1 // Copyright (C) The Arvados Authors. All rights reserved.
3 // SPDX-License-Identifier: AGPL-3.0
30 "git.curoverse.com/arvados.git/lib/crunchstat"
31 "git.curoverse.com/arvados.git/sdk/go/arvados"
32 "git.curoverse.com/arvados.git/sdk/go/arvadosclient"
33 "git.curoverse.com/arvados.git/sdk/go/keepclient"
34 "git.curoverse.com/arvados.git/sdk/go/manifest"
35 "github.com/shirou/gopsutil/process"
36 "golang.org/x/net/context"
38 dockertypes "github.com/docker/docker/api/types"
39 dockercontainer "github.com/docker/docker/api/types/container"
40 dockernetwork "github.com/docker/docker/api/types/network"
41 dockerclient "github.com/docker/docker/client"
46 // IArvadosClient is the minimal Arvados API methods used by crunch-run.
47 type IArvadosClient interface {
48 Create(resourceType string, parameters arvadosclient.Dict, output interface{}) error
49 Get(resourceType string, uuid string, parameters arvadosclient.Dict, output interface{}) error
50 Update(resourceType string, uuid string, parameters arvadosclient.Dict, output interface{}) error
51 Call(method, resourceType, uuid, action string, parameters arvadosclient.Dict, output interface{}) error
52 CallRaw(method string, resourceType string, uuid string, action string, parameters arvadosclient.Dict) (reader io.ReadCloser, err error)
53 Discovery(key string) (interface{}, error)
56 // ErrCancelled is the error returned when the container is cancelled.
57 var ErrCancelled = errors.New("Cancelled")
59 // IKeepClient is the minimal Keep API methods used by crunch-run.
60 type IKeepClient interface {
61 PutB(buf []byte) (string, int, error)
62 ReadAt(locator string, p []byte, off int) (int, error)
63 ManifestFileReader(m manifest.Manifest, filename string) (arvados.File, error)
67 // NewLogWriter is a factory function to create a new log writer.
68 type NewLogWriter func(name string) (io.WriteCloser, error)
70 type RunArvMount func(args []string, tok string) (*exec.Cmd, error)
72 type MkTempDir func(string, string) (string, error)
74 // ThinDockerClient is the minimal Docker client interface used by crunch-run.
75 type ThinDockerClient interface {
76 ContainerAttach(ctx context.Context, container string, options dockertypes.ContainerAttachOptions) (dockertypes.HijackedResponse, error)
77 ContainerCreate(ctx context.Context, config *dockercontainer.Config, hostConfig *dockercontainer.HostConfig,
78 networkingConfig *dockernetwork.NetworkingConfig, containerName string) (dockercontainer.ContainerCreateCreatedBody, error)
79 ContainerStart(ctx context.Context, container string, options dockertypes.ContainerStartOptions) error
80 ContainerRemove(ctx context.Context, container string, options dockertypes.ContainerRemoveOptions) error
81 ContainerWait(ctx context.Context, container string, condition dockercontainer.WaitCondition) (<-chan dockercontainer.ContainerWaitOKBody, <-chan error)
82 ImageInspectWithRaw(ctx context.Context, image string) (dockertypes.ImageInspect, []byte, error)
83 ImageLoad(ctx context.Context, input io.Reader, quiet bool) (dockertypes.ImageLoadResponse, error)
84 ImageRemove(ctx context.Context, image string, options dockertypes.ImageRemoveOptions) ([]dockertypes.ImageDeleteResponseItem, error)
87 type PsProcess interface {
88 CmdlineSlice() ([]string, error)
91 // ContainerRunner is the main stateful struct used for a single execution of a
93 type ContainerRunner struct {
94 Docker ThinDockerClient
95 client *arvados.Client
96 ArvClient IArvadosClient
99 ContainerConfig dockercontainer.Config
100 dockercontainer.HostConfig
105 loggingDone chan bool
106 CrunchLog *ThrottledLogger
107 Stdout io.WriteCloser
108 Stderr io.WriteCloser
109 LogCollection arvados.CollectionFileSystem
117 Volumes map[string]struct{}
119 SigChan chan os.Signal
120 ArvMountExit chan error
121 SecretMounts map[string]arvados.Mount
122 MkArvClient func(token string) (IArvadosClient, error)
126 ListProcesses func() ([]PsProcess, error)
128 statLogger io.WriteCloser
129 statReporter *crunchstat.Reporter
130 hoststatLogger io.WriteCloser
131 hoststatReporter *crunchstat.Reporter
132 statInterval time.Duration
134 // What we expect the container's cgroup parent to be.
135 expectCgroupParent string
136 // What we tell docker to use as the container's cgroup
137 // parent. Note: Ideally we would use the same field for both
138 // expectCgroupParent and setCgroupParent, and just make it
139 // default to "docker". However, when using docker < 1.10 with
140 // systemd, specifying a non-empty cgroup parent (even the
141 // default value "docker") hits a docker bug
142 // (https://github.com/docker/docker/issues/17126). Using two
143 // separate fields makes it possible to use the "expect cgroup
144 // parent to be X" feature even on sites where the "specify
145 // cgroup parent" feature breaks.
146 setCgroupParent string
148 cStateLock sync.Mutex
149 cCancelled bool // StopContainer() invoked
151 enableNetwork string // one of "default" or "always"
152 networkMode string // passed through to HostConfig.NetworkMode
153 arvMountLog *ThrottledLogger
154 checkContainerd time.Duration
157 // setupSignals sets up signal handling to gracefully terminate the underlying
158 // Docker container and update state when receiving a TERM, INT or QUIT signal.
159 func (runner *ContainerRunner) setupSignals() {
160 runner.SigChan = make(chan os.Signal, 1)
161 signal.Notify(runner.SigChan, syscall.SIGTERM)
162 signal.Notify(runner.SigChan, syscall.SIGINT)
163 signal.Notify(runner.SigChan, syscall.SIGQUIT)
165 go func(sig chan os.Signal) {
172 // stop the underlying Docker container.
173 func (runner *ContainerRunner) stop(sig os.Signal) {
174 runner.cStateLock.Lock()
175 defer runner.cStateLock.Unlock()
177 runner.CrunchLog.Printf("caught signal: %v", sig)
179 if runner.ContainerID == "" {
182 runner.cCancelled = true
183 runner.CrunchLog.Printf("removing container")
184 err := runner.Docker.ContainerRemove(context.TODO(), runner.ContainerID, dockertypes.ContainerRemoveOptions{Force: true})
186 runner.CrunchLog.Printf("error removing container: %s", err)
190 var errorBlacklist = []string{
191 "(?ms).*[Cc]annot connect to the Docker daemon.*",
192 "(?ms).*oci runtime error.*starting container process.*container init.*mounting.*to rootfs.*no such file or directory.*",
194 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)")
196 func (runner *ContainerRunner) runBrokenNodeHook() {
197 if *brokenNodeHook == "" {
198 runner.CrunchLog.Printf("No broken node hook provided, cannot mark node as broken.")
200 runner.CrunchLog.Printf("Running broken node hook %q", *brokenNodeHook)
202 c := exec.Command(*brokenNodeHook)
203 c.Stdout = runner.CrunchLog
204 c.Stderr = runner.CrunchLog
207 runner.CrunchLog.Printf("Error running broken node hook: %v", err)
212 func (runner *ContainerRunner) checkBrokenNode(goterr error) bool {
213 for _, d := range errorBlacklist {
214 if m, e := regexp.MatchString(d, goterr.Error()); m && e == nil {
215 runner.CrunchLog.Printf("Error suggests node is unable to run containers: %v", goterr)
216 runner.runBrokenNodeHook()
223 // LoadImage determines the docker image id from the container record and
224 // checks if it is available in the local Docker image store. If not, it loads
225 // the image from Keep.
226 func (runner *ContainerRunner) LoadImage() (err error) {
228 runner.CrunchLog.Printf("Fetching Docker image from collection '%s'", runner.Container.ContainerImage)
230 var collection arvados.Collection
231 err = runner.ArvClient.Get("collections", runner.Container.ContainerImage, nil, &collection)
233 return fmt.Errorf("While getting container image collection: %v", err)
235 manifest := manifest.Manifest{Text: collection.ManifestText}
236 var img, imageID string
237 for ms := range manifest.StreamIter() {
238 img = ms.FileStreamSegments[0].Name
239 if !strings.HasSuffix(img, ".tar") {
240 return fmt.Errorf("First file in the container image collection does not end in .tar")
242 imageID = img[:len(img)-4]
245 runner.CrunchLog.Printf("Using Docker image id '%s'", imageID)
247 _, _, err = runner.Docker.ImageInspectWithRaw(context.TODO(), imageID)
249 runner.CrunchLog.Print("Loading Docker image from keep")
251 var readCloser io.ReadCloser
252 readCloser, err = runner.Kc.ManifestFileReader(manifest, img)
254 return fmt.Errorf("While creating ManifestFileReader for container image: %v", err)
257 response, err := runner.Docker.ImageLoad(context.TODO(), readCloser, true)
259 return fmt.Errorf("While loading container image into Docker: %v", err)
262 defer response.Body.Close()
263 rbody, err := ioutil.ReadAll(response.Body)
265 return fmt.Errorf("Reading response to image load: %v", err)
267 runner.CrunchLog.Printf("Docker response: %s", rbody)
269 runner.CrunchLog.Print("Docker image is available")
272 runner.ContainerConfig.Image = imageID
274 runner.Kc.ClearBlockCache()
279 func (runner *ContainerRunner) ArvMountCmd(arvMountCmd []string, token string) (c *exec.Cmd, err error) {
280 c = exec.Command("arv-mount", arvMountCmd...)
282 // Copy our environment, but override ARVADOS_API_TOKEN with
283 // the container auth token.
285 for _, s := range os.Environ() {
286 if !strings.HasPrefix(s, "ARVADOS_API_TOKEN=") {
287 c.Env = append(c.Env, s)
290 c.Env = append(c.Env, "ARVADOS_API_TOKEN="+token)
292 w, err := runner.NewLogWriter("arv-mount")
296 runner.arvMountLog = NewThrottledLogger(w)
297 c.Stdout = runner.arvMountLog
298 c.Stderr = runner.arvMountLog
300 runner.CrunchLog.Printf("Running %v", c.Args)
307 statReadme := make(chan bool)
308 runner.ArvMountExit = make(chan error)
313 time.Sleep(100 * time.Millisecond)
314 _, err = os.Stat(fmt.Sprintf("%s/by_id/README", runner.ArvMountPoint))
326 runner.CrunchLog.Printf("Arv-mount exit error: %v", mnterr)
328 runner.ArvMountExit <- mnterr
329 close(runner.ArvMountExit)
335 case err := <-runner.ArvMountExit:
336 runner.ArvMount = nil
344 func (runner *ContainerRunner) SetupArvMountPoint(prefix string) (err error) {
345 if runner.ArvMountPoint == "" {
346 runner.ArvMountPoint, err = runner.MkTempDir(runner.parentTemp, prefix)
351 func copyfile(src string, dst string) (err error) {
352 srcfile, err := os.Open(src)
357 os.MkdirAll(path.Dir(dst), 0777)
359 dstfile, err := os.Create(dst)
363 _, err = io.Copy(dstfile, srcfile)
368 err = srcfile.Close()
369 err2 := dstfile.Close()
382 func (runner *ContainerRunner) SetupMounts() (err error) {
383 err = runner.SetupArvMountPoint("keep")
385 return fmt.Errorf("While creating keep mount temp dir: %v", err)
388 token, err := runner.ContainerToken()
390 return fmt.Errorf("could not get container token: %s", err)
395 arvMountCmd := []string{
399 fmt.Sprintf("--crunchstat-interval=%v", runner.statInterval.Seconds())}
401 if runner.Container.RuntimeConstraints.KeepCacheRAM > 0 {
402 arvMountCmd = append(arvMountCmd, "--file-cache", fmt.Sprintf("%d", runner.Container.RuntimeConstraints.KeepCacheRAM))
405 collectionPaths := []string{}
407 runner.Volumes = make(map[string]struct{})
408 needCertMount := true
409 type copyFile struct {
413 var copyFiles []copyFile
416 for bind := range runner.Container.Mounts {
417 binds = append(binds, bind)
419 for bind := range runner.SecretMounts {
420 if _, ok := runner.Container.Mounts[bind]; ok {
421 return fmt.Errorf("Secret mount %q conflicts with regular mount", bind)
423 if runner.SecretMounts[bind].Kind != "json" &&
424 runner.SecretMounts[bind].Kind != "text" {
425 return fmt.Errorf("Secret mount %q type is %q but only 'json' and 'text' are permitted.",
426 bind, runner.SecretMounts[bind].Kind)
428 binds = append(binds, bind)
432 for _, bind := range binds {
433 mnt, ok := runner.Container.Mounts[bind]
435 mnt = runner.SecretMounts[bind]
437 if bind == "stdout" || bind == "stderr" {
438 // Is it a "file" mount kind?
439 if mnt.Kind != "file" {
440 return fmt.Errorf("Unsupported mount kind '%s' for %s. Only 'file' is supported.", mnt.Kind, bind)
443 // Does path start with OutputPath?
