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
109 CleanupTempDir []string
111 Volumes map[string]struct{}
113 SigChan chan os.Signal
114 ArvMountExit chan error
117 statLogger io.WriteCloser
118 statReporter *crunchstat.Reporter
119 statInterval time.Duration
121 // What we expect the container's cgroup parent to be.
122 expectCgroupParent string
123 // What we tell docker to use as the container's cgroup
124 // parent. Note: Ideally we would use the same field for both
125 // expectCgroupParent and setCgroupParent, and just make it
126 // default to "docker". However, when using docker < 1.10 with
127 // systemd, specifying a non-empty cgroup parent (even the
128 // default value "docker") hits a docker bug
129 // (https://github.com/docker/docker/issues/17126). Using two
130 // separate fields makes it possible to use the "expect cgroup
131 // parent to be X" feature even on sites where the "specify
132 // cgroup parent" feature breaks.
133 setCgroupParent string
135 cStateLock sync.Mutex
136 cCancelled bool // StopContainer() invoked
138 enableNetwork string // one of "default" or "always"
139 networkMode string // passed through to HostConfig.NetworkMode
140 arvMountLog *ThrottledLogger
143 // setupSignals sets up signal handling to gracefully terminate the underlying
144 // Docker container and update state when receiving a TERM, INT or QUIT signal.
145 func (runner *ContainerRunner) setupSignals() {
146 runner.SigChan = make(chan os.Signal, 1)
147 signal.Notify(runner.SigChan, syscall.SIGTERM)
148 signal.Notify(runner.SigChan, syscall.SIGINT)
149 signal.Notify(runner.SigChan, syscall.SIGQUIT)
151 go func(sig chan os.Signal) {
153 runner.CrunchLog.Printf("caught signal: %v", s)
159 // stop the underlying Docker container.
160 func (runner *ContainerRunner) stop() {
161 runner.cStateLock.Lock()
162 defer runner.cStateLock.Unlock()
163 if runner.ContainerID == "" {
166 runner.cCancelled = true
167 runner.CrunchLog.Printf("removing container")
168 err := runner.Docker.ContainerRemove(context.TODO(), runner.ContainerID, dockertypes.ContainerRemoveOptions{Force: true})
170 runner.CrunchLog.Printf("error removing container: %s", err)
174 func (runner *ContainerRunner) stopSignals() {
175 if runner.SigChan != nil {
176 signal.Stop(runner.SigChan)
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("", prefix)
333 func copyfile(src string, dst string) (err error) {
334 srcfile, err := os.Open(src)
339 os.MkdirAll(path.Dir(dst), 0770)
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)
403 for _, bind := range binds {
404 mnt := runner.Container.Mounts[bind]
405 if bind == "stdout" || bind == "stderr" {
406 // Is it a "file" mount kind?
407 if mnt.Kind != "file" {
408 return fmt.Errorf("Unsupported mount kind '%s' for %s. Only 'file' is supported.", mnt.Kind, bind)
411 // Does path start with OutputPath?
412 prefix := runner.Container.OutputPath
413 if !strings.HasSuffix(prefix, "/") {
416 if !strings.HasPrefix(mnt.Path, prefix) {
417 return fmt.Errorf("%s path does not start with OutputPath: %s, %s", strings.Title(bind), mnt.Path, prefix)
422 // Is it a "collection" mount kind?
423 if mnt.Kind != "collection" && mnt.Kind != "json" {
424 return fmt.Errorf("Unsupported mount kind '%s' for stdin. Only 'collection' or 'json' are supported.", mnt.Kind)
428 if bind == "/etc/arvados/ca-certificates.crt" {
429 needCertMount = false
432 if strings.HasPrefix(bind, runner.Container.OutputPath+"/") && bind != runner.Container.OutputPath+"/" {
433 if mnt.Kind != "collection" {
434 return fmt.Errorf("Only mount points of kind 'collection' are supported underneath the output_path: %v", bind)
439 case mnt.Kind == "collection" && bind != "stdin":
441 if mnt.UUID != "" && mnt.PortableDataHash != "" {
442 return fmt.Errorf("Cannot specify both 'uuid' and 'portable_data_hash' for a collection mount")
446 return fmt.Errorf("Writing to existing collections currently not permitted.")
449 src = fmt.Sprintf("%s/by_id/%s", runner.ArvMountPoint, mnt.UUID)
450 } else if mnt.PortableDataHash != "" {
451 if mnt.Writable && !strings.HasPrefix(bind, runner.Container.OutputPath+"/") {
452 return fmt.Errorf("Can never write to a collection specified by portable data hash")
454 idx := strings.Index(mnt.PortableDataHash, "/")
456 mnt.Path = path.Clean(mnt.PortableDataHash[idx:])
457 mnt.PortableDataHash = mnt.PortableDataHash[0:idx]
458 runner.Container.Mounts[bind] = mnt
460 src = fmt.Sprintf("%s/by_id/%s", runner.ArvMountPoint, mnt.PortableDataHash)
461 if mnt.Path != "" && mnt.Path != "." {
462 if strings.HasPrefix(mnt.Path, "./") {
463 mnt.Path = mnt.Path[2:]
464 } else if strings.HasPrefix(mnt.Path, "/") {
465 mnt.Path = mnt.Path[1:]
467 src += "/" + mnt.Path
470 src = fmt.Sprintf("%s/tmp%d", runner.ArvMountPoint, tmpcount)
471 arvMountCmd = append(arvMountCmd, "--mount-tmp")
472 arvMountCmd = append(arvMountCmd, fmt.Sprintf("tmp%d", tmpcount))
476 if bind == runner.Container.OutputPath {
477 runner.HostOutputDir = src
478 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s", src, bind))
479 } else if strings.HasPrefix(bind, runner.Container.OutputPath+"/") {
480 copyFiles = append(copyFiles, copyFile{src, runner.HostOutputDir + bind[len(runner.Container.OutputPath):]})
482 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s", src, bind))
485 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s:ro", src, bind))
487 collectionPaths = append(collectionPaths, src)
489 case mnt.Kind == "tmp":
491 tmpdir, err = runner.MkTempDir("", "")
493 return fmt.Errorf("While creating mount temp dir: %v", err)
495 st, staterr := os.Stat(tmpdir)
497 return fmt.Errorf("While Stat on temp dir: %v", staterr)
499 err = os.Chmod(tmpdir, st.Mode()|os.ModeSetgid|0777)
501 return fmt.Errorf("While Chmod temp dir: %v", err)
503 runner.CleanupTempDir = append(runner.CleanupTempDir, tmpdir)
504 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s", tmpdir, bind))
505 if bind == runner.Container.OutputPath {
506 runner.HostOutputDir = tmpdir
509 case mnt.Kind == "json":
510 jsondata, err := json.Marshal(mnt.Content)
512 return fmt.Errorf("encoding json data: %v", err)
514 // Create a tempdir with a single file
515 // (instead of just a tempfile): this way we
516 // can ensure the file is world-readable
517 // inside the container, without having to
518 // make it world-readable on the docker host.
519 tmpdir, err := runner.MkTempDir("", "")
521 return fmt.Errorf("creating temp dir: %v", err)
523 runner.CleanupTempDir = append(runner.CleanupTempDir, tmpdir)
524 tmpfn := filepath.Join(tmpdir, "mountdata.json")
525 err = ioutil.WriteFile(tmpfn, jsondata, 0644)
527 return fmt.Errorf("writing temp file: %v", err)
529 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s:ro", tmpfn, bind))
531 case mnt.Kind == "git_tree":
532 tmpdir, err := runner.MkTempDir("", "")
534 return fmt.Errorf("creating temp dir: %v", err)
536 runner.CleanupTempDir = append(runner.CleanupTempDir, tmpdir)
537 err = gitMount(mnt).extractTree(runner.ArvClient, tmpdir, token)
541 runner.Binds = append(runner.Binds, tmpdir+":"+bind+":ro")
545 if runner.HostOutputDir == "" {
546 return fmt.Errorf("Output path does not correspond to a writable mount point")
549 if wantAPI := runner.Container.RuntimeConstraints.API; needCertMount && wantAPI != nil && *wantAPI {
550 for _, certfile := range arvadosclient.CertFiles {
551 _, err := os.Stat(certfile)
553 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:/etc/arvados/ca-certificates.crt:ro", certfile))
560 arvMountCmd = append(arvMountCmd, "--mount-by-pdh", "by_id")
562 arvMountCmd = append(arvMountCmd, "--mount-by-id", "by_id")
564 arvMountCmd = append(arvMountCmd, runner.ArvMountPoint)
566 runner.ArvMount, err = runner.RunArvMount(arvMountCmd, token)
568 return fmt.Errorf("While trying to start arv-mount: %v", err)
571 for _, p := range collectionPaths {
574 return fmt.Errorf("While checking that input files exist: %v", err)
578 for _, cp := range copyFiles {
579 dir, err := os.Stat(cp.src)
581 return fmt.Errorf("While staging writable file from %q to %q: %v", cp.src, cp.bind, err)
584 err = filepath.Walk(cp.src, func(walkpath string, walkinfo os.FileInfo, walkerr error) error {
588 if walkinfo.Mode().IsRegular() {
589 return copyfile(walkpath, path.Join(cp.bind, walkpath[len(cp.src):]))
590 } else if walkinfo.Mode().IsDir() {
591 // will be visited by Walk()
594 return fmt.Errorf("Source %q is not a regular file or directory", cp.src)
598 err = copyfile(cp.src, cp.bind)
601 return fmt.Errorf("While staging writable file from %q to %q: %v", cp.src, cp.bind, err)
608 func (runner *ContainerRunner) ProcessDockerAttach(containerReader io.Reader) {
609 // Handle docker log protocol
610 // https://docs.docker.com/engine/reference/api/docker_remote_api_v1.15/#attach-to-a-container
612 header := make([]byte, 8)
614 _, readerr := io.ReadAtLeast(containerReader, header, 8)
617 readsize := int64(header[7]) | (int64(header[6]) << 8) | (int64(header[5]) << 16) | (int64(header[4]) << 24)
620 _, readerr = io.CopyN(runner.Stdout, containerReader, readsize)
623 _, readerr = io.CopyN(runner.Stderr, containerReader, readsize)
628 if readerr != io.EOF {
629 runner.CrunchLog.Printf("While reading docker logs: %v", readerr)
632 closeerr := runner.Stdout.Close()
634 runner.CrunchLog.Printf("While closing stdout logs: %v", closeerr)
637 closeerr = runner.Stderr.Close()
639 runner.CrunchLog.Printf("While closing stderr logs: %v", closeerr)
642 if runner.statReporter != nil {
643 runner.statReporter.Stop()
644 closeerr = runner.statLogger.Close()
646 runner.CrunchLog.Printf("While closing crunchstat logs: %v", closeerr)
650 close(runner.loggingDone)
656 func (runner *ContainerRunner) StartCrunchstat() {
657 runner.statLogger = NewThrottledLogger(runner.NewLogWriter("crunchstat"))
658 runner.statReporter = &crunchstat.Reporter{
659 CID: runner.ContainerID,
660 Logger: log.New(runner.statLogger, "", 0),
661 CgroupParent: runner.expectCgroupParent,
662 CgroupRoot: runner.cgroupRoot,
663 PollPeriod: runner.statInterval,
665 runner.statReporter.Start()
668 type infoCommand struct {
673 // LogHostInfo logs info about the current host, for debugging and
674 // accounting purposes. Although it's logged as "node-info", this is
675 // about the environment where crunch-run is actually running, which
676 // might differ from what's described in the node record (see
678 func (runner *ContainerRunner) LogHostInfo() (err error) {
679 w := runner.NewLogWriter("node-info")
681 commands := []infoCommand{
683 label: "Host Information",
684 cmd: []string{"uname", "-a"},
687 label: "CPU Information",
688 cmd: []string{"cat", "/proc/cpuinfo"},
691 label: "Memory Information",
692 cmd: []string{"cat", "/proc/meminfo"},
696 cmd: []string{"df", "-m", "/", os.TempDir()},
699 label: "Disk INodes",
700 cmd: []string{"df", "-i", "/", os.TempDir()},
704 // Run commands with informational output to be logged.
705 for _, command := range commands {
706 fmt.Fprintln(w, command.label)
707 cmd := exec.Command(command.cmd[0], command.cmd[1:]...)
710 if err := cmd.Run(); err != nil {
711 err = fmt.Errorf("While running command %q: %v", command.cmd, err)
720 return fmt.Errorf("While closing node-info logs: %v", err)
725 // LogContainerRecord gets and saves the raw JSON container record from the API server
726 func (runner *ContainerRunner) LogContainerRecord() error {
727 logged, err := runner.logAPIResponse("container", "containers", map[string]interface{}{"filters": [][]string{{"uuid", "=", runner.Container.UUID}}}, nil)
728 if !logged && err == nil {
729 err = fmt.Errorf("error: no container record found for %s", runner.Container.UUID)
734 // LogNodeRecord logs arvados#node record corresponding to the current host.
735 func (runner *ContainerRunner) LogNodeRecord() error {
736 hostname := os.Getenv("SLURMD_NODENAME")
738 hostname, _ = os.Hostname()
740 _, err := runner.logAPIResponse("node", "nodes", map[string]interface{}{"filters": [][]string{{"hostname", "=", hostname}}}, func(resp interface{}) {
741 // The "info" field has admin-only info when obtained
742 // with a privileged token, and should not be logged.
743 node, ok := resp.(map[string]interface{})
751 func (runner *ContainerRunner) logAPIResponse(label, path string, params map[string]interface{}, munge func(interface{})) (logged bool, err error) {
753 ArvClient: runner.ArvClient,
754 UUID: runner.Container.UUID,
755 loggingStream: label,
756 writeCloser: runner.LogCollection.Open(label + ".json"),
759 reader, err := runner.ArvClient.CallRaw("GET", path, "", "", arvadosclient.Dict(params))
761 return false, fmt.Errorf("error getting %s record: %v", label, err)
765 dec := json.NewDecoder(reader)
767 var resp map[string]interface{}
768 if err = dec.Decode(&resp); err != nil {
769 return false, fmt.Errorf("error decoding %s list response: %v", label, err)
771 items, ok := resp["items"].([]interface{})
773 return false, fmt.Errorf("error decoding %s list response: no \"items\" key in API list response", label)
774 } else if len(items) < 1 {
780 // Re-encode it using indentation to improve readability
781 enc := json.NewEncoder(w)
782 enc.SetIndent("", " ")
783 if err = enc.Encode(items[0]); err != nil {
784 return false, fmt.Errorf("error logging %s record: %v", label, err)
788 return false, fmt.Errorf("error closing %s.json in log collection: %v", label, err)
793 // AttachStreams connects the docker container stdin, stdout and stderr logs
794 // to the Arvados logger which logs to Keep and the API server logs table.
795 func (runner *ContainerRunner) AttachStreams() (err error) {
797 runner.CrunchLog.Print("Attaching container streams")
799 // If stdin mount is provided, attach it to the docker container
800 var stdinRdr arvados.File
802 if stdinMnt, ok := runner.Container.Mounts["stdin"]; ok {
803 if stdinMnt.Kind == "collection" {
804 var stdinColl arvados.Collection
805 collId := stdinMnt.UUID
807 collId = stdinMnt.PortableDataHash
809 err = runner.ArvClient.Get("collections", collId, nil, &stdinColl)
811 return fmt.Errorf("While getting stding collection: %v", err)
814 stdinRdr, err = runner.Kc.ManifestFileReader(manifest.Manifest{Text: stdinColl.ManifestText}, stdinMnt.Path)
815 if os.IsNotExist(err) {
816 return fmt.Errorf("stdin collection path not found: %v", stdinMnt.Path)
817 } else if err != nil {
818 return fmt.Errorf("While getting stdin collection path %v: %v", stdinMnt.Path, err)
820 } else if stdinMnt.Kind == "json" {
821 stdinJson, err = json.Marshal(stdinMnt.Content)
823 return fmt.Errorf("While encoding stdin json data: %v", err)
828 stdinUsed := stdinRdr != nil || len(stdinJson) != 0
829 response, err := runner.Docker.ContainerAttach(context.TODO(), runner.ContainerID,
830 dockertypes.ContainerAttachOptions{Stream: true, Stdin: stdinUsed, Stdout: true, Stderr: true})
832 return fmt.Errorf("While attaching container stdout/stderr streams: %v", err)
835 runner.loggingDone = make(chan bool)
837 if stdoutMnt, ok := runner.Container.Mounts["stdout"]; ok {
838 stdoutFile, err := runner.getStdoutFile(stdoutMnt.Path)
842 runner.Stdout = stdoutFile
844 runner.Stdout = NewThrottledLogger(runner.NewLogWriter("stdout"))
847 if stderrMnt, ok := runner.Container.Mounts["stderr"]; ok {
848 stderrFile, err := runner.getStdoutFile(stderrMnt.Path)
852 runner.Stderr = stderrFile
854 runner.Stderr = NewThrottledLogger(runner.NewLogWriter("stderr"))
859 _, err := io.Copy(response.Conn, stdinRdr)
861 runner.CrunchLog.Print("While writing stdin collection to docker container %q", err)
865 response.CloseWrite()
867 } else if len(stdinJson) != 0 {
869 _, err := io.Copy(response.Conn, bytes.NewReader(stdinJson))
871 runner.CrunchLog.Print("While writing stdin json to docker container %q", err)
874 response.CloseWrite()
878 go runner.ProcessDockerAttach(response.Reader)
883 func (runner *ContainerRunner) getStdoutFile(mntPath string) (*os.File, error) {
884 stdoutPath := mntPath[len(runner.Container.OutputPath):]
885 index := strings.LastIndex(stdoutPath, "/")
887 subdirs := stdoutPath[:index]
889 st, err := os.Stat(runner.HostOutputDir)
891 return nil, fmt.Errorf("While Stat on temp dir: %v", err)
893 stdoutPath := filepath.Join(runner.HostOutputDir, subdirs)
894 err = os.MkdirAll(stdoutPath, st.Mode()|os.ModeSetgid|0777)
896 return nil, fmt.Errorf("While MkdirAll %q: %v", stdoutPath, err)
900 stdoutFile, err := os.Create(filepath.Join(runner.HostOutputDir, stdoutPath))
902 return nil, fmt.Errorf("While creating file %q: %v", stdoutPath, err)
905 return stdoutFile, nil
908 // CreateContainer creates the docker container.
909 func (runner *ContainerRunner) CreateContainer() error {
910 runner.CrunchLog.Print("Creating Docker container")
912 runner.ContainerConfig.Cmd = runner.Container.Command
913 if runner.Container.Cwd != "." {
914 runner.ContainerConfig.WorkingDir = runner.Container.Cwd
917 for k, v := range runner.Container.Environment {
918 runner.ContainerConfig.Env = append(runner.ContainerConfig.Env, k+"="+v)
921 runner.ContainerConfig.Volumes = runner.Volumes
923 runner.HostConfig = dockercontainer.HostConfig{
925 LogConfig: dockercontainer.LogConfig{
928 Resources: dockercontainer.Resources{
929 CgroupParent: runner.setCgroupParent,
933 if wantAPI := runner.Container.RuntimeConstraints.API; wantAPI != nil && *wantAPI {
934 tok, err := runner.ContainerToken()
938 runner.ContainerConfig.Env = append(runner.ContainerConfig.Env,
939 "ARVADOS_API_TOKEN="+tok,
940 "ARVADOS_API_HOST="+os.Getenv("ARVADOS_API_HOST"),
941 "ARVADOS_API_HOST_INSECURE="+os.Getenv("ARVADOS_API_HOST_INSECURE"),
943 runner.HostConfig.NetworkMode = dockercontainer.NetworkMode(runner.networkMode)
945 if runner.enableNetwork == "always" {
946 runner.HostConfig.NetworkMode = dockercontainer.NetworkMode(runner.networkMode)
948 runner.HostConfig.NetworkMode = dockercontainer.NetworkMode("none")
952 _, stdinUsed := runner.Container.Mounts["stdin"]
953 runner.ContainerConfig.OpenStdin = stdinUsed
954 runner.ContainerConfig.StdinOnce = stdinUsed
955 runner.ContainerConfig.AttachStdin = stdinUsed
956 runner.ContainerConfig.AttachStdout = true
957 runner.ContainerConfig.AttachStderr = true
959 createdBody, err := runner.Docker.ContainerCreate(context.TODO(), &runner.ContainerConfig, &runner.HostConfig, nil, runner.Container.UUID)
961 return fmt.Errorf("While creating container: %v", err)
964 runner.ContainerID = createdBody.ID
966 return runner.AttachStreams()
969 // StartContainer starts the docker container created by CreateContainer.
970 func (runner *ContainerRunner) StartContainer() error {
971 runner.CrunchLog.Printf("Starting Docker container id '%s'", runner.ContainerID)
972 runner.cStateLock.Lock()
973 defer runner.cStateLock.Unlock()
974 if runner.cCancelled {
977 err := runner.Docker.ContainerStart(context.TODO(), runner.ContainerID,
978 dockertypes.ContainerStartOptions{})
981 if m, e := regexp.MatchString("(?ms).*(exec|System error).*(no such file or directory|file not found).*", err.Error()); m && e == nil {
982 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])
984 return fmt.Errorf("could not start container: %v%s", err, advice)
989 // WaitFinish waits for the container to terminate, capture the exit code, and
990 // close the stdout/stderr logging.
991 func (runner *ContainerRunner) WaitFinish() error {
992 runner.CrunchLog.Print("Waiting for container to finish")
994 waitOk, waitErr := runner.Docker.ContainerWait(context.TODO(), runner.ContainerID, dockercontainer.WaitConditionNotRunning)
995 arvMountExit := runner.ArvMountExit
998 case waitBody := <-waitOk:
999 runner.CrunchLog.Printf("Container exited with code: %v", waitBody.StatusCode)
1000 code := int(waitBody.StatusCode)
1001 runner.ExitCode = &code
1003 // wait for stdout/stderr to complete
1004 <-runner.loggingDone
1007 case err := <-waitErr:
1008 return fmt.Errorf("container wait: %v", err)
1010 case <-arvMountExit:
1011 runner.CrunchLog.Printf("arv-mount exited while container is still running. Stopping container.")
1013 // arvMountExit will always be ready now that
1014 // it's closed, but that doesn't interest us.
1020 var ErrNotInOutputDir = fmt.Errorf("Must point to path within the output directory")
1022 func (runner *ContainerRunner) derefOutputSymlink(path string, startinfo os.FileInfo) (tgt string, readlinktgt string, info os.FileInfo, err error) {
1023 // Follow symlinks if necessary
1028 for followed := 0; info.Mode()&os.ModeSymlink != 0; followed++ {
1029 if followed >= limitFollowSymlinks {
1030 // Got stuck in a loop or just a pathological number of links, give up.
1031 err = fmt.Errorf("Followed more than %v symlinks from path %q", limitFollowSymlinks, path)
1035 readlinktgt, err = os.Readlink(nextlink)
1041 if !strings.HasPrefix(tgt, "/") {
1042 // Relative symlink, resolve it to host path
1043 tgt = filepath.Join(filepath.Dir(path), tgt)
1045 if strings.HasPrefix(tgt, runner.Container.OutputPath+"/") && !strings.HasPrefix(tgt, runner.HostOutputDir+"/") {
1046 // Absolute symlink to container output path, adjust it to host output path.
1047 tgt = filepath.Join(runner.HostOutputDir, tgt[len(runner.Container.OutputPath):])
1049 if !strings.HasPrefix(tgt, runner.HostOutputDir+"/") {
1050 // After dereferencing, symlink target must either be
1051 // within output directory, or must point to a
1052 // collection mount.
1053 err = ErrNotInOutputDir
1057 info, err = os.Lstat(tgt)
1060 err = fmt.Errorf("Symlink in output %q points to invalid location %q: %v",
1061 path[len(runner.HostOutputDir):], readlinktgt, err)
1071 var limitFollowSymlinks = 10
1073 // UploadFile uploads files within the output directory, with special handling
1074 // for symlinks. If the symlink leads to a keep mount, copy the manifest text
1075 // from the keep mount into the output manifestText. Ensure that whether
1076 // symlinks are relative or absolute, every symlink target (even targets that
1077 // are symlinks themselves) must point to a path in either the output directory
1078 // or a collection mount.
1080 // Assumes initial value of "path" is absolute, and located within runner.HostOutputDir.
1081 func (runner *ContainerRunner) UploadOutputFile(
1086 walkUpload *WalkUpload,
1087 relocateFrom string,
1089 followed int) (manifestText string, err error) {
1095 if info.Mode().IsDir() {
1096 // if empty, need to create a .keep file
1097 dir, direrr := os.Open(path)
1102 names, eof := dir.Readdirnames(1)
1103 if len(names) == 0 && eof == io.EOF && path != runner.HostOutputDir {
1104 containerPath := runner.OutputPath + path[len(runner.HostOutputDir):]
1105 for _, bind := range binds {
1106 mnt := runner.Container.Mounts[bind]
1107 // Check if there is a bind for this
1108 // directory, in which case assume we don't need .keep
1109 if (containerPath == bind || strings.HasPrefix(containerPath, bind+"/")) && mnt.PortableDataHash != "d41d8cd98f00b204e9800998ecf8427e+0" {
1113 outputSuffix := path[len(runner.HostOutputDir)+1:]
1114 return fmt.Sprintf("./%v d41d8cd98f00b204e9800998ecf8427e+0 0:0:.keep\n", outputSuffix), nil
1119 if followed >= limitFollowSymlinks {
1120 // Got stuck in a loop or just a pathological number of
1121 // directory links, give up.
1122 err = fmt.Errorf("Followed more than %v symlinks from path %q", limitFollowSymlinks, path)
1126 // "path" is the actual path we are visiting
1127 // "tgt" is the target of "path" (a non-symlink) after following symlinks
1128 // "relocated" is the path in the output manifest where the file should be placed,
1129 // but has HostOutputDir as a prefix.
1131 // The destination path in the output manifest may need to be
1132 // logically relocated to some other path in order to appear
1133 // in the correct location as a result of following a symlink.
1134 // Remove the relocateFrom prefix and replace it with
1136 relocated := relocateTo + path[len(relocateFrom):]
1138 tgt, readlinktgt, info, derefErr := runner.derefOutputSymlink(path, info)
1139 if derefErr != nil && derefErr != ErrNotInOutputDir {
1143 // go through mounts and try reverse map to collection reference
1144 for _, bind := range binds {
1145 mnt := runner.Container.Mounts[bind]
1146 if (tgt == bind || strings.HasPrefix(tgt, bind+"/")) && !mnt.Writable {
1147 // get path relative to bind
1148 targetSuffix := tgt[len(bind):]
1150 // Copy mount and adjust the path to add path relative to the bind
1151 adjustedMount := mnt
1152 adjustedMount.Path = filepath.Join(adjustedMount.Path, targetSuffix)
1154 // Terminates in this keep mount, so add the
1155 // manifest text at appropriate location.
1156 outputSuffix := relocated[len(runner.HostOutputDir):]
1157 manifestText, err = runner.getCollectionManifestForPath(adjustedMount, outputSuffix)
1162 // If target is not a collection mount, it must be located within the
1163 // output directory, otherwise it is an error.
1164 if derefErr == ErrNotInOutputDir {
1165 err = fmt.Errorf("Symlink in output %q points to invalid location %q, must point to path within the output directory.",
1166 path[len(runner.HostOutputDir):], readlinktgt)
1170 if info.Mode().IsRegular() {
1171 return "", walkUpload.UploadFile(relocated, tgt)
1174 if info.Mode().IsDir() {
1175 // Symlink leads to directory. Walk() doesn't follow
1176 // directory symlinks, so we walk the target directory
1177 // instead. Within the walk, file paths are relocated
1178 // so they appear under the original symlink path.
1179 err = filepath.Walk(tgt, func(walkpath string, walkinfo os.FileInfo, walkerr error) error {
1181 m, walkerr = runner.UploadOutputFile(walkpath, walkinfo, walkerr,
1182 binds, walkUpload, tgt, relocated, followed+1)
1184 manifestText = manifestText + m
1194 // HandleOutput sets the output, unmounts the FUSE mount, and deletes temporary directories
1195 func (runner *ContainerRunner) CaptureOutput() error {
1196 if wantAPI := runner.Container.RuntimeConstraints.API; wantAPI != nil && *wantAPI {
1197 // Output may have been set directly by the container, so
1198 // refresh the container record to check.
1199 err := runner.ArvClient.Get("containers", runner.Container.UUID,
1200 nil, &runner.Container)
1204 if runner.Container.Output != "" {
1205 // Container output is already set.
1206 runner.OutputPDH = &runner.Container.Output
1211 if runner.HostOutputDir == "" {
1215 _, err := os.Stat(runner.HostOutputDir)
1217 return fmt.Errorf("While checking host output path: %v", err)
1220 // Pre-populate output from the configured mount points
1222 for bind, mnt := range runner.Container.Mounts {
1223 if mnt.Kind == "collection" {
1224 binds = append(binds, bind)
1229 var manifestText string
1231 collectionMetafile := fmt.Sprintf("%s/.arvados#collection", runner.HostOutputDir)
1232 _, err = os.Stat(collectionMetafile)
1234 // Regular directory
1236 cw := CollectionWriter{0, runner.Kc, nil, nil, sync.Mutex{}}
1237 walkUpload := cw.BeginUpload(runner.HostOutputDir, runner.CrunchLog.Logger)
1240 err = filepath.Walk(runner.HostOutputDir, func(path string, info os.FileInfo, err error) error {
1241 m, err = runner.UploadOutputFile(path, info, err, binds, walkUpload, "", "", 0)
1243 manifestText = manifestText + m
1248 cw.EndUpload(walkUpload)
1251 return fmt.Errorf("While uploading output files: %v", err)
1254 m, err = cw.ManifestText()
1255 manifestText = manifestText + m
1257 return fmt.Errorf("While uploading output files: %v", err)
1260 // FUSE mount directory
1261 file, openerr := os.Open(collectionMetafile)
1263 return fmt.Errorf("While opening FUSE metafile: %v", err)
1267 var rec arvados.Collection
1268 err = json.NewDecoder(file).Decode(&rec)
1270 return fmt.Errorf("While reading FUSE metafile: %v", err)
1272 manifestText = rec.ManifestText
1275 for _, bind := range binds {
1276 mnt := runner.Container.Mounts[bind]
1278 bindSuffix := strings.TrimPrefix(bind, runner.Container.OutputPath)
1280 if bindSuffix == bind || len(bindSuffix) <= 0 {
1281 // either does not start with OutputPath or is OutputPath itself
1285 if mnt.ExcludeFromOutput == true || mnt.Writable {
1289 // append to manifest_text
1290 m, err := runner.getCollectionManifestForPath(mnt, bindSuffix)
1295 manifestText = manifestText + m
1299 var response arvados.Collection
1300 manifest := manifest.Manifest{Text: manifestText}
1301 manifestText = manifest.Extract(".", ".").Text
1302 err = runner.ArvClient.Create("collections",
1304 "ensure_unique_name": true,
1305 "collection": arvadosclient.Dict{
1307 "name": "output for " + runner.Container.UUID,
1308 "manifest_text": manifestText}},
1311 return fmt.Errorf("While creating output collection: %v", err)
1313 runner.OutputPDH = &response.PortableDataHash
1317 var outputCollections = make(map[string]arvados.Collection)
1319 // Fetch the collection for the mnt.PortableDataHash
1320 // Return the manifest_text fragment corresponding to the specified mnt.Path
1321 // after making any required updates.
1323 // If mnt.Path is not specified,
1324 // return the entire manifest_text after replacing any "." with bindSuffix
1325 // If mnt.Path corresponds to one stream,
1326 // return the manifest_text for that stream after replacing that stream name with bindSuffix
1327 // Otherwise, check if a filename in any one stream is being sought. Return the manifest_text
1328 // for that stream after replacing stream name with bindSuffix minus the last word
1329 // and the file name with last word of the bindSuffix
1330 // Allowed path examples:
1332 // "path":"/subdir1"
1333 // "path":"/subdir1/subdir2"
1334 // "path":"/subdir/filename" etc
1335 func (runner *ContainerRunner) getCollectionManifestForPath(mnt arvados.Mount, bindSuffix string) (string, error) {
1336 collection := outputCollections[mnt.PortableDataHash]
1337 if collection.PortableDataHash == "" {
1338 err := runner.ArvClient.Get("collections", mnt.PortableDataHash, nil, &collection)
1340 return "", fmt.Errorf("While getting collection for %v: %v", mnt.PortableDataHash, err)
1342 outputCollections[mnt.PortableDataHash] = collection
1345 if collection.ManifestText == "" {
1346 runner.CrunchLog.Printf("No manifest text for collection %v", collection.PortableDataHash)
1350 mft := manifest.Manifest{Text: collection.ManifestText}
1351 extracted := mft.Extract(mnt.Path, bindSuffix)
1352 if extracted.Err != nil {
1353 return "", fmt.Errorf("Error parsing manifest for %v: %v", mnt.PortableDataHash, extracted.Err.Error())
1355 return extracted.Text, nil
1358 func (runner *ContainerRunner) CleanupDirs() {
1359 if runner.ArvMount != nil {
1361 umount := exec.Command("arv-mount", fmt.Sprintf("--unmount-timeout=%d", delay), "--unmount", runner.ArvMountPoint)
1362 umount.Stdout = runner.CrunchLog
1363 umount.Stderr = runner.CrunchLog
1364 runner.CrunchLog.Printf("Running %v", umount.Args)
1365 umnterr := umount.Start()
1368 runner.CrunchLog.Printf("Error unmounting: %v", umnterr)
1370 // If arv-mount --unmount gets stuck for any reason, we
1371 // don't want to wait for it forever. Do Wait() in a goroutine
1372 // so it doesn't block crunch-run.
1373 umountExit := make(chan error)
1375 mnterr := umount.Wait()
1377 runner.CrunchLog.Printf("Error unmounting: %v", mnterr)
1379 umountExit <- mnterr
1382 for again := true; again; {
1388 case <-runner.ArvMountExit:
1390 case <-time.After(time.Duration((delay + 1) * int64(time.Second))):
1391 runner.CrunchLog.Printf("Timed out waiting for unmount")
1393 umount.Process.Kill()
1395 runner.ArvMount.Process.Kill()
1401 if runner.ArvMountPoint != "" {
1402 if rmerr := os.Remove(runner.ArvMountPoint); rmerr != nil {
1403 runner.CrunchLog.Printf("While cleaning up arv-mount directory %s: %v", runner.ArvMountPoint, rmerr)
1407 for _, tmpdir := range runner.CleanupTempDir {
1408 if rmerr := os.RemoveAll(tmpdir); rmerr != nil {
1409 runner.CrunchLog.Printf("While cleaning up temporary directory %s: %v", tmpdir, rmerr)
1414 // CommitLogs posts the collection containing the final container logs.
1415 func (runner *ContainerRunner) CommitLogs() error {
1416 runner.CrunchLog.Print(runner.finalState)
1418 if runner.arvMountLog != nil {
1419 runner.arvMountLog.Close()
1421 runner.CrunchLog.Close()
1423 // Closing CrunchLog above allows them to be committed to Keep at this
1424 // point, but re-open crunch log with ArvClient in case there are any
1425 // other further errors (such as failing to write the log to Keep!)
1426 // while shutting down
1427 runner.CrunchLog = NewThrottledLogger(&ArvLogWriter{ArvClient: runner.ArvClient,
1428 UUID: runner.Container.UUID, loggingStream: "crunch-run", writeCloser: nil})
1429 runner.CrunchLog.Immediate = log.New(os.Stderr, runner.Container.UUID+" ", 0)
1431 if runner.LogsPDH != nil {
1432 // If we have already assigned something to LogsPDH,
1433 // we must be closing the re-opened log, which won't
1434 // end up getting attached to the container record and
1435 // therefore doesn't need to be saved as a collection
1436 // -- it exists only to send logs to other channels.
1440 mt, err := runner.LogCollection.ManifestText()
1442 return fmt.Errorf("While creating log manifest: %v", err)
1445 var response arvados.Collection
1446 err = runner.ArvClient.Create("collections",
1448 "ensure_unique_name": true,
1449 "collection": arvadosclient.Dict{
1451 "name": "logs for " + runner.Container.UUID,
1452 "manifest_text": mt}},
1455 return fmt.Errorf("While creating log collection: %v", err)
1457 runner.LogsPDH = &response.PortableDataHash
1461 // UpdateContainerRunning updates the container state to "Running"
1462 func (runner *ContainerRunner) UpdateContainerRunning() error {
1463 runner.cStateLock.Lock()
1464 defer runner.cStateLock.Unlock()
1465 if runner.cCancelled {
1468 return runner.ArvClient.Update("containers", runner.Container.UUID,
1469 arvadosclient.Dict{"container": arvadosclient.Dict{"state": "Running"}}, nil)
1472 // ContainerToken returns the api_token the container (and any
1473 // arv-mount processes) are allowed to use.
1474 func (runner *ContainerRunner) ContainerToken() (string, error) {
1475 if runner.token != "" {
1476 return runner.token, nil
1479 var auth arvados.APIClientAuthorization
1480 err := runner.ArvClient.Call("GET", "containers", runner.Container.UUID, "auth", nil, &auth)
1484 runner.token = auth.APIToken
1485 return runner.token, nil
1488 // UpdateContainerComplete updates the container record state on API
1489 // server to "Complete" or "Cancelled"
1490 func (runner *ContainerRunner) UpdateContainerFinal() error {
1491 update := arvadosclient.Dict{}
1492 update["state"] = runner.finalState
1493 if runner.LogsPDH != nil {
1494 update["log"] = *runner.LogsPDH
1496 if runner.finalState == "Complete" {
1497 if runner.ExitCode != nil {
1498 update["exit_code"] = *runner.ExitCode
1500 if runner.OutputPDH != nil {
1501 update["output"] = *runner.OutputPDH
1504 return runner.ArvClient.Update("containers", runner.Container.UUID, arvadosclient.Dict{"container": update}, nil)
1507 // IsCancelled returns the value of Cancelled, with goroutine safety.
1508 func (runner *ContainerRunner) IsCancelled() bool {
1509 runner.cStateLock.Lock()
1510 defer runner.cStateLock.Unlock()
1511 return runner.cCancelled
1514 // NewArvLogWriter creates an ArvLogWriter
1515 func (runner *ContainerRunner) NewArvLogWriter(name string) io.WriteCloser {
1516 return &ArvLogWriter{
1517 ArvClient: runner.ArvClient,
1518 UUID: runner.Container.UUID,
1519 loggingStream: name,
1520 writeCloser: runner.LogCollection.Open(name + ".txt")}
1523 // Run the full container lifecycle.
1524 func (runner *ContainerRunner) Run() (err error) {
1525 runner.CrunchLog.Printf("crunch-run %s started", version)
1526 runner.CrunchLog.Printf("Executing container '%s'", runner.Container.UUID)
1528 hostname, hosterr := os.Hostname()
1530 runner.CrunchLog.Printf("Error getting hostname '%v'", hosterr)
1532 runner.CrunchLog.Printf("Executing on host '%s'", hostname)
1535 runner.finalState = "Queued"
1538 runner.stopSignals()
1539 runner.CleanupDirs()
1541 runner.CrunchLog.Printf("crunch-run finished")
1542 runner.CrunchLog.Close()
1546 // checkErr prints e (unless it's nil) and sets err to
1547 // e (unless err is already non-nil). Thus, if err
1548 // hasn't already been assigned when Run() returns,
1549 // this cleanup func will cause Run() to return the
1550 // first non-nil error that is passed to checkErr().
1551 checkErr := func(e error) {
1555 runner.CrunchLog.Print(e)
1559 if runner.finalState == "Complete" {
1560 // There was an error in the finalization.
1561 runner.finalState = "Cancelled"
1565 // Log the error encountered in Run(), if any
1568 if runner.finalState == "Queued" {
1569 runner.UpdateContainerFinal()
1573 if runner.IsCancelled() {
1574 runner.finalState = "Cancelled"
1575 // but don't return yet -- we still want to
1576 // capture partial output and write logs
1579 checkErr(runner.CaptureOutput())
1580 checkErr(runner.CommitLogs())
1581 checkErr(runner.UpdateContainerFinal())
1584 err = runner.fetchContainerRecord()
1589 // setup signal handling
1590 runner.setupSignals()
1592 // check for and/or load image
1593 err = runner.LoadImage()
1595 if !runner.checkBrokenNode(err) {
1596 // Failed to load image but not due to a "broken node"
1597 // condition, probably user error.
1598 runner.finalState = "Cancelled"
1600 err = fmt.Errorf("While loading container image: %v", err)
1604 // set up FUSE mount and binds
1605 err = runner.SetupMounts()
1607 runner.finalState = "Cancelled"
1608 err = fmt.Errorf("While setting up mounts: %v", err)
1612 err = runner.CreateContainer()
1616 err = runner.LogHostInfo()
1620 err = runner.LogNodeRecord()
1624 err = runner.LogContainerRecord()
1629 if runner.IsCancelled() {
1633 err = runner.UpdateContainerRunning()
1637 runner.finalState = "Cancelled"
1639 runner.StartCrunchstat()
1641 err = runner.StartContainer()
1643 runner.checkBrokenNode(err)
1647 err = runner.WaitFinish()
1648 if err == nil && !runner.IsCancelled() {
1649 runner.finalState = "Complete"
1654 // Fetch the current container record (uuid = runner.Container.UUID)
1655 // into runner.Container.
1656 func (runner *ContainerRunner) fetchContainerRecord() error {
1657 reader, err := runner.ArvClient.CallRaw("GET", "containers", runner.Container.UUID, "", nil)
1659 return fmt.Errorf("error fetching container record: %v", err)
1661 defer reader.Close()
1663 dec := json.NewDecoder(reader)
1665 err = dec.Decode(&runner.Container)
1667 return fmt.Errorf("error decoding container record: %v", err)
1672 // NewContainerRunner creates a new container runner.
1673 func NewContainerRunner(api IArvadosClient,
1675 docker ThinDockerClient,
1676 containerUUID string) *ContainerRunner {
1678 cr := &ContainerRunner{ArvClient: api, Kc: kc, Docker: docker}
1679 cr.NewLogWriter = cr.NewArvLogWriter
1680 cr.RunArvMount = cr.ArvMountCmd
1681 cr.MkTempDir = ioutil.TempDir
1682 cr.LogCollection = &CollectionWriter{0, kc, nil, nil, sync.Mutex{}}
1683 cr.Container.UUID = containerUUID
1684 cr.CrunchLog = NewThrottledLogger(cr.NewLogWriter("crunch-run"))
1685 cr.CrunchLog.Immediate = log.New(os.Stderr, containerUUID+" ", 0)
1687 loadLogThrottleParams(api)
1693 statInterval := flag.Duration("crunchstat-interval", 10*time.Second, "sampling period for periodic resource usage reporting")
1694 cgroupRoot := flag.String("cgroup-root", "/sys/fs/cgroup", "path to sysfs cgroup tree")
1695 cgroupParent := flag.String("cgroup-parent", "docker", "name of container's parent cgroup (ignored if -cgroup-parent-subsystem is used)")
1696 cgroupParentSubsystem := flag.String("cgroup-parent-subsystem", "", "use current cgroup for given subsystem as parent cgroup for container")
1697 caCertsPath := flag.String("ca-certs", "", "Path to TLS root certificates")
1698 enableNetwork := flag.String("container-enable-networking", "default",
1699 `Specify if networking should be enabled for container. One of 'default', 'always':
1700 default: only enable networking if container requests it.
1701 always: containers always have networking enabled
1703 networkMode := flag.String("container-network-mode", "default",
1704 `Set networking mode for container. Corresponds to Docker network mode (--net).
1706 memprofile := flag.String("memprofile", "", "write memory profile to `file` after running container")
1707 getVersion := flag.Bool("version", false, "Print version information and exit.")
1710 // Print version information if requested
1712 fmt.Printf("crunch-run %s\n", version)
1716 log.Printf("crunch-run %s started", version)
1718 containerId := flag.Arg(0)
1720 if *caCertsPath != "" {
1721 arvadosclient.CertFiles = []string{*caCertsPath}
1724 api, err := arvadosclient.MakeArvadosClient()
1726 log.Fatalf("%s: %v", containerId, err)
1730 kc, kcerr := keepclient.MakeKeepClient(api)
1732 log.Fatalf("%s: %v", containerId, kcerr)
1734 kc.BlockCache = &keepclient.BlockCache{MaxBlocks: 2}
1737 // API version 1.21 corresponds to Docker 1.9, which is currently the
1738 // minimum version we want to support.
1739 docker, dockererr := dockerclient.NewClient(dockerclient.DefaultDockerHost, "1.21", nil, nil)
1741 cr := NewContainerRunner(api, kc, docker, containerId)
1742 if dockererr != nil {
1743 cr.CrunchLog.Printf("%s: %v", containerId, dockererr)
1744 cr.checkBrokenNode(dockererr)
1745 cr.CrunchLog.Close()
1749 cr.statInterval = *statInterval
1750 cr.cgroupRoot = *cgroupRoot
1751 cr.expectCgroupParent = *cgroupParent
1752 cr.enableNetwork = *enableNetwork
1753 cr.networkMode = *networkMode
1754 if *cgroupParentSubsystem != "" {
1755 p := findCgroup(*cgroupParentSubsystem)
1756 cr.setCgroupParent = p
1757 cr.expectCgroupParent = p
1762 if *memprofile != "" {
1763 f, err := os.Create(*memprofile)
1765 log.Printf("could not create memory profile: ", err)
1767 runtime.GC() // get up-to-date statistics
1768 if err := pprof.WriteHeapProfile(f); err != nil {
1769 log.Printf("could not write memory profile: ", err)
1771 closeerr := f.Close()
1772 if closeerr != nil {
1773 log.Printf("closing memprofile file: ", err)
1778 log.Fatalf("%s: %v", containerId, runerr)