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 "golang.org/x/net/context"
37 dockertypes "github.com/docker/docker/api/types"
38 dockercontainer "github.com/docker/docker/api/types/container"
39 dockernetwork "github.com/docker/docker/api/types/network"
40 dockerclient "github.com/docker/docker/client"
45 // IArvadosClient is the minimal Arvados API methods used by crunch-run.
46 type IArvadosClient interface {
47 Create(resourceType string, parameters arvadosclient.Dict, output interface{}) error
48 Get(resourceType string, uuid string, parameters arvadosclient.Dict, output interface{}) error
49 Update(resourceType string, uuid string, parameters arvadosclient.Dict, output interface{}) error
50 Call(method, resourceType, uuid, action string, parameters arvadosclient.Dict, output interface{}) error
51 CallRaw(method string, resourceType string, uuid string, action string, parameters arvadosclient.Dict) (reader io.ReadCloser, err error)
52 Discovery(key string) (interface{}, error)
55 // ErrCancelled is the error returned when the container is cancelled.
56 var ErrCancelled = errors.New("Cancelled")
58 // IKeepClient is the minimal Keep API methods used by crunch-run.
59 type IKeepClient interface {
60 PutHB(hash string, buf []byte) (string, int, error)
61 ManifestFileReader(m manifest.Manifest, filename string) (arvados.File, error)
65 // NewLogWriter is a factory function to create a new log writer.
66 type NewLogWriter func(name string) io.WriteCloser
68 type RunArvMount func(args []string, tok string) (*exec.Cmd, error)
70 type MkTempDir func(string, string) (string, error)
72 // ThinDockerClient is the minimal Docker client interface used by crunch-run.
73 type ThinDockerClient interface {
74 ContainerAttach(ctx context.Context, container string, options dockertypes.ContainerAttachOptions) (dockertypes.HijackedResponse, error)
75 ContainerCreate(ctx context.Context, config *dockercontainer.Config, hostConfig *dockercontainer.HostConfig,
76 networkingConfig *dockernetwork.NetworkingConfig, containerName string) (dockercontainer.ContainerCreateCreatedBody, error)
77 ContainerStart(ctx context.Context, container string, options dockertypes.ContainerStartOptions) error
78 ContainerRemove(ctx context.Context, container string, options dockertypes.ContainerRemoveOptions) error
79 ContainerWait(ctx context.Context, container string, condition dockercontainer.WaitCondition) (<-chan dockercontainer.ContainerWaitOKBody, <-chan error)
80 ImageInspectWithRaw(ctx context.Context, image string) (dockertypes.ImageInspect, []byte, error)
81 ImageLoad(ctx context.Context, input io.Reader, quiet bool) (dockertypes.ImageLoadResponse, error)
82 ImageRemove(ctx context.Context, image string, options dockertypes.ImageRemoveOptions) ([]dockertypes.ImageDeleteResponseItem, error)
85 // ContainerRunner is the main stateful struct used for a single execution of a
87 type ContainerRunner struct {
88 Docker ThinDockerClient
89 ArvClient IArvadosClient
92 ContainerConfig dockercontainer.Config
93 dockercontainer.HostConfig
99 CrunchLog *ThrottledLogger
100 Stdout io.WriteCloser
101 Stderr io.WriteCloser
102 LogCollection *CollectionWriter
110 Volumes map[string]struct{}
112 SigChan chan os.Signal
113 ArvMountExit chan error
114 SecretMounts map[string]arvados.Mount
115 MkArvClient func(token string) (IArvadosClient, error)
119 statLogger io.WriteCloser
120 statReporter *crunchstat.Reporter
121 hoststatLogger io.WriteCloser
122 hoststatReporter *crunchstat.Reporter
123 statInterval time.Duration
125 // What we expect the container's cgroup parent to be.
126 expectCgroupParent string
127 // What we tell docker to use as the container's cgroup
128 // parent. Note: Ideally we would use the same field for both
129 // expectCgroupParent and setCgroupParent, and just make it
130 // default to "docker". However, when using docker < 1.10 with
131 // systemd, specifying a non-empty cgroup parent (even the
132 // default value "docker") hits a docker bug
133 // (https://github.com/docker/docker/issues/17126). Using two
134 // separate fields makes it possible to use the "expect cgroup
135 // parent to be X" feature even on sites where the "specify
136 // cgroup parent" feature breaks.
137 setCgroupParent string
139 cStateLock sync.Mutex
140 cCancelled bool // StopContainer() invoked
142 enableNetwork string // one of "default" or "always"
143 networkMode string // passed through to HostConfig.NetworkMode
144 arvMountLog *ThrottledLogger
147 // setupSignals sets up signal handling to gracefully terminate the underlying
148 // Docker container and update state when receiving a TERM, INT or QUIT signal.
149 func (runner *ContainerRunner) setupSignals() {
150 runner.SigChan = make(chan os.Signal, 1)
151 signal.Notify(runner.SigChan, syscall.SIGTERM)
152 signal.Notify(runner.SigChan, syscall.SIGINT)
153 signal.Notify(runner.SigChan, syscall.SIGQUIT)
155 go func(sig chan os.Signal) {
162 // stop the underlying Docker container.
163 func (runner *ContainerRunner) stop(sig os.Signal) {
164 runner.cStateLock.Lock()
165 defer runner.cStateLock.Unlock()
167 runner.CrunchLog.Printf("caught signal: %v", sig)
169 if runner.ContainerID == "" {
172 runner.cCancelled = true
173 runner.CrunchLog.Printf("removing container")
174 err := runner.Docker.ContainerRemove(context.TODO(), runner.ContainerID, dockertypes.ContainerRemoveOptions{Force: true})
176 runner.CrunchLog.Printf("error removing container: %s", err)
180 var errorBlacklist = []string{
181 "(?ms).*[Cc]annot connect to the Docker daemon.*",
182 "(?ms).*oci runtime error.*starting container process.*container init.*mounting.*to rootfs.*no such file or directory.*",
184 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)")
186 func (runner *ContainerRunner) checkBrokenNode(goterr error) bool {
187 for _, d := range errorBlacklist {
188 if m, e := regexp.MatchString(d, goterr.Error()); m && e == nil {
189 runner.CrunchLog.Printf("Error suggests node is unable to run containers: %v", goterr)
190 if *brokenNodeHook == "" {
191 runner.CrunchLog.Printf("No broken node hook provided, cannot mark node as broken.")
193 runner.CrunchLog.Printf("Running broken node hook %q", *brokenNodeHook)
195 c := exec.Command(*brokenNodeHook)
196 c.Stdout = runner.CrunchLog
197 c.Stderr = runner.CrunchLog
200 runner.CrunchLog.Printf("Error running broken node hook: %v", err)
209 // LoadImage determines the docker image id from the container record and
210 // checks if it is available in the local Docker image store. If not, it loads
211 // the image from Keep.