444 prefix := runner.Container.OutputPath
445 if !strings.HasSuffix(prefix, "/") {
448 if !strings.HasPrefix(mnt.Path, prefix) {
449 return fmt.Errorf("%s path does not start with OutputPath: %s, %s", strings.Title(bind), mnt.Path, prefix)
454 // Is it a "collection" mount kind?
455 if mnt.Kind != "collection" && mnt.Kind != "json" {
456 return fmt.Errorf("Unsupported mount kind '%s' for stdin. Only 'collection' or 'json' are supported.", mnt.Kind)
460 if bind == "/etc/arvados/ca-certificates.crt" {
461 needCertMount = false
464 if strings.HasPrefix(bind, runner.Container.OutputPath+"/") && bind != runner.Container.OutputPath+"/" {
465 if mnt.Kind != "collection" && mnt.Kind != "text" && mnt.Kind != "json" {
466 return fmt.Errorf("Only mount points of kind 'collection', 'text' or 'json' are supported underneath the output_path for %q, was %q", bind, mnt.Kind)
471 case mnt.Kind == "collection" && bind != "stdin":
473 if mnt.UUID != "" && mnt.PortableDataHash != "" {
474 return fmt.Errorf("Cannot specify both 'uuid' and 'portable_data_hash' for a collection mount")
478 return fmt.Errorf("Writing to existing collections currently not permitted.")
481 src = fmt.Sprintf("%s/by_id/%s", runner.ArvMountPoint, mnt.UUID)
482 } else if mnt.PortableDataHash != "" {
483 if mnt.Writable && !strings.HasPrefix(bind, runner.Container.OutputPath+"/") {
484 return fmt.Errorf("Can never write to a collection specified by portable data hash")
486 idx := strings.Index(mnt.PortableDataHash, "/")
488 mnt.Path = path.Clean(mnt.PortableDataHash[idx:])
489 mnt.PortableDataHash = mnt.PortableDataHash[0:idx]
490 runner.Container.Mounts[bind] = mnt
492 src = fmt.Sprintf("%s/by_id/%s", runner.ArvMountPoint, mnt.PortableDataHash)
493 if mnt.Path != "" && mnt.Path != "." {
494 if strings.HasPrefix(mnt.Path, "./") {
495 mnt.Path = mnt.Path[2:]
496 } else if strings.HasPrefix(mnt.Path, "/") {
497 mnt.Path = mnt.Path[1:]
499 src += "/" + mnt.Path
502 src = fmt.Sprintf("%s/tmp%d", runner.ArvMountPoint, tmpcount)
503 arvMountCmd = append(arvMountCmd, "--mount-tmp")
504 arvMountCmd = append(arvMountCmd, fmt.Sprintf("tmp%d", tmpcount))
508 if bind == runner.Container.OutputPath {
509 runner.HostOutputDir = src
510 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s", src, bind))
511 } else if strings.HasPrefix(bind, runner.Container.OutputPath+"/") {
512 copyFiles = append(copyFiles, copyFile{src, runner.HostOutputDir + bind[len(runner.Container.OutputPath):]})
514 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s", src, bind))
517 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s:ro", src, bind))
519 collectionPaths = append(collectionPaths, src)
521 case mnt.Kind == "tmp":
523 tmpdir, err = runner.MkTempDir(runner.parentTemp, "tmp")
525 return fmt.Errorf("While creating mount temp dir: %v", err)
527 st, staterr := os.Stat(tmpdir)
529 return fmt.Errorf("While Stat on temp dir: %v", staterr)
531 err = os.Chmod(tmpdir, st.Mode()|os.ModeSetgid|0777)
533 return fmt.Errorf("While Chmod temp dir: %v", err)
535 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s", tmpdir, bind))
536 if bind == runner.Container.OutputPath {
537 runner.HostOutputDir = tmpdir
540 case mnt.Kind == "json" || mnt.Kind == "text":
542 if mnt.Kind == "json" {
543 filedata, err = json.Marshal(mnt.Content)
545 return fmt.Errorf("encoding json data: %v", err)
548 text, ok := mnt.Content.(string)
550 return fmt.Errorf("content for mount %q must be a string", bind)
552 filedata = []byte(text)
555 tmpdir, err := runner.MkTempDir(runner.parentTemp, mnt.Kind)
557 return fmt.Errorf("creating temp dir: %v", err)
559 tmpfn := filepath.Join(tmpdir, "mountdata."+mnt.Kind)
560 err = ioutil.WriteFile(tmpfn, filedata, 0444)
562 return fmt.Errorf("writing temp file: %v", err)
564 if strings.HasPrefix(bind, runner.Container.OutputPath+"/") {
565 copyFiles = append(copyFiles, copyFile{tmpfn, runner.HostOutputDir + bind[len(runner.Container.OutputPath):]})
567 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s:ro", tmpfn, bind))
570 case mnt.Kind == "git_tree":
571 tmpdir, err := runner.MkTempDir(runner.parentTemp, "git_tree")
573 return fmt.Errorf("creating temp dir: %v", err)
575 err = gitMount(mnt).extractTree(runner.ArvClient, tmpdir, token)
579 runner.Binds = append(runner.Binds, tmpdir+":"+bind+":ro")
583 if runner.HostOutputDir == "" {
584 return fmt.Errorf("Output path does not correspond to a writable mount point")
587 if wantAPI := runner.Container.RuntimeConstraints.API; needCertMount && wantAPI != nil && *wantAPI {
588 for _, certfile := range arvadosclient.CertFiles {
589 _, err := os.Stat(certfile)
591 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:/etc/arvados/ca-certificates.crt:ro", certfile))
598 arvMountCmd = append(arvMountCmd, "--mount-by-pdh", "by_id")
600 arvMountCmd = append(arvMountCmd, "--mount-by-id", "by_id")
602 arvMountCmd = append(arvMountCmd, runner.ArvMountPoint)
604 runner.ArvMount, err = runner.RunArvMount(arvMountCmd, token)
606 return fmt.Errorf("While trying to start arv-mount: %v", err)
609 for _, p := range collectionPaths {
612 return fmt.Errorf("While checking that input files exist: %v", err)
616 for _, cp := range copyFiles {
617 st, err := os.Stat(cp.src)
619 return fmt.Errorf("While staging writable file from %q to %q: %v", cp.src, cp.bind, err)
622 err = filepath.Walk(cp.src, func(walkpath string, walkinfo os.FileInfo, walkerr error) error {
626 target := path.Join(cp.bind, walkpath[len(cp.src):])
627 if walkinfo.Mode().IsRegular() {
628 copyerr := copyfile(walkpath, target)
632 return os.Chmod(target, walkinfo.Mode()|0777)
633 } else if walkinfo.Mode().IsDir() {
634 mkerr := os.MkdirAll(target, 0777)
638 return os.Chmod(target, walkinfo.Mode()|os.ModeSetgid|0777)
640 return fmt.Errorf("Source %q is not a regular file or directory", cp.src)
643 } else if st.Mode().IsRegular() {
644 err = copyfile(cp.src, cp.bind)
646 err = os.Chmod(cp.bind, st.Mode()|0777)
650 return fmt.Errorf("While staging writable file from %q to %q: %v", cp.src, cp.bind, err)
657 func (runner *ContainerRunner) ProcessDockerAttach(containerReader io.Reader) {
658 // Handle docker log protocol
659 // https://docs.docker.com/engine/reference/api/docker_remote_api_v1.15/#attach-to-a-container
660 defer close(runner.loggingDone)
662 header := make([]byte, 8)
665 _, err = io.ReadAtLeast(containerReader, header, 8)
672 readsize := int64(header[7]) | (int64(header[6]) << 8) | (int64(header[5]) << 16) | (int64(header[4]) << 24)
675 _, err = io.CopyN(runner.Stdout, containerReader, readsize)
678 _, err = io.CopyN(runner.Stderr, containerReader, readsize)
683 runner.CrunchLog.Printf("error reading docker logs: %v", err)
686 err = runner.Stdout.Close()
688 runner.CrunchLog.Printf("error closing stdout logs: %v", err)
691 err = runner.Stderr.Close()
693 runner.CrunchLog.Printf("error closing stderr logs: %v", err)
696 if runner.statReporter != nil {
697 runner.statReporter.Stop()
698 err = runner.statLogger.Close()
700 runner.CrunchLog.Printf("error closing crunchstat logs: %v", err)
705 func (runner *ContainerRunner) stopHoststat() error {
706 if runner.hoststatReporter == nil {
709 runner.hoststatReporter.Stop()
710 err := runner.hoststatLogger.Close()
712 return fmt.Errorf("error closing hoststat logs: %v", err)
717 func (runner *ContainerRunner) startHoststat() error {
718 w, err := runner.NewLogWriter("hoststat")
722 runner.hoststatLogger = NewThrottledLogger(w)
723 runner.hoststatReporter = &crunchstat.Reporter{
724 Logger: log.New(runner.hoststatLogger, "", 0),
725 CgroupRoot: runner.cgroupRoot,
726 PollPeriod: runner.statInterval,
728 runner.hoststatReporter.Start()
732 func (runner *ContainerRunner) startCrunchstat() error {
733 w, err := runner.NewLogWriter("crunchstat")
737 runner.statLogger = NewThrottledLogger(w)
738 runner.statReporter = &crunchstat.Reporter{
739 CID: runner.ContainerID,
740 Logger: log.New(runner.statLogger, "", 0),
741 CgroupParent: runner.expectCgroupParent,
742 CgroupRoot: runner.cgroupRoot,
743 PollPeriod: runner.statInterval,
745 runner.statReporter.Start()
749 type infoCommand struct {
754 // LogHostInfo logs info about the current host, for debugging and
755 // accounting purposes. Although it's logged as "node-info", this is
756 // about the environment where crunch-run is actually running, which
757 // might differ from what's described in the node record (see
759 func (runner *ContainerRunner) LogHostInfo() (err error) {
760 w, err := runner.NewLogWriter("node-info")
765 commands := []infoCommand{
767 label: "Host Information",
768 cmd: []string{"uname", "-a"},
771 label: "CPU Information",
772 cmd: []string{"cat", "/proc/cpuinfo"},
775 label: "Memory Information",
776 cmd: []string{"cat", "/proc/meminfo"},
780 cmd: []string{"df", "-m", "/", os.TempDir()},
783 label: "Disk INodes",
784 cmd: []string{"df", "-i", "/", os.TempDir()},
788 // Run commands with informational output to be logged.
789 for _, command := range commands {
790 fmt.Fprintln(w, command.label)
791 cmd := exec.Command(command.cmd[0], command.cmd[1:]...)
794 if err := cmd.Run(); err != nil {
795 err = fmt.Errorf("While running command %q: %v", command.cmd, err)
804 return fmt.Errorf("While closing node-info logs: %v", err)
809 // LogContainerRecord gets and saves the raw JSON container record from the API server
810 func (runner *ContainerRunner) LogContainerRecord() error {
811 logged, err := runner.logAPIResponse("container", "containers", map[string]interface{}{"filters": [][]string{{"uuid", "=", runner.Container.UUID}}}, nil)
812 if !logged && err == nil {
813 err = fmt.Errorf("error: no container record found for %s", runner.Container.UUID)
818 // LogNodeRecord logs arvados#node record corresponding to the current host.
819 func (runner *ContainerRunner) LogNodeRecord() error {
820 hostname := os.Getenv("SLURMD_NODENAME")
822 hostname, _ = os.Hostname()
824 _, err := runner.logAPIResponse("node", "nodes", map[string]interface{}{"filters": [][]string{{"hostname", "=", hostname}}}, func(resp interface{}) {
825 // The "info" field has admin-only info when obtained
826 // with a privileged token, and should not be logged.
827 node, ok := resp.(map[string]interface{})
835 func (runner *ContainerRunner) logAPIResponse(label, path string, params map[string]interface{}, munge func(interface{})) (logged bool, err error) {
836 writer, err := runner.LogCollection.OpenFile(label+".json", os.O_CREATE|os.O_WRONLY, 0666)
841 ArvClient: runner.ArvClient,
842 UUID: runner.Container.UUID,
843 loggingStream: label,
847 reader, err := runner.ArvClient.CallRaw("GET", path, "", "", arvadosclient.Dict(params))
849 return false, fmt.Errorf("error getting %s record: %v", label, err)
853 dec := json.NewDecoder(reader)
855 var resp map[string]interface{}
856 if err = dec.Decode(&resp); err != nil {
857 return false, fmt.Errorf("error decoding %s list response: %v", label, err)
859 items, ok := resp["items"].([]interface{})
861 return false, fmt.Errorf("error decoding %s list response: no \"items\" key in API list response", label)
862 } else if len(items) < 1 {
868 // Re-encode it using indentation to improve readability
869 enc := json.NewEncoder(w)
870 enc.SetIndent("", " ")
871 if err = enc.Encode(items[0]); err != nil {
872 return false, fmt.Errorf("error logging %s record: %v", label, err)
876 return false, fmt.Errorf("error closing %s.json in log collection: %v", label, err)
881 // AttachStreams connects the docker container stdin, stdout and stderr logs
882 // to the Arvados logger which logs to Keep and the API server logs table.
883 func (runner *ContainerRunner) AttachStreams() (err error) {
885 runner.CrunchLog.Print("Attaching container streams")
887 // If stdin mount is provided, attach it to the docker container
888 var stdinRdr arvados.File
890 if stdinMnt, ok := runner.Container.Mounts["stdin"]; ok {
891 if stdinMnt.Kind == "collection" {
892 var stdinColl arvados.Collection
893 collId := stdinMnt.UUID
895 collId = stdinMnt.PortableDataHash
897 err = runner.ArvClient.Get("collections", collId, nil, &stdinColl)
899 return fmt.Errorf("While getting stding collection: %v", err)
902 stdinRdr, err = runner.Kc.ManifestFileReader(manifest.Manifest{Text: stdinColl.ManifestText}, stdinMnt.Path)
903 if os.IsNotExist(err) {
904 return fmt.Errorf("stdin collection path not found: %v", stdinMnt.Path)
905 } else if err != nil {
906 return fmt.Errorf("While getting stdin collection path %v: %v", stdinMnt.Path, err)
908 } else if stdinMnt.Kind == "json" {
909 stdinJson, err = json.Marshal(stdinMnt.Content)
911 return fmt.Errorf("While encoding stdin json data: %v", err)
916 stdinUsed := stdinRdr != nil || len(stdinJson) != 0
917 response, err := runner.Docker.ContainerAttach(context.TODO(), runner.ContainerID,
918 dockertypes.ContainerAttachOptions{Stream: true, Stdin: stdinUsed, Stdout: true, Stderr: true})
920 return fmt.Errorf("While attaching container stdout/stderr streams: %v", err)
923 runner.loggingDone = make(chan bool)
925 if stdoutMnt, ok := runner.Container.Mounts["stdout"]; ok {
926 stdoutFile, err := runner.getStdoutFile(stdoutMnt.Path)
930 runner.Stdout = stdoutFile
931 } else if w, err := runner.NewLogWriter("stdout"); err != nil {
934 runner.Stdout = NewThrottledLogger(w)
937 if stderrMnt, ok := runner.Container.Mounts["stderr"]; ok {
938 stderrFile, err := runner.getStdoutFile(stderrMnt.Path)
942 runner.Stderr = stderrFile
943 } else if w, err := runner.NewLogWriter("stderr"); err != nil {
946 runner.Stderr = NewThrottledLogger(w)
951 _, err := io.Copy(response.Conn, stdinRdr)
953 runner.CrunchLog.Print("While writing stdin collection to docker container %q", err)
957 response.CloseWrite()
959 } else if len(stdinJson) != 0 {
961 _, err := io.Copy(response.Conn, bytes.NewReader(stdinJson))
963 runner.CrunchLog.Print("While writing stdin json to docker container %q", err)
966 response.CloseWrite()
970 go runner.ProcessDockerAttach(response.Reader)
975 func (runner *ContainerRunner) getStdoutFile(mntPath string) (*os.File, error) {
976 stdoutPath := mntPath[len(runner.Container.OutputPath):]
977 index := strings.LastIndex(stdoutPath, "/")
979 subdirs := stdoutPath[:index]
981 st, err := os.Stat(runner.HostOutputDir)
983 return nil, fmt.Errorf("While Stat on temp dir: %v", err)
985 stdoutPath := filepath.Join(runner.HostOutputDir, subdirs)
986 err = os.MkdirAll(stdoutPath, st.Mode()|os.ModeSetgid|0777)
988 return nil, fmt.Errorf("While MkdirAll %q: %v", stdoutPath, err)
992 stdoutFile, err := os.Create(filepath.Join(runner.HostOutputDir, stdoutPath))
994 return nil, fmt.Errorf("While creating file %q: %v", stdoutPath, err)
997 return stdoutFile, nil
1000 // CreateContainer creates the docker container.
1001 func (runner *ContainerRunner) CreateContainer() error {
1002 runner.CrunchLog.Print("Creating Docker container")
1004 runner.ContainerConfig.Cmd = runner.Container.Command
1005 if runner.Container.Cwd != "." {
1006 runner.ContainerConfig.WorkingDir = runner.Container.Cwd
1009 for k, v := range runner.Container.Environment {
1010 runner.ContainerConfig.Env = append(runner.ContainerConfig.Env, k+"="+v)
1013 runner.ContainerConfig.Volumes = runner.Volumes
1015 maxRAM := int64(runner.Container.RuntimeConstraints.RAM)
1016 if maxRAM < 4*1024*1024 {
1017 // Docker daemon won't let you set a limit less than 4 MiB
1018 maxRAM = 4 * 1024 * 1024
1020 runner.HostConfig = dockercontainer.HostConfig{
1021 Binds: runner.Binds,
1022 LogConfig: dockercontainer.LogConfig{
1025 Resources: dockercontainer.Resources{
1026 CgroupParent: runner.setCgroupParent,
1027 NanoCPUs: int64(runner.Container.RuntimeConstraints.VCPUs) * 1000000000,
1028 Memory: maxRAM, // RAM
1029 MemorySwap: maxRAM, // RAM+swap
1030 KernelMemory: maxRAM, // kernel portion
1034 if wantAPI := runner.Container.RuntimeConstraints.API; wantAPI != nil && *wantAPI {
1035 tok, err := runner.ContainerToken()
1039 runner.ContainerConfig.Env = append(runner.ContainerConfig.Env,
1040 "ARVADOS_API_TOKEN="+tok,
1041 "ARVADOS_API_HOST="+os.Getenv("ARVADOS_API_HOST"),
1042 "ARVADOS_API_HOST_INSECURE="+os.Getenv("ARVADOS_API_HOST_INSECURE"),
1044 runner.HostConfig.NetworkMode = dockercontainer.NetworkMode(runner.networkMode)
1046 if runner.enableNetwork == "always" {
1047 runner.HostConfig.NetworkMode = dockercontainer.NetworkMode(runner.networkMode)
1049 runner.HostConfig.NetworkMode = dockercontainer.NetworkMode("none")
1053 _, stdinUsed := runner.Container.Mounts["stdin"]
1054 runner.ContainerConfig.OpenStdin = stdinUsed
1055 runner.ContainerConfig.StdinOnce = stdinUsed
1056 runner.ContainerConfig.AttachStdin = stdinUsed
1057 runner.ContainerConfig.AttachStdout = true
1058 runner.ContainerConfig.AttachStderr = true
1060 createdBody, err := runner.Docker.ContainerCreate(context.TODO(), &runner.ContainerConfig, &runner.HostConfig, nil, runner.Container.UUID)
1062 return fmt.Errorf("While creating container: %v", err)
1065 runner.ContainerID = createdBody.ID
1067 return runner.AttachStreams()
1070 // StartContainer starts the docker container created by CreateContainer.
1071 func (runner *ContainerRunner) StartContainer() error {
1072 runner.CrunchLog.Printf("Starting Docker container id '%s'", runner.ContainerID)
1073 runner.cStateLock.Lock()
1074 defer runner.cStateLock.Unlock()
1075 if runner.cCancelled {
1078 err := runner.Docker.ContainerStart(context.TODO(), runner.ContainerID,
1079 dockertypes.ContainerStartOptions{})
1082 if m, e := regexp.MatchString("(?ms).*(exec|System error).*(no such file or directory|file not found).*", err.Error()); m && e == nil {
1083 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])
1085 return fmt.Errorf("could not start container: %v%s", err, advice)
1090 // checkContainerd checks if "containerd" is present in the process list.
1091 func (runner *ContainerRunner) CheckContainerd() error {
1092 if runner.checkContainerd == 0 {
1095 p, _ := runner.ListProcesses()
1096 for _, i := range p {
1097 e, _ := i.CmdlineSlice()
1099 if strings.Index(e[0], "containerd") > -1 {
1106 runner.runBrokenNodeHook()
1108 return fmt.Errorf("'containerd' not found in process list.")
1111 // WaitFinish waits for the container to terminate, capture the exit code, and
1112 // close the stdout/stderr logging.
1113 func (runner *ContainerRunner) WaitFinish() error {
1114 var runTimeExceeded <-chan time.Time
1115 runner.CrunchLog.Print("Waiting for container to finish")
1117 waitOk, waitErr := runner.Docker.ContainerWait(context.TODO(), runner.ContainerID, dockercontainer.WaitConditionNotRunning)
1118 arvMountExit := runner.ArvMountExit
1119 if timeout := runner.Container.SchedulingParameters.MaxRunTime; timeout > 0 {
1120 runTimeExceeded = time.After(time.Duration(timeout) * time.Second)
1123 containerdGone := make(chan error)
1125 close(containerdGone)
1127 if runner.checkContainerd > 0 {
1129 ticker := time.NewTicker(time.Duration(runner.checkContainerd))
1134 if ck := runner.CheckContainerd(); ck != nil {
1135 containerdGone <- ck
1138 case <-containerdGone:
1147 case waitBody := <-waitOk:
1148 runner.CrunchLog.Printf("Container exited with code: %v", waitBody.StatusCode)
1149 code := int(waitBody.StatusCode)
1150 runner.ExitCode = &code
1152 // wait for stdout/stderr to complete
1153 <-runner.loggingDone
1156 case err := <-waitErr:
1157 return fmt.Errorf("container wait: %v", err)
1159 case <-arvMountExit:
1160 runner.CrunchLog.Printf("arv-mount exited while container is still running. Stopping container.")
1162 // arvMountExit will always be ready now that
1163 // it's closed, but that doesn't interest us.
1166 case <-runTimeExceeded:
1167 runner.CrunchLog.Printf("maximum run time exceeded. Stopping container.")
1169 runTimeExceeded = nil
1171 case err := <-containerdGone:
1177 // CaptureOutput saves data from the container's output directory if
1178 // needed, and updates the container output accordingly.
1179 func (runner *ContainerRunner) CaptureOutput() error {
1180 if wantAPI := runner.Container.RuntimeConstraints.API; wantAPI != nil && *wantAPI {
1181 // Output may have been set directly by the container, so
1182 // refresh the container record to check.
1183 err := runner.ArvClient.Get("containers", runner.Container.UUID,
1184 nil, &runner.Container)
1188 if runner.Container.Output != "" {
1189 // Container output is already set.
1190 runner.OutputPDH = &runner.Container.Output
1195 txt, err := (&copier{
1196 client: runner.client,
1197 arvClient: runner.ArvClient,
1198 keepClient: runner.Kc,
1199 hostOutputDir: runner.HostOutputDir,
1200 ctrOutputDir: runner.Container.OutputPath,
1201 binds: runner.Binds,
1202 mounts: runner.Container.Mounts,
1203 secretMounts: runner.SecretMounts,
1204 logger: runner.CrunchLog,
1209 var resp arvados.Collection
1210 err = runner.ArvClient.Create("collections", arvadosclient.Dict{
1211 "ensure_unique_name": true,
1212 "collection": arvadosclient.Dict{
1214 "name": "output for " + runner.Container.UUID,
1215 "manifest_text": txt,
1219 return fmt.Errorf("error creating output collection: %v", err)
1221 runner.OutputPDH = &resp.PortableDataHash
1225 func (runner *ContainerRunner) CleanupDirs() {
1226 if runner.ArvMount != nil {
1228 umount := exec.Command("arv-mount", fmt.Sprintf("--unmount-timeout=%d", delay), "--unmount", runner.ArvMountPoint)
1229 umount.Stdout = runner.CrunchLog
1230 umount.Stderr = runner.CrunchLog
1231 runner.CrunchLog.Printf("Running %v", umount.Args)
1232 umnterr := umount.Start()
1235 runner.CrunchLog.Printf("Error unmounting: %v", umnterr)
1237 // If arv-mount --unmount gets stuck for any reason, we
1238 // don't want to wait for it forever. Do Wait() in a goroutine
1239 // so it doesn't block crunch-run.
1240 umountExit := make(chan error)
1242 mnterr := umount.Wait()
1244 runner.CrunchLog.Printf("Error unmounting: %v", mnterr)
1246 umountExit <- mnterr
1249 for again := true; again; {
1255 case <-runner.ArvMountExit:
1257 case <-time.After(time.Duration((delay + 1) * int64(time.Second))):
1258 runner.CrunchLog.Printf("Timed out waiting for unmount")
1260 umount.Process.Kill()
1262 runner.ArvMount.Process.Kill()
1268 if runner.ArvMountPoint != "" {
1269 if rmerr := os.Remove(runner.ArvMountPoint); rmerr != nil {
1270 runner.CrunchLog.Printf("While cleaning up arv-mount directory %s: %v", runner.ArvMountPoint, rmerr)
1274 if rmerr := os.RemoveAll(runner.parentTemp); rmerr != nil {
1275 runner.CrunchLog.Printf("While cleaning up temporary directory %s: %v", runner.parentTemp, rmerr)
1279 // CommitLogs posts the collection containing the final container logs.
1280 func (runner *ContainerRunner) CommitLogs() error {
1282 // Hold cStateLock to prevent races on CrunchLog (e.g., stop()).
1283 runner.cStateLock.Lock()
1284 defer runner.cStateLock.Unlock()
1286 runner.CrunchLog.Print(runner.finalState)
1288 if runner.arvMountLog != nil {
1289 runner.arvMountLog.Close()
1291 runner.CrunchLog.Close()
1293 // Closing CrunchLog above allows them to be committed to Keep at this
1294 // point, but re-open crunch log with ArvClient in case there are any
1295 // other further errors (such as failing to write the log to Keep!)
1296 // while shutting down
1297 runner.CrunchLog = NewThrottledLogger(&ArvLogWriter{
1298 ArvClient: runner.ArvClient,
1299 UUID: runner.Container.UUID,
1300 loggingStream: "crunch-run",
1303 runner.CrunchLog.Immediate = log.New(os.Stderr, runner.Container.UUID+" ", 0)
1306 if runner.LogsPDH != nil {
1307 // If we have already assigned something to LogsPDH,
1308 // we must be closing the re-opened log, which won't
1309 // end up getting attached to the container record and
1310 // therefore doesn't need to be saved as a collection
1311 // -- it exists only to send logs to other channels.
1315 mt, err := runner.LogCollection.MarshalManifest(".")
1317 return fmt.Errorf("While creating log manifest: %v", err)
1320 var response arvados.Collection
1321 err = runner.ArvClient.Create("collections",
1323 "ensure_unique_name": true,
1324 "collection": arvadosclient.Dict{
1326 "name": "logs for " + runner.Container.UUID,
1327 "manifest_text": mt}},
1330 return fmt.Errorf("While creating log collection: %v", err)
1332 runner.LogsPDH = &response.PortableDataHash
1336 // UpdateContainerRunning updates the container state to "Running"
1337 func (runner *ContainerRunner) UpdateContainerRunning() error {
1338 runner.cStateLock.Lock()
1339 defer runner.cStateLock.Unlock()
1340 if runner.cCancelled {
1343 return runner.ArvClient.Update("containers", runner.Container.UUID,
1344 arvadosclient.Dict{"container": arvadosclient.Dict{"state": "Running"}}, nil)
1347 // ContainerToken returns the api_token the container (and any
1348 // arv-mount processes) are allowed to use.
1349 func (runner *ContainerRunner) ContainerToken() (string, error) {
1350 if runner.token != "" {
1351 return runner.token, nil
1354 var auth arvados.APIClientAuthorization
1355 err := runner.ArvClient.Call("GET", "containers", runner.Container.UUID, "auth", nil, &auth)
1359 runner.token = auth.APIToken
1360 return runner.token, nil
1363 // UpdateContainerComplete updates the container record state on API
1364 // server to "Complete" or "Cancelled"
1365 func (runner *ContainerRunner) UpdateContainerFinal() error {
1366 update := arvadosclient.Dict{}
1367 update["state"] = runner.finalState
1368 if runner.LogsPDH != nil {
1369 update["log"] = *runner.LogsPDH
1371 if runner.finalState == "Complete" {
1372 if runner.ExitCode != nil {
1373 update["exit_code"] = *runner.ExitCode
1375 if runner.OutputPDH != nil {
1376 update["output"] = *runner.OutputPDH
1379 return runner.ArvClient.Update("containers", runner.Container.UUID, arvadosclient.Dict{"container": update}, nil)
1382 // IsCancelled returns the value of Cancelled, with goroutine safety.
1383 func (runner *ContainerRunner) IsCancelled() bool {
1384 runner.cStateLock.Lock()
1385 defer runner.cStateLock.Unlock()
1386 return runner.cCancelled
1389 // NewArvLogWriter creates an ArvLogWriter
1390 func (runner *ContainerRunner) NewArvLogWriter(name string) (io.WriteCloser, error) {
1391 writer, err := runner.LogCollection.OpenFile(name+".txt", os.O_CREATE|os.O_WRONLY, 0666)
1395 return &ArvLogWriter{
1396 ArvClient: runner.ArvClient,
1397 UUID: runner.Container.UUID,
1398 loggingStream: name,
1399 writeCloser: writer,
1403 // Run the full container lifecycle.
1404 func (runner *ContainerRunner) Run() (err error) {
1405 runner.CrunchLog.Printf("crunch-run %s started", version)
1406 runner.CrunchLog.Printf("Executing container '%s'", runner.Container.UUID)
1408 hostname, hosterr := os.Hostname()
1410 runner.CrunchLog.Printf("Error getting hostname '%v'", hosterr)
1412 runner.CrunchLog.Printf("Executing on host '%s'", hostname)
1415 runner.finalState = "Queued"
1418 runner.CleanupDirs()
1420 runner.CrunchLog.Printf("crunch-run finished")
1421 runner.CrunchLog.Close()
1425 // checkErr prints e (unless it's nil) and sets err to
1426 // e (unless err is already non-nil). Thus, if err
1427 // hasn't already been assigned when Run() returns,
1428 // this cleanup func will cause Run() to return the
1429 // first non-nil error that is passed to checkErr().
1430 checkErr := func(e error) {
1434 runner.CrunchLog.Print(e)
1438 if runner.finalState == "Complete" {
1439 // There was an error in the finalization.
1440 runner.finalState = "Cancelled"
1444 // Log the error encountered in Run(), if any
1447 if runner.finalState == "Queued" {
1448 runner.UpdateContainerFinal()
1452 if runner.IsCancelled() {
1453 runner.finalState = "Cancelled"
1454 // but don't return yet -- we still want to
1455 // capture partial output and write logs
1458 checkErr(runner.CaptureOutput())
1459 checkErr(runner.stopHoststat())
1460 checkErr(runner.CommitLogs())
1461 checkErr(runner.UpdateContainerFinal())
1464 err = runner.fetchContainerRecord()
1468 runner.setupSignals()
1469 err = runner.startHoststat()
1474 // Sanity check that containerd is running.
1475 err = runner.CheckContainerd()
1480 // check for and/or load image
1481 err = runner.LoadImage()
1483 if !runner.checkBrokenNode(err) {
1484 // Failed to load image but not due to a "broken node"
1485 // condition, probably user error.
1486 runner.finalState = "Cancelled"
1488 err = fmt.Errorf("While loading container image: %v", err)
1492 // set up FUSE mount and binds
1493 err = runner.SetupMounts()
1495 runner.finalState = "Cancelled"
1496 err = fmt.Errorf("While setting up mounts: %v", err)
1500 err = runner.CreateContainer()
1504 err = runner.LogHostInfo()
1508 err = runner.LogNodeRecord()
1512 err = runner.LogContainerRecord()
1517 if runner.IsCancelled() {
1521 err = runner.UpdateContainerRunning()
1525 runner.finalState = "Cancelled"
1527 err = runner.startCrunchstat()
1532 err = runner.StartContainer()
1534 runner.checkBrokenNode(err)
1538 err = runner.WaitFinish()
1539 if err == nil && !runner.IsCancelled() {
1540 runner.finalState = "Complete"
1545 // Fetch the current container record (uuid = runner.Container.UUID)
1546 // into runner.Container.
1547 func (runner *ContainerRunner) fetchContainerRecord() error {
1548 reader, err := runner.ArvClient.CallRaw("GET", "containers", runner.Container.UUID, "", nil)
1550 return fmt.Errorf("error fetching container record: %v", err)
1552 defer reader.Close()
1554 dec := json.NewDecoder(reader)
1556 err = dec.Decode(&runner.Container)
1558 return fmt.Errorf("error decoding container record: %v", err)
1562 SecretMounts map[string]arvados.Mount `json:"secret_mounts"`
1565 containerToken, err := runner.ContainerToken()
1567 return fmt.Errorf("error getting container token: %v", err)
1570 containerClient, err := runner.MkArvClient(containerToken)
1572 return fmt.Errorf("error creating container API client: %v", err)
1575 err = containerClient.Call("GET", "containers", runner.Container.UUID, "secret_mounts", nil, &sm)
1577 if apierr, ok := err.(arvadosclient.APIServerError); !ok || apierr.HttpStatusCode != 404 {
1578 return fmt.Errorf("error fetching secret_mounts: %v", err)
1580 // ok && apierr.HttpStatusCode == 404, which means
1581 // secret_mounts isn't supported by this API server.
1583 runner.SecretMounts = sm.SecretMounts
1588 // NewContainerRunner creates a new container runner.
1589 func NewContainerRunner(client *arvados.Client, api IArvadosClient, kc IKeepClient, docker ThinDockerClient, containerUUID string) (*ContainerRunner, error) {
1590 cr := &ContainerRunner{
1596 cr.NewLogWriter = cr.NewArvLogWriter
1597 cr.RunArvMount = cr.ArvMountCmd
1598 cr.MkTempDir = ioutil.TempDir
1599 cr.ListProcesses = func() ([]PsProcess, error) {
1600 pr, err := process.Processes()
1604 ps := make([]PsProcess, len(pr))
1605 for i, j := range pr {
1610 cr.MkArvClient = func(token string) (IArvadosClient, error) {
1611 cl, err := arvadosclient.MakeArvadosClient()
1619 cr.LogCollection, err = (&arvados.Collection{}).FileSystem(cr.client, cr.Kc)
1623 cr.Container.UUID = containerUUID
1624 w, err := cr.NewLogWriter("crunch-run")
1628 cr.CrunchLog = NewThrottledLogger(w)
1629 cr.CrunchLog.Immediate = log.New(os.Stderr, containerUUID+" ", 0)
1631 loadLogThrottleParams(api)
1637 statInterval := flag.Duration("crunchstat-interval", 10*time.Second, "sampling period for periodic resource usage reporting")
1638 cgroupRoot := flag.String("cgroup-root", "/sys/fs/cgroup", "path to sysfs cgroup tree")
1639 cgroupParent := flag.String("cgroup-parent", "docker", "name of container's parent cgroup (ignored if -cgroup-parent-subsystem is used)")
1640 cgroupParentSubsystem := flag.String("cgroup-parent-subsystem", "", "use current cgroup for given subsystem as parent cgroup for container")
1641 caCertsPath := flag.String("ca-certs", "", "Path to TLS root certificates")
1642 enableNetwork := flag.String("container-enable-networking", "default",
1643 `Specify if networking should be enabled for container. One of 'default', 'always':
1644 default: only enable networking if container requests it.
1645 always: containers always have networking enabled
1647 networkMode := flag.String("container-network-mode", "default",
1648 `Set networking mode for container. Corresponds to Docker network mode (--net).
1650 memprofile := flag.String("memprofile", "", "write memory profile to `file` after running container")
1651 getVersion := flag.Bool("version", false, "Print version information and exit.")
1652 checkContainerd := flag.Duration("check-containerd", 60*time.Second, "Periodic check if (docker-)containerd is running, period in seconds.")
1655 // Print version information if requested
1657 fmt.Printf("crunch-run %s\n", version)
1661 log.Printf("crunch-run %s started", version)
1663 containerId := flag.Arg(0)
1665 if *caCertsPath != "" {
1666 arvadosclient.CertFiles = []string{*caCertsPath}
1669 api, err := arvadosclient.MakeArvadosClient()
1671 log.Fatalf("%s: %v", containerId, err)
1675 kc, kcerr := keepclient.MakeKeepClient(api)
1677 log.Fatalf("%s: %v", containerId, kcerr)
1679 kc.BlockCache = &keepclient.BlockCache{MaxBlocks: 2}
1682 // API version 1.21 corresponds to Docker 1.9, which is currently the
1683 // minimum version we want to support.
1684 docker, dockererr := dockerclient.NewClient(dockerclient.DefaultDockerHost, "1.21", nil, nil)
1686 cr, err := NewContainerRunner(arvados.NewClientFromEnv(), api, kc, docker, containerId)
1690 if dockererr != nil {
1691 cr.CrunchLog.Printf("%s: %v", containerId, dockererr)
1692 cr.checkBrokenNode(dockererr)
1693 cr.CrunchLog.Close()
1697 parentTemp, tmperr := cr.MkTempDir("", "crunch-run."+containerId+".")
1699 log.Fatalf("%s: %v", containerId, tmperr)
1702 cr.parentTemp = parentTemp
1703 cr.statInterval = *statInterval
1704 cr.cgroupRoot = *cgroupRoot
1705 cr.expectCgroupParent = *cgroupParent
1706 cr.enableNetwork = *enableNetwork
1707 cr.networkMode = *networkMode
1708 cr.checkContainerd = *checkContainerd
1709 if *cgroupParentSubsystem != "" {
1710 p := findCgroup(*cgroupParentSubsystem)
1711 cr.setCgroupParent = p
1712 cr.expectCgroupParent = p
1717 if *memprofile != "" {
1718 f, err := os.Create(*memprofile)
1720 log.Printf("could not create memory profile: %s", err)
1722 runtime.GC() // get up-to-date statistics
1723 if err := pprof.WriteHeapProfile(f); err != nil {
1724 log.Printf("could not write memory profile: %s", err)
1726 closeerr := f.Close()
1727 if closeerr != nil {
1728 log.Printf("closing memprofile file: %s", err)
1733 log.Fatalf("%s: %v", containerId, runerr)