
874 return stdoutFile, nil
877 // CreateContainer creates the docker container.
878 func (runner *ContainerRunner) CreateContainer(imageID string, bindmounts map[string]bindmount) error {
879 var stdin io.ReadCloser = ioutil.NopCloser(bytes.NewReader(nil))
880 if mnt, ok := runner.Container.Mounts["stdin"]; ok {
887 collID = mnt.PortableDataHash
889 path := runner.ArvMountPoint + "/by_id/" + collID + "/" + mnt.Path
890 f, err := os.Open(path)
896 j, err := json.Marshal(mnt.Content)
898 return fmt.Errorf("error encoding stdin json data: %v", err)
900 stdin = ioutil.NopCloser(bytes.NewReader(j))
902 return fmt.Errorf("stdin mount has unsupported kind %q", mnt.Kind)
906 var stdout, stderr io.WriteCloser
907 if mnt, ok := runner.Container.Mounts["stdout"]; ok {
908 f, err := runner.getStdoutFile(mnt.Path)
913 } else if w, err := runner.NewLogWriter("stdout"); err != nil {
916 stdout = NewThrottledLogger(w)
919 if mnt, ok := runner.Container.Mounts["stderr"]; ok {
920 f, err := runner.getStdoutFile(mnt.Path)
925 } else if w, err := runner.NewLogWriter("stderr"); err != nil {
928 stderr = NewThrottledLogger(w)
931 env := runner.Container.Environment
932 enableNetwork := runner.enableNetwork == "always"
933 if runner.Container.RuntimeConstraints.API {
935 tok, err := runner.ContainerToken()
939 env = map[string]string{}
940 for k, v := range runner.Container.Environment {
943 env["ARVADOS_API_TOKEN"] = tok
944 env["ARVADOS_API_HOST"] = os.Getenv("ARVADOS_API_HOST")
945 env["ARVADOS_API_HOST_INSECURE"] = os.Getenv("ARVADOS_API_HOST_INSECURE")
947 workdir := runner.Container.Cwd
949 // both "" and "." mean default
952 ram := runner.Container.RuntimeConstraints.RAM
953 if !runner.enableMemoryLimit {
956 runner.executorStdin = stdin
957 runner.executorStdout = stdout
958 runner.executorStderr = stderr
959 return runner.executor.Create(containerSpec{
961 VCPUs: runner.Container.RuntimeConstraints.VCPUs,
965 BindMounts: bindmounts,
966 Command: runner.Container.Command,
967 EnableNetwork: enableNetwork,
968 NetworkMode: runner.networkMode,
969 CgroupParent: runner.setCgroupParent,
976 // StartContainer starts the docker container created by CreateContainer.
977 func (runner *ContainerRunner) StartContainer() error {
978 runner.CrunchLog.Printf("Starting container")
979 runner.cStateLock.Lock()
980 defer runner.cStateLock.Unlock()
981 if runner.cCancelled {
984 err := runner.executor.Start()
987 if m, e := regexp.MatchString("(?ms).*(exec|System error).*(no such file or directory|file not found).*", err.Error()); m && e == nil {
988 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])
990 return fmt.Errorf("could not start container: %v%s", err, advice)
995 // WaitFinish waits for the container to terminate, capture the exit code, and
996 // close the stdout/stderr logging.
997 func (runner *ContainerRunner) WaitFinish() error {
998 runner.CrunchLog.Print("Waiting for container to finish")
999 var timeout <-chan time.Time
1000 if s := runner.Container.SchedulingParameters.MaxRunTime; s > 0 {
1001 timeout = time.After(time.Duration(s) * time.Second)
1003 ctx, cancel := context.WithCancel(context.Background())
1008 runner.CrunchLog.Printf("maximum run time exceeded. Stopping container.")
1010 case <-runner.ArvMountExit:
1011 runner.CrunchLog.Printf("arv-mount exited while container is still running. Stopping container.")
1016 exitcode, err := runner.executor.Wait(ctx)
1018 runner.checkBrokenNode(err)
1021 runner.ExitCode = &exitcode
1024 if err = runner.executorStdin.Close(); err != nil {
1025 err = fmt.Errorf("error closing container stdin: %s", err)
1026 runner.CrunchLog.Printf("%s", err)
1029 if err = runner.executorStdout.Close(); err != nil {
1030 err = fmt.Errorf("error closing container stdout: %s", err)
1031 runner.CrunchLog.Printf("%s", err)
1032 if returnErr == nil {
1036 if err = runner.executorStderr.Close(); err != nil {
1037 err = fmt.Errorf("error closing container stderr: %s", err)
1038 runner.CrunchLog.Printf("%s", err)
1039 if returnErr == nil {
1044 if runner.statReporter != nil {
1045 runner.statReporter.Stop()
1046 err = runner.statLogger.Close()
1048 runner.CrunchLog.Printf("error closing crunchstat logs: %v", err)
1054 func (runner *ContainerRunner) updateLogs() {
1055 ticker := time.NewTicker(crunchLogUpdatePeriod / 360)
1058 sigusr1 := make(chan os.Signal, 1)
1059 signal.Notify(sigusr1, syscall.SIGUSR1)
1060 defer signal.Stop(sigusr1)
1062 saveAtTime := time.Now().Add(crunchLogUpdatePeriod)
1063 saveAtSize := crunchLogUpdateSize
1069 saveAtTime = time.Now()
1071 runner.logMtx.Lock()
1072 done := runner.LogsPDH != nil
1073 runner.logMtx.Unlock()
1077 size := runner.LogCollection.Size()
1078 if size == savedSize || (time.Now().Before(saveAtTime) && size < saveAtSize) {
1081 saveAtTime = time.Now().Add(crunchLogUpdatePeriod)
1082 saveAtSize = runner.LogCollection.Size() + crunchLogUpdateSize
1083 saved, err := runner.saveLogCollection(false)
1085 runner.CrunchLog.Printf("error updating log collection: %s", err)
1089 var updated arvados.Container
1090 err = runner.DispatcherArvClient.Update("containers", runner.Container.UUID, arvadosclient.Dict{
1091 "container": arvadosclient.Dict{"log": saved.PortableDataHash},
1094 runner.CrunchLog.Printf("error updating container log to %s: %s", saved.PortableDataHash, err)
1102 // CaptureOutput saves data from the container's output directory if
1103 // needed, and updates the container output accordingly.
1104 func (runner *ContainerRunner) CaptureOutput(bindmounts map[string]bindmount) error {
1105 if runner.Container.RuntimeConstraints.API {
1106 // Output may have been set directly by the container, so
1107 // refresh the container record to check.
1108 err := runner.DispatcherArvClient.Get("containers", runner.Container.UUID,
1109 nil, &runner.Container)
1113 if runner.Container.Output != "" {
1114 // Container output is already set.
1115 runner.OutputPDH = &runner.Container.Output
1120 txt, err := (&copier{
1121 client: runner.containerClient,
1122 arvClient: runner.ContainerArvClient,
1123 keepClient: runner.ContainerKeepClient,
1124 hostOutputDir: runner.HostOutputDir,
1125 ctrOutputDir: runner.Container.OutputPath,
1126 bindmounts: bindmounts,
1127 mounts: runner.Container.Mounts,
1128 secretMounts: runner.SecretMounts,
1129 logger: runner.CrunchLog,
1134 if n := len(regexp.MustCompile(` [0-9a-f]+\+\S*\+R`).FindAllStringIndex(txt, -1)); n > 0 {
1135 runner.CrunchLog.Printf("Copying %d data blocks from remote input collections...", n)
1136 fs, err := (&arvados.Collection{ManifestText: txt}).FileSystem(runner.containerClient, runner.ContainerKeepClient)
1140 txt, err = fs.MarshalManifest(".")
1145 var resp arvados.Collection
1146 err = runner.ContainerArvClient.Create("collections", arvadosclient.Dict{
1147 "ensure_unique_name": true,
1148 "collection": arvadosclient.Dict{
1150 "name": "output for " + runner.Container.UUID,
1151 "manifest_text": txt,
1155 return fmt.Errorf("error creating output collection: %v", err)
1157 runner.OutputPDH = &resp.PortableDataHash
1161 func (runner *ContainerRunner) CleanupDirs() {
1162 if runner.ArvMount != nil {
1164 umount := exec.Command("arv-mount", fmt.Sprintf("--unmount-timeout=%d", delay), "--unmount", runner.ArvMountPoint)
1165 umount.Stdout = runner.CrunchLog
1166 umount.Stderr = runner.CrunchLog
1167 runner.CrunchLog.Printf("Running %v", umount.Args)
1168 umnterr := umount.Start()
1171 runner.CrunchLog.Printf("Error unmounting: %v", umnterr)
1173 // If arv-mount --unmount gets stuck for any reason, we
1174 // don't want to wait for it forever. Do Wait() in a goroutine
1175 // so it doesn't block crunch-run.
1176 umountExit := make(chan error)
1178 mnterr := umount.Wait()
1180 runner.CrunchLog.Printf("Error unmounting: %v", mnterr)
1182 umountExit <- mnterr
1185 for again := true; again; {
1191 case <-runner.ArvMountExit:
1193 case <-time.After(time.Duration((delay + 1) * int64(time.Second))):
1194 runner.CrunchLog.Printf("Timed out waiting for unmount")
1196 umount.Process.Kill()
1198 runner.ArvMount.Process.Kill()
1202 runner.ArvMount = nil
1205 if runner.ArvMountPoint != "" {
1206 if rmerr := os.Remove(runner.ArvMountPoint); rmerr != nil {
1207 runner.CrunchLog.Printf("While cleaning up arv-mount directory %s: %v", runner.ArvMountPoint, rmerr)
1209 runner.ArvMountPoint = ""
1212 if rmerr := os.RemoveAll(runner.parentTemp); rmerr != nil {
1213 runner.CrunchLog.Printf("While cleaning up temporary directory %s: %v", runner.parentTemp, rmerr)
1217 // CommitLogs posts the collection containing the final container logs.
1218 func (runner *ContainerRunner) CommitLogs() error {
1220 // Hold cStateLock to prevent races on CrunchLog (e.g., stop()).
1221 runner.cStateLock.Lock()
1222 defer runner.cStateLock.Unlock()
1224 runner.CrunchLog.Print(runner.finalState)
1226 if runner.arvMountLog != nil {
1227 runner.arvMountLog.Close()
1229 runner.CrunchLog.Close()
1231 // Closing CrunchLog above allows them to be committed to Keep at this
1232 // point, but re-open crunch log with ArvClient in case there are any
1233 // other further errors (such as failing to write the log to Keep!)
1234 // while shutting down
1235 runner.CrunchLog = NewThrottledLogger(&ArvLogWriter{
1236 ArvClient: runner.DispatcherArvClient,
1237 UUID: runner.Container.UUID,
1238 loggingStream: "crunch-run",
1241 runner.CrunchLog.Immediate = log.New(os.Stderr, runner.Container.UUID+" ", 0)
1244 if runner.LogsPDH != nil {
1245 // If we have already assigned something to LogsPDH,
1246 // we must be closing the re-opened log, which won't
1247 // end up getting attached to the container record and
1248 // therefore doesn't need to be saved as a collection
1249 // -- it exists only to send logs to other channels.
1252 saved, err := runner.saveLogCollection(true)
1254 return fmt.Errorf("error saving log collection: %s", err)
1256 runner.logMtx.Lock()
1257 defer runner.logMtx.Unlock()
1258 runner.LogsPDH = &saved.PortableDataHash
1262 func (runner *ContainerRunner) saveLogCollection(final bool) (response arvados.Collection, err error) {
1263 runner.logMtx.Lock()
1264 defer runner.logMtx.Unlock()
1265 if runner.LogsPDH != nil {
1266 // Already finalized.
1269 updates := arvadosclient.Dict{
1270 "name": "logs for " + runner.Container.UUID,
1272 mt, err1 := runner.LogCollection.MarshalManifest(".")
1274 // Only send updated manifest text if there was no
1276 updates["manifest_text"] = mt
1279 // Even if flushing the manifest had an error, we still want
1280 // to update the log record, if possible, to push the trash_at
1281 // and delete_at times into the future. Details on bug
1284 updates["is_trashed"] = true
1286 exp := time.Now().Add(crunchLogUpdatePeriod * 24)
1287 updates["trash_at"] = exp
1288 updates["delete_at"] = exp
1290 reqBody := arvadosclient.Dict{"collection": updates}
1292 if runner.logUUID == "" {
1293 reqBody["ensure_unique_name"] = true
1294 err2 = runner.DispatcherArvClient.Create("collections", reqBody, &response)
1296 err2 = runner.DispatcherArvClient.Update("collections", runner.logUUID, reqBody, &response)
1299 runner.logUUID = response.UUID
1302 if err1 != nil || err2 != nil {
1303 err = fmt.Errorf("error recording logs: %q, %q", err1, err2)
1308 // UpdateContainerRunning updates the container state to "Running"
1309 func (runner *ContainerRunner) UpdateContainerRunning() error {
1310 runner.cStateLock.Lock()
1311 defer runner.cStateLock.Unlock()
1312 if runner.cCancelled {
1315 return runner.DispatcherArvClient.Update("containers", runner.Container.UUID,
1316 arvadosclient.Dict{"container": arvadosclient.Dict{"state": "Running", "gateway_address": runner.gateway.Address}}, nil)
1319 // ContainerToken returns the api_token the container (and any
1320 // arv-mount processes) are allowed to use.
1321 func (runner *ContainerRunner) ContainerToken() (string, error) {
1322 if runner.token != "" {
1323 return runner.token, nil
1326 var auth arvados.APIClientAuthorization
1327 err := runner.DispatcherArvClient.Call("GET", "containers", runner.Container.UUID, "auth", nil, &auth)
1331 runner.token = fmt.Sprintf("v2/%s/%s/%s", auth.UUID, auth.APIToken, runner.Container.UUID)
1332 return runner.token, nil
1335 // UpdateContainerFinal updates the container record state on API
1336 // server to "Complete" or "Cancelled"
1337 func (runner *ContainerRunner) UpdateContainerFinal() error {
1338 update := arvadosclient.Dict{}
1339 update["state"] = runner.finalState
1340 if runner.LogsPDH != nil {
1341 update["log"] = *runner.LogsPDH
1343 if runner.finalState == "Complete" {
1344 if runner.ExitCode != nil {
1345 update["exit_code"] = *runner.ExitCode
1347 if runner.OutputPDH != nil {
1348 update["output"] = *runner.OutputPDH
1351 return runner.DispatcherArvClient.Update("containers", runner.Container.UUID, arvadosclient.Dict{"container": update}, nil)
1354 // IsCancelled returns the value of Cancelled, with goroutine safety.
1355 func (runner *ContainerRunner) IsCancelled() bool {
1356 runner.cStateLock.Lock()
1357 defer runner.cStateLock.Unlock()
1358 return runner.cCancelled
1361 // NewArvLogWriter creates an ArvLogWriter
1362 func (runner *ContainerRunner) NewArvLogWriter(name string) (io.WriteCloser, error) {
1363 writer, err := runner.LogCollection.OpenFile(name+".txt", os.O_CREATE|os.O_WRONLY, 0666)
1367 return &ArvLogWriter{
1368 ArvClient: runner.DispatcherArvClient,
1369 UUID: runner.Container.UUID,
1370 loggingStream: name,
1371 writeCloser: writer,
1375 // Run the full container lifecycle.
1376 func (runner *ContainerRunner) Run() (err error) {
1377 runner.CrunchLog.Printf("crunch-run %s started", cmd.Version.String())
1378 runner.CrunchLog.Printf("Executing container '%s'", runner.Container.UUID)
1380 hostname, hosterr := os.Hostname()
1382 runner.CrunchLog.Printf("Error getting hostname '%v'", hosterr)
1384 runner.CrunchLog.Printf("Executing on host '%s'", hostname)
1387 runner.finalState = "Queued"
1390 runner.CleanupDirs()
1392 runner.CrunchLog.Printf("crunch-run finished")
1393 runner.CrunchLog.Close()
1396 err = runner.fetchContainerRecord()
1400 if runner.Container.State != "Locked" {
1401 return fmt.Errorf("dispatch error detected: container %q has state %q", runner.Container.UUID, runner.Container.State)
1404 var bindmounts map[string]bindmount
1406 // checkErr prints e (unless it's nil) and sets err to
1407 // e (unless err is already non-nil). Thus, if err
1408 // hasn't already been assigned when Run() returns,
1409 // this cleanup func will cause Run() to return the
1410 // first non-nil error that is passed to checkErr().
1411 checkErr := func(errorIn string, e error) {
1415 runner.CrunchLog.Printf("error in %s: %v", errorIn, e)
1419 if runner.finalState == "Complete" {
1420 // There was an error in the finalization.
1421 runner.finalState = "Cancelled"
1425 // Log the error encountered in Run(), if any
1426 checkErr("Run", err)
1428 if runner.finalState == "Queued" {
1429 runner.UpdateContainerFinal()
1433 if runner.IsCancelled() {
1434 runner.finalState = "Cancelled"
1435 // but don't return yet -- we still want to
1436 // capture partial output and write logs
1439 if bindmounts != nil {
1440 checkErr("CaptureOutput", runner.CaptureOutput(bindmounts))
1442 checkErr("stopHoststat", runner.stopHoststat())
1443 checkErr("CommitLogs", runner.CommitLogs())
1444 runner.CleanupDirs()
1445 checkErr("UpdateContainerFinal", runner.UpdateContainerFinal())
1448 runner.setupSignals()
1449 err = runner.startHoststat()
1454 // set up FUSE mount and binds
1455 bindmounts, err = runner.SetupMounts()
1457 runner.finalState = "Cancelled"
1458 err = fmt.Errorf("While setting up mounts: %v", err)
1462 // check for and/or load image
1463 imageID, err := runner.LoadImage()
1465 if !runner.checkBrokenNode(err) {
1466 // Failed to load image but not due to a "broken node"
1467 // condition, probably user error.
1468 runner.finalState = "Cancelled"
1470 err = fmt.Errorf("While loading container image: %v", err)
1474 err = runner.CreateContainer(imageID, bindmounts)
1478 err = runner.LogHostInfo()
1482 err = runner.LogNodeRecord()
1486 err = runner.LogContainerRecord()
1491 if runner.IsCancelled() {
1495 err = runner.UpdateContainerRunning()
1499 runner.finalState = "Cancelled"
1501 err = runner.startCrunchstat()
1506 err = runner.StartContainer()
1508 runner.checkBrokenNode(err)
1512 err = runner.WaitFinish()
1513 if err == nil && !runner.IsCancelled() {
1514 runner.finalState = "Complete"
1519 // Fetch the current container record (uuid = runner.Container.UUID)
1520 // into runner.Container.
1521 func (runner *ContainerRunner) fetchContainerRecord() error {
1522 reader, err := runner.DispatcherArvClient.CallRaw("GET", "containers", runner.Container.UUID, "", nil)
1524 return fmt.Errorf("error fetching container record: %v", err)
1526 defer reader.Close()
1528 dec := json.NewDecoder(reader)
1530 err = dec.Decode(&runner.Container)
1532 return fmt.Errorf("error decoding container record: %v", err)
1536 SecretMounts map[string]arvados.Mount `json:"secret_mounts"`
1539 containerToken, err := runner.ContainerToken()
1541 return fmt.Errorf("error getting container token: %v", err)
1544 runner.ContainerArvClient, runner.ContainerKeepClient,
1545 runner.containerClient, err = runner.MkArvClient(containerToken)
1547 return fmt.Errorf("error creating container API client: %v", err)
1550 runner.ContainerKeepClient.SetStorageClasses(runner.Container.OutputStorageClasses)
1551 runner.DispatcherKeepClient.SetStorageClasses(runner.Container.OutputStorageClasses)
1553 err = runner.ContainerArvClient.Call("GET", "containers", runner.Container.UUID, "secret_mounts", nil, &sm)
1555 if apierr, ok := err.(arvadosclient.APIServerError); !ok || apierr.HttpStatusCode != 404 {
1556 return fmt.Errorf("error fetching secret_mounts: %v", err)
1558 // ok && apierr.HttpStatusCode == 404, which means
1559 // secret_mounts isn't supported by this API server.
1561 runner.SecretMounts = sm.SecretMounts
1566 // NewContainerRunner creates a new container runner.
1567 func NewContainerRunner(dispatcherClient *arvados.Client,
1568 dispatcherArvClient IArvadosClient,
1569 dispatcherKeepClient IKeepClient,
1570 containerUUID string) (*ContainerRunner, error) {
1572 cr := &ContainerRunner{
1573 dispatcherClient: dispatcherClient,
1574 DispatcherArvClient: dispatcherArvClient,
1575 DispatcherKeepClient: dispatcherKeepClient,
1577 cr.NewLogWriter = cr.NewArvLogWriter
1578 cr.RunArvMount = cr.ArvMountCmd
1579 cr.MkTempDir = ioutil.TempDir
1580 cr.MkArvClient = func(token string) (IArvadosClient, IKeepClient, *arvados.Client, error) {
1581 cl, err := arvadosclient.MakeArvadosClient()
1583 return nil, nil, nil, err
1586 kc, err := keepclient.MakeKeepClient(cl)
1588 return nil, nil, nil, err
1590 c2 := arvados.NewClientFromEnv()
1591 c2.AuthToken = token
1592 return cl, kc, c2, nil
1595 cr.LogCollection, err = (&arvados.Collection{}).FileSystem(cr.dispatcherClient, cr.DispatcherKeepClient)
1599 cr.Container.UUID = containerUUID
1600 w, err := cr.NewLogWriter("crunch-run")
1604 cr.CrunchLog = NewThrottledLogger(w)
1605 cr.CrunchLog.Immediate = log.New(os.Stderr, containerUUID+" ", 0)
1607 loadLogThrottleParams(dispatcherArvClient)
1613 func (command) RunCommand(prog string, args []string, stdin io.Reader, stdout, stderr io.Writer) int {
1614 flags := flag.NewFlagSet(prog, flag.ContinueOnError)
1615 statInterval := flags.Duration("crunchstat-interval", 10*time.Second, "sampling period for periodic resource usage reporting")
1616 cgroupRoot := flags.String("cgroup-root", "/sys/fs/cgroup", "path to sysfs cgroup tree")
1617 cgroupParent := flags.String("cgroup-parent", "docker", "name of container's parent cgroup (ignored if -cgroup-parent-subsystem is used)")
1618 cgroupParentSubsystem := flags.String("cgroup-parent-subsystem", "", "use current cgroup for given subsystem as parent cgroup for container")
1619 caCertsPath := flags.String("ca-certs", "", "Path to TLS root certificates")
1620 detach := flags.Bool("detach", false, "Detach from parent process and run in the background")
1621 stdinEnv := flags.Bool("stdin-env", false, "Load environment variables from JSON message on stdin")
1622 sleep := flags.Duration("sleep", 0, "Delay before starting (testing use only)")
1623 kill := flags.Int("kill", -1, "Send signal to an existing crunch-run process for given UUID")
1624 list := flags.Bool("list", false, "List UUIDs of existing crunch-run processes")
1625 enableMemoryLimit := flags.Bool("enable-memory-limit", true, "tell container runtime to limit container's memory usage")
1626 enableNetwork := flags.String("container-enable-networking", "default", "enable networking \"always\" (for all containers) or \"default\" (for containers that request it)")
1627 networkMode := flags.String("container-network-mode", "default", `Docker network mode for container (use any argument valid for docker --net)`)
1628 memprofile := flags.String("memprofile", "", "write memory profile to `file` after running container")
1629 runtimeEngine := flags.String("runtime-engine", "docker", "container runtime: docker or singularity")
1630 flags.Duration("check-containerd", 0, "Ignored. Exists for compatibility with older versions.")
1632 ignoreDetachFlag := false
1633 if len(args) > 0 && args[0] == "-no-detach" {
1634 // This process was invoked by a parent process, which
1635 // has passed along its own arguments, including
1636 // -detach, after the leading -no-detach flag. Strip
1637 // the leading -no-detach flag (it's not recognized by
1638 // flags.Parse()) and ignore the -detach flag that
1641 ignoreDetachFlag = true
1644 if err := flags.Parse(args); err == flag.ErrHelp {
1646 } else if err != nil {
1651 if *stdinEnv && !ignoreDetachFlag {
1652 // Load env vars on stdin if asked (but not in a
1653 // detached child process, in which case stdin is
1655 err := loadEnv(os.Stdin)
1662 containerUUID := flags.Arg(0)
1665 case *detach && !ignoreDetachFlag:
1666 return Detach(containerUUID, prog, args, os.Stdout, os.Stderr)
1668 return KillProcess(containerUUID, syscall.Signal(*kill), os.Stdout, os.Stderr)
1670 return ListProcesses(os.Stdout, os.Stderr)
1673 if containerUUID == "" {
1674 log.Printf("usage: %s [options] UUID", prog)
1678 log.Printf("crunch-run %s started", cmd.Version.String())
1681 if *caCertsPath != "" {
1682 arvadosclient.CertFiles = []string{*caCertsPath}
1685 api, err := arvadosclient.MakeArvadosClient()
1687 log.Printf("%s: %v", containerUUID, err)
1692 kc, kcerr := keepclient.MakeKeepClient(api)
1694 log.Printf("%s: %v", containerUUID, kcerr)
1697 kc.BlockCache = &keepclient.BlockCache{MaxBlocks: 2}
1700 cr, err := NewContainerRunner(arvados.NewClientFromEnv(), api, kc, containerUUID)
1706 switch *runtimeEngine {
1708 cr.executor, err = newDockerExecutor(containerUUID, cr.CrunchLog.Printf, cr.containerWatchdogInterval)
1710 cr.executor, err = newSingularityExecutor(cr.CrunchLog.Printf)
1712 cr.CrunchLog.Printf("%s: unsupported RuntimeEngine %q", containerUUID, *runtimeEngine)
1713 cr.CrunchLog.Close()
1717 cr.CrunchLog.Printf("%s: %v", containerUUID, err)
1718 cr.checkBrokenNode(err)
1719 cr.CrunchLog.Close()
1722 defer cr.executor.Close()
1724 gwAuthSecret := os.Getenv("GatewayAuthSecret")
1725 os.Unsetenv("GatewayAuthSecret")
1726 if gwAuthSecret == "" {
1727 // not safe to run a gateway service without an auth
1729 cr.CrunchLog.Printf("Not starting a gateway server (GatewayAuthSecret was not provided by dispatcher)")
1730 } else if gwListen := os.Getenv("GatewayAddress"); gwListen == "" {
1731 // dispatcher did not tell us which external IP
1732 // address to advertise --> no gateway service
1733 cr.CrunchLog.Printf("Not starting a gateway server (GatewayAddress was not provided by dispatcher)")
1734 } else if de, ok := cr.executor.(*dockerExecutor); ok {
1735 cr.gateway = Gateway{
1737 AuthSecret: gwAuthSecret,
1738 ContainerUUID: containerUUID,
1739 DockerContainerID: &de.containerID,
1741 ContainerIPAddress: dockerContainerIPAddress(&de.containerID),
1743 err = cr.gateway.Start()
1745 log.Printf("error starting gateway server: %s", err)
1750 parentTemp, tmperr := cr.MkTempDir("", "crunch-run."+containerUUID+".")
1752 log.Printf("%s: %v", containerUUID, tmperr)
1756 cr.parentTemp = parentTemp
1757 cr.statInterval = *statInterval
1758 cr.cgroupRoot = *cgroupRoot
1759 cr.expectCgroupParent = *cgroupParent
1760 cr.enableMemoryLimit = *enableMemoryLimit
1761 cr.enableNetwork = *enableNetwork
1762 cr.networkMode = *networkMode
1763 if *cgroupParentSubsystem != "" {
1764 p := findCgroup(*cgroupParentSubsystem)
1765 cr.setCgroupParent = p
1766 cr.expectCgroupParent = p
1771 if *memprofile != "" {
1772 f, err := os.Create(*memprofile)
1774 log.Printf("could not create memory profile: %s", err)
1776 runtime.GC() // get up-to-date statistics
1777 if err := pprof.WriteHeapProfile(f); err != nil {
1778 log.Printf("could not write memory profile: %s", err)
1780 closeerr := f.Close()
1781 if closeerr != nil {
1782 log.Printf("closing memprofile file: %s", err)
1787 log.Printf("%s: %v", containerUUID, runerr)
1793 func loadEnv(rdr io.Reader) error {
1794 buf, err := ioutil.ReadAll(rdr)
1796 return fmt.Errorf("read stdin: %s", err)
1798 var env map[string]string
1799 err = json.Unmarshal(buf, &env)
1801 return fmt.Errorf("decode stdin: %s", err)
1803 for k, v := range env {
1804 err = os.Setenv(k, v)
1806 return fmt.Errorf("setenv(%q): %s", k, err)