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
117 statLogger io.WriteCloser
118 statReporter *crunchstat.Reporter
119 hoststatLogger io.WriteCloser
120 hoststatReporter *crunchstat.Reporter
121 statInterval time.Duration
123 // What we expect the container's cgroup parent to be.
124 expectCgroupParent string
125 // What we tell docker to use as the container's cgroup
126 // parent. Note: Ideally we would use the same field for both
127 // expectCgroupParent and setCgroupParent, and just make it
128 // default to "docker". However, when using docker < 1.10 with
129 // systemd, specifying a non-empty cgroup parent (even the
130 // default value "docker") hits a docker bug
131 // (https://github.com/docker/docker/issues/17126). Using two
132 // separate fields makes it possible to use the "expect cgroup
133 // parent to be X" feature even on sites where the "specify
134 // cgroup parent" feature breaks.
135 setCgroupParent string
137 cStateLock sync.Mutex
138 cCancelled bool // StopContainer() invoked
140 enableNetwork string // one of "default" or "always"
141 networkMode string // passed through to HostConfig.NetworkMode
142 arvMountLog *ThrottledLogger
145 // setupSignals sets up signal handling to gracefully terminate the underlying
146 // Docker container and update state when receiving a TERM, INT or QUIT signal.
147 func (runner *ContainerRunner) setupSignals() {
148 runner.SigChan = make(chan os.Signal, 1)
149 signal.Notify(runner.SigChan, syscall.SIGTERM)
150 signal.Notify(runner.SigChan, syscall.SIGINT)
151 signal.Notify(runner.SigChan, syscall.SIGQUIT)
153 go func(sig chan os.Signal) {
155 runner.CrunchLog.Printf("caught signal: %v", s)
161 // stop the underlying Docker container.
162 func (runner *ContainerRunner) stop() {
163 runner.cStateLock.Lock()
164 defer runner.cStateLock.Unlock()
165 if runner.ContainerID == "" {
168 runner.cCancelled = true
169 runner.CrunchLog.Printf("removing container")
170 err := runner.Docker.ContainerRemove(context.TODO(), runner.ContainerID, dockertypes.ContainerRemoveOptions{Force: true})
172 runner.CrunchLog.Printf("error removing container: %s", err)
176 func (runner *ContainerRunner) stopSignals() {
177 if runner.SigChan != nil {
178 signal.Stop(runner.SigChan)
182 var errorBlacklist = []string{
183 "(?ms).*[Cc]annot connect to the Docker daemon.*",
184 "(?ms).*oci runtime error.*starting container process.*container init.*mounting.*to rootfs.*no such file or directory.*",
186 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)")
188 func (runner *ContainerRunner) checkBrokenNode(goterr error) bool {
189 for _, d := range errorBlacklist {
190 if m, e := regexp.MatchString(d, goterr.Error()); m && e == nil {
191 runner.CrunchLog.Printf("Error suggests node is unable to run containers: %v", goterr)
192 if *brokenNodeHook == "" {
193 runner.CrunchLog.Printf("No broken node hook provided, cannot mark node as broken.")
195 runner.CrunchLog.Printf("Running broken node hook %q", *brokenNodeHook)
197 c := exec.Command(*brokenNodeHook)
198 c.Stdout = runner.CrunchLog
199 c.Stderr = runner.CrunchLog
202 runner.CrunchLog.Printf("Error running broken node hook: %v", err)
211 // LoadImage determines the docker image id from the container record and
212 // checks if it is available in the local Docker image store. If not, it loads
213 // the image from Keep.
214 func (runner *ContainerRunner) LoadImage() (err error) {
216 runner.CrunchLog.Printf("Fetching Docker image from collection '%s'", runner.Container.ContainerImage)
218 var collection arvados.Collection
219 err = runner.ArvClient.Get("collections", runner.Container.ContainerImage, nil, &collection)
221 return fmt.Errorf("While getting container image collection: %v", err)
223 manifest := manifest.Manifest{Text: collection.ManifestText}
224 var img, imageID string
225 for ms := range manifest.StreamIter() {
226 img = ms.FileStreamSegments[0].Name
227 if !strings.HasSuffix(img, ".tar") {
228 return fmt.Errorf("First file in the container image collection does not end in .tar")
230 imageID = img[:len(img)-4]
233 runner.CrunchLog.Printf("Using Docker image id '%s'", imageID)
235 _, _, err = runner.Docker.ImageInspectWithRaw(context.TODO(), imageID)
237 runner.CrunchLog.Print("Loading Docker image from keep")
239 var readCloser io.ReadCloser
240 readCloser, err = runner.Kc.ManifestFileReader(manifest, img)
242 return fmt.Errorf("While creating ManifestFileReader for container image: %v", err)
245 response, err := runner.Docker.ImageLoad(context.TODO(), readCloser, true)
247 return fmt.Errorf("While loading container image into Docker: %v", err)
250 defer response.Body.Close()
251 rbody, err := ioutil.ReadAll(response.Body)
253 return fmt.Errorf("Reading response to image load: %v", err)
255 runner.CrunchLog.Printf("Docker response: %s", rbody)
257 runner.CrunchLog.Print("Docker image is available")
260 runner.ContainerConfig.Image = imageID
262 runner.Kc.ClearBlockCache()
267 func (runner *ContainerRunner) ArvMountCmd(arvMountCmd []string, token string) (c *exec.Cmd, err error) {
268 c = exec.Command("arv-mount", arvMountCmd...)
270 // Copy our environment, but override ARVADOS_API_TOKEN with
271 // the container auth token.
273 for _, s := range os.Environ() {
274 if !strings.HasPrefix(s, "ARVADOS_API_TOKEN=") {
275 c.Env = append(c.Env, s)
278 c.Env = append(c.Env, "ARVADOS_API_TOKEN="+token)
280 runner.arvMountLog = NewThrottledLogger(runner.NewLogWriter("arv-mount"))
281 c.Stdout = runner.arvMountLog
282 c.Stderr = runner.arvMountLog
284 runner.CrunchLog.Printf("Running %v", c.Args)
291 statReadme := make(chan bool)
292 runner.ArvMountExit = make(chan error)
297 time.Sleep(100 * time.Millisecond)
298 _, err = os.Stat(fmt.Sprintf("%s/by_id/README", runner.ArvMountPoint))
310 runner.CrunchLog.Printf("Arv-mount exit error: %v", mnterr)
312 runner.ArvMountExit <- mnterr
313 close(runner.ArvMountExit)
319 case err := <-runner.ArvMountExit:
320 runner.ArvMount = nil
328 func (runner *ContainerRunner) SetupArvMountPoint(prefix string) (err error) {
329 if runner.ArvMountPoint == "" {
330 runner.ArvMountPoint, err = runner.MkTempDir(runner.parentTemp, prefix)
335 func copyfile(src string, dst string) (err error) {
336 srcfile, err := os.Open(src)
341 os.MkdirAll(path.Dir(dst), 0777)
343 dstfile, err := os.Create(dst)
347 _, err = io.Copy(dstfile, srcfile)
352 err = srcfile.Close()
353 err2 := dstfile.Close()
366 func (runner *ContainerRunner) SetupMounts() (err error) {
367 err = runner.SetupArvMountPoint("keep")
369 return fmt.Errorf("While creating keep mount temp dir: %v", err)
372 token, err := runner.ContainerToken()
374 return fmt.Errorf("could not get container token: %s", err)
379 arvMountCmd := []string{
383 fmt.Sprintf("--crunchstat-interval=%v", runner.statInterval.Seconds())}
385 if runner.Container.RuntimeConstraints.KeepCacheRAM > 0 {
386 arvMountCmd = append(arvMountCmd, "--file-cache", fmt.Sprintf("%d", runner.Container.RuntimeConstraints.KeepCacheRAM))
389 collectionPaths := []string{}
391 runner.Volumes = make(map[string]struct{})
392 needCertMount := true
393 type copyFile struct {
397 var copyFiles []copyFile
400 for bind := range runner.Container.Mounts {
401 binds = append(binds, bind)
405 for _, bind := range binds {
406 mnt := runner.Container.Mounts[bind]
407 if bind == "stdout" || bind == "stderr" {
408 // Is it a "file" mount kind?
409 if mnt.Kind != "file" {
410 return fmt.Errorf("Unsupported mount kind '%s' for %s. Only 'file' is supported.", mnt.Kind, bind)
413 // Does path start with OutputPath?
414 prefix := runner.Container.OutputPath
415 if !strings.HasSuffix(prefix, "/") {
418 if !strings.HasPrefix(mnt.Path, prefix) {
419 return fmt.Errorf("%s path does not start with OutputPath: %s, %s", strings.Title(bind), mnt.Path, prefix)
424 // Is it a "collection" mount kind?
425 if mnt.Kind != "collection" && mnt.Kind != "json" {
426 return fmt.Errorf("Unsupported mount kind '%s' for stdin. Only 'collection' or 'json' are supported.", mnt.Kind)
430 if bind == "/etc/arvados/ca-certificates.crt" {
431 needCertMount = false
434 if strings.HasPrefix(bind, runner.Container.OutputPath+"/") && bind != runner.Container.OutputPath+"/" {
435 if mnt.Kind != "collection" {
436 return fmt.Errorf("Only mount points of kind 'collection' are supported underneath the output_path: %v", bind)
441 case mnt.Kind == "collection" && bind != "stdin":
443 if mnt.UUID != "" && mnt.PortableDataHash != "" {
444 return fmt.Errorf("Cannot specify both 'uuid' and 'portable_data_hash' for a collection mount")
448 return fmt.Errorf("Writing to existing collections currently not permitted.")
451 src = fmt.Sprintf("%s/by_id/%s", runner.ArvMountPoint, mnt.UUID)
452 } else if mnt.PortableDataHash != "" {
453 if mnt.Writable && !strings.HasPrefix(bind, runner.Container.OutputPath+"/") {
454 return fmt.Errorf("Can never write to a collection specified by portable data hash")
456 idx := strings.Index(mnt.PortableDataHash, "/")
458 mnt.Path = path.Clean(mnt.PortableDataHash[idx:])
459 mnt.PortableDataHash = mnt.PortableDataHash[0:idx]
460 runner.Container.Mounts[bind] = mnt
462 src = fmt.Sprintf("%s/by_id/%s", runner.ArvMountPoint, mnt.PortableDataHash)
463 if mnt.Path != "" && mnt.Path != "." {
464 if strings.HasPrefix(mnt.Path, "./") {
465 mnt.Path = mnt.Path[2:]
466 } else if strings.HasPrefix(mnt.Path, "/") {
467 mnt.Path = mnt.Path[1:]
469 src += "/" + mnt.Path
472 src = fmt.Sprintf("%s/tmp%d", runner.ArvMountPoint, tmpcount)
473 arvMountCmd = append(arvMountCmd, "--mount-tmp")
474 arvMountCmd = append(arvMountCmd, fmt.Sprintf("tmp%d", tmpcount))
478 if bind == runner.Container.OutputPath {
479 runner.HostOutputDir = src
480 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s", src, bind))
481 } else if strings.HasPrefix(bind, runner.Container.OutputPath+"/") {
482 copyFiles = append(copyFiles, copyFile{src, runner.HostOutputDir + bind[len(runner.Container.OutputPath):]})
484 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s", src, bind))
487 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s:ro", src, bind))
489 collectionPaths = append(collectionPaths, src)
491 case mnt.Kind == "tmp":
493 tmpdir, err = runner.MkTempDir(runner.parentTemp, "tmp")
495 return fmt.Errorf("While creating mount temp dir: %v", err)
497 st, staterr := os.Stat(tmpdir)
499 return fmt.Errorf("While Stat on temp dir: %v", staterr)
501 err = os.Chmod(tmpdir, st.Mode()|os.ModeSetgid|0777)
503 return fmt.Errorf("While Chmod temp dir: %v", err)
505 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s", tmpdir, bind))
506 if bind == runner.Container.OutputPath {
507 runner.HostOutputDir = tmpdir
510 case mnt.Kind == "json":
511 jsondata, err := json.Marshal(mnt.Content)
513 return fmt.Errorf("encoding json data: %v", err)
515 // Create a tempdir with a single file
516 // (instead of just a tempfile): this way we
517 // can ensure the file is world-readable
518 // inside the container, without having to
519 // make it world-readable on the docker host.
520 tmpdir, err := runner.MkTempDir(runner.parentTemp, "json")
522 return fmt.Errorf("creating temp dir: %v", err)
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(runner.parentTemp, "git_tree")
534 return fmt.Errorf("creating temp dir: %v", err)
536 err = gitMount(mnt).extractTree(runner.ArvClient, tmpdir, token)
540 runner.Binds = append(runner.Binds, tmpdir+":"+bind+":ro")
544 if runner.HostOutputDir == "" {
545 return fmt.Errorf("Output path does not correspond to a writable mount point")
548 if wantAPI := runner.Container.RuntimeConstraints.API; needCertMount && wantAPI != nil && *wantAPI {
549 for _, certfile := range arvadosclient.CertFiles {
550 _, err := os.Stat(certfile)
552 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:/etc/arvados/ca-certificates.crt:ro", certfile))
559 arvMountCmd = append(arvMountCmd, "--mount-by-pdh", "by_id")
561 arvMountCmd = append(arvMountCmd, "--mount-by-id", "by_id")
563 arvMountCmd = append(arvMountCmd, runner.ArvMountPoint)
565 runner.ArvMount, err = runner.RunArvMount(arvMountCmd, token)
567 return fmt.Errorf("While trying to start arv-mount: %v", err)
570 for _, p := range collectionPaths {
573 return fmt.Errorf("While checking that input files exist: %v", err)
577 for _, cp := range copyFiles {
578 dir, err := os.Stat(cp.src)
580 return fmt.Errorf("While staging writable file from %q to %q: %v", cp.src, cp.bind, err)
583 err = filepath.Walk(cp.src, func(walkpath string, walkinfo os.FileInfo, walkerr error) error {
587 if walkinfo.Mode().IsRegular() {
588 return copyfile(walkpath, path.Join(cp.bind, walkpath[len(cp.src):]))
589 } else if walkinfo.Mode().IsDir() {
590 return os.MkdirAll(path.Join(cp.bind, walkpath[len(cp.src):]), 0777)
592 return fmt.Errorf("Source %q is not a regular file or directory", cp.src)
596 err = copyfile(cp.src, cp.bind)
599 return fmt.Errorf("While staging writable file from %q to %q: %v", cp.src, cp.bind, err)
606 func (runner *ContainerRunner) ProcessDockerAttach(containerReader io.Reader) {
607 // Handle docker log protocol
608 // https://docs.docker.com/engine/reference/api/docker_remote_api_v1.15/#attach-to-a-container
609 defer close(runner.loggingDone)
611 header := make([]byte, 8)
614 _, err = io.ReadAtLeast(containerReader, header, 8)
621 readsize := int64(header[7]) | (int64(header[6]) << 8) | (int64(header[5]) << 16) | (int64(header[4]) << 24)
624 _, err = io.CopyN(runner.Stdout, containerReader, readsize)
627 _, err = io.CopyN(runner.Stderr, containerReader, readsize)
632 runner.CrunchLog.Printf("error reading docker logs: %v", err)
635 err = runner.Stdout.Close()
637 runner.CrunchLog.Printf("error closing stdout logs: %v", err)
640 err = runner.Stderr.Close()
642 runner.CrunchLog.Printf("error closing stderr logs: %v", err)
645 if runner.statReporter != nil {
646 runner.statReporter.Stop()
647 err = runner.statLogger.Close()
649 runner.CrunchLog.Printf("error closing crunchstat logs: %v", err)
654 func (runner *ContainerRunner) stopHoststat() error {
655 if runner.hoststatReporter == nil {
658 runner.hoststatReporter.Stop()
659 err := runner.hoststatLogger.Close()
661 return fmt.Errorf("error closing hoststat logs: %v", err)
666 func (runner *ContainerRunner) startHoststat() {
667 runner.hoststatLogger = NewThrottledLogger(runner.NewLogWriter("hoststat"))
668 runner.hoststatReporter = &crunchstat.Reporter{
669 Logger: log.New(runner.hoststatLogger, "", 0),
670 CgroupRoot: runner.cgroupRoot,
671 PollPeriod: runner.statInterval,
673 runner.hoststatReporter.Start()
676 func (runner *ContainerRunner) startCrunchstat() {
677 runner.statLogger = NewThrottledLogger(runner.NewLogWriter("crunchstat"))
678 runner.statReporter = &crunchstat.Reporter{
679 CID: runner.ContainerID,
680 Logger: log.New(runner.statLogger, "", 0),
681 CgroupParent: runner.expectCgroupParent,
682 CgroupRoot: runner.cgroupRoot,
683 PollPeriod: runner.statInterval,
685 runner.statReporter.Start()
688 type infoCommand struct {
693 // LogHostInfo logs info about the current host, for debugging and
694 // accounting purposes. Although it's logged as "node-info", this is
695 // about the environment where crunch-run is actually running, which
696 // might differ from what's described in the node record (see
698 func (runner *ContainerRunner) LogHostInfo() (err error) {
699 w := runner.NewLogWriter("node-info")
701 commands := []infoCommand{
703 label: "Host Information",
704 cmd: []string{"uname", "-a"},
707 label: "CPU Information",
708 cmd: []string{"cat", "/proc/cpuinfo"},
711 label: "Memory Information",
712 cmd: []string{"cat", "/proc/meminfo"},
716 cmd: []string{"df", "-m", "/", os.TempDir()},
719 label: "Disk INodes",
720 cmd: []string{"df", "-i", "/", os.TempDir()},
724 // Run commands with informational output to be logged.
725 for _, command := range commands {
726 fmt.Fprintln(w, command.label)
727 cmd := exec.Command(command.cmd[0], command.cmd[1:]...)
730 if err := cmd.Run(); err != nil {
731 err = fmt.Errorf("While running command %q: %v", command.cmd, err)
740 return fmt.Errorf("While closing node-info logs: %v", err)
745 // LogContainerRecord gets and saves the raw JSON container record from the API server
746 func (runner *ContainerRunner) LogContainerRecord() error {
747 logged, err := runner.logAPIResponse("container", "containers", map[string]interface{}{"filters": [][]string{{"uuid", "=", runner.Container.UUID}}}, nil)
748 if !logged && err == nil {
749 err = fmt.Errorf("error: no container record found for %s", runner.Container.UUID)
754 // LogNodeRecord logs arvados#node record corresponding to the current host.
755 func (runner *ContainerRunner) LogNodeRecord() error {
756 hostname := os.Getenv("SLURMD_NODENAME")
758 hostname, _ = os.Hostname()
760 _, err := runner.logAPIResponse("node", "nodes", map[string]interface{}{"filters": [][]string{{"hostname", "=", hostname}}}, func(resp interface{}) {
761 // The "info" field has admin-only info when obtained
762 // with a privileged token, and should not be logged.
763 node, ok := resp.(map[string]interface{})
771 func (runner *ContainerRunner) logAPIResponse(label, path string, params map[string]interface{}, munge func(interface{})) (logged bool, err error) {
773 ArvClient: runner.ArvClient,
774 UUID: runner.Container.UUID,
775 loggingStream: label,
776 writeCloser: runner.LogCollection.Open(label + ".json"),
779 reader, err := runner.ArvClient.CallRaw("GET", path, "", "", arvadosclient.Dict(params))
781 return false, fmt.Errorf("error getting %s record: %v", label, err)
785 dec := json.NewDecoder(reader)
787 var resp map[string]interface{}
788 if err = dec.Decode(&resp); err != nil {
789 return false, fmt.Errorf("error decoding %s list response: %v", label, err)
791 items, ok := resp["items"].([]interface{})
793 return false, fmt.Errorf("error decoding %s list response: no \"items\" key in API list response", label)
794 } else if len(items) < 1 {
800 // Re-encode it using indentation to improve readability
801 enc := json.NewEncoder(w)
802 enc.SetIndent("", " ")
803 if err = enc.Encode(items[0]); err != nil {
804 return false, fmt.Errorf("error logging %s record: %v", label, err)
808 return false, fmt.Errorf("error closing %s.json in log collection: %v", label, err)
813 // AttachStreams connects the docker container stdin, stdout and stderr logs
814 // to the Arvados logger which logs to Keep and the API server logs table.
815 func (runner *ContainerRunner) AttachStreams() (err error) {
817 runner.CrunchLog.Print("Attaching container streams")
819 // If stdin mount is provided, attach it to the docker container
820 var stdinRdr arvados.File
822 if stdinMnt, ok := runner.Container.Mounts["stdin"]; ok {
823 if stdinMnt.Kind == "collection" {
824 var stdinColl arvados.Collection
825 collId := stdinMnt.UUID
827 collId = stdinMnt.PortableDataHash
829 err = runner.ArvClient.Get("collections", collId, nil, &stdinColl)
831 return fmt.Errorf("While getting stding collection: %v", err)
834 stdinRdr, err = runner.Kc.ManifestFileReader(manifest.Manifest{Text: stdinColl.ManifestText}, stdinMnt.Path)
835 if os.IsNotExist(err) {
836 return fmt.Errorf("stdin collection path not found: %v", stdinMnt.Path)
837 } else if err != nil {
838 return fmt.Errorf("While getting stdin collection path %v: %v", stdinMnt.Path, err)
840 } else if stdinMnt.Kind == "json" {
841 stdinJson, err = json.Marshal(stdinMnt.Content)
843 return fmt.Errorf("While encoding stdin json data: %v", err)
848 stdinUsed := stdinRdr != nil || len(stdinJson) != 0
849 response, err := runner.Docker.ContainerAttach(context.TODO(), runner.ContainerID,
850 dockertypes.ContainerAttachOptions{Stream: true, Stdin: stdinUsed, Stdout: true, Stderr: true})
852 return fmt.Errorf("While attaching container stdout/stderr streams: %v", err)
855 runner.loggingDone = make(chan bool)
857 if stdoutMnt, ok := runner.Container.Mounts["stdout"]; ok {
858 stdoutFile, err := runner.getStdoutFile(stdoutMnt.Path)
862 runner.Stdout = stdoutFile
864 runner.Stdout = NewThrottledLogger(runner.NewLogWriter("stdout"))
867 if stderrMnt, ok := runner.Container.Mounts["stderr"]; ok {
868 stderrFile, err := runner.getStdoutFile(stderrMnt.Path)
872 runner.Stderr = stderrFile
874 runner.Stderr = NewThrottledLogger(runner.NewLogWriter("stderr"))
879 _, err := io.Copy(response.Conn, stdinRdr)
881 runner.CrunchLog.Print("While writing stdin collection to docker container %q", err)
885 response.CloseWrite()
887 } else if len(stdinJson) != 0 {
889 _, err := io.Copy(response.Conn, bytes.NewReader(stdinJson))
891 runner.CrunchLog.Print("While writing stdin json to docker container %q", err)
894 response.CloseWrite()
898 go runner.ProcessDockerAttach(response.Reader)
903 func (runner *ContainerRunner) getStdoutFile(mntPath string) (*os.File, error) {
904 stdoutPath := mntPath[len(runner.Container.OutputPath):]
905 index := strings.LastIndex(stdoutPath, "/")
907 subdirs := stdoutPath[:index]
909 st, err := os.Stat(runner.HostOutputDir)
911 return nil, fmt.Errorf("While Stat on temp dir: %v", err)
913 stdoutPath := filepath.Join(runner.HostOutputDir, subdirs)
914 err = os.MkdirAll(stdoutPath, st.Mode()|os.ModeSetgid|0777)
916 return nil, fmt.Errorf("While MkdirAll %q: %v", stdoutPath, err)
920 stdoutFile, err := os.Create(filepath.Join(runner.HostOutputDir, stdoutPath))
922 return nil, fmt.Errorf("While creating file %q: %v", stdoutPath, err)
925 return stdoutFile, nil
928 // CreateContainer creates the docker container.
929 func (runner *ContainerRunner) CreateContainer() error {
930 runner.CrunchLog.Print("Creating Docker container")
932 runner.ContainerConfig.Cmd = runner.Container.Command
933 if runner.Container.Cwd != "." {
934 runner.ContainerConfig.WorkingDir = runner.Container.Cwd
937 for k, v := range runner.Container.Environment {
938 runner.ContainerConfig.Env = append(runner.ContainerConfig.Env, k+"="+v)
941 runner.ContainerConfig.Volumes = runner.Volumes
943 runner.HostConfig = dockercontainer.HostConfig{
945 LogConfig: dockercontainer.LogConfig{
948 Resources: dockercontainer.Resources{
949 CgroupParent: runner.setCgroupParent,
953 if wantAPI := runner.Container.RuntimeConstraints.API; wantAPI != nil && *wantAPI {
954 tok, err := runner.ContainerToken()
958 runner.ContainerConfig.Env = append(runner.ContainerConfig.Env,
959 "ARVADOS_API_TOKEN="+tok,
960 "ARVADOS_API_HOST="+os.Getenv("ARVADOS_API_HOST"),
961 "ARVADOS_API_HOST_INSECURE="+os.Getenv("ARVADOS_API_HOST_INSECURE"),
963 runner.HostConfig.NetworkMode = dockercontainer.NetworkMode(runner.networkMode)
965 if runner.enableNetwork == "always" {
966 runner.HostConfig.NetworkMode = dockercontainer.NetworkMode(runner.networkMode)
968 runner.HostConfig.NetworkMode = dockercontainer.NetworkMode("none")
972 _, stdinUsed := runner.Container.Mounts["stdin"]
973 runner.ContainerConfig.OpenStdin = stdinUsed
974 runner.ContainerConfig.StdinOnce = stdinUsed
975 runner.ContainerConfig.AttachStdin = stdinUsed
976 runner.ContainerConfig.AttachStdout = true
977 runner.ContainerConfig.AttachStderr = true
979 createdBody, err := runner.Docker.ContainerCreate(context.TODO(), &runner.ContainerConfig, &runner.HostConfig, nil, runner.Container.UUID)
981 return fmt.Errorf("While creating container: %v", err)
984 runner.ContainerID = createdBody.ID
986 return runner.AttachStreams()
989 // StartContainer starts the docker container created by CreateContainer.
990 func (runner *ContainerRunner) StartContainer() error {
991 runner.CrunchLog.Printf("Starting Docker container id '%s'", runner.ContainerID)
992 runner.cStateLock.Lock()
993 defer runner.cStateLock.Unlock()
994 if runner.cCancelled {
997 err := runner.Docker.ContainerStart(context.TODO(), runner.ContainerID,
998 dockertypes.ContainerStartOptions{})
1001 if m, e := regexp.MatchString("(?ms).*(exec|System error).*(no such file or directory|file not found).*", err.Error()); m && e == nil {
1002 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])
1004 return fmt.Errorf("could not start container: %v%s", err, advice)
1009 // WaitFinish waits for the container to terminate, capture the exit code, and
1010 // close the stdout/stderr logging.
1011 func (runner *ContainerRunner) WaitFinish() error {
1012 runner.CrunchLog.Print("Waiting for container to finish")
1014 waitOk, waitErr := runner.Docker.ContainerWait(context.TODO(), runner.ContainerID, dockercontainer.WaitConditionNotRunning)
1015 arvMountExit := runner.ArvMountExit
1018 case waitBody := <-waitOk:
1019 runner.CrunchLog.Printf("Container exited with code: %v", waitBody.StatusCode)
1020 code := int(waitBody.StatusCode)
1021 runner.ExitCode = &code
1023 // wait for stdout/stderr to complete
1024 <-runner.loggingDone
1027 case err := <-waitErr:
1028 return fmt.Errorf("container wait: %v", err)
1030 case <-arvMountExit:
1031 runner.CrunchLog.Printf("arv-mount exited while container is still running. Stopping container.")
1033 // arvMountExit will always be ready now that
1034 // it's closed, but that doesn't interest us.
1040 var ErrNotInOutputDir = fmt.Errorf("Must point to path within the output directory")
1042 func (runner *ContainerRunner) derefOutputSymlink(path string, startinfo os.FileInfo) (tgt string, readlinktgt string, info os.FileInfo, err error) {
1043 // Follow symlinks if necessary
1048 for followed := 0; info.Mode()&os.ModeSymlink != 0; followed++ {
1049 if followed >= limitFollowSymlinks {
1050 // Got stuck in a loop or just a pathological number of links, give up.
1051 err = fmt.Errorf("Followed more than %v symlinks from path %q", limitFollowSymlinks, path)
1055 readlinktgt, err = os.Readlink(nextlink)
1061 if !strings.HasPrefix(tgt, "/") {
1062 // Relative symlink, resolve it to host path
1063 tgt = filepath.Join(filepath.Dir(path), tgt)
1065 if strings.HasPrefix(tgt, runner.Container.OutputPath+"/") && !strings.HasPrefix(tgt, runner.HostOutputDir+"/") {
1066 // Absolute symlink to container output path, adjust it to host output path.
1067 tgt = filepath.Join(runner.HostOutputDir, tgt[len(runner.Container.OutputPath):])
1069 if !strings.HasPrefix(tgt, runner.HostOutputDir+"/") {
1070 // After dereferencing, symlink target must either be
1071 // within output directory, or must point to a
1072 // collection mount.
1073 err = ErrNotInOutputDir
1077 info, err = os.Lstat(tgt)
1080 err = fmt.Errorf("Symlink in output %q points to invalid location %q: %v",
1081 path[len(runner.HostOutputDir):], readlinktgt, err)
1091 var limitFollowSymlinks = 10
1093 // UploadFile uploads files within the output directory, with special handling
1094 // for symlinks. If the symlink leads to a keep mount, copy the manifest text
1095 // from the keep mount into the output manifestText. Ensure that whether
1096 // symlinks are relative or absolute, every symlink target (even targets that
1097 // are symlinks themselves) must point to a path in either the output directory
1098 // or a collection mount.
1100 // Assumes initial value of "path" is absolute, and located within runner.HostOutputDir.
1101 func (runner *ContainerRunner) UploadOutputFile(
1106 walkUpload *WalkUpload,
1107 relocateFrom string,
1109 followed int) (manifestText string, err error) {
1115 if info.Mode().IsDir() {
1116 // if empty, need to create a .keep file
1117 dir, direrr := os.Open(path)
1122 names, eof := dir.Readdirnames(1)
1123 if len(names) == 0 && eof == io.EOF && path != runner.HostOutputDir {
1124 containerPath := runner.OutputPath + path[len(runner.HostOutputDir):]
1125 for _, bind := range binds {
1126 mnt := runner.Container.Mounts[bind]
1127 // Check if there is a bind for this
1128 // directory, in which case assume we don't need .keep
1129 if (containerPath == bind || strings.HasPrefix(containerPath, bind+"/")) && mnt.PortableDataHash != "d41d8cd98f00b204e9800998ecf8427e+0" {
1133 outputSuffix := path[len(runner.HostOutputDir)+1:]
1134 return fmt.Sprintf("./%v d41d8cd98f00b204e9800998ecf8427e+0 0:0:.keep\n", outputSuffix), nil
1139 if followed >= limitFollowSymlinks {
1140 // Got stuck in a loop or just a pathological number of
1141 // directory links, give up.
1142 err = fmt.Errorf("Followed more than %v symlinks from path %q", limitFollowSymlinks, path)
1146 // "path" is the actual path we are visiting
1147 // "tgt" is the target of "path" (a non-symlink) after following symlinks
1148 // "relocated" is the path in the output manifest where the file should be placed,
1149 // but has HostOutputDir as a prefix.
1151 // The destination path in the output manifest may need to be
1152 // logically relocated to some other path in order to appear
1153 // in the correct location as a result of following a symlink.
1154 // Remove the relocateFrom prefix and replace it with
1156 relocated := relocateTo + path[len(relocateFrom):]
1158 tgt, readlinktgt, info, derefErr := runner.derefOutputSymlink(path, info)
1159 if derefErr != nil && derefErr != ErrNotInOutputDir {
1163 // go through mounts and try reverse map to collection reference
1164 for _, bind := range binds {
1165 mnt := runner.Container.Mounts[bind]
1166 if (tgt == bind || strings.HasPrefix(tgt, bind+"/")) && !mnt.Writable {
1167 // get path relative to bind
1168 targetSuffix := tgt[len(bind):]
1170 // Copy mount and adjust the path to add path relative to the bind
1171 adjustedMount := mnt
1172 adjustedMount.Path = filepath.Join(adjustedMount.Path, targetSuffix)
1174 // Terminates in this keep mount, so add the
1175 // manifest text at appropriate location.
1176 outputSuffix := relocated[len(runner.HostOutputDir):]
1177 manifestText, err = runner.getCollectionManifestForPath(adjustedMount, outputSuffix)
1182 // If target is not a collection mount, it must be located within the
1183 // output directory, otherwise it is an error.
1184 if derefErr == ErrNotInOutputDir {
1185 err = fmt.Errorf("Symlink in output %q points to invalid location %q, must point to path within the output directory.",
1186 path[len(runner.HostOutputDir):], readlinktgt)
1190 if info.Mode().IsRegular() {
1191 return "", walkUpload.UploadFile(relocated, tgt)
1194 if info.Mode().IsDir() {
1195 // Symlink leads to directory. Walk() doesn't follow
1196 // directory symlinks, so we walk the target directory
1197 // instead. Within the walk, file paths are relocated
1198 // so they appear under the original symlink path.
1199 err = filepath.Walk(tgt, func(walkpath string, walkinfo os.FileInfo, walkerr error) error {
1201 m, walkerr = runner.UploadOutputFile(walkpath, walkinfo, walkerr,
1202 binds, walkUpload, tgt, relocated, followed+1)
1204 manifestText = manifestText + m
1214 // HandleOutput sets the output, unmounts the FUSE mount, and deletes temporary directories
1215 func (runner *ContainerRunner) CaptureOutput() error {
1216 if wantAPI := runner.Container.RuntimeConstraints.API; wantAPI != nil && *wantAPI {
1217 // Output may have been set directly by the container, so
1218 // refresh the container record to check.
1219 err := runner.ArvClient.Get("containers", runner.Container.UUID,
1220 nil, &runner.Container)
1224 if runner.Container.Output != "" {
1225 // Container output is already set.
1226 runner.OutputPDH = &runner.Container.Output
1231 if runner.HostOutputDir == "" {
1235 _, err := os.Stat(runner.HostOutputDir)
1237 return fmt.Errorf("While checking host output path: %v", err)
1240 // Pre-populate output from the configured mount points
1242 for bind, mnt := range runner.Container.Mounts {
1243 if mnt.Kind == "collection" {
1244 binds = append(binds, bind)
1249 var manifestText string
1251 collectionMetafile := fmt.Sprintf("%s/.arvados#collection", runner.HostOutputDir)
1252 _, err = os.Stat(collectionMetafile)
1254 // Regular directory
1256 cw := CollectionWriter{0, runner.Kc, nil, nil, sync.Mutex{}}
1257 walkUpload := cw.BeginUpload(runner.HostOutputDir, runner.CrunchLog.Logger)
1260 err = filepath.Walk(runner.HostOutputDir, func(path string, info os.FileInfo, err error) error {
1261 m, err = runner.UploadOutputFile(path, info, err, binds, walkUpload, "", "", 0)
1263 manifestText = manifestText + m
1268 cw.EndUpload(walkUpload)
1271 return fmt.Errorf("While uploading output files: %v", err)
1274 m, err = cw.ManifestText()
1275 manifestText = manifestText + m
1277 return fmt.Errorf("While uploading output files: %v", err)
1280 // FUSE mount directory
1281 file, openerr := os.Open(collectionMetafile)
1283 return fmt.Errorf("While opening FUSE metafile: %v", err)
1287 var rec arvados.Collection
1288 err = json.NewDecoder(file).Decode(&rec)
1290 return fmt.Errorf("While reading FUSE metafile: %v", err)
1292 manifestText = rec.ManifestText
1295 for _, bind := range binds {
1296 mnt := runner.Container.Mounts[bind]
1298 bindSuffix := strings.TrimPrefix(bind, runner.Container.OutputPath)
1300 if bindSuffix == bind || len(bindSuffix) <= 0 {
1301 // either does not start with OutputPath or is OutputPath itself
1305 if mnt.ExcludeFromOutput == true || mnt.Writable {
1309 // append to manifest_text
1310 m, err := runner.getCollectionManifestForPath(mnt, bindSuffix)
1315 manifestText = manifestText + m
1319 var response arvados.Collection
1320 manifest := manifest.Manifest{Text: manifestText}
1321 manifestText = manifest.Extract(".", ".").Text
1322 err = runner.ArvClient.Create("collections",
1324 "ensure_unique_name": true,
1325 "collection": arvadosclient.Dict{
1327 "name": "output for " + runner.Container.UUID,
1328 "manifest_text": manifestText}},
1331 return fmt.Errorf("While creating output collection: %v", err)
1333 runner.OutputPDH = &response.PortableDataHash
1337 var outputCollections = make(map[string]arvados.Collection)
1339 // Fetch the collection for the mnt.PortableDataHash
1340 // Return the manifest_text fragment corresponding to the specified mnt.Path
1341 // after making any required updates.
1343 // If mnt.Path is not specified,
1344 // return the entire manifest_text after replacing any "." with bindSuffix
1345 // If mnt.Path corresponds to one stream,
1346 // return the manifest_text for that stream after replacing that stream name with bindSuffix
1347 // Otherwise, check if a filename in any one stream is being sought. Return the manifest_text
1348 // for that stream after replacing stream name with bindSuffix minus the last word
1349 // and the file name with last word of the bindSuffix
1350 // Allowed path examples:
1352 // "path":"/subdir1"
1353 // "path":"/subdir1/subdir2"
1354 // "path":"/subdir/filename" etc
1355 func (runner *ContainerRunner) getCollectionManifestForPath(mnt arvados.Mount, bindSuffix string) (string, error) {
1356 collection := outputCollections[mnt.PortableDataHash]
1357 if collection.PortableDataHash == "" {
1358 err := runner.ArvClient.Get("collections", mnt.PortableDataHash, nil, &collection)
1360 return "", fmt.Errorf("While getting collection for %v: %v", mnt.PortableDataHash, err)
1362 outputCollections[mnt.PortableDataHash] = collection
1365 if collection.ManifestText == "" {
1366 runner.CrunchLog.Printf("No manifest text for collection %v", collection.PortableDataHash)
1370 mft := manifest.Manifest{Text: collection.ManifestText}
1371 extracted := mft.Extract(mnt.Path, bindSuffix)
1372 if extracted.Err != nil {
1373 return "", fmt.Errorf("Error parsing manifest for %v: %v", mnt.PortableDataHash, extracted.Err.Error())
1375 return extracted.Text, nil
1378 func (runner *ContainerRunner) CleanupDirs() {
1379 if runner.ArvMount != nil {
1381 umount := exec.Command("arv-mount", fmt.Sprintf("--unmount-timeout=%d", delay), "--unmount", runner.ArvMountPoint)
1382 umount.Stdout = runner.CrunchLog
1383 umount.Stderr = runner.CrunchLog
1384 runner.CrunchLog.Printf("Running %v", umount.Args)
1385 umnterr := umount.Start()
1388 runner.CrunchLog.Printf("Error unmounting: %v", umnterr)
1390 // If arv-mount --unmount gets stuck for any reason, we
1391 // don't want to wait for it forever. Do Wait() in a goroutine
1392 // so it doesn't block crunch-run.
1393 umountExit := make(chan error)
1395 mnterr := umount.Wait()
1397 runner.CrunchLog.Printf("Error unmounting: %v", mnterr)
1399 umountExit <- mnterr
1402 for again := true; again; {
1408 case <-runner.ArvMountExit:
1410 case <-time.After(time.Duration((delay + 1) * int64(time.Second))):
1411 runner.CrunchLog.Printf("Timed out waiting for unmount")
1413 umount.Process.Kill()
1415 runner.ArvMount.Process.Kill()
1421 if runner.ArvMountPoint != "" {
1422 if rmerr := os.Remove(runner.ArvMountPoint); rmerr != nil {
1423 runner.CrunchLog.Printf("While cleaning up arv-mount directory %s: %v", runner.ArvMountPoint, rmerr)
1427 if rmerr := os.RemoveAll(runner.parentTemp); rmerr != nil {
1428 runner.CrunchLog.Printf("While cleaning up temporary directory %s: %v", runner.parentTemp, rmerr)
1432 // CommitLogs posts the collection containing the final container logs.
1433 func (runner *ContainerRunner) CommitLogs() error {
1434 runner.CrunchLog.Print(runner.finalState)
1436 if runner.arvMountLog != nil {
1437 runner.arvMountLog.Close()
1439 runner.CrunchLog.Close()
1441 // Closing CrunchLog above allows them to be committed to Keep at this
1442 // point, but re-open crunch log with ArvClient in case there are any
1443 // other further errors (such as failing to write the log to Keep!)
1444 // while shutting down
1445 runner.CrunchLog = NewThrottledLogger(&ArvLogWriter{ArvClient: runner.ArvClient,
1446 UUID: runner.Container.UUID, loggingStream: "crunch-run", writeCloser: nil})
1447 runner.CrunchLog.Immediate = log.New(os.Stderr, runner.Container.UUID+" ", 0)
1449 if runner.LogsPDH != nil {
1450 // If we have already assigned something to LogsPDH,
1451 // we must be closing the re-opened log, which won't
1452 // end up getting attached to the container record and
1453 // therefore doesn't need to be saved as a collection
1454 // -- it exists only to send logs to other channels.
1458 mt, err := runner.LogCollection.ManifestText()
1460 return fmt.Errorf("While creating log manifest: %v", err)
1463 var response arvados.Collection
1464 err = runner.ArvClient.Create("collections",
1466 "ensure_unique_name": true,
1467 "collection": arvadosclient.Dict{
1469 "name": "logs for " + runner.Container.UUID,
1470 "manifest_text": mt}},
1473 return fmt.Errorf("While creating log collection: %v", err)
1475 runner.LogsPDH = &response.PortableDataHash
1479 // UpdateContainerRunning updates the container state to "Running"
1480 func (runner *ContainerRunner) UpdateContainerRunning() error {
1481 runner.cStateLock.Lock()
1482 defer runner.cStateLock.Unlock()
1483 if runner.cCancelled {
1486 return runner.ArvClient.Update("containers", runner.Container.UUID,
1487 arvadosclient.Dict{"container": arvadosclient.Dict{"state": "Running"}}, nil)
1490 // ContainerToken returns the api_token the container (and any
1491 // arv-mount processes) are allowed to use.
1492 func (runner *ContainerRunner) ContainerToken() (string, error) {
1493 if runner.token != "" {
1494 return runner.token, nil
1497 var auth arvados.APIClientAuthorization
1498 err := runner.ArvClient.Call("GET", "containers", runner.Container.UUID, "auth", nil, &auth)
1502 runner.token = auth.APIToken
1503 return runner.token, nil
1506 // UpdateContainerComplete updates the container record state on API
1507 // server to "Complete" or "Cancelled"
1508 func (runner *ContainerRunner) UpdateContainerFinal() error {
1509 update := arvadosclient.Dict{}
1510 update["state"] = runner.finalState
1511 if runner.LogsPDH != nil {
1512 update["log"] = *runner.LogsPDH
1514 if runner.finalState == "Complete" {
1515 if runner.ExitCode != nil {
1516 update["exit_code"] = *runner.ExitCode
1518 if runner.OutputPDH != nil {
1519 update["output"] = *runner.OutputPDH
1522 return runner.ArvClient.Update("containers", runner.Container.UUID, arvadosclient.Dict{"container": update}, nil)
1525 // IsCancelled returns the value of Cancelled, with goroutine safety.
1526 func (runner *ContainerRunner) IsCancelled() bool {
1527 runner.cStateLock.Lock()
1528 defer runner.cStateLock.Unlock()
1529 return runner.cCancelled
1532 // NewArvLogWriter creates an ArvLogWriter
1533 func (runner *ContainerRunner) NewArvLogWriter(name string) io.WriteCloser {
1534 return &ArvLogWriter{
1535 ArvClient: runner.ArvClient,
1536 UUID: runner.Container.UUID,
1537 loggingStream: name,
1538 writeCloser: runner.LogCollection.Open(name + ".txt")}
1541 // Run the full container lifecycle.
1542 func (runner *ContainerRunner) Run() (err error) {
1543 runner.CrunchLog.Printf("crunch-run %s started", version)
1544 runner.CrunchLog.Printf("Executing container '%s'", runner.Container.UUID)
1546 hostname, hosterr := os.Hostname()
1548 runner.CrunchLog.Printf("Error getting hostname '%v'", hosterr)
1550 runner.CrunchLog.Printf("Executing on host '%s'", hostname)
1553 runner.finalState = "Queued"
1556 runner.stopSignals()
1557 runner.CleanupDirs()
1559 runner.CrunchLog.Printf("crunch-run finished")
1560 runner.CrunchLog.Close()
1564 // checkErr prints e (unless it's nil) and sets err to
1565 // e (unless err is already non-nil). Thus, if err
1566 // hasn't already been assigned when Run() returns,
1567 // this cleanup func will cause Run() to return the
1568 // first non-nil error that is passed to checkErr().
1569 checkErr := func(e error) {
1573 runner.CrunchLog.Print(e)
1577 if runner.finalState == "Complete" {
1578 // There was an error in the finalization.
1579 runner.finalState = "Cancelled"
1583 // Log the error encountered in Run(), if any
1586 if runner.finalState == "Queued" {
1587 runner.UpdateContainerFinal()
1591 if runner.IsCancelled() {
1592 runner.finalState = "Cancelled"
1593 // but don't return yet -- we still want to
1594 // capture partial output and write logs
1597 checkErr(runner.CaptureOutput())
1598 checkErr(runner.stopHoststat())
1599 checkErr(runner.CommitLogs())
1600 checkErr(runner.UpdateContainerFinal())
1603 err = runner.fetchContainerRecord()
1607 runner.setupSignals()
1608 runner.startHoststat()
1610 // check for and/or load image
1611 err = runner.LoadImage()
1613 if !runner.checkBrokenNode(err) {
1614 // Failed to load image but not due to a "broken node"
1615 // condition, probably user error.
1616 runner.finalState = "Cancelled"
1618 err = fmt.Errorf("While loading container image: %v", err)
1622 // set up FUSE mount and binds
1623 err = runner.SetupMounts()
1625 runner.finalState = "Cancelled"
1626 err = fmt.Errorf("While setting up mounts: %v", err)
1630 err = runner.CreateContainer()
1634 err = runner.LogHostInfo()
1638 err = runner.LogNodeRecord()
1642 err = runner.LogContainerRecord()
1647 if runner.IsCancelled() {
1651 err = runner.UpdateContainerRunning()
1655 runner.finalState = "Cancelled"
1657 runner.startCrunchstat()
1659 err = runner.StartContainer()
1661 runner.checkBrokenNode(err)
1665 err = runner.WaitFinish()
1666 if err == nil && !runner.IsCancelled() {
1667 runner.finalState = "Complete"
1672 // Fetch the current container record (uuid = runner.Container.UUID)
1673 // into runner.Container.
1674 func (runner *ContainerRunner) fetchContainerRecord() error {
1675 reader, err := runner.ArvClient.CallRaw("GET", "containers", runner.Container.UUID, "", nil)
1677 return fmt.Errorf("error fetching container record: %v", err)
1679 defer reader.Close()
1681 dec := json.NewDecoder(reader)
1683 err = dec.Decode(&runner.Container)
1685 return fmt.Errorf("error decoding container record: %v", err)
1690 // NewContainerRunner creates a new container runner.
1691 func NewContainerRunner(api IArvadosClient,
1693 docker ThinDockerClient,
1694 containerUUID string) *ContainerRunner {
1696 cr := &ContainerRunner{ArvClient: api, Kc: kc, Docker: docker}
1697 cr.NewLogWriter = cr.NewArvLogWriter
1698 cr.RunArvMount = cr.ArvMountCmd
1699 cr.MkTempDir = ioutil.TempDir
1700 cr.LogCollection = &CollectionWriter{0, kc, nil, nil, sync.Mutex{}}
1701 cr.Container.UUID = containerUUID
1702 cr.CrunchLog = NewThrottledLogger(cr.NewLogWriter("crunch-run"))
1703 cr.CrunchLog.Immediate = log.New(os.Stderr, containerUUID+" ", 0)
1705 loadLogThrottleParams(api)
1711 statInterval := flag.Duration("crunchstat-interval", 10*time.Second, "sampling period for periodic resource usage reporting")
1712 cgroupRoot := flag.String("cgroup-root", "/sys/fs/cgroup", "path to sysfs cgroup tree")
1713 cgroupParent := flag.String("cgroup-parent", "docker", "name of container's parent cgroup (ignored if -cgroup-parent-subsystem is used)")
1714 cgroupParentSubsystem := flag.String("cgroup-parent-subsystem", "", "use current cgroup for given subsystem as parent cgroup for container")
1715 caCertsPath := flag.String("ca-certs", "", "Path to TLS root certificates")
1716 enableNetwork := flag.String("container-enable-networking", "default",
1717 `Specify if networking should be enabled for container. One of 'default', 'always':
1718 default: only enable networking if container requests it.
1719 always: containers always have networking enabled
1721 networkMode := flag.String("container-network-mode", "default",
1722 `Set networking mode for container. Corresponds to Docker network mode (--net).
1724 memprofile := flag.String("memprofile", "", "write memory profile to `file` after running container")
1725 getVersion := flag.Bool("version", false, "Print version information and exit.")
1728 // Print version information if requested
1730 fmt.Printf("crunch-run %s\n", version)
1734 log.Printf("crunch-run %s started", version)
1736 containerId := flag.Arg(0)
1738 if *caCertsPath != "" {
1739 arvadosclient.CertFiles = []string{*caCertsPath}
1742 api, err := arvadosclient.MakeArvadosClient()
1744 log.Fatalf("%s: %v", containerId, err)
1748 kc, kcerr := keepclient.MakeKeepClient(api)
1750 log.Fatalf("%s: %v", containerId, kcerr)
1752 kc.BlockCache = &keepclient.BlockCache{MaxBlocks: 2}
1755 // API version 1.21 corresponds to Docker 1.9, which is currently the
1756 // minimum version we want to support.
1757 docker, dockererr := dockerclient.NewClient(dockerclient.DefaultDockerHost, "1.21", nil, nil)
1759 cr := NewContainerRunner(api, kc, docker, containerId)
1760 if dockererr != nil {
1761 cr.CrunchLog.Printf("%s: %v", containerId, dockererr)
1762 cr.checkBrokenNode(dockererr)
1763 cr.CrunchLog.Close()
1767 parentTemp, tmperr := cr.MkTempDir("", "crunch-run."+containerId+".")
1769 log.Fatalf("%s: %v", containerId, tmperr)
1772 cr.parentTemp = parentTemp
1773 cr.statInterval = *statInterval
1774 cr.cgroupRoot = *cgroupRoot
1775 cr.expectCgroupParent = *cgroupParent
1776 cr.enableNetwork = *enableNetwork
1777 cr.networkMode = *networkMode
1778 if *cgroupParentSubsystem != "" {
1779 p := findCgroup(*cgroupParentSubsystem)
1780 cr.setCgroupParent = p
1781 cr.expectCgroupParent = p
1786 if *memprofile != "" {
1787 f, err := os.Create(*memprofile)
1789 log.Printf("could not create memory profile: ", err)
1791 runtime.GC() // get up-to-date statistics
1792 if err := pprof.WriteHeapProfile(f); err != nil {
1793 log.Printf("could not write memory profile: ", err)
1795 closeerr := f.Close()
1796 if closeerr != nil {
1797 log.Printf("closing memprofile file: ", err)
1802 log.Fatalf("%s: %v", containerId, runerr)