212 func (runner *ContainerRunner) LoadImage() (err error) {
214 runner.CrunchLog.Printf("Fetching Docker image from collection '%s'", runner.Container.ContainerImage)
216 var collection arvados.Collection
217 err = runner.ArvClient.Get("collections", runner.Container.ContainerImage, nil, &collection)
219 return fmt.Errorf("While getting container image collection: %v", err)
221 manifest := manifest.Manifest{Text: collection.ManifestText}
222 var img, imageID string
223 for ms := range manifest.StreamIter() {
224 img = ms.FileStreamSegments[0].Name
225 if !strings.HasSuffix(img, ".tar") {
226 return fmt.Errorf("First file in the container image collection does not end in .tar")
228 imageID = img[:len(img)-4]
231 runner.CrunchLog.Printf("Using Docker image id '%s'", imageID)
233 _, _, err = runner.Docker.ImageInspectWithRaw(context.TODO(), imageID)
235 runner.CrunchLog.Print("Loading Docker image from keep")
237 var readCloser io.ReadCloser
238 readCloser, err = runner.Kc.ManifestFileReader(manifest, img)
240 return fmt.Errorf("While creating ManifestFileReader for container image: %v", err)
243 response, err := runner.Docker.ImageLoad(context.TODO(), readCloser, true)
245 return fmt.Errorf("While loading container image into Docker: %v", err)
248 defer response.Body.Close()
249 rbody, err := ioutil.ReadAll(response.Body)
251 return fmt.Errorf("Reading response to image load: %v", err)
253 runner.CrunchLog.Printf("Docker response: %s", rbody)
255 runner.CrunchLog.Print("Docker image is available")
258 runner.ContainerConfig.Image = imageID
260 runner.Kc.ClearBlockCache()
265 func (runner *ContainerRunner) ArvMountCmd(arvMountCmd []string, token string) (c *exec.Cmd, err error) {
266 c = exec.Command("arv-mount", arvMountCmd...)
268 // Copy our environment, but override ARVADOS_API_TOKEN with
269 // the container auth token.
271 for _, s := range os.Environ() {
272 if !strings.HasPrefix(s, "ARVADOS_API_TOKEN=") {
273 c.Env = append(c.Env, s)
276 c.Env = append(c.Env, "ARVADOS_API_TOKEN="+token)
278 runner.arvMountLog = NewThrottledLogger(runner.NewLogWriter("arv-mount"))
279 c.Stdout = runner.arvMountLog
280 c.Stderr = runner.arvMountLog
282 runner.CrunchLog.Printf("Running %v", c.Args)
289 statReadme := make(chan bool)
290 runner.ArvMountExit = make(chan error)
295 time.Sleep(100 * time.Millisecond)
296 _, err = os.Stat(fmt.Sprintf("%s/by_id/README", runner.ArvMountPoint))
308 runner.CrunchLog.Printf("Arv-mount exit error: %v", mnterr)
310 runner.ArvMountExit <- mnterr
311 close(runner.ArvMountExit)
317 case err := <-runner.ArvMountExit:
318 runner.ArvMount = nil
326 func (runner *ContainerRunner) SetupArvMountPoint(prefix string) (err error) {
327 if runner.ArvMountPoint == "" {
328 runner.ArvMountPoint, err = runner.MkTempDir(runner.parentTemp, prefix)
333 func copyfile(src string, dst string) (err error) {
334 srcfile, err := os.Open(src)
339 os.MkdirAll(path.Dir(dst), 0777)
341 dstfile, err := os.Create(dst)
345 _, err = io.Copy(dstfile, srcfile)
350 err = srcfile.Close()
351 err2 := dstfile.Close()
364 func (runner *ContainerRunner) SetupMounts() (err error) {
365 err = runner.SetupArvMountPoint("keep")
367 return fmt.Errorf("While creating keep mount temp dir: %v", err)
370 token, err := runner.ContainerToken()
372 return fmt.Errorf("could not get container token: %s", err)
377 arvMountCmd := []string{
381 fmt.Sprintf("--crunchstat-interval=%v", runner.statInterval.Seconds())}
383 if runner.Container.RuntimeConstraints.KeepCacheRAM > 0 {
384 arvMountCmd = append(arvMountCmd, "--file-cache", fmt.Sprintf("%d", runner.Container.RuntimeConstraints.KeepCacheRAM))
387 collectionPaths := []string{}
389 runner.Volumes = make(map[string]struct{})
390 needCertMount := true
391 type copyFile struct {
395 var copyFiles []copyFile
398 for bind := range runner.Container.Mounts {
399 binds = append(binds, bind)
401 for bind := range runner.SecretMounts {
402 if runner.SecretMounts[bind].Kind != "json" &&
403 runner.SecretMounts[bind].Kind != "text" {
404 return fmt.Errorf("Secret mount %q type is %q but only 'json' and 'text' are permitted.",
405 bind, runner.SecretMounts[bind].Kind)
407 binds = append(binds, bind)
411 for _, bind := range binds {
412 mnt, ok := runner.Container.Mounts[bind]
414 mnt = runner.SecretMounts[bind]
416 if bind == "stdout" || bind == "stderr" {
417 // Is it a "file" mount kind?
418 if mnt.Kind != "file" {
419 return fmt.Errorf("Unsupported mount kind '%s' for %s. Only 'file' is supported.", mnt.Kind, bind)
422 // Does path start with OutputPath?
423 prefix := runner.Container.OutputPath
424 if !strings.HasSuffix(prefix, "/") {
427 if !strings.HasPrefix(mnt.Path, prefix) {
428 return fmt.Errorf("%s path does not start with OutputPath: %s, %s", strings.Title(bind), mnt.Path, prefix)
433 // Is it a "collection" mount kind?
434 if mnt.Kind != "collection" && mnt.Kind != "json" {
435 return fmt.Errorf("Unsupported mount kind '%s' for stdin. Only 'collection' or 'json' are supported.", mnt.Kind)
439 if bind == "/etc/arvados/ca-certificates.crt" {
440 needCertMount = false
443 if strings.HasPrefix(bind, runner.Container.OutputPath+"/") && bind != runner.Container.OutputPath+"/" {
444 if mnt.Kind != "collection" && mnt.Kind != "text" && mnt.Kind != "json" {
445 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)
450 case mnt.Kind == "collection" && bind != "stdin":
452 if mnt.UUID != "" && mnt.PortableDataHash != "" {
453 return fmt.Errorf("Cannot specify both 'uuid' and 'portable_data_hash' for a collection mount")
457 return fmt.Errorf("Writing to existing collections currently not permitted.")
460 src = fmt.Sprintf("%s/by_id/%s", runner.ArvMountPoint, mnt.UUID)
461 } else if mnt.PortableDataHash != "" {
462 if mnt.Writable && !strings.HasPrefix(bind, runner.Container.OutputPath+"/") {
463 return fmt.Errorf("Can never write to a collection specified by portable data hash")
465 idx := strings.Index(mnt.PortableDataHash, "/")
467 mnt.Path = path.Clean(mnt.PortableDataHash[idx:])
468 mnt.PortableDataHash = mnt.PortableDataHash[0:idx]
469 runner.Container.Mounts[bind] = mnt
471 src = fmt.Sprintf("%s/by_id/%s", runner.ArvMountPoint, mnt.PortableDataHash)
472 if mnt.Path != "" && mnt.Path != "." {
473 if strings.HasPrefix(mnt.Path, "./") {
474 mnt.Path = mnt.Path[2:]
475 } else if strings.HasPrefix(mnt.Path, "/") {
476 mnt.Path = mnt.Path[1:]
478 src += "/" + mnt.Path
481 src = fmt.Sprintf("%s/tmp%d", runner.ArvMountPoint, tmpcount)
482 arvMountCmd = append(arvMountCmd, "--mount-tmp")
483 arvMountCmd = append(arvMountCmd, fmt.Sprintf("tmp%d", tmpcount))
487 if bind == runner.Container.OutputPath {
488 runner.HostOutputDir = src
489 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s", src, bind))
490 } else if strings.HasPrefix(bind, runner.Container.OutputPath+"/") {
491 copyFiles = append(copyFiles, copyFile{src, runner.HostOutputDir + bind[len(runner.Container.OutputPath):]})
493 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s", src, bind))
496 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s:ro", src, bind))
498 collectionPaths = append(collectionPaths, src)
500 case mnt.Kind == "tmp":
502 tmpdir, err = runner.MkTempDir(runner.parentTemp, "tmp")
504 return fmt.Errorf("While creating mount temp dir: %v", err)
506 st, staterr := os.Stat(tmpdir)
508 return fmt.Errorf("While Stat on temp dir: %v", staterr)
510 err = os.Chmod(tmpdir, st.Mode()|os.ModeSetgid|0777)
512 return fmt.Errorf("While Chmod temp dir: %v", err)
514 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s", tmpdir, bind))
515 if bind == runner.Container.OutputPath {
516 runner.HostOutputDir = tmpdir
519 case mnt.Kind == "json" || mnt.Kind == "text":
521 if mnt.Kind == "json" {
522 filedata, err = json.Marshal(mnt.Content)
524 return fmt.Errorf("encoding json data: %v", err)
527 text, ok := mnt.Content.(string)
529 return fmt.Errorf("content for mount %q must be a string", bind)
531 filedata = []byte(text)
534 tmpdir, err := runner.MkTempDir(runner.parentTemp, mnt.Kind)
536 return fmt.Errorf("creating temp dir: %v", err)
538 tmpfn := filepath.Join(tmpdir, "mountdata."+mnt.Kind)
539 err = ioutil.WriteFile(tmpfn, filedata, 0444)
541 return fmt.Errorf("writing temp file: %v", err)
543 if strings.HasPrefix(bind, runner.Container.OutputPath+"/") {
544 copyFiles = append(copyFiles, copyFile{tmpfn, runner.HostOutputDir + bind[len(runner.Container.OutputPath):]})
546 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s:ro", tmpfn, bind))
549 case mnt.Kind == "git_tree":
550 tmpdir, err := runner.MkTempDir(runner.parentTemp, "git_tree")
552 return fmt.Errorf("creating temp dir: %v", err)
554 err = gitMount(mnt).extractTree(runner.ArvClient, tmpdir, token)
558 runner.Binds = append(runner.Binds, tmpdir+":"+bind+":ro")
562 if runner.HostOutputDir == "" {
563 return fmt.Errorf("Output path does not correspond to a writable mount point")
566 if wantAPI := runner.Container.RuntimeConstraints.API; needCertMount && wantAPI != nil && *wantAPI {
567 for _, certfile := range arvadosclient.CertFiles {
568 _, err := os.Stat(certfile)
570 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:/etc/arvados/ca-certificates.crt:ro", certfile))
577 arvMountCmd = append(arvMountCmd, "--mount-by-pdh", "by_id")
579 arvMountCmd = append(arvMountCmd, "--mount-by-id", "by_id")
581 arvMountCmd = append(arvMountCmd, runner.ArvMountPoint)
583 runner.ArvMount, err = runner.RunArvMount(arvMountCmd, token)
585 return fmt.Errorf("While trying to start arv-mount: %v", err)
588 for _, p := range collectionPaths {
591 return fmt.Errorf("While checking that input files exist: %v", err)
595 for _, cp := range copyFiles {
596 st, err := os.Stat(cp.src)
598 return fmt.Errorf("While staging writable file from %q to %q: %v", cp.src, cp.bind, err)
601 err = filepath.Walk(cp.src, func(walkpath string, walkinfo os.FileInfo, walkerr error) error {
605 target := path.Join(cp.bind, walkpath[len(cp.src):])
606 if walkinfo.Mode().IsRegular() {
607 copyerr := copyfile(walkpath, target)
611 return os.Chmod(target, walkinfo.Mode()|0777)
612 } else if walkinfo.Mode().IsDir() {
613 mkerr := os.MkdirAll(target, 0777)
617 return os.Chmod(target, walkinfo.Mode()|os.ModeSetgid|0777)
619 return fmt.Errorf("Source %q is not a regular file or directory", cp.src)
622 } else if st.Mode().IsRegular() {
623 err = copyfile(cp.src, cp.bind)
625 err = os.Chmod(cp.bind, st.Mode()|0777)
629 return fmt.Errorf("While staging writable file from %q to %q: %v", cp.src, cp.bind, err)
636 func (runner *ContainerRunner) ProcessDockerAttach(containerReader io.Reader) {
637 // Handle docker log protocol
638 // https://docs.docker.com/engine/reference/api/docker_remote_api_v1.15/#attach-to-a-container
639 defer close(runner.loggingDone)
641 header := make([]byte, 8)
644 _, err = io.ReadAtLeast(containerReader, header, 8)
651 readsize := int64(header[7]) | (int64(header[6]) << 8) | (int64(header[5]) << 16) | (int64(header[4]) << 24)
654 _, err = io.CopyN(runner.Stdout, containerReader, readsize)
657 _, err = io.CopyN(runner.Stderr, containerReader, readsize)
662 runner.CrunchLog.Printf("error reading docker logs: %v", err)
665 err = runner.Stdout.Close()
667 runner.CrunchLog.Printf("error closing stdout logs: %v", err)
670 err = runner.Stderr.Close()
672 runner.CrunchLog.Printf("error closing stderr logs: %v", err)
675 if runner.statReporter != nil {
676 runner.statReporter.Stop()
677 err = runner.statLogger.Close()
679 runner.CrunchLog.Printf("error closing crunchstat logs: %v", err)
684 func (runner *ContainerRunner) stopHoststat() error {
685 if runner.hoststatReporter == nil {
688 runner.hoststatReporter.Stop()
689 err := runner.hoststatLogger.Close()
691 return fmt.Errorf("error closing hoststat logs: %v", err)
696 func (runner *ContainerRunner) startHoststat() {
697 runner.hoststatLogger = NewThrottledLogger(runner.NewLogWriter("hoststat"))
698 runner.hoststatReporter = &crunchstat.Reporter{
699 Logger: log.New(runner.hoststatLogger, "", 0),
700 CgroupRoot: runner.cgroupRoot,
701 PollPeriod: runner.statInterval,
703 runner.hoststatReporter.Start()
706 func (runner *ContainerRunner) startCrunchstat() {
707 runner.statLogger = NewThrottledLogger(runner.NewLogWriter("crunchstat"))
708 runner.statReporter = &crunchstat.Reporter{
709 CID: runner.ContainerID,
710 Logger: log.New(runner.statLogger, "", 0),
711 CgroupParent: runner.expectCgroupParent,
712 CgroupRoot: runner.cgroupRoot,
713 PollPeriod: runner.statInterval,
715 runner.statReporter.Start()
718 type infoCommand struct {
723 // LogHostInfo logs info about the current host, for debugging and
724 // accounting purposes. Although it's logged as "node-info", this is
725 // about the environment where crunch-run is actually running, which
726 // might differ from what's described in the node record (see
728 func (runner *ContainerRunner) LogHostInfo() (err error) {
729 w := runner.NewLogWriter("node-info")
731 commands := []infoCommand{
733 label: "Host Information",
734 cmd: []string{"uname", "-a"},
737 label: "CPU Information",
738 cmd: []string{"cat", "/proc/cpuinfo"},
741 label: "Memory Information",
742 cmd: []string{"cat", "/proc/meminfo"},
746 cmd: []string{"df", "-m", "/", os.TempDir()},
749 label: "Disk INodes",
750 cmd: []string{"df", "-i", "/", os.TempDir()},
754 // Run commands with informational output to be logged.
755 for _, command := range commands {
756 fmt.Fprintln(w, command.label)
757 cmd := exec.Command(command.cmd[0], command.cmd[1:]...)
760 if err := cmd.Run(); err != nil {
761 err = fmt.Errorf("While running command %q: %v", command.cmd, err)
770 return fmt.Errorf("While closing node-info logs: %v", err)
775 // LogContainerRecord gets and saves the raw JSON container record from the API server
776 func (runner *ContainerRunner) LogContainerRecord() error {
777 logged, err := runner.logAPIResponse("container", "containers", map[string]interface{}{"filters": [][]string{{"uuid", "=", runner.Container.UUID}}}, nil)
778 if !logged && err == nil {
779 err = fmt.Errorf("error: no container record found for %s", runner.Container.UUID)
784 // LogNodeRecord logs arvados#node record corresponding to the current host.
785 func (runner *ContainerRunner) LogNodeRecord() error {
786 hostname := os.Getenv("SLURMD_NODENAME")
788 hostname, _ = os.Hostname()
790 _, err := runner.logAPIResponse("node", "nodes", map[string]interface{}{"filters": [][]string{{"hostname", "=", hostname}}}, func(resp interface{}) {
791 // The "info" field has admin-only info when obtained
792 // with a privileged token, and should not be logged.
793 node, ok := resp.(map[string]interface{})
801 func (runner *ContainerRunner) logAPIResponse(label, path string, params map[string]interface{}, munge func(interface{})) (logged bool, err error) {
803 ArvClient: runner.ArvClient,
804 UUID: runner.Container.UUID,
805 loggingStream: label,
806 writeCloser: runner.LogCollection.Open(label + ".json"),
809 reader, err := runner.ArvClient.CallRaw("GET", path, "", "", arvadosclient.Dict(params))
811 return false, fmt.Errorf("error getting %s record: %v", label, err)
815 dec := json.NewDecoder(reader)
817 var resp map[string]interface{}
818 if err = dec.Decode(&resp); err != nil {
819 return false, fmt.Errorf("error decoding %s list response: %v", label, err)
821 items, ok := resp["items"].([]interface{})
823 return false, fmt.Errorf("error decoding %s list response: no \"items\" key in API list response", label)
824 } else if len(items) < 1 {
830 // Re-encode it using indentation to improve readability
831 enc := json.NewEncoder(w)
832 enc.SetIndent("", " ")
833 if err = enc.Encode(items[0]); err != nil {
834 return false, fmt.Errorf("error logging %s record: %v", label, err)
838 return false, fmt.Errorf("error closing %s.json in log collection: %v", label, err)
843 // AttachStreams connects the docker container stdin, stdout and stderr logs
844 // to the Arvados logger which logs to Keep and the API server logs table.
845 func (runner *ContainerRunner) AttachStreams() (err error) {
847 runner.CrunchLog.Print("Attaching container streams")
849 // If stdin mount is provided, attach it to the docker container
850 var stdinRdr arvados.File
852 if stdinMnt, ok := runner.Container.Mounts["stdin"]; ok {
853 if stdinMnt.Kind == "collection" {
854 var stdinColl arvados.Collection
855 collId := stdinMnt.UUID
857 collId = stdinMnt.PortableDataHash
859 err = runner.ArvClient.Get("collections", collId, nil, &stdinColl)
861 return fmt.Errorf("While getting stding collection: %v", err)
864 stdinRdr, err = runner.Kc.ManifestFileReader(manifest.Manifest{Text: stdinColl.ManifestText}, stdinMnt.Path)
865 if os.IsNotExist(err) {
866 return fmt.Errorf("stdin collection path not found: %v", stdinMnt.Path)
867 } else if err != nil {
868 return fmt.Errorf("While getting stdin collection path %v: %v", stdinMnt.Path, err)
870 } else if stdinMnt.Kind == "json" {
871 stdinJson, err = json.Marshal(stdinMnt.Content)
873 return fmt.Errorf("While encoding stdin json data: %v", err)
878 stdinUsed := stdinRdr != nil || len(stdinJson) != 0
879 response, err := runner.Docker.ContainerAttach(context.TODO(), runner.ContainerID,
880 dockertypes.ContainerAttachOptions{Stream: true, Stdin: stdinUsed, Stdout: true, Stderr: true})
882 return fmt.Errorf("While attaching container stdout/stderr streams: %v", err)
885 runner.loggingDone = make(chan bool)
887 if stdoutMnt, ok := runner.Container.Mounts["stdout"]; ok {
888 stdoutFile, err := runner.getStdoutFile(stdoutMnt.Path)
892 runner.Stdout = stdoutFile
894 runner.Stdout = NewThrottledLogger(runner.NewLogWriter("stdout"))
897 if stderrMnt, ok := runner.Container.Mounts["stderr"]; ok {
898 stderrFile, err := runner.getStdoutFile(stderrMnt.Path)
902 runner.Stderr = stderrFile
904 runner.Stderr = NewThrottledLogger(runner.NewLogWriter("stderr"))
909 _, err := io.Copy(response.Conn, stdinRdr)
911 runner.CrunchLog.Print("While writing stdin collection to docker container %q", err)
915 response.CloseWrite()
917 } else if len(stdinJson) != 0 {
919 _, err := io.Copy(response.Conn, bytes.NewReader(stdinJson))
921 runner.CrunchLog.Print("While writing stdin json to docker container %q", err)
924 response.CloseWrite()
928 go runner.ProcessDockerAttach(response.Reader)
933 func (runner *ContainerRunner) getStdoutFile(mntPath string) (*os.File, error) {
934 stdoutPath := mntPath[len(runner.Container.OutputPath):]
935 index := strings.LastIndex(stdoutPath, "/")
937 subdirs := stdoutPath[:index]
939 st, err := os.Stat(runner.HostOutputDir)
941 return nil, fmt.Errorf("While Stat on temp dir: %v", err)
943 stdoutPath := filepath.Join(runner.HostOutputDir, subdirs)
944 err = os.MkdirAll(stdoutPath, st.Mode()|os.ModeSetgid|0777)
946 return nil, fmt.Errorf("While MkdirAll %q: %v", stdoutPath, err)
950 stdoutFile, err := os.Create(filepath.Join(runner.HostOutputDir, stdoutPath))
952 return nil, fmt.Errorf("While creating file %q: %v", stdoutPath, err)
955 return stdoutFile, nil
958 // CreateContainer creates the docker container.
959 func (runner *ContainerRunner) CreateContainer() error {
960 runner.CrunchLog.Print("Creating Docker container")
962 runner.ContainerConfig.Cmd = runner.Container.Command
963 if runner.Container.Cwd != "." {
964 runner.ContainerConfig.WorkingDir = runner.Container.Cwd
967 for k, v := range runner.Container.Environment {
968 runner.ContainerConfig.Env = append(runner.ContainerConfig.Env, k+"="+v)
971 runner.ContainerConfig.Volumes = runner.Volumes
973 maxRAM := int64(runner.Container.RuntimeConstraints.RAM)
974 runner.HostConfig = dockercontainer.HostConfig{
976 LogConfig: dockercontainer.LogConfig{
979 Resources: dockercontainer.Resources{
980 CgroupParent: runner.setCgroupParent,
981 NanoCPUs: int64(runner.Container.RuntimeConstraints.VCPUs) * 1000000000,
982 Memory: maxRAM, // RAM
983 MemorySwap: maxRAM, // RAM+swap
984 KernelMemory: maxRAM, // kernel portion
988 if wantAPI := runner.Container.RuntimeConstraints.API; wantAPI != nil && *wantAPI {
989 tok, err := runner.ContainerToken()
993 runner.ContainerConfig.Env = append(runner.ContainerConfig.Env,
994 "ARVADOS_API_TOKEN="+tok,
995 "ARVADOS_API_HOST="+os.Getenv("ARVADOS_API_HOST"),
996 "ARVADOS_API_HOST_INSECURE="+os.Getenv("ARVADOS_API_HOST_INSECURE"),
998 runner.HostConfig.NetworkMode = dockercontainer.NetworkMode(runner.networkMode)
1000 if runner.enableNetwork == "always" {
1001 runner.HostConfig.NetworkMode = dockercontainer.NetworkMode(runner.networkMode)
1003 runner.HostConfig.NetworkMode = dockercontainer.NetworkMode("none")
1007 _, stdinUsed := runner.Container.Mounts["stdin"]
1008 runner.ContainerConfig.OpenStdin = stdinUsed
1009 runner.ContainerConfig.StdinOnce = stdinUsed
1010 runner.ContainerConfig.AttachStdin = stdinUsed
1011 runner.ContainerConfig.AttachStdout = true
1012 runner.ContainerConfig.AttachStderr = true
1014 createdBody, err := runner.Docker.ContainerCreate(context.TODO(), &runner.ContainerConfig, &runner.HostConfig, nil, runner.Container.UUID)
1016 return fmt.Errorf("While creating container: %v", err)
1019 runner.ContainerID = createdBody.ID
1021 return runner.AttachStreams()
1024 // StartContainer starts the docker container created by CreateContainer.
1025 func (runner *ContainerRunner) StartContainer() error {
1026 runner.CrunchLog.Printf("Starting Docker container id '%s'", runner.ContainerID)
1027 runner.cStateLock.Lock()
1028 defer runner.cStateLock.Unlock()
1029 if runner.cCancelled {
1032 err := runner.Docker.ContainerStart(context.TODO(), runner.ContainerID,
1033 dockertypes.ContainerStartOptions{})
1036 if m, e := regexp.MatchString("(?ms).*(exec|System error).*(no such file or directory|file not found).*", err.Error()); m && e == nil {
1037 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])
1039 return fmt.Errorf("could not start container: %v%s", err, advice)
1044 // WaitFinish waits for the container to terminate, capture the exit code, and
1045 // close the stdout/stderr logging.
1046 func (runner *ContainerRunner) WaitFinish() error {
1047 runner.CrunchLog.Print("Waiting for container to finish")
1049 waitOk, waitErr := runner.Docker.ContainerWait(context.TODO(), runner.ContainerID, dockercontainer.WaitConditionNotRunning)
1050 arvMountExit := runner.ArvMountExit
1053 case waitBody := <-waitOk:
1054 runner.CrunchLog.Printf("Container exited with code: %v", waitBody.StatusCode)
1055 code := int(waitBody.StatusCode)
1056 runner.ExitCode = &code
1058 // wait for stdout/stderr to complete
1059 <-runner.loggingDone
1062 case err := <-waitErr:
1063 return fmt.Errorf("container wait: %v", err)
1065 case <-arvMountExit:
1066 runner.CrunchLog.Printf("arv-mount exited while container is still running. Stopping container.")
1068 // arvMountExit will always be ready now that
1069 // it's closed, but that doesn't interest us.
1075 var ErrNotInOutputDir = fmt.Errorf("Must point to path within the output directory")
1077 func (runner *ContainerRunner) derefOutputSymlink(path string, startinfo os.FileInfo) (tgt string, readlinktgt string, info os.FileInfo, err error) {
1078 // Follow symlinks if necessary
1083 for followed := 0; info.Mode()&os.ModeSymlink != 0; followed++ {
1084 if followed >= limitFollowSymlinks {
1085 // Got stuck in a loop or just a pathological number of links, give up.
1086 err = fmt.Errorf("Followed more than %v symlinks from path %q", limitFollowSymlinks, path)
1090 readlinktgt, err = os.Readlink(nextlink)
1096 if !strings.HasPrefix(tgt, "/") {
1097 // Relative symlink, resolve it to host path
1098 tgt = filepath.Join(filepath.Dir(path), tgt)
1100 if strings.HasPrefix(tgt, runner.Container.OutputPath+"/") && !strings.HasPrefix(tgt, runner.HostOutputDir+"/") {
1101 // Absolute symlink to container output path, adjust it to host output path.
1102 tgt = filepath.Join(runner.HostOutputDir, tgt[len(runner.Container.OutputPath):])
1104 if !strings.HasPrefix(tgt, runner.HostOutputDir+"/") {
1105 // After dereferencing, symlink target must either be
1106 // within output directory, or must point to a
1107 // collection mount.
1108 err = ErrNotInOutputDir
1112 info, err = os.Lstat(tgt)
1115 err = fmt.Errorf("Symlink in output %q points to invalid location %q: %v",
1116 path[len(runner.HostOutputDir):], readlinktgt, err)
1126 var limitFollowSymlinks = 10
1128 // UploadFile uploads files within the output directory, with special handling
1129 // for symlinks. If the symlink leads to a keep mount, copy the manifest text
1130 // from the keep mount into the output manifestText. Ensure that whether
1131 // symlinks are relative or absolute, every symlink target (even targets that
1132 // are symlinks themselves) must point to a path in either the output directory
1133 // or a collection mount.
1135 // Assumes initial value of "path" is absolute, and located within runner.HostOutputDir.
1136 func (runner *ContainerRunner) UploadOutputFile(
1141 walkUpload *WalkUpload,
1142 relocateFrom string,
1144 followed int) (manifestText string, err error) {
1150 if info.Mode().IsDir() {
1151 // if empty, need to create a .keep file
1152 dir, direrr := os.Open(path)
1157 names, eof := dir.Readdirnames(1)
1158 if len(names) == 0 && eof == io.EOF && path != runner.HostOutputDir {
1159 containerPath := runner.OutputPath + path[len(runner.HostOutputDir):]
1160 for _, bind := range binds {
1161 mnt := runner.Container.Mounts[bind]
1162 // Check if there is a bind for this
1163 // directory, in which case assume we don't need .keep
1164 if (containerPath == bind || strings.HasPrefix(containerPath, bind+"/")) && mnt.PortableDataHash != "d41d8cd98f00b204e9800998ecf8427e+0" {
1168 outputSuffix := path[len(runner.HostOutputDir)+1:]
1169 return fmt.Sprintf("./%v d41d8cd98f00b204e9800998ecf8427e+0 0:0:.keep\n", outputSuffix), nil
1174 if followed >= limitFollowSymlinks {
1175 // Got stuck in a loop or just a pathological number of
1176 // directory links, give up.
1177 err = fmt.Errorf("Followed more than %v symlinks from path %q", limitFollowSymlinks, path)
1181 // "path" is the actual path we are visiting
1182 // "tgt" is the target of "path" (a non-symlink) after following symlinks
1183 // "relocated" is the path in the output manifest where the file should be placed,
1184 // but has HostOutputDir as a prefix.
1186 // The destination path in the output manifest may need to be
1187 // logically relocated to some other path in order to appear
1188 // in the correct location as a result of following a symlink.
1189 // Remove the relocateFrom prefix and replace it with
1191 relocated := relocateTo + path[len(relocateFrom):]
1193 tgt, readlinktgt, info, derefErr := runner.derefOutputSymlink(path, info)
1194 if derefErr != nil && derefErr != ErrNotInOutputDir {
1198 // go through mounts and try reverse map to collection reference
1199 for _, bind := range binds {
1200 mnt := runner.Container.Mounts[bind]
1201 if (tgt == bind || strings.HasPrefix(tgt, bind+"/")) && !mnt.Writable {
1202 // get path relative to bind
1203 targetSuffix := tgt[len(bind):]
1205 // Copy mount and adjust the path to add path relative to the bind
1206 adjustedMount := mnt
1207 adjustedMount.Path = filepath.Join(adjustedMount.Path, targetSuffix)
1209 // Terminates in this keep mount, so add the
1210 // manifest text at appropriate location.
1211 outputSuffix := relocated[len(runner.HostOutputDir):]
1212 manifestText, err = runner.getCollectionManifestForPath(adjustedMount, outputSuffix)
1217 // If target is not a collection mount, it must be located within the
1218 // output directory, otherwise it is an error.
1219 if derefErr == ErrNotInOutputDir {
1220 err = fmt.Errorf("Symlink in output %q points to invalid location %q, must point to path within the output directory.",
1221 path[len(runner.HostOutputDir):], readlinktgt)
1225 if info.Mode().IsRegular() {
1226 return "", walkUpload.UploadFile(relocated, tgt)
1229 if info.Mode().IsDir() {
1230 // Symlink leads to directory. Walk() doesn't follow
1231 // directory symlinks, so we walk the target directory
1232 // instead. Within the walk, file paths are relocated
1233 // so they appear under the original symlink path.
1234 err = filepath.Walk(tgt, func(walkpath string, walkinfo os.FileInfo, walkerr error) error {
1236 m, walkerr = runner.UploadOutputFile(walkpath, walkinfo, walkerr,
1237 binds, walkUpload, tgt, relocated, followed+1)
1239 manifestText = manifestText + m
1249 // HandleOutput sets the output, unmounts the FUSE mount, and deletes temporary directories
1250 func (runner *ContainerRunner) CaptureOutput() error {
1251 if wantAPI := runner.Container.RuntimeConstraints.API; wantAPI != nil && *wantAPI {
1252 // Output may have been set directly by the container, so
1253 // refresh the container record to check.
1254 err := runner.ArvClient.Get("containers", runner.Container.UUID,
1255 nil, &runner.Container)
1259 if runner.Container.Output != "" {
1260 // Container output is already set.
1261 runner.OutputPDH = &runner.Container.Output
1266 if runner.HostOutputDir == "" {
1270 _, err := os.Stat(runner.HostOutputDir)
1272 return fmt.Errorf("While checking host output path: %v", err)
1275 // Pre-populate output from the configured mount points
1277 for bind, mnt := range runner.Container.Mounts {
1278 if mnt.Kind == "collection" {
1279 binds = append(binds, bind)
1284 // Delete secret mounts so they don't get saved to the output collection.
1285 for bind := range runner.SecretMounts {
1286 if strings.HasPrefix(bind, runner.Container.OutputPath+"/") {
1287 err = os.Remove(runner.HostOutputDir + bind[len(runner.Container.OutputPath):])
1289 return fmt.Errorf("Unable to remove secret mount: %v", err)
1294 var manifestText string
1296 collectionMetafile := fmt.Sprintf("%s/.arvados#collection", runner.HostOutputDir)
1297 _, err = os.Stat(collectionMetafile)
1299 // Regular directory
1301 cw := CollectionWriter{0, runner.Kc, nil, nil, sync.Mutex{}}
1302 walkUpload := cw.BeginUpload(runner.HostOutputDir, runner.CrunchLog.Logger)
1305 err = filepath.Walk(runner.HostOutputDir, func(path string, info os.FileInfo, err error) error {
1306 m, err = runner.UploadOutputFile(path, info, err, binds, walkUpload, "", "", 0)
1308 manifestText = manifestText + m
1313 cw.EndUpload(walkUpload)
1316 return fmt.Errorf("While uploading output files: %v", err)
1319 m, err = cw.ManifestText()
1320 manifestText = manifestText + m
1322 return fmt.Errorf("While uploading output files: %v", err)
1325 // FUSE mount directory
1326 file, openerr := os.Open(collectionMetafile)
1328 return fmt.Errorf("While opening FUSE metafile: %v", err)
1332 var rec arvados.Collection
1333 err = json.NewDecoder(file).Decode(&rec)
1335 return fmt.Errorf("While reading FUSE metafile: %v", err)
1337 manifestText = rec.ManifestText
1340 for _, bind := range binds {
1341 mnt := runner.Container.Mounts[bind]
1343 bindSuffix := strings.TrimPrefix(bind, runner.Container.OutputPath)
1345 if bindSuffix == bind || len(bindSuffix) <= 0 {
1346 // either does not start with OutputPath or is OutputPath itself
1350 if mnt.ExcludeFromOutput == true || mnt.Writable {
1354 // append to manifest_text
1355 m, err := runner.getCollectionManifestForPath(mnt, bindSuffix)
1360 manifestText = manifestText + m
1364 var response arvados.Collection
1365 manifest := manifest.Manifest{Text: manifestText}
1366 manifestText = manifest.Extract(".", ".").Text
1367 err = runner.ArvClient.Create("collections",
1369 "ensure_unique_name": true,
1370 "collection": arvadosclient.Dict{
1372 "name": "output for " + runner.Container.UUID,
1373 "manifest_text": manifestText}},
1376 return fmt.Errorf("While creating output collection: %v", err)
1378 runner.OutputPDH = &response.PortableDataHash
1382 var outputCollections = make(map[string]arvados.Collection)
1384 // Fetch the collection for the mnt.PortableDataHash
1385 // Return the manifest_text fragment corresponding to the specified mnt.Path
1386 // after making any required updates.
1388 // If mnt.Path is not specified,
1389 // return the entire manifest_text after replacing any "." with bindSuffix
1390 // If mnt.Path corresponds to one stream,
1391 // return the manifest_text for that stream after replacing that stream name with bindSuffix
1392 // Otherwise, check if a filename in any one stream is being sought. Return the manifest_text
1393 // for that stream after replacing stream name with bindSuffix minus the last word
1394 // and the file name with last word of the bindSuffix
1395 // Allowed path examples:
1397 // "path":"/subdir1"
1398 // "path":"/subdir1/subdir2"
1399 // "path":"/subdir/filename" etc
1400 func (runner *ContainerRunner) getCollectionManifestForPath(mnt arvados.Mount, bindSuffix string) (string, error) {
1401 collection := outputCollections[mnt.PortableDataHash]
1402 if collection.PortableDataHash == "" {
1403 err := runner.ArvClient.Get("collections", mnt.PortableDataHash, nil, &collection)
1405 return "", fmt.Errorf("While getting collection for %v: %v", mnt.PortableDataHash, err)
1407 outputCollections[mnt.PortableDataHash] = collection
1410 if collection.ManifestText == "" {
1411 runner.CrunchLog.Printf("No manifest text for collection %v", collection.PortableDataHash)
1415 mft := manifest.Manifest{Text: collection.ManifestText}
1416 extracted := mft.Extract(mnt.Path, bindSuffix)
1417 if extracted.Err != nil {
1418 return "", fmt.Errorf("Error parsing manifest for %v: %v", mnt.PortableDataHash, extracted.Err.Error())
1420 return extracted.Text, nil
1423 func (runner *ContainerRunner) CleanupDirs() {
1424 if runner.ArvMount != nil {
1426 umount := exec.Command("arv-mount", fmt.Sprintf("--unmount-timeout=%d", delay), "--unmount", runner.ArvMountPoint)
1427 umount.Stdout = runner.CrunchLog
1428 umount.Stderr = runner.CrunchLog
1429 runner.CrunchLog.Printf("Running %v", umount.Args)
1430 umnterr := umount.Start()
1433 runner.CrunchLog.Printf("Error unmounting: %v", umnterr)
1435 // If arv-mount --unmount gets stuck for any reason, we
1436 // don't want to wait for it forever. Do Wait() in a goroutine
1437 // so it doesn't block crunch-run.
1438 umountExit := make(chan error)
1440 mnterr := umount.Wait()
1442 runner.CrunchLog.Printf("Error unmounting: %v", mnterr)
1444 umountExit <- mnterr
1447 for again := true; again; {
1453 case <-runner.ArvMountExit:
1455 case <-time.After(time.Duration((delay + 1) * int64(time.Second))):
1456 runner.CrunchLog.Printf("Timed out waiting for unmount")
1458 umount.Process.Kill()
1460 runner.ArvMount.Process.Kill()
1466 if runner.ArvMountPoint != "" {
1467 if rmerr := os.Remove(runner.ArvMountPoint); rmerr != nil {
1468 runner.CrunchLog.Printf("While cleaning up arv-mount directory %s: %v", runner.ArvMountPoint, rmerr)
1472 if rmerr := os.RemoveAll(runner.parentTemp); rmerr != nil {
1473 runner.CrunchLog.Printf("While cleaning up temporary directory %s: %v", runner.parentTemp, rmerr)
1477 // CommitLogs posts the collection containing the final container logs.
1478 func (runner *ContainerRunner) CommitLogs() error {
1480 // Hold cStateLock to prevent races on CrunchLog (e.g., stop()).
1481 runner.cStateLock.Lock()
1482 defer runner.cStateLock.Unlock()
1484 runner.CrunchLog.Print(runner.finalState)
1486 if runner.arvMountLog != nil {
1487 runner.arvMountLog.Close()
1489 runner.CrunchLog.Close()
1491 // Closing CrunchLog above allows them to be committed to Keep at this
1492 // point, but re-open crunch log with ArvClient in case there are any
1493 // other further errors (such as failing to write the log to Keep!)
1494 // while shutting down
1495 runner.CrunchLog = NewThrottledLogger(&ArvLogWriter{ArvClient: runner.ArvClient,
1496 UUID: runner.Container.UUID, loggingStream: "crunch-run", writeCloser: nil})
1497 runner.CrunchLog.Immediate = log.New(os.Stderr, runner.Container.UUID+" ", 0)
1500 if runner.LogsPDH != nil {
1501 // If we have already assigned something to LogsPDH,
1502 // we must be closing the re-opened log, which won't
1503 // end up getting attached to the container record and
1504 // therefore doesn't need to be saved as a collection
1505 // -- it exists only to send logs to other channels.
1509 mt, err := runner.LogCollection.ManifestText()
1511 return fmt.Errorf("While creating log manifest: %v", err)
1514 var response arvados.Collection
1515 err = runner.ArvClient.Create("collections",
1517 "ensure_unique_name": true,
1518 "collection": arvadosclient.Dict{
1520 "name": "logs for " + runner.Container.UUID,
1521 "manifest_text": mt}},
1524 return fmt.Errorf("While creating log collection: %v", err)
1526 runner.LogsPDH = &response.PortableDataHash
1530 // UpdateContainerRunning updates the container state to "Running"
1531 func (runner *ContainerRunner) UpdateContainerRunning() error {
1532 runner.cStateLock.Lock()
1533 defer runner.cStateLock.Unlock()
1534 if runner.cCancelled {
1537 return runner.ArvClient.Update("containers", runner.Container.UUID,
1538 arvadosclient.Dict{"container": arvadosclient.Dict{"state": "Running"}}, nil)
1541 // ContainerToken returns the api_token the container (and any
1542 // arv-mount processes) are allowed to use.
1543 func (runner *ContainerRunner) ContainerToken() (string, error) {
1544 if runner.token != "" {
1545 return runner.token, nil
1548 var auth arvados.APIClientAuthorization
1549 err := runner.ArvClient.Call("GET", "containers", runner.Container.UUID, "auth", nil, &auth)
1553 runner.token = auth.APIToken
1554 return runner.token, nil
1557 // UpdateContainerComplete updates the container record state on API
1558 // server to "Complete" or "Cancelled"
1559 func (runner *ContainerRunner) UpdateContainerFinal() error {
1560 update := arvadosclient.Dict{}
1561 update["state"] = runner.finalState
1562 if runner.LogsPDH != nil {
1563 update["log"] = *runner.LogsPDH
1565 if runner.finalState == "Complete" {
1566 if runner.ExitCode != nil {
1567 update["exit_code"] = *runner.ExitCode
1569 if runner.OutputPDH != nil {
1570 update["output"] = *runner.OutputPDH
1573 return runner.ArvClient.Update("containers", runner.Container.UUID, arvadosclient.Dict{"container": update}, nil)
1576 // IsCancelled returns the value of Cancelled, with goroutine safety.
1577 func (runner *ContainerRunner) IsCancelled() bool {
1578 runner.cStateLock.Lock()
1579 defer runner.cStateLock.Unlock()
1580 return runner.cCancelled
1583 // NewArvLogWriter creates an ArvLogWriter
1584 func (runner *ContainerRunner) NewArvLogWriter(name string) io.WriteCloser {
1585 return &ArvLogWriter{
1586 ArvClient: runner.ArvClient,
1587 UUID: runner.Container.UUID,
1588 loggingStream: name,
1589 writeCloser: runner.LogCollection.Open(name + ".txt")}
1592 // Run the full container lifecycle.
1593 func (runner *ContainerRunner) Run() (err error) {
1594 runner.CrunchLog.Printf("crunch-run %s started", version)
1595 runner.CrunchLog.Printf("Executing container '%s'", runner.Container.UUID)
1597 hostname, hosterr := os.Hostname()
1599 runner.CrunchLog.Printf("Error getting hostname '%v'", hosterr)
1601 runner.CrunchLog.Printf("Executing on host '%s'", hostname)
1604 runner.finalState = "Queued"
1607 runner.CleanupDirs()
1609 runner.CrunchLog.Printf("crunch-run finished")
1610 runner.CrunchLog.Close()
1614 // checkErr prints e (unless it's nil) and sets err to
1615 // e (unless err is already non-nil). Thus, if err
1616 // hasn't already been assigned when Run() returns,
1617 // this cleanup func will cause Run() to return the
1618 // first non-nil error that is passed to checkErr().
1619 checkErr := func(e error) {
1623 runner.CrunchLog.Print(e)
1627 if runner.finalState == "Complete" {
1628 // There was an error in the finalization.
1629 runner.finalState = "Cancelled"
1633 // Log the error encountered in Run(), if any
1636 if runner.finalState == "Queued" {
1637 runner.UpdateContainerFinal()
1641 if runner.IsCancelled() {
1642 runner.finalState = "Cancelled"
1643 // but don't return yet -- we still want to
1644 // capture partial output and write logs
1647 checkErr(runner.CaptureOutput())
1648 checkErr(runner.stopHoststat())
1649 checkErr(runner.CommitLogs())
1650 checkErr(runner.UpdateContainerFinal())
1653 err = runner.fetchContainerRecord()
1657 runner.setupSignals()
1658 runner.startHoststat()
1660 // check for and/or load image
1661 err = runner.LoadImage()
1663 if !runner.checkBrokenNode(err) {
1664 // Failed to load image but not due to a "broken node"
1665 // condition, probably user error.
1666 runner.finalState = "Cancelled"
1668 err = fmt.Errorf("While loading container image: %v", err)
1672 // set up FUSE mount and binds
1673 err = runner.SetupMounts()
1675 runner.finalState = "Cancelled"
1676 err = fmt.Errorf("While setting up mounts: %v", err)
1680 err = runner.CreateContainer()
1684 err = runner.LogHostInfo()
1688 err = runner.LogNodeRecord()
1692 err = runner.LogContainerRecord()
1697 if runner.IsCancelled() {
1701 err = runner.UpdateContainerRunning()
1705 runner.finalState = "Cancelled"
1707 runner.startCrunchstat()
1709 err = runner.StartContainer()
1711 runner.checkBrokenNode(err)
1715 err = runner.WaitFinish()
1716 if err == nil && !runner.IsCancelled() {
1717 runner.finalState = "Complete"
1722 // Fetch the current container record (uuid = runner.Container.UUID)
1723 // into runner.Container.
1724 func (runner *ContainerRunner) fetchContainerRecord() error {
1725 reader, err := runner.ArvClient.CallRaw("GET", "containers", runner.Container.UUID, "", nil)
1727 return fmt.Errorf("error fetching container record: %v", err)
1729 defer reader.Close()
1731 dec := json.NewDecoder(reader)
1733 err = dec.Decode(&runner.Container)
1735 return fmt.Errorf("error decoding container record: %v", err)
1739 SecretMounts map[string]arvados.Mount `json:"secret_mounts"`
1742 containerToken, err := runner.ContainerToken()
1744 return fmt.Errorf("error getting container token: %v", err)
1747 containerClient, err := runner.MkArvClient(containerToken)
1749 return fmt.Errorf("error creating container API client: %v", err)
1752 err = containerClient.Call("GET", "containers", runner.Container.UUID, "secret_mounts", nil, &sm)
1754 if apierr, ok := err.(arvadosclient.APIServerError); !ok || apierr.HttpStatusCode != 404 {
1755 return fmt.Errorf("error fetching secret_mounts: %v", err)
1757 // ok && apierr.HttpStatusCode == 404, which means
1758 // secret_mounts isn't supported by this API server.
1760 runner.SecretMounts = sm.SecretMounts
1765 // NewContainerRunner creates a new container runner.
1766 func NewContainerRunner(api IArvadosClient,
1768 docker ThinDockerClient,
1769 containerUUID string) *ContainerRunner {
1771 cr := &ContainerRunner{ArvClient: api, Kc: kc, Docker: docker}
1772 cr.NewLogWriter = cr.NewArvLogWriter
1773 cr.RunArvMount = cr.ArvMountCmd
1774 cr.MkTempDir = ioutil.TempDir
1775 cr.MkArvClient = func(token string) (IArvadosClient, error) {
1776 cl, err := arvadosclient.MakeArvadosClient()
1783 cr.LogCollection = &CollectionWriter{0, kc, nil, nil, sync.Mutex{}}
1784 cr.Container.UUID = containerUUID
1785 cr.CrunchLog = NewThrottledLogger(cr.NewLogWriter("crunch-run"))
1786 cr.CrunchLog.Immediate = log.New(os.Stderr, containerUUID+" ", 0)
1788 loadLogThrottleParams(api)
1794 statInterval := flag.Duration("crunchstat-interval", 10*time.Second, "sampling period for periodic resource usage reporting")
1795 cgroupRoot := flag.String("cgroup-root", "/sys/fs/cgroup", "path to sysfs cgroup tree")
1796 cgroupParent := flag.String("cgroup-parent", "docker", "name of container's parent cgroup (ignored if -cgroup-parent-subsystem is used)")
1797 cgroupParentSubsystem := flag.String("cgroup-parent-subsystem", "", "use current cgroup for given subsystem as parent cgroup for container")
1798 caCertsPath := flag.String("ca-certs", "", "Path to TLS root certificates")
1799 enableNetwork := flag.String("container-enable-networking", "default",
1800 `Specify if networking should be enabled for container. One of 'default', 'always':
1801 default: only enable networking if container requests it.
1802 always: containers always have networking enabled
1804 networkMode := flag.String("container-network-mode", "default",
1805 `Set networking mode for container. Corresponds to Docker network mode (--net).
1807 memprofile := flag.String("memprofile", "", "write memory profile to `file` after running container")
1808 getVersion := flag.Bool("version", false, "Print version information and exit.")
1811 // Print version information if requested
1813 fmt.Printf("crunch-run %s\n", version)
1817 log.Printf("crunch-run %s started", version)
1819 containerId := flag.Arg(0)
1821 if *caCertsPath != "" {
1822 arvadosclient.CertFiles = []string{*caCertsPath}
1825 api, err := arvadosclient.MakeArvadosClient()
1827 log.Fatalf("%s: %v", containerId, err)
1831 kc, kcerr := keepclient.MakeKeepClient(api)
1833 log.Fatalf("%s: %v", containerId, kcerr)
1835 kc.BlockCache = &keepclient.BlockCache{MaxBlocks: 2}
1838 // API version 1.21 corresponds to Docker 1.9, which is currently the
1839 // minimum version we want to support.
1840 docker, dockererr := dockerclient.NewClient(dockerclient.DefaultDockerHost, "1.21", nil, nil)
1842 cr := NewContainerRunner(api, kc, docker, containerId)
1843 if dockererr != nil {
1844 cr.CrunchLog.Printf("%s: %v", containerId, dockererr)
1845 cr.checkBrokenNode(dockererr)
1846 cr.CrunchLog.Close()
1850 parentTemp, tmperr := cr.MkTempDir("", "crunch-run."+containerId+".")
1852 log.Fatalf("%s: %v", containerId, tmperr)
1855 cr.parentTemp = parentTemp
1856 cr.statInterval = *statInterval
1857 cr.cgroupRoot = *cgroupRoot
1858 cr.expectCgroupParent = *cgroupParent
1859 cr.enableNetwork = *enableNetwork
1860 cr.networkMode = *networkMode
1861 if *cgroupParentSubsystem != "" {
1862 p := findCgroup(*cgroupParentSubsystem)
1863 cr.setCgroupParent = p
1864 cr.expectCgroupParent = p
1869 if *memprofile != "" {
1870 f, err := os.Create(*memprofile)
1872 log.Printf("could not create memory profile: ", err)
1874 runtime.GC() // get up-to-date statistics
1875 if err := pprof.WriteHeapProfile(f); err != nil {
1876 log.Printf("could not write memory profile: ", err)
1878 closeerr := f.Close()
1879 if closeerr != nil {
1880 log.Printf("closing memprofile file: ", err)
1885 log.Fatalf("%s: %v", containerId, runerr